4 mberiah SPSS ASSIGNMENT 1
Introduction to Statistical Analysis Using IBM SPSS Statistics Student Guide Course Code: 0G517 ERC 1.0 Introduction to Statistical Analysis Using IBM SPSS Statistics Licensed Materials - Property of IBM © Copyright IBM Corp. 2010 0G517 Published October 2010 US Government Users Restricted Rights - Use, duplication or disclosure restricted by GSA ADP Schedule Contract with IBM Corp. IBM, the IBM logo and ibm.com are trademarks of International Business Machines Corp., registered in many jurisdictions worldwide. SPSS, SamplePower, and PASW are trademarks of SPSS Inc., an IBM Company, registered in many jurisdictions worldwide. Other product and service names might be trademarks of IBM or other companies.
This guide contains proprietary information which is protected by copyright. No part of this document may be photocopied, reproduced, or translated into another language without a legal license agreement from IBM Corporation. Any references in this information to non- IBM Web sites are provided for convenience only and do not in any manner serve as an endorsement of those W eb sites. The materials at those Web sites are not part of the materials for this IBM product and use of those Web sites is at your own risk. TABLE OF CONTENTS i Table of Contents LESSON 0: COURSE INTRODUCTION .................................................. 0-1 LESSON 1: INTRODUCTI ON TO STATISTICAL ANALYSIS ............. 1-1 LESSON 2: UNDERSTANDING DATA DISTRIBUTIONS – THEORY 2 -1 LESSON 3: DATA DISTR IBUTIONS FOR CATEGOR ICAL VARIABLES ............................................................................................ 3-1 0.1 INTRODUCTION ................................ ................................ ................................ ................... 0-1 0.2 COURSE OBJECTIVES .......................................................................................................... 0-1 0.3 ABOUT SPSS ...................................................................................................................... 0-1 0.4 SUPPORTING MATERIALS ................................................................................................... 0-2 0.5 COURSE ASSUMPTIONS ................................ ................................ ................................ ...... 0-2 1.1 OBJECTIVES ................................ ................................ ................................ ........................ 1-1 1.2 INTRODUCTION ................................................................................................................... 1-1 1.3 BASIC STEPS OF THE RESEARCH PROCESS ......................................................................... 1-1 1.4 POPULATIONS AND SAMPLES ............................................................................................. 1-3 1.5 RESEARCH DESIGN ............................................................................................................. 1-3 1.6 INDEPENDENT AND DEPENDENT VARIABLES ..................................................................... 1-4 1.7 NOTE ABOUT DEFAULT STARTUP FOLDER AND VARIABLE DISPLAY IN DIALOG BOXES .. 1- 4 1.8 LESSON SUMMARY ............................................................................................................. 1-5 1.9 LEARNING ACTIVITY ................................ ................................ ................................ .......... 1-6 2.1 OBJECTIVES ................................ ................................ ................................ ........................ 2-1 INTRODUCTION ......................................................................................................................... 2-1 2.2 LEVELS OF MEASUREMENT AND STATISTICAL METHODS ................................................. 2-1 2.3 MEASURES OF CENTRAL TENDENCY AND DISPERSION ..................................................... 2-5 2.4 NORMAL DISTRIBUTIONS ................................................................................................... 2-7 2.5 STANDARDIZED (Z -) SCORES ............................................................................................. 2-8 2.6 REQUESTING STANDARDIZED (Z -) SCORES ...................................................................... 2-10 2.7 STANDARDIZED (Z -) SCORES OUTPUT ............................................................................. 2-10 2.8 PROCEDURE : DESCRIPTIVES FOR STANDARDIZED (Z -) SCORES ...................................... 2-10 2.9 DEMONSTRATION : DESCRIPTIVES FOR Z -S CORES ............................................................ 2-11 2.10 LESSON SUMMARY ......................................................................................................... 2-12 2.11 LEARNING ACTIVITY ................................ ................................ ................................ ...... 2-13 3.1 OBJECTIVES ................................ ................................ ................................ ........................ 3-1 3.2 INTRODUCTION ................................................................................................................... 3-1 3.3 USING FREQUENCIES TO SUMMARIZE NOMINAL AND ORDINAL VARIABLES .................... 3-2 3.4 REQUESTING FREQUENCIES ............................................................................................... 3-3 3.5 FREQUENCIES OUTPUT ....................................................................................................... 3-3 3.6 PROCEDURE : FREQUENCIES ............................................................................................... 3-4 3.7 DEMONSTRATION : FREQUENCIES ....................................................................................... 3-6 3.8 LESSON SUMMARY ........................................................................................................... 3-10 3.9 LEARNING ACTIVITY ................................ ................................ ................................ ........ 3-10 INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS ii LESSON 4: DATA DISTR IBUTIONS FOR SCALE VARIABLES ......... 4-1 LESSON 5: MAKING INF ERENCES ABOUT POPULATIONS FROM SAMPLES ............................................................................................ 5-1 LESSON 6: RELATIONSHIPS BETWEEN CATEGORI CAL VARIABLES ........................................................................................... 6-1 4.1 OBJECTIVES ................................ ................................ ................................ ........................ 4-1 4.2 INTRODUCTION ................................................................................................................... 4-1 4.3 SUMMARIZING SCALE VARIABLES USING FREQUE NCIES ................................................... 4-1 4.4 REQUESTING FREQUENCIES ................................................................................................ 4-2 4.5 FREQUENCIES OUTPUT ....................................................................................................... 4-2 4.6 PROCEDURE : FREQUENCIES ................................................................................................ 4-4 4.7 DEMONSTRATION : FREQUENCIES ....................................................................................... 4-6 4.8 SUMMARIZING SCALE VARIABLES USING DESCRIPTIVES ................................................. 4-11 4.9 REQUESTING DESCRIPTIVES ............................................................................................. 4-11 4.10 DESCRIPTIVES OUTPUT ................................................................................................... 4-11 4.11 PROCEDURE : DESCRIPTIVES ........................................................................................... 4-11 4.12 DEMONSTRATION : DESCRIPTIVES ................................................................................... 4-12 4.13 SUMMARIZING SCALE VARIABLES USING THE EXPLORE PROCEDURE ........................... 4-13 4.14 REQUESTING EXPLORE ................................................................................................... 4-13 4.15 PROCEDURE : EXPLORE ................................................................................................... 4-16 4.16 DEMONSTRATION : EXPLORE ........................................................................................... 4-19 4.17 LESSON SUMMARY ......................................................................................................... 4-24 4.18 LEARNING ACTIVITY ................................ ................................ ................................ ...... 4-25 5.1 OBJECTIVES ................................ ................................ ................................ ........................ 5-1 5.2 INTRODUCTION ................................................................................................................... 5-1 5.3 BASICS OF MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES .......................... 5-1 5.4 INFLUENCE OF SAMPLE SIZE ............................................................................................... 5-2 5.5 HYPOTHESIS TESTING ....................................................................................................... 5-10 5.6 THE NATURE OF PROBABILITY ......................................................................................... 5-11 5.7 TYPES OF STATISTICAL ERRORS ....................................................................................... 5-11 5.8 STATISTICAL SIGNIFICANCE AND PRACTICAL IMPORTANCE ............................................ 5-12 5.9 LESSON SUMMARY ........................................................................................................... 5-13 5.10 LEARNING ACTIVITY ................................ ................................ ................................ ...... 5-13 TABLE OF CONTENTS iii LESSON 7: THE INDEPENDENT- SAMPLES T TEST .......................... 7-1 LESSON 8: THE PAIRED-SAMPLES T TEST ........................................ 8-1 6.1 OBJECTIVES ................................ ................................ ................................ ........................ 6-1 6.2 INTRODUCTION ................................................................................................................... 6-1 6.3 CROSSTABS ......................................................................................................................... 6-2 6.4 CROSSTABS ASSUMPTIONS ................................................................................................. 6-3 6.5 REQUESTING CROSSTABS ................................................................................................... 6-3 6.6 CROSSTABS OUTPUT .......................................................................................................... 6-3 6.7 PROCEDURE : CROSSTABS ................................................................................................... 6-4 6.8 EXAMPLE : CROSSTABS ....................................................................................................... 6-5 6.9 CHI- SQUARE TEST .............................................................................................................. 6-7 6.10 REQUESTING THE CHI- SQUARE TEST ............................................................................... 6-8 6.11 CHI- SQUARE OUTPUT ....................................................................................................... 6-8 6.12 PROCEDURE : CHI- SQUARE TEST ...................................................................................... 6-9 6.13 EXAMPLE : CHI- SQUARE TEST ........................................................................................ 6-10 6.14 CLUSTERED BAR CHART ................................................................................................ 6-11 6.15 REQUESTING A CLUSTERED BAR CHART WITH CHART BUILDER .................................. 6-12 6.16 CLUSTERED BAR CHART FROM CHART BUILDER OUTPUT ............................................ 6-12 6.17 PROCEDURE : CLUSTERED BAR CHART WITH CHART BUILDER ..................................... 6-13 6.18 EXAMPLE : CLUSTERED BAR CHART WITH CHART BUILDER ......................................... 6-15 6.19 ADDING A CONTROL VARIABLE ..................................................................................... 6-16 6.20 REQUESTING A CONTROL VARIABLE ............................................................................. 6-17 6.21 CONTROL VARIABLE OUTPUT ........................................................................................ 6-17 6.22 PROCEDURE : ADDING A CONTROL VARIABLE ............................................................... 6-18 6.23 EXAMPLE : ADDING A CONTROL VARIABLE ................................................................... 6-19 6.24 EXTENSIONS : BEYOND CROSSTABS ............................................................................... 6-22 6.25 ASSOCIATION MEASURES ............................................................................................... 6-23 6.26 LESSON SUMMARY ......................................................................................................... 6-23 6.27 LEARNING ACTIVITY ................................ ................................ ................................ ...... 6-24 7.1 OBJECTIVES ................................ ................................ ................................ ........................ 7-1 7.2 INTRODUCTION ................................................................................................................... 7-1 7.3 THE INDEPENDENT - SAMPLES T TEST ................................................................................ 7-1 7.4 INDEPENDENT - SAMPLES T TEST ASSUMPTIONS ................................................................ 7-2 7.5 REQUESTING THE INDEPENDENT - SAMPLES T TEST ........................................................... 7-2 7.6 INDEPENDENT - SAMPLES T TEST OUTPUT .......................................................................... 7-3 7.7 PROCEDURE : INDEPENDENT - SAMPLES T TEST .................................................................. 7-5 7.8 DEMONSTRATION : INDEPENDENT - SAMPLES T TEST .......................................................... 7-6 7.9 ERROR BAR CHART .......................................................................................................... 7-10 7.10 REQUESTING AN ERROR BAR CHART WITH CHART BUILDER ........................................ 7-11 7.11 ERROR BAR CHART OUTPUT .......................................................................................... 7-11 7.12 DEMONSTRATION : ERROR BAR CHART WITH CHART BUILDER .................................... 7-12 7.13 LESSON SUMMARY ......................................................................................................... 7-14 7.14 LEARNING ACTIVITY ................................ ................................ ................................ ...... 7-14 INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS iv LESSON 9: ONE- WAY ANOVA ............................................................... 9-1 LESSON 10: BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES .......................................................................................... 10-1 8.1 OBJECTIVES ................................ ................................ ................................ ........................ 8-1 8.2 INTRODUCTION ................................................................................................................... 8-1 8.3 THE PAIRED - SAMPLES T TEST ............................................................................................ 8-1 8.4 ASSUMPTIONS FOR THE PAIRED - SAMPLES T TEST ............................................................. 8-2 8.5 REQUESTING A PAIRED - SAMPLES T TEST .......................................................................... 8-3 8.6 PAIRED - SAMPLES T TEST OUTPUT ..................................................................................... 8-3 8.7 PROCEDURE : PAIRED - SAMPLES T TEST .............................................................................. 8-4 8.8 DEMONSTRATION : PAIRED - SAMPLES T TEST ..................................................................... 8-4 8.9 LESSON SUMMARY ............................................................................................................. 8-6 8.10 LEARNING ACTIVITY ................................ ................................ ................................ ........ 8-6 9.1 OBJECTIVES ................................ ................................ ................................ ........................ 9-1 9.2 INTRODUCTION ................................................................................................................... 9-1 9.3 ONE - W AY ANOVA .............................................................................................................. 9-1 9.4 ASSUMPTIONS OF ONE - W AY ANOVA ............................................................................... 9-2 9.5 REQUESTING ONE - W AY ANOVA ...................................................................................... 9-2 9.6 ONE - W AY ANOVA OUTPUT .............................................................................................. 9-3 9.7 PROCEDURE : ONE - W AY ANOVA ...................................................................................... 9-4 9.8 DEMONSTRATION : ONE - W AY ANOVA ............................................................................. 9-6 9.9 POST HOC TESTS WITH A ONE - W AY ANOVA ................................................................... 9-8 9.10 REQUESTING POST HOC TESTS WITH A ONE - W AY ANOVA ........................................... 9-9 9.11 POST HOC TESTS OUTPUT ................................................................................................. 9-9 9.12 PROCEDURE : POST HOC TESTS WITH A ONE - W AY ANOVA.......................................... 9-10 9.13 DEMONSTRATION : POST HOC TESTS WITH A ONE - W AY ANOVA ................................. 9-12 9.14 ERROR BAR CHART WITH CHART BUILDER ................................................................... 9-14 9.15 REQUESTING AN ERROR BAR CHART WITH CHART BUILDER ........................................ 9-14 9.16 ERROR BAR CHART OUTPUT .......................................................................................... 9-14 9.17 PROCEDURE : ERROR BAR CHART WITH CHART BUILDER .............................................. 9-15 9.18 DEMONSTRATION : ERROR BAR CHART WITH CHART BUILDER ..................................... 9-16 9.19 LESSON SUMMARY ......................................................................................................... 9-18 9.20 LEARNING ACTIVITY ................................ ................................ ................................ ...... 9-18 10.1 OBJECTIVES ................................ ................................ ................................ .................... 10 -1 10.2 INTRODUCTION ............................................................................................................... 10-1 10.3 SCATTERPLOTS ............................................................................................................... 10-1 10.4 REQUESTING A SCATTERPLOT ........................................................................................ 10-2 10.5 SCATTERPLOT OUTPUT ................................................................................................... 10-3 10.6 PROCEDURE : SCATTERPLOT ........................................................................................... 10-3 10.7 DEMONSTRATION : SCATTERPLOT ................................................................................... 10-4 10.8 ADDING A BEST FIT STRAIGHT LINE TO THE SCATTERPLOT .......................................... 10-5 10.9 PEARSON CORRELATION COEFFICIENT ........................................................................... 10-7 10.10 REQUESTING A PEARSON CORRELATION COEFFICIENT ................................................ 10-8 10.11 BIVARIATE CORRELATION OUTPUT .............................................................................. 10-8 10.12 PROCEDURE : PEARSON CORRELATION WITH BIVARIATE CORRELATIONS ................... 10-9 10.13 DEMONSTRATION : PEARSON CORRELATION WITH BIVARIATE CORRELATIONS ........ 10-10 10.14 LESSON SUMMARY ..................................................................................................... 10-11 10.15 LEARNING ACTIVITY ................................ ................................ ................................ .. 10 -12 TABLE OF CONTENTS v LESSON 11: REGRESSION ANALYSIS ................................................ 11-1 LESSON 12: NONPARAMETRIC TESTS .............................................. 12-1 LESSON 13: COURSE SUMMARY ........................................................ 13-1 APPENDIX A: INTRODUC TION TO STATISTICAL ANALYSIS REFERENCES 1 11.1 OBJECTIVES ................................ ................................ ................................ .................... 11 -1 11.2 INTRODUCTION ............................................................................................................... 11-1 11.3 SIMPLE LINEAR REGRESSION ......................................................................................... 11-1 11.4 SIMPLE LINEAR REGRESSION ASSUMPTIONS ................................................................. 11-3 11.5 REQUESTING SIMPLE LINEAR REGRESSION ................................................................... 11-4 11.6 SIMPLE LINEAR REGRESSION OUTPUT ........................................................................... 11-4 11.7 PROCEDURE : SIMPLE LINEAR REGRESSION ................................................................... 11-5 11.8 DEMONSTRATION : SIMPLE LINEAR REGRESSION ........................................................... 11-7 11.9 MULTIPLE REGRESSION ................................................................................................ 11-11 11.10 MULTIPLE LINEAR REGRESSION ASSUMPTIONS ........................................................ 11-11 11.11 REQUESTING MULTIPLE LINEAR REGRESSION ........................................................... 11-11 11.12 MULTIPLE LINEAR REGRESSION OUTPUT .................................................................. 11-11 11.13 PROCEDURE : MULTIPLE LINEAR REGRESSION ........................................................... 11-14 11.14 DEMONSTRATION : MULTIPLE LINEAR REGRESSION .................................................. 11-16 11.15 LESSON SUMMARY ..................................................................................................... 11-22 11.16 LEARNING ACTIVITY ................................ ................................ ................................ .. 11 -22 12.1 OBJECTIVES ................................ ................................ ................................ .................... 12 -1 12.2 INTRODUCTION ............................................................................................................... 12-1 12.3 NONPARAMETRIC ANALYSES ......................................................................................... 12-2 12.4 THE INDEP ENDENT SAMPLES NONPARAMETRIC ANALYSIS .......................................... 12-2 12.5 REQUESTING AN INDEPENDENT SAMPLES NONPARAMETRIC ANALYSIS ....................... 12-3 12.6 INDEPENDENT SAMPLES NONPARAMETRIC TESTS OUTPUT .......................................... 12-3 12.7 PROCEDURE : INDEPENDENT SAMPLES NONPARAMETRIC TESTS ................................... 12-5 12.8 DEMONSTRATION : INDEPENDENT SAMPLES NONPARAMETRIC TESTS .......................... 12-8 12.9 THE RELATED SAMPLES NONPARAMETRIC ANALYSIS ................................................ 12-11 12.10 REQUESTING A RELATED SAMPLES NONPARAMETRIC ANALYSIS ............................. 12-12 12.11 RELATED SAMPLES NONPARAMETRIC TESTS OUTPUT .............................................. 12-12 12.12 PROCEDURE : RELATED SAMPLES NONPARAMETRIC TESTS ...................................... 12-13 12.13 DEMONSTRATION : RELATED SAMPLES NONPARAMETRIC TESTS .............................. 12-16 12.14 LESSON SUMMARY ..................................................................................................... 12-19 12.15 LEARNING ACTIVITY ................................ ................................ ................................ .. 12 -20 13.1 COURSE OBJECTIVES REVIEW ................................ ................................ ........................ 13 -1 13.2 COURSE REVIEW : DISCUSSION QUESTIONS ................................................................... 13-1 13.3 NEXT STEPS ................................ ................................ ................................ .................... 13 -2 1.1 INTRODUCTION ................................ ................................ ................................ .................. A-1 1.2 REFERENCES ...................................................................................................................... A-1 INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS vi COURSE INTRODUCTION 0 -1 Lesson 0: Course Introduction 0.1 Introduction The focus of this two- day course is an introduction to the statistical component of IBM ® SPSS ® Statistics . This is an application -oriented course and the approach is practical. You'll take a look at several statistical techniques and discuss situations in which you would use each technique, the assumptions made by each method, how to set up the analysis using PASW ® 0.2 Course Objectives Statistics, as well as how to interpret the results. This includes a broad range of techniques for exploring and summarizing data, as well as investigating and testing underlying relationships. You will gain an understanding of when and why to use t hese various techniques as well as how to apply them with confidence, and interpret their output, and graphi cally display the results.
After completing this course students will be able to:
• Perform basic statistical analysis using select ed statistical techniques with PASW Statistics To support the achievement of this primary objective, students will also be able to: • Explain the basic elements of quantitative research and issues that should be considered in data analysis • Determine the level of measurement of variables and obtain appropriate summary statistics based on the level of measurement • Run the Frequencies procedure to obtain appropriate summary statistics for categorical variables • Request and interpret appropriate summary statistics for scale variables • Explain how to make inferences about populations from samples • Perform crosstab analysis on categorical variables • Perform a statistical test to determine whether there is a statistically significant relationship between categorical variables • Perform a statistical test to determine whether there is a statistically significant difference between two groups on a scale variable • Perform a statistical test to determine whether there is a statistically significant difference between the means of two scale variables • Perform a statistical test to determine whether there is a statistically significant difference among three or more groups on a scale dependent variable • Perform a statistical test to determine whether two scale variables are correlated (related) • Perform linear regression to determine whether one or more variables can significantly predict or explain a dependent variable • Perform non- parametric tests on data that don’t meet the assumptions for standard statistical tests 0.3 About SPSS SPSS ® Inc. , an IBM ® Company is a leading global provider of predictive analytics software and solutions. The Company’s complete portfolio of products - data collection, statistics, modeling and deployment - captures people's attitudes and opinions, predicts outcomes of future customer interactions, and then acts on these insights by embedding analytics into business processes. SPSS solutions address interconnected business objectives across an entire organization by focusing on the conver gence of analytics, IT architecture and business process. Commercial, government and academic customers worldwide rely on SPSS technology as a competitive advantage in attracting, INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 0 -2 retaining and growing customers, while reducing fraud and mitigating risk. S PSS was acquired by IBM ® 0.4 Supporting Materials in October 2009. For more information, visit http://www.spss.com. We use several datasets in the course because no one data file contains all the types of variables and relationships between them that are ideal for every technique we discuss. As much as possible, we try to minimize the need within one lesson to switch between datasets, but the first priority is to use appropriate data for each method. The following data files are used in this course: • Bank.sav • Drinks .sav • Census.sav • Employee data.sav • SPSS_CUST.sav 0.5 Course Assumptions General computer literacy. Completion of the "Introduction to PASW Statistics" and/or "Data Management and Manipulation with PASW Statistics " courses or experience with PASW Statistics including familiarity with, opening, defining, and saving data files and manipulating and saving output.
Basic statistical knowledge or at least one introductory level course in s tatistics is recommended.
Note about Default Startup Folder and Variable Display in Dialog Boxes In this course, all of the files used for the demonstrations and exercises are located in the folder c:\Train \Statistics _IntroAnalysis .
Note: If the course files are stored in a different location, your instructor will give you instructions specific to that location. Either variable names or longer variable labels will appear in list boxes in dialog boxes. Additionally, variables in list boxes can be ordered alphabetically or by their position in the file. In this course, we will display variable names in alphabetical order within list boxes. 1) Select Edit...Options 2) Select the General tab (if necessary) 3) Select Display names in the Variable Lists group on the General tab 4) Select Alphabetical 5) Select OK and OK in the information box to confirm the change INTRODUCTION TO STATISTICAL ANALYSIS 1 -1 Lesson 1: Introduction to Statistical Analysis 1.1 Objectives After completing this lesson students will be able to: • Explain the basic elements of quantitative research and issues that should be considered in data analysis To support the achievement of the primary objective, students will also be able to: • Explain the basic steps of the research process • Explain differences between populations and samples • Explain differences between experimental and non- experimental research designs • Explain differences between independent and dependent variables 1.2 Introduction The goal of this course is to enable you to perform useful analyses on your data using PASW Statistics . Keeping this in mind, these lessons demonstrate how to perform descriptive and inferential statistical analyses and creat e charts to support these analyses . This course guide will focus on the elements necessary for you to answer questions from your data. In this chapter , we begin by briefly reviewing the basic elements of quantitative research and issues that should be considered in data analysis. We will then discuss a number of statistical procedures that PASW Statistics performs. This is an application- oriented course and the approach will be practical. We will discuss: 1) The situations in which you would use each technique. 2) The assumptions made by the method. 3) How to set up the ana lysis using PASW Statistics. 4) Interpretation of the results. We will not derive proofs, but rather focus on the practical matters of data analysis in support of answering research questions. For example, we will discuss what correlation coefficients are, when to use them, and how to produce and interpret them, but will not formally derive their properties. This course is not a substitute for a course in statistics. You will benefit if you have had such a course in the past, but even if not, you will understand the basics of each technique after completion of this course. We will cover descriptive statistics and exploratory data analysis, and then examine relationships between categorical variables using crosstabulation tables and chi -square tests. Testing for mean differences between groups using T Tests and analysis of variance (ANOVA ) will be considered.
Correlation and regression will be used to investigate the relationships between interval /scale variables and we will also discuss some non-parametric techniques. Graphs comprise an integral part of the analyses and we will demonstrat e how to create and interpret these as well. 1.3 Basic Steps of the Research Process All research projects, whether analyzing a survey, doing program evaluations, assessing marketing campaigns, doing pharmaceutical research, etc., can be broken down into a number of discrete INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 1 -2 components . These components can be categorized in a variety of ways. We might summarize the main steps as : 1) Specify exactly the aims and objectives of the research along with the main hypotheses .
2) Define the p opulation and sample design. 3) Choose a method of data collection, design the research and decide upon an appropriate sampling strategy .
4) Collect the data. 5) Prepare the data for analysis .
6) Analyse the data .
7) Report the findings .
Some of these points may seem obvious, but it is surprising how often some of the most basic principles are overlooked, potentially resulting in data that is impossible to analy ze with any confidence. Each step is crucial for a successful research project and it is never too early i n the process to consider the methods that you intend to use for your data analysis. In order to place the statistical techniques that we will discuss in this course in the broader framework of research design, we will briefly review some of the consider ations of the first steps. Statistics and research design are highly interconnected disciplines and you should have a thorough grasp of both before embarking on a research project . This introductory chapter merely skims the surface of the issues involved i n research design. If you are unfamiliar with these principles, we recommend that you refer to the research methodology literature for more thorough coverage of the issues. Research Objectives It is important that a research project begin with a set of well -defined objectives. Yet, this step is often overlooked or not well defined. The specific aims and objectives may not be addressed because those commissioning the research do not know exactly which questions they would like answered.
This rather vague approach can be a recipe for disaster and may result in a completely wasted opportunity as the most interesting aspects of the subject matter under investigation could well be missed. If you do not identify the specific objectives, you will fail to collect the necessary information or ask the necessary question in the correct form. You can end up with a data file that does not contain the information that you need for your data analysis step. For example, you may be asked to conduct a survey "to find out a bout alcohol consumption and driving". This general objective could lead to a number of possible survey questions. Rather than proceeding with this general objective, you need to uncover more specific hypotheses that are of interest to your organization. T his example could lead to a number of very specific research questions, such as: “What proportion of people admits to driving while above the legal alcohol limit?” “What demographic factors (e.g. , age/sex/social class) are linked with a propensity to drunk -driving?” “Does having a conviction for drunk -driving affect attitudes towards driving while over the legal limit?” These specific research questions would then define the questionnaire items. Additionally, the research questions will affect the definition of the population and the sampling strategy. For example, the third question above requires that the responder have a drunk -driving conviction. Given that a INTRODUCTION TO STATISTICAL ANALYSIS 1 -3 relatively small proportion of the general population has such a conviction, you would need t o take that into consideration when defining the population and sampling design. Therefore, it is essential to state formally the main aims and objectives at the outset of the research so the subsequent stages can be done with these specific questions in mind.
1.4 Populations and Samples In studies involving statistical analysis it is important to be able to characterize accurately the population under investigation. The population is the group to which you wish to generalize your conclusions, while the sampl e is the group you directly study. In some instances the sample and population are identical or nearly identical; consider the Census of any country. In the majority of studies, the sample represents a small proportion of the population. In the example a bove, the population might be defined as those people with registered drivers' licenses. We could select a sample from the drivers' license registration list for our survey. Other common examples are: membership surveys in which a small percentage of members are sent questionnaires, medical experiments in which samples of patients with a disease are given different treatments, marketing studies in which users and non users of a product are compared, and political polling. The problem is to draw valid inferences from data summaries in the sample so that they apply to the larger population. In some sense you have complete information about the sample, but you want conclusions that are valid for the population. An important component of statistics and a large part of what we cover in the course involves statistical tests used in making such inferences. Because the findings can only be generalized to the population under investigation, you should give careful thought to defining the population of interest to yo u and making certain that the sample reflects this population. To state it in a simple way, statistical inference provides a method of drawing conclusions about a population of interest based on sample results. 1.5 Research Design With specific research goals and a target population in mind, it is then possible to begin the design stage of the research. There are many things to consider at the design stage. We will consider a few issues that relate specifically to data analysis and statistical techniques. This is not meant as a complete list of issues to consider. For example, for survey projects, the mode of data collection, question selection and wording, and questionnaire design are all important considerations. Refer to the survey research literature as wel l as general research methodology literature for discussion of these and other research design issues. First, you must consider the type of research that will be most appropriate to the research aims and objectives. Two main alternatives are experimental and non- experimental research. The data may be recorded using either objective or subjective techniques. The former includes items measured by an instrument and by computer such as physiological measures (e.g. heart -rate) while the latter includes observat ional techniques such as recordings of a specific behavior and responses to questionnaire surveys. Most research goals lend themselves to one particular form of research, although there are cases where more than one technique may be used. For example, a q uestionnaire survey would be inappropriate if the aim of the research was to test the effectiveness of different levels of a new drug to relieve high blood pressure. This type of work would be more suited to a tightly controlled experimental study in which the levels of the drug administered could be carefully controlled and objective measures of blood pressure could be accurately recorded. On the other hand, this type of laboratory -based work would not be a suitable means of uncovering people’s voting inte ntions.
INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 1 -4 The classic experimental design consists of two groups: the experimental group and the control group . They should be equivalent in all respects other than that those in the former group are subjected to an effect or treatment and the latter is not . Therefore, any differences between the two groups can be directly attributed to the effect of this treatment. The treatment variables are usually referred to as independent variables, and the quantity being measured as the effect is the dependent variable. There are many other research designs, but most are more elaborate variations on this basic theme. In non- experimental research, you rarely have the opportunity to implement such a rigorously controlled design. For example, we cannot randomly assign students to schools, however, the same general principles apply to many of the analyses you perform. 1.6 Independent and Dependent Variables In general, the dependent (sometimes referred to as the outcome ) variable is the one we wish to study as a function of other variables. Within an experiment, the dependent variable is the measure expected to change as a result of the experimental manipulation. For example, a drug experiment designed to test the effectiveness of different sleeping pills might employ the num ber of hours of sleep as the dependent variable. In surveys and other non- experimental studies, the dependent variable is also studied as a function of other variables. However, no direct experimental manipulation is performed; rather the dependent variabl e is hypothesized to vary as a result of changes in the other (independent) variables. Correspondingly, independent (sometimes referred to as predictor ) variables are those used to measure features manipulated by the experimenter in an experiment. In a no n-experimental study, they represent variables believed to influence or predict a dependent measure. Thus terms (dependent, independent) reasonably applied to experiments have taken on more general meanings within statistics. Whether such relations are v iewed causally, or as merely predictive, is a matter of belief and reasoning. As such, it is not something that statistical analysis alone can resolve.
To illustrate, we might investigate the relationship between starting salary (dependent) and years of education, based on survey data, and then develop an equation predicting starting salary from years of education. Here starting salary would be considered the dependent variable although no experimental manipulation of education has been performed. One way t o think of the distinction is to ask yourself which variable is likely to influence the other? In summary, the dependent variable is believed to be influenced by, or be predicted by, the independent variable(s). Finally, in some studies, or parts of studi es, the emphasis is on exploring and characterizing relationships among variables with no causal view or focus on prediction. In such situations there is no designation of dependent and independent variables. For example, in crosstabulation tables and corr elation matrices the distinction between dependent and independent variables is not necessary. It rather resides in the eye of the beholder (researcher). 1.7 Note about Default Startup Folder and Variable Display in Dialog Boxes In this course, all of the file s used for the demonstrations and exercises are located in the folder c: \Train \Statistics_IntroAnalysis .You can set the startup folder that will appear in all Open and Save dialog boxes. We will use this option to set the startup folder. Select Edit...Opt ions, and then select the File Locations tab Select the Browse button to the right of the Data Files text box Select Train from the Look In: drop down list, then select Statistics_IntroAnalysis from the list of folders and select Set button Click the Browse button to the right of the Other Files text box and repeat the process to set this folder to Train \Statistics_IntroAnalysis INTRODUCTION TO STATISTICAL ANALYSIS 1 -5 Figure 1.1 Set Default File Location in the Edit Options Dialog Box Note: If the course files are stored in a different location, your instructor will give you instructions specific to that location. Either variable names or longer variable labels will appear in list boxes in dialog boxes. Additionally, variables in list boxes can be ordered alphabetically or by their position in the file. In this course, we will display variable names in alphabetical order within list boxes. Select General tab Select Display names in the Variable Lists group on the General tab Select Alphabetical (Not shown) Select OK and then OK in the information box to confirm the change 1.8 Lesson Summary In this lesson, we reviewed the basic elements of quantitative research and issues that should be considered in data analysis. Lesson Objectives Review Students who have completed this lesson should now be able to: • Explain the basic elements of quantitative research and issues that should be considered in data analysis To support the achievement of the primary objective, students should now also be able to:
• Explain the basic steps of research process • Explain differences between populations and samples • Explain differences between experimental and non- experimental research designs • Explain differences between independent and dependent variables INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 1 -6 1.9 Lea rning Activity In this set of learning activities you won’t need any supporting material. 1. In each of the following scenarios, state the possible goals of the research, the type of design you can use, and the independent and dependent variables: a. The relat ionship between gender and whether a product was purchased. b. The difference between income categories (e.g., low, medium, and high) and number of years of education. c. The effect of two different marketing campaigns on number of items purchased. 2. In your own organization/field, are experimental studies ever done? If not, can you imagine how an experiment might be done to study a topic of interest to you or your organization?
Describe that and the challenges such an experimental design would encounter. UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -1 Lesson 2: Understanding Data Distributions – Theory 2.1 Objectives After completing this lesson students will be able to: • Determine the level of measurement of variables and obtain appropriate summary statistics based on the level of measurement To support the achievement of this primary objective, students will also be able to: • Describe the levels of measurement used in PASW Statistics • Use measures of central tendency and dispersion • Use normal distributions and z -scores Introduction Ideally, we would like to obtain as much information as possible from our data. In practice however, given the measurement level of our variab les, only some information is meaningful. In this lesson we will discuss level of measurement and see how this determines the summary statistics we can request. Business Context Understanding how level of measurement impacts the kind of information we can obtain is an important step before we collect our data. In addition, level of measurement also determines the kind of research questions we can answer, and so this is a critical step in the research process.
2.2 Levels of Measurement and Statistical Methods The term levels of measurement refers to the properties and meaning of numbers assigned to observations for each item. Many statistical techniques are only appropriate for data measured at particular levels or combinations of levels. Therefore, when possible, you should determine the a nalyses you will be using before deciding upon the level of measurement to use for each of your variables. For example, if you want to report and test the mean age of your customers , you will need to ask their age in years (or year of birth) rather than as king them to choose an age group into which their age falls. Because measurement type is important when choosing test statistics, we briefly review the common taxonomy of level of measurement. The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -2 The four major classifications that follow are found in many introductory statistics texts. They are presented beginning with the weakest and ending with those having the strongest measurement properties. Each successive level can be said to contain the properties of the preceding types and to record information at a higher level. • Nominal — In nominal measurement each numeric value represents a category or group identifier, only. The categories cannot be ranked and hav e no underlying numeric value. An example would be marital status, coded 1 (Married), 2 (Widowed), 3 (Divorced), 4 (Separated) and 5 (Never Married); each number represents a category and the matching of specific numbers to categories is arbitrary. Counts and percentages of observations falling into each category are appropriate summary statistics. S uch statistics as the mean (the average marital status?) would not be appropriate, but the mode would be appropriate (the most frequent category). • Ordinal — For ordinal measures the data values represent ranking or ordering information.
However, the difference between the data values along the scale is not equal. An example would be specifying how happy you are with your life, coded 1 (Very Happy), 2 (Happy ), and 3 (Not Happy). There are specific statistics associated with ranks; PASW Statistics provides a number of them mostly within the Crosstabs, Nonparametric and Ordinal Regression procedures. The mode and median can be used as summary statistics. • Inter val — In interval measurement, a unit increase in numeric value represents the same change in quantity regardless of where it occurs on the scale. For interval scale variables such summaries as means and standard deviations are appropriate. Statistical tec hniques such as regression and analysis of variance assume that the dependent (or outcome) variable is measured on an interval scale. Examples might be temperature in degrees Fahrenheit or SAT score.
• Ratio — Ratio measures have interval scale properties w ith the addition of a meaningful zero point; that is, zero indicates complete absence of the characteristic measured. For statistics such as ANOVA and regression only interval scale properties are assumed, so ratio scales have stronger properties than necessary for most statistical analyses. Health care researchers often use ratio scale variables (number of deaths, admissions, discharges) to calculate rates. The ratio of two variables with ratio scale properties can thus be directly interpreted. Money is an example of a ratio scale, so someone with $10,000 has ten times the amount as someone with $1,000. The distinction between the four types is summarized below .
Table 2.1 Level of Measurement Properties Level of Measurement Property Categories Ranks Equal Intervals True Zero Point Nominal Ordinal Interval Ratio These four levels of measurement are often combined into two main types, categorical consisting of nominal and ordinal measurement levels and scale (or continuous) consisting of interval and ratio measurement levels. UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -3 The measurement level variable attribute in PASW Statistics recognizes three measurement levels :
Nominal, Ordinal and Scale. The icon indicating the measurement level is displayed preceding the variable name or label in the variable lists of all dialog boxes. The following table shows the most common icons used for the measurement levels. Special dat a types, such as Date and Time variables have distinct icons not shown in this table. Table 2.2 Variable List Icons Measurement Level Data Type Numeric String Nominal Ordinal Not Applicable Scale Not Applicable Rating Scales and Dichotomous Variables A common scale used in surveys and market research is an ordered rating scale usually consisting of five- or seven- point scales. Such ordered scales are also called Likert scales and might be coded 1 (Strongly Agree, or Very Satisfied), 2 (Agree, or Satisf ied), 3 (Neither agree nor disagree, or Neutral), 4 (Disagree, or Dissatisfied), and 5 (Strongly Disagree, or Very Dissatisfied). There is an ongoing debate among researchers as to whether such scales should be considered ordinal or interval. PASW Statisti cs contains procedures capable of handling such variables under either assumption. When in doubt about the measurement scale, some researchers run their analyses using two separate methods, since each make different assumptions about the nature of the meas urement. If the results agree, the researcher has greater confidence in the conclusion. Dichotomous (binary) variables containing two possible responses (often coded 0 and 1) are often considered to fall into all of the measurement levels except ratio (at least as independent variables) .
As we will see, this flexibility allows them to be used in a wide range of statistical procedures Implications of Measurement Level As we have discussed, t he level of m easurement of a variable is important because it determines the appr opriate summary statistics, tables, and graphs to describe the data. The following table summarizes the most common summary measures and graphs for each of the measurement levels and PASW Statistics procedures that can produce them. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -4 Table 2.3 Summary of Descriptive Statistics and Graphs NOMINAL ORDINAL SCALE Definition Unordered Categories Ordered Categories Metric/Numeric Values Examples Labor force status, gender, marital status Satisfaction ratings, degree of education Income, height, weight Measures of Central Tendency Mode Mode Median Mode Median Mean Measures of Dispersion N/A Min/Max/Range, InterQuartile Range (IQR) Min/Max/Range, IQR, Standard Deviation/Variance Graph Pie or Bar Pie or Bar Histogram, Box & W hisker, Stem & Leaf Procedures Frequencies Frequencies Frequencies, Descriptives, Explore Measurement Level and Statistical Methods Statistics are available for variables at all levels of measurement for more advanced analysis. In practice, your choice of method depends on the questions you are interested in asking of the data and the nature of the measurements you make. The table below suggests which statistical techniques are most appropriate, based on t he measurement level of the dependent and independent variable.
Much more extensive diagrams and discussion are found in Andrews et al. (1981), or other standard statistical texts. Table 2.4 L evel of Measurem ent and Appropriate Statistical Methods Dependent Variable Independent Variables Nominal Ordinal Interval/Ratio Nominal Crosstabs Crosstabs Discriminant , Logistic Regression Ordinal Nonparametric tests, Ordinal Regression Nonparametric correlation, Optimal Scaling Regression Ordinal Regression Interval/ Ratio T Test, ANOVA Nonparametric Correlation Correlation , Linear Regression UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -5 Apply Your Knowledge 1. PASW Statistics distinguishes three levels of measurement . Which of these is not one of those levels? a. Categorical b. Scale c. Nominal d. Ordinal 2. True or false? An ordinal variable has all properties of a nominal variable? 3. Consider the dataset depicted below. Which statements are correct ?
a. The variable r egion is a n ordinal variable b. The variable a ge is a scale variable c. The variable a gecategory is an ordinal variab le d. The variable s alarycategory is a scale variable 2.3 Measures of Central Tendency and Dispersion Measures of central tendency and dispersion are the most common measures used to summarize the distribution of variables. We give a brief description of each of these measures below. Measures of Central Tendency Statistical measures of central tendency gi ve that one number that is often used to summarize the distribution of a variable. They may be referred to generically as the "average ." There are three main If in doubt about the measurement properties of your variables, you can apply a statistical technique that assumes weaker measurement properties and compare the results to methods making stronger assumptions. A consistent answer provides greater confidence in the conclusions. Best Practice INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -6 central tendency measures: mode, median, and mean. In addition, Tukey devised the 5% trimmed mean. • Mode : The mode for any variable is merely the group or class that contains the most cases.
If two or more groups contain the same highest number of cases, the distribution is said to be “ multimodal. ” This measure is more typically used on nominal or ordinal data and can easily be determined by examining a frequency table. • Median - If all the cases for a variable are arranged in order according to their value, the median is that value that splits the c ases into two equally sized groups. The median is the same as the 50th percentile. Medians are resistant to extreme scores, and so are considered robust measures of central tendency. • Mean : - The mean is the simple arithmetic average of all the values in the distribution (i.e. , the sum of the values of all cases divided by the total number of cases). It is the most commonly reported measure of central tendency. The mean along with the associated measures of dispersion are the basis for many statistical techniques. • 5% trimmed mean - The 5% trimmed mean is the mean calculated after the extreme upper 5% and the extreme lower 5% of the data values are dropped. Such a measure is resistant to extreme values. The specific measure that you choose will depend on a number of factors, most importantly the level of measurement of the variable. The mean is considered the most "powerful" measure of the three classic measures of central tendenc y. However, it is good practice to compare the median, mean, and 5% trimmed mean to get a more complete understanding of a distribution. Measures of Dispersion Measures of dispersion or variability describe the degree of spread, dispersion, o r variability around the central tendency measure. You might think of this as a measure of the extent to which observations cluster within the distribution. There are a number of measures of dispersion, including simple measures such as maximum, minimum, and range, common statistical measures , such as standard deviation and variance, as well as the interquartile range (IQR). • Maximum: Simply the highest value observed for a particular variable. By itself, it can tell us nothing about the shape of the distribution, merely how high the top value is. • Minimum: The lowest value in the distribution and, li ke the maximum, is only useful when reported in conjunction with other statistics. • Range: The difference between the maximum and minimum values gives a general impression of how broad the distribution is. It says nothing about the shape of a distribution a nd can give a distorted impression of the data if just one case has an extreme value. • Variance: Both the variance and s tandard deviation provide information about the amount of spread around the mean value. They are overall measures of how clustered around the mean the data values are. The variance is calculated by summing the square of the difference between the value and the mean for each case and dividing this quantity by the number of cases minus 1. If all cases had the same value, the variance (and sta ndard deviation) would be zero. The variance measure is expressed in the units of the variable squared. This can cause difficulty in interpretation , so more often the standard deviation is used. In general terms, the larger the variance, the more spread there is in the data, the smaller the variance, the more the data values are clustered around the mean. • Standard Deviation: The standard deviation is the square root of the variance which restores the value of variability to the units of measurement of the original variable. It is therefore easier to interpret. Either the variance or standard deviation is often used in conjunction with the mean as a basis for a wide variety of statistical techniques. • Interquartile Range (IQR) - This measure of variation is the range of values between the 25th and 75th percentile values. Thus, the IQR represents the range of the middle 50 percent of the sample and is more resistant to extreme values than the standard deviation. UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -7 Like the measures of central tendency, these meas ures differ in their usefulness with variables of different measurement levels. The variability measures, variance and standard deviation, are used in conjunction with the mean for statistical evaluation of the distribution of a scale variable. The other m easures of dispersion, although less useful statistically, can provide useful descriptive information about a variable. Apply Your Knowledge 1. True or false? The mode is that value that splits the cases into two equally sized groups. 2. True or false? Consider the table depicted below. The salaries of men are clustered tighter around their mean than the salaries of women around their mean? 2.4 Normal Distributions An important statistical concept is that of the normal distribution. This is a frequency (or probability) distribution which is symmetrical and is often referred to as the normal bell -shaped curve. The histogram below illustrates a normal distribution. The mean, median and mode exactly coincide in a perfectly normal distribution. And the proport ion of cases contained within any portion of the normal curve can be exactly calculated mathematically. Its symmetry means that 50% of cases lie to either side of the central point as defined by the mean.
Two of the other most frequently -used representat ions are the portions lying between plus and minus one standard deviation of the mean (containing approximately 68% of cases) and that between plus and minus 1.96 standard deviations (containing approximately 95% of cases, sometimes rounded up to 2.00 for convenience) . Thus, if a variable is normally distributed, we expect 95% of the cases to be within roughly 2 standard deviations from the mean. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -8 Figure 2.1 The Normal Distribution Many naturally occurring phenomena, such as height, weight and blood pressure, are distributed normally. Random errors also tend to conform to this type of distribution. It is important to understand the properties of normal distributions and how to assess the normality of particular distributions because of their theoretical importance in many inferential statistical procedures. 2.5 Standardized (Z-) Scores The properties of the normal distribution allow us to calculate a standardized score, often referred to as a z-score , which indi cates the number of standard deviations above or below the sample mean for each value. Standardized scores can be used to calculate the relative position of each value in the distribution. Z -scores are most often used in statistics to standardize variables of unequal scale units for statistical comparisons or for use in multivariate procedures. For example, if you obtain a score of 68 out of 100 on a test of verbal ability , this information alone is not enough to tell how well you did in relation to other s taking the test. However, if you know the mean score is 52.32, the standard deviation 8.00 and the scores are normally distributed, you can calculate the proportion of people who achieved a score at least as high as your own. The standardized score is calculated by subtracting the mean from the value of the observation in question (68- 52.32 = 15.68) and dividing by the standard deviation for the sample (15.68/8 = 1.96). Standardized Score = Standard Deviation Case Score - Sample Mean Therefore, the mean of a standardized distribution is 0 and the standard deviation is 1. In this case, your score of 68 is 1.96 standard deviations above the mean. UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -9 The histogram of the normal distribution above displays the distribution as a Z -score so the values on the x -axis are standard deviation units. From this figure, we can see only 2.5% of the cases are likely to have a score above 68 on the verbal ability test (1.96 standardized score). The normal distribution table (see below), found in an appendix of most statistics books, show s proportions for z -score values. Table 2.5 Normal Distribution Table A score of 1.96, for example, corresponds to a value of .025 in the “ one-tailed” column and .050 in the “ two -tailed” column. The former means that the probability of obtaining a z -score at least as large as +1.96 is .025 (or 2.5%), the latter that the probability of obtaining a z -score of more than +1.96 or less than -1.96 is .05 (or 5%) or 2.5% at each end of the distribution. You can see these cutoffs in the histogram above. As we mentioned, another advantage of standardized scores is that they allow for comparisons on variables measured in different units. For example, in addition to the verbal test score, you might have a mathematics test score of 150 out of 200 (or 75%). Although it appears that you did better on the mathematics test from the percentages alone, y ou would need to calculate the z-score for the mathematics test and compare the z -scores in order to answer the question. You might want to compute z -scores for a series of variables and determine whether certain subgroups of your sample are, on average, above or below the mean on these variables by requesting descriptive statistics or using the Case Summaries procedure. For example, you might want to compare a customer’s yearly revenue using z-scores. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -10 2.6 Requesting Standardized (Z- ) Scores The Descriptives procedure has an option to calculate standardized score variables. A new variable containing the standardized values is calculated for the specified variables. Creating standardized scores is accomplished by following these steps: 1) Choose variables to transform into standardized- scores. 2) Review the new variables that were created. 2.7 Standardized (Z-) Scores Output The Descriptives procedure provides descriptive statistics of the original variables. The standardized variables will appear in the Data Editor. Figure 2.2 Example of Descriptives Output 2.8 Procedure: Descriptives for Standardized (Z-) Scores The Descriptives procedure is accessed from the Analyze…Descriptive Statistics…Descriptives menu choice. W ith t he Descriptives dialog box open: 1) Place one or more scale variables in the Variable(s) box .
2) Select the Save standardized values as variables b ox.
Figure 2.3 Descriptives Dialog Box to Create Z -Scores UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -11 2.9 Demonstration : Descriptives for Z -Scores We will work with the Census.sav data file in this example. W e create standardized scores for number of years of education ( educ) and age of respondent ( age). W e would like to determine where respondents fall on the distribution of these variables. Detailed Steps for Z-Scores 1) Place the variable educ and age in the Variable(s) box .
2) Select the Save stand ardized values as variables box. Results from Z -Scores By default, the new variable name is the old variable name prefixed with the letter "Z". Two new variables, zeduc and zage, containing the z -scores of the two variables , are created at the end of the data file. These variables can be saved in your file and used in any statistical procedure. We observe that: • The f irst person (row) in the data file is below the average on education but above the average on age. Figure 2.4 Two Z -score Variables in the Data Editor Apply Your Knowledge 1. True or false? Only for variables of measurement level scale in PASW Statistics is it meaningful to calculate standardized scores ?
2. Consider the data below, where we computed standardized values for the variables educ (highest year of education) and salary (salary in dollars). Which of the following statements are correct ?
a. The observation with employee_id=49 has a salary very close to the mean salary. b. The observation with employee_id=50 has a salary that is more than one standard deviation above the mean. c. The observation with employee_id= 46 is more extreme i n her education than in salary. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -12 Additional Resources 2.10 Lesson Summary We explored the concept of the level of measurement and the appropriate summary statistics given level of measurement. We also discussed the normal distribution and z-s cores. Lesson Objectives Review Students who have completed this lesson should now be able to: • Determine the level of measurement of variables and obtain appropriate summary statistics based on the level of measurement To support the achievement of the primary objective, students should n ow also be able to:
• Describe the levels of measurement used in PASW Statistics • Use measures of central tendency and dispersion • Use normal distributions and z -scores For additional information on Level of Measurement and Statistical Tests, see: Andrews, Frank M, Klem, L., Davidson, T.N., O’Malley, P.M. and Rodgers, W.L. 1981. A Guide for Selecting Statistical Techniques for Analyzing Social Science Data. Ann Arbor, MI: Institute for Social Research, University of Michigan. Velleman, Paul F. and Wilkinson, L. 1993. “Nominal, Ordinal and Ratio Typologies are Misleading for Classifying Statistical Methodology,” The American Statistician , vol. 47, pp. 65 -72. Further Info UNDERSTANDING DATA DISTRIBUTIONS - THEORY 2 -13 2.11 Learning Activity The overall goal of this learning activity is to create standardized (Z -) scores for several variables. In this set of learning activities you will use the Drinks .sav data file.
1. Create standardized s cores for all scale variables ( price through alcohol). Which beverages have positive standardized s cores on every variable? What does this mean?
2. What is the most extreme z -score on each variable? What is the most extreme z -score across all variables? 3. What beverage is most typical of all beverages, that is, has z -score values closest to 0 for these variables ?
4. If the variable is normally distributed, what percentage of cases should be above 1 standard deviation from the mean or below 1 standard deviation from the mean? Calculate this percentage for a couple of the variables. Is the percentage of beverages with a n absolute z- score above 1 close to the theoretical value? The file Drinks .sav, a PASW Statistics data file that contains hypothetical data on 35 beverages . Included is information on their characteristics (e.g., % alcohol), price, origin, and a rating of quality. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 2 -14 DATA DISTRIBUTIONS FOR CATEGORICAL VARIABLES 3 -1 Lesson 3: Data Distributions for Categorical Variables 3.1 Objectives After completing this lesson students will be able to: • Run the Frequencies procedure to obtain appropriate summary statistics for categorical variables To support the achievement of this primary objective, students will also be able to: • Use the options in the Frequencies procedure • Interpret the results of the Frequencies procedure 3.2 Introduction As a first step in analyzing data, one must gain knowledge of the overall distribution of the individual variables and check for any unusual or unexpected values. You often want to examine the values that occur in a variable and the number of cases in each. For some variables, you want to summarize the distribution of the variable by examining si mple summary measures including the mode, median, and minimum and maximum values. In this chapter, we will review tables and graphs appropriate for describing categorical (nominal and ordinal) variables. Business Context Summaries of individual variables provide the basis for more complex analyses. There are a number of reasons for performing single variable analyses. One would be to establish base rates for the population sampled. These rates may be of immediate interest: W hat percentage of our customers is satisfied with services this year? In addition, studying a frequency table containing many categories might suggest ways of collapsing groups f or a more succinct and statistically appropriate table. When studying relationships between variables, the base rates of the separate variables indicate whether there is a sufficient sample size in each group to proceed with the analysis. A second use of s uch summaries would be as a data- checking device—unusual values would be apparent in a frequency table. The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 3 -2 3.3 Using Frequencies to Summarize Nominal and Ordinal Variables The most common technique for describing categorical data is a frequency analysis which provides a summary table indicating the number and percentage of cases falling into each category of a variable, as well as the number of valid and missing cases. We can also use the mode, which indicates the category with the highest frequency , and , if there is a large number of categories, the median (for ordinal variables), which is the value above and below which half the cases fall . Figure 3.1 Typical Frequencies Table To represent the frequencies graphically we use bar or pie charts. • A pie chart displays the contribution of parts to a whole. Each slice of a pie chart corresponds to a group that is defined by a s ingle grouping variable. • A bar chart displays the count for each distinct value or category as a separate bar, allowing you to compare categories vertically. Figure 3.2 P ie Chart illustrated DATA DISTRIBUTIONS FOR CATEGORICAL VARIABLES 3 -3 Figure 3.3 Bar Chart Illustrated 3.4 Requesting Frequencies Requesting Frequencies is accomplished by following these steps: 1) Choose variables for the Frequencies procedure. 2) Request additional summary statistics and graphs.
3) Review the procedure output to investigate the distribution of the variables including: a. Frequency Tables b. Graphs 3.5 Frequencies Output The information in the frequency table is comprised of counts and percentages :
• The Frequency column contains counts , i.e. , the number of occurrences of each data value. • The Percent column shows the percentage of cases in each category relative to the number of cases in the entire data set, including those with missing values. • The Valid Percent column contains the percentage of cases in each category relative to the number of valid (non- missing) cases. • The Cumulative percentage column contains the percentage of cases whose values are less than or equal to the indicated value. Cumulative percent is only useful for variables that are ordinal. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 3 -4 Figure 3.4 Example of Frequency Output 3.6 Procedure: Frequencies The Frequencies procedure is accessed from the Analyze…Descriptive Statistics…Frequencies menu choice. W ith the Frequencies dialog box open: 1) Place one or more variables in the Variable(s) box .
2) Open the Statistics dialog to request summary statistics .
3) Open the Charts dialog to request graphs .
Figure 3.5 Frequencies Dialog Box In the Statistics dialog box :
1) Ask for the appropriate measures of central tendency and dispersion. DATA DISTRIBUTIONS FOR CATEGORICAL VARIABLES 3 -5 Figure 3.6 Frequencies: Statistics Dialog Box In the Chart s dialog:
1) Ask for the appropriate chart based on the scale of measurement of the variable. Figure 3.7 Frequencies: Charts Dialog Box INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 3 -6 3.7 Demonstration : Frequencies We will work with the Census.sav data file in this lesson. In this example we exam ine the distribution of the variables marital and happy . These variables are either nominal or ordinal in scale of measurement. Detailed Steps for Frequencies 1) Place the variables marital and happy in the Variable(s) box 2) In the Statistics dialog, s elect Mode and Median in the Statistics dialog 3) In the Charts dialog, s elect Bar Chart in the Chart Types area and Percentages in the Chart Value area Results from Frequencies The first table produced is the table labeled Statistics. Figure 3.8 Statistics for Marital Status and General Happiness This table shows the number of cases having a valid value on Marital Status (2018) and General Happiness (2015), the number of cases having a (user - or system -) missing value (5 and 8, respectively) and the Mode and Median. The mode, the category that has the highest frequency, is a value of 1 and 2 respectively, and represents the category of “Married” for marital and the “Pretty Happy ” group for happy . The median, the middle point of the distribution (50th percentile), is a value of 2 for both variables. The second table shows the frequencies and percentages for each variable. This table confirms that almost half of the respondents are married. Since there is almost no missing data for marital status, the percentages in the Percent column and in the Valid Percent column are almost identical. Figure 3.9 Frequency Table of M arital Status Examine the table. Note the disparate category sizes. About half of the sample is married, and there is one category that has less than 5% of the cases. Before using this variable in a crosstabulation DATA DISTRIBUTIONS FOR CATEGORICAL VARIABLES 3 -7 analysis, should you consider combining some of the categories with fewer cases? Decisions about collapsing categories usually have to do with which groups need to be kept distinct in order to answer the research question asked, and the sample sizes for the groups. For example, could we create a “was previously married” group? The bar chart summarizes the distribution that we observed in the frequency table and allows us to “see” the distribution. Figure 3.10 Bar Chart of Marital Status For the variable happy, over half of the people fall into one category, pretty happy. Might it be interesting to look at the relationship between this variable and marital status: to what ext ent is general happiness related to marital status ? For a nominal variable (where the order of the categories is arbitrary) sorting the table and graph descending on counts gives better insight in what the main categories are (use the Format subdialog box to sort descending on counts). Tip INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 3 -8 Figure 3.11 Frequency Table of General Happiness Next we view a bar chart based on the general happiness variable. Does the picture make it easier to understand the distribution?
Figure 3.12 Bar Chart of General Happiness For an ordinal variable, sorting the categories on descending/ascending counts (which was useful for nominal variables) will disturb the natural order of categories and so is not as useful for an ordinal variable. Note DATA DISTRIBUTIONS FOR CATEGORICAL VARIABLES 3 -9 Apply Your Knowledge 1. See the output belo w. Which statements are correct? a. The median is an appropriate statistic to report for the variable region .
b. The region that has the highest frequency is the North. c. The cumulativ e percent is meaningful for region.
d. The columns Percent and Valid Percent are identical because there are no missing values on region 2. See the output below . Which bar chart is best to present the distribut ion of the ordinal variable age of employees , in categories? Bars sorted descending on count (A) or sort ed on ascending value (B)? 3. See the table below (a frequency table of HAPPINESS OF MARRIAGE, with those not married defined as missing). Which statements are correct ?
a. 29.5% of those married are very h appy in their marriage. b. The mode is the category pretty happy.
c. 96.9% of those married are pretty happy or very happy INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 3 -10 Additional Resources 3.8 Lesson Summary In this lesson we used the Frequencies procedure to explore the distribution of categorical variables, via both tables and graphs. Lesson Objectives Review Students who have completed this lesson should now be able to: • Run the Frequencies procedure to obtain appropriate summary statistics for categorical variables To support the achievement of the primary objective, students should now also be able to: • Use the options in the Frequencies procedure • Interpret the results of the Frequencies procedure 3.9 Learning Activity The overall goal of this learning activity is to run Frequencies to explore the distributions of several variables. In the exercises you will use the data file Census.sav.
The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials For additional information on how to present data in tables and graphs, see:
Few, Stephen. 2004. Show Me the Numbers:
Designing Tables and Graphs to Enlighten . Analytics Press Further Info DATA DISTRIBUTIONS FOR CATEGORICAL VARIABLES 3 -11 1. Run the Frequencies procedure on the following variables: sex , wrkstat (Labor Force Status) , paeduc (Father’s highest degree) , and satjob (Job or Housework) . What is the scale of measurement for each? Request appropriat e summary statistics and charts. 2. For wh ich of these variables is it appropriate to use the median? What conclusions can you draw about the distributions of these variables? 3. What percent of respondents have a bachelor’s degree, or higher? What percent of respondents are working? 4. How might you combine some of the categories of wrkstat to insure that there are a sufficient number of respondents in each category? INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 3 -12 DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -1 Lesson 4: Data Distributions for Scale Variables 4.1 Objectives After completing this lesson students will be able to: • Request and interpret appropriate summary statistics for scale variables To support the achievement of this primary objective, students will also be able to: • Use the options in the Frequencies , Descriptives , and Explore procedures • Interpret the results of the Frequencies, Descriptives, and Explore procedures 4.2 Introduction As a first step in analyzing your data, you must first gain knowledge of the overall distribution of the individual variables and check for any unusual or unexpected values. You often want to examine the values that occur in a variable and the number of cases in each. For some variables, you want to summarize the distribution of the variable by examining simple summary measures including minimum and maximum values for the range. Frequently used summary measures describe the central tendency of t he distribution, such as the arithmetic mean, and dispersion, the spread around the central point. In this lesson, we will review tables and graphs appropriate for describing scale (interval and ratio) variables. Business Context Summaries of individual v ariables provide the basis for more complex analyses. There are a number of reasons for performing single variable analyses. One would be to establish base rates for the population sampled. These rates may be of immediate interest: W hat is the average cust omer satisfaction? In addition, studying distributions might suggest ways of collapsing information for a more succinct and statistically appropriate table. When studying relationships between variables, the base rates of the separate variables indicate whether there is a sufficient sample size in each group to proceed with the analysis. A second use of such summaries would be as a data- checking device, as unusual values would be apparent in tables. 4.3 Summarizing Scale Variables Using Frequencies When working with categorical variables, frequency tables containing counts and percentages are appropriate summaries. For a scale variable, c ounts and percentages may still be of interest, especially when the variables can take only a limited number of distinct values. For example, when working with a one to five point rating scale we might be very interested in knowing the percentage of respondents who reply “Strongly Agree.” However, as the number of possible response values The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -2 increases, frequency tables based on interval scale variables become less useful. Suppose we asked respondents for their family income to the nearest dollar? It is likely that each response would have a different value and so a frequency table would be quite lengthy and not particularly helpful as a summary of the variable. In data cleaning, you might find a frequency table useful for examining possible clustering of cases on specific values or looking at cumulative percentages. But, beware of using frequency tables for scale variables with many values as they can be very long. If the variables of interest are scale we can expand the summaries to include means, standard deviations and other statistical measures. You will want to spend some time looking over the sum mary statistics you requested. Do they make sense, or is something unusual? For a categorical variable, we request a pie chart or a bar chart to graphically display the distribution of the v ariable. For a scale variable, a histogram is used to display the distribution. 4.4 Requesting Frequencies Requesting statistics and a graphical display is accomplished by following these steps: 1) Select variables in the Frequencies procedure. 2) Request additional summary statistics and graphs. 3) Review th e procedure output to investigate the distribution of the variables including: a. Frequency t ables (if requested) b. Statistics tables c. Graphs 4.5 Frequencies Output Statistics for the variable are presented in a separate table. A normal curve can be superimposed on the histogram and helps you to judge whether the variable is normally distributed. Tip DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -3 Figure 4.1 Example of Summary Statistics for Frequencies Output A histogram shows the distribution graphically. A histogram has bars, but, unlike the bar chart, they are plotted along an equal interval scale. The height of each bar i s the count of values of a quantitative variable falling within the interval. A histogram shows the shape, center, and spread of the distribution. Figure 4.2 Example of Histogram for Frequencies Output INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -4 4.6 Procedure: Frequencies The Frequencies procedure is accessed from the Analyze…Descriptive Statistics…Frequencies menu choice. W ith the Frequencies dialog box open: 1) Place one or more variables in the Variable(s) box .
2) Deselect the Display frequency tables check box for variables with many values .
3) Open the Statistics dialog to request summary statistics .
4) Open the Charts dialog to request graphs .
Figure 4.3 Frequencies Dialog Box In the Statistics dialog: 1) Select appropriate measures of central tendency 2) Select appropriate measures of dispersion DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -5 Figure 4.4 Frequencies: Statistics Dialog Box In the Charts dialog: 1) Ask for a histogram for scale variables. 2) Op tionally, superimpose a normal curve on the histogram .
INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -6 Figure 4.5 Frequencies: Charts Dialog Box 4.7 Demonst ration : Frequencies We will work with the Census.sav data file in this lesson. In this demonstrati on we examine the distribution of number of brothers and sisters ( sibs ) and res pondent’s age . W e would like to see the distribution of these variables. Detailed Steps for Frequencies 1) Place the variables sibs and age in the Variable(s) box 2) Deselect the Display frequency tables check box for variables with many values 3) Select Mode, Median , Mean , Minimum , Maximum and Standard Deviation in the Statistics dialog 4) Select Histogram s and Show n ormal curve on histogram in the Charts dialog If you request hi stograms and summary statistics for scale variables with many categories , you might want to uncheck (turn off) Display frequency tables in the Frequencies dialog box, as there may be almost as many distinct values as there are cases in the data file. Note DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -7 Results from Frequencies The table labeled Statistics shows the requested statistics. Figure 4.6 Summary Statistics for Number of Brothers and Sisters and Age of Res pondent This table shows the number of cases having a valid value on sibs (2021) and age (2013), the n umber of cases having a (user - or system -) missing value (2 and 10, respectively) and measures of central tendency and dispersion. The minimum value is 0 and the maximum value is 55 (seems unusual) for number of siblings. For age, the minimum value is 18 and the maximum value 89. Note, that the means and medians within each variable are similar, indicating that the variables are roughly normally distribu ted within the defined range. We can visually check the distribution of these variables with a histogram. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -8 Figure 4.7 Histogram of Number of Brothers and Sisters We can see that the lower range of values is truncated at 0 and the number of people is greatest between 0 to 6 siblings, although we do have some extreme values. The distribution is not normal. DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -9 Figure 4.8 Histogram of Age of Respondent We can see that the lower range of values is truncated at 18 and the number of people is highest in the middle age values (the "baby boomers") with the number of cases tapering off at the higher ages as we would expect. Thus, the age variable for respondents of this sa mple of adults is roughly normally distributed. Apply Your Knowledge 1. Suppose we have a variable region (with the categories north/east/south/west). Which of these statements is true? a. The mean is a meaningful statistic for region b. The standard deviation is a meaningful statistic for region c. The median is a meaningful statistic for region INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -10 2. See output below, with statistics for two variables : Current Salary and Beginning Salary (data collected on employees) . Which statements are correct?
a. There are 474 cases in the dataset b. Both variables are skewed to the right (meaning: there are employees with some large salaries compared to the average c. Half of the employees have a current salary below 30,750. d. The highest current salary is 135, 000.
3. See the histogram below for Current Salary (data collected on employees ). W hich of these statements is correct?
a. The variable seems normally distributed b. The variable is skewed to the right c. The standard deviation would be smaller, if the case with salary of 135, 000 would be removed from the histogram . DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -11 4.8 Summarizing Scale Variables using Descriptives The Descriptive procedure is a good alternative to Frequencies when the objective is to summarize scale variables. Descriptives is usually used to provide a table of st atistical summaries (means, standard deviations, variance, minimum, maximum, etc.) for several scale variables. The Descriptives procedure also provides a succinct summary of the number of cases with valid values for each variable included in the table as well as the number of cases with valid values for all variables included in the table. These summaries are quite useful in evaluating the extent of missing values in your data and in identifying variables with missing values for a large proportion of the data. 4.9 Requesting Descriptives Running Descriptives is accomplished by following these steps: 1) Select variables for the Descriptives procedure.
2) Review the procedure output to investigate the distribution of the variables. 4.10 Descriptives Output The figure below shows the Descriptives output table for a few variables. Figure 4.9 Example Descriptives Output The minimum and maximum provide an efficient way to check for values outside the expected range.
In general, this is a useful check for categorical variables as well. Thus, although mean and standard deviation are not relevant for respondent’s sex, minimum and maximum for this variable show that there are no values outside the expected range. The last row in the table labeled Valid N (listwise) gives the number of cases that have a valid value on all of variables appearing in the table. In this example, 1333 cases have valid values for all three variables listed. Although this number is not par ticularly useful for this set of variables, it would be useful for a set of variables that you intended to use for a specific multivariate analysis. As you p roceed with your analysis plans, it is helpful to know how many cases have complete information and which variables are likely to be the main sources of potential problem s. 4.11 Procedure: Descriptives The Descriptiv es procedure is accessed from the Analyze…Descriptive Statistics…D escriptives menu choice. W ith the Descriptives dialog box open: 1) Place one or more variables in the Variable(s) box .
INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -12 Figure 4.10 Descriptives Dialog Box Only numeric variables appear in the Descriptives dialog box. The Save standardized values as variables feature creates new variables that are standardized forms of the original variables. These new variables, referred to as z -scores, have values standardized to a mean of 0 and standard deviation of 1. The Options button allows you to select additional summary statistics to display. You can also select the display order of the variables in the table (for example, by ascending or descending mean value). 4.12 Demonstration : Descriptives We will work with the Census.sav data file in this lesson. In this example we examine the summary statistics of number of siblings, respondent’s age, education and respondent ´s gender. W e would like to see the summary statistics for these variables, as well as how much missing data there is , and if there are unusual cases. Detailed Steps for Descriptives 1) Place the variables sex, s ibs , educ , and age in the Variable(s) box . Results from Descriptives The table labeled Descriptive Statistics contains the statistics. Figure 4.11 Descriptives Output DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -13 The column labeled N show s the number of valid observations for each variable in the table . W e see there is little variation in the number of valid observations. The number of valid cases can be a useful check on the data and help us determine which variables might be appropriate for specific analyses. Here 2006 cases have valid values for the entire set of questions. The minimum and maximum provide an efficient way to check for values outside the expected range.
Here the maximu m for the variable sibs seems high and deserves further investigation. 4.13 Summarizing Scale Variables using the Explore Procedure Exploratory data analysis (EDA ) was primarily developed by John Tukey. He devised several statistical measu res and plots designed to reveal data features that might not be readily apparent from standard statistical summaries. Exploratory data analysis can be viewed either as an analysis in its own right, or as a set of data checks that investigators perform bef ore applying inferential testing procedures. These methods are best applied to variables with at least ordinal (more commonly interval) or scale properties and which can take many different values. The plots and summaries would be less helpful for a vari able that takes on only a few values (for example, a five- point scale). 4.14 Requesting Explore Running Explore is accomplished with these steps: 1) Select variables on which to report statistics in the Dependent List box 2) Select grouping variables in the Factor box 3) Request additional summary statistics and graphs. 4) Review the procedure output to investigate the summary statistics and distribution of the variables including tables and graphs Explore Output The Descriptives table displays a series of descriptive statistics for a ge. From the previous table (not shown), we know that these statistics are based on 1763 respondents. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -14 Figure 4.12 Summaries for Age of Respondent First, several measures of central tendency appear: the Mean, 5% Trimmed Mean, and Median.
These statistics attempt to describe with a single number where data values are typically found, or the center of the distribution. Useful information about the dist ribution can be gained by comparing these values to each other. If the mean were considerably above or below the median and trimmed mean, it would suggest a skewed or asymmetric distribution. The measures of central tendency are followed in the table by several measures of dispersion or variability. These indicate to what degree observations tend to cluster or be widely separated. Both the standard deviation (Std.Dev iation) and Variance (standard deviation squared) appear. The standard error ( Std.E rror) is an estimate of the standard deviation of the mean if repeated samples of the same size (here 1763) were taken. It is used in calculating the 95% confidence interval for the sample mean. Technically speaking, if we would draw 100 samples of this size (1763) and construct a 95% confidence for the mean for each of the 100 samples, then the expectation is that 95 out of these 100 intervals will contain the (unknown) population mean. Also appearing is the Interquartile R ange (often abbreviated to IQR ) which is essentially the range between the 25th and the 75th percentile values. It is a variability measure more resistant to extreme scores than the standard deviation. We also see the Minimum, Maximum and Range . The final two statistical measures, S kewness and Kurtosis , provide numeric summaries about the shape of the distribution of the data. Skewness is a measure of the symmetry of a distribution. It measures the degree to which cases are clustered towards one end of the distribution. It is normed so that a symmetric distribution has zero skewness. A positive skewness value indicates bunching on the left and a longer tail on the right (for example, income distribution); negative skewness follows the reverse pattern. The standard error of skewness al so appears in the Descriptives table and we can use it to determine if the data are significantly skewed. One method is to use the standard errors to calculate the 95% confidence interval around the skewness. If zero is not in this range, we could conclude that the distribution was skewed. A second method is to compare the skewness value to 1.96* (S tandard error of skewness) from zero. Kurtosis also has to do with the shape of a distribution and is a measure of how much of the data is concentrated near the center, as opposed to the tails, of the distribution. It is normed to the normal curve (for which kurtosis is zero). As an example, a distribution with longer tails and more peaked in the middle than a normal is referred to as a leptokurtic distribution and would have a positive kurtosis DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -15 measure. On the other hand, a platykurtic distribution is a flattened distribution and has negative kurtosis values. A standard error for kurtosis also appears. The same methods used for evaluating skewness can be used to e valuate the kurtosis values. Since most analysts are content to view histograms in order to make judgments regarding the distribution of a variable, skewness and kurtosis are infrequently used. Figure 4.13 Histogram of Age of Respondent The histogram shows the shape of the distribution. For this sample of adults, age is roughly normally distributed , except that the distribution is truncated below 18 years .
Boxplots, also referred to as box & whisker plots, are a more easily interpreted plot to convey the same information about the distribution of a variable. In addition, the boxplot graphically identifies outliers. Below we see the boxplot for hours worked. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -16 Figure 4.14 Boxplot of Hours Worked The vertical axis represents the scale for the number of hours worked. The solid line inside the box represents the median or 50th percentile. The top and bottom borders (referred to as “ hinges”) of the box correspond to the 75th and 25th percentile values of hours worked and thus define the interquartile range (IQR). In other words, the middle 50% of data values fall within the box. The “whiskers” (vertical lines extending from the top and bottom of the box) are the last data values that lie within 1.5 box lengths (or IQRs) of the respective hinges (borders of box). Tukey considers data points more than 1.5 box lengths from a hinge to be "outliers ." These points are marked with a circle.
Points more than 3 box lengths (IQR) from a hinge are considered by Tukey to be “far out” points and are marked with an asterisk symbol (there are none here). This plot has many outliers. If a single outlier exists at a data value, the case sequence number appears bes ide it (an ID variable can be substituted), which aids data checking. If the distribution were symmetric, the median would be centered within the box. In the plot above, the median is toward the bottom of the box, indicating a positively skewed distribut ion. 4.15 Procedure: Explore The Explore procedure is accessed from the Analyze…Descriptive Statistics …Explore menu choice. With the Explore dialog box open: 1) Place one or more scale variables to be summarized in the Dependent list box .
2) The Factor list box can contain one or more categorical variables, and if used would cause the procedure to present summaries for each cat egory of the factor variable(s).
3) We can request specific statistical summaries and plots using the Statistics and Plots buttons .
4) The Options button specifies how missing data will be handled. DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -17 Figure 4.15 Explore Dialog Box In the Plots dialog: 1) Request a histogram rather than a stem and leaf plot. The stem & leaf plot (devised by Tukey) is modeled after the histogram, but contains more information. For most purposes, the histogram is easier to interpret and more useful. By default, a boxplot will be displayed for each scale variable. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -18 Figure 4.16 Explore: Plots Dialog Box In the Options dialog the user specifies how to deal with missing values :
1) Request pairwise rather than listwise deletion. Figure 4.17 Explore: Options Dialog Box When several variables are used you have a choice as to whether the analysis should be based on only those observations with valid values for all variables in the analysis (called listwise deletion), or whether missing values should be excluded separately for each variable (called pairwise deletion).
When only a single variable is considered both methods yield the same result, but they will not give identical answers when multiple variables are analyzed in the presence of missing values. DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -19 Rarely used, the Report values option includes cases with user -defined missing values in frequency analyses, but excludes them from summary statistics and charts.
4.16 Demonstration : Explore We will work with the Census.sav data file in this lesson. In this example we examine the summar y statistics of age, and educ. W e would like to see summary statistics for the mentioned variables, as well as how much missing data we have, and if there are unusual cases. Detailed Steps for Explore 1) Place the variables age, and educ in the Dependent list box 2) Deselect the Stem and leaf and select Histogram in the Plots dialog 3) Request Exclude cases pairwise method in the Options dialog Results from Explore The Explore procedure produces two tables followed by the requested charts for each variable. The first table, Case Processing Summary, displays the number of valid and missing cases for each variable. Each variable has little missing data. For example, 2013 cases (r espondents) had valid values for age, while 0.5% were missing. Figure 4.18 Explore Case Processing Table The Descriptives table displays a series of descriptive statistics for a ge and educ . From the previous table, we know that these statistics are based on 2013 and 2018 respondents, respectively. Comparing measures of central tendency can provide useful information about the distribution. Here the mean, median and 5% trimmed mean are very close within each variable and this suggests either that there are not many extreme scores, or that the number of high and low scores is balanced. If the mean were considerably above or below the median and trimmed mean, it would suggest a skewed or asymmetric distr ibution. A perfectly symmetric distribution, the normal distribution, would produce identical means, medians and trimmed means. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -20 Figure 4.19 Summaries for Age of Respondent and Highest Year of School Complete d The standard deviation of age, 17.351, indicates a variation around the mean of plus or minus 17 years. The standard error is used in calculating the 95% confidence interval for the sample mean.
The interquartile range of 26 indicates that the middle 50% of the sample lies within a range of 26 years. DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -21 The shape of the distribution can be of interest in its own right. Also, assumptions are made about the shape of the data distribution within each group when performing significance tests on mean diffe rences between groups. The shape of the distributions of both these variables does not appear problematic. Figure 4.20 Histogram of Age of Respondent Figure 4.21 Histogram of Highest Year of School Completed INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -22 Below we see the boxplots for age and educ . Boxplots are particularly useful for obtaining an overall “ feel ” for a distribution. The median tells us the location or central tendency of the data. The length of the box indicates the amount of spread within the data, and the position of the median in relation to the box tells us something of the nature of the distr ibution. Boxplots are also useful when comparing several groups. Figure 4.22 Boxplot of Age of Respondent The plot for age has no outliers while the plot for educ has a few outliers on the lower end of the distribution. Figure 4.23 Boxplot of Highest Year of School Completed If suspicious outliers appear in your data you should check whether they are data errors. If not, you need to consider whether you wish them included in your analysis. This is especially problematic when dealing with a small sample (not the case here), since an outlier can substantially influence the analysis. We would not argue that something of interest always appears through us e of the methods of exploratory data analysis. However, you can quickly glance over these results, and if anything strikes DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -23 your attention, pursue it in more detail. The possibility of detecting something unusual encourages the use of these techniques. Apply Your Knowledge 1. See the output below, with statistics for Current Salary (data collected on employees ). Which statements are correct? a. If repeated samples of the same sample size (here 467) were taken, and we record the sample mean for each sample, then we expect a standard deviation of 795.399 for these sample means .
b. The m edian is lower than the mean, indicating that the dis tribution is skewed to the left. c. 50% of the salaries are in a range of salary of 13,500. 2. True or false? See the boxplot for variable X. The boxplot of this variable indicates a normal distribution? INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -24 Additional Resources 4.17 Lesson Summary In this lesson we explored how to obtain summary statistics for scale variables. We reviewed three procedures—Frequencies, Descriptives and Explore —that can help us obtain that information. Lesson Objectives Review Students who have completed this lesson should now be able to: • Request and interpret appropriate summary statistics for scale variables To support the achievement of the primary objective, students should now also be able to: • Use the options in the Frequencies , Descriptives , and Explore procedures • Interpret the results of the Frequencies, Descriptives, and Explore procedures For additional information on Exploratory data analysis, see: Tukey, John W. 1977. Exploratory Data Analysis .
Reading, MA: Addison- W esley.
Further Info DATA DISTRIBUTIONS FOR SCALE VARIABLES 4 -25 4.18 Learning Activity In this set of learning activities you will use the Drinks .sav data file. The overall goal of t his learning activity is to obtain summary statistics for scale variables. 1. Run Frequencies on the variable alcohol, requesting the summary statistics median and mean, plus a histogram with a superimposed normal curve. Suppress the display of the frequency table. 2. What is the value of value of alcohol that splits the distribution in half? Is the median the same as the mean? Which value is lower? What does that tell you about the shape of the distribution of alcohol? 3. Does the histogram verify your description of the distribution of alcohol ? How does it differ from a normal distribution? 4. Run Descriptives to obtain default statistics for price and calories . On which variable is there more dispersion? Is it even realistic to compare these two variables since they are on different scales? 5. Continuing your analysis of price and calories , run the Explore procedure for these two variables. Request a histogram in addition to the defaults. 6. Does the standard error of each variable help you better determine which variable has more dispersion? 7. Review th e boxplots and histogram for each variable. Which one has more outliers? What are the outliers on each? Which variable now appears to have more dispersion, based on these graphs? Does that match what you expected based on the statistics? The file Drinks .sav, a PASW Statistics data file that contains data on 35 beverages . Included is information on their characteristics (e.g., % alcohol), price, origin, and a rating of quality. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 4 -26 MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -1 Lesson 5: Making Inferences about Populations from Samples 5.1 Objectives After completing this lesson students will be able to: • Explain how to make inferences about populations from samples To support the achievement of the primary objective, students will also be able to: • Explain the influence of sample size • Explain the nature of probability • Explain hypothesis testing • Explain different types of statistical errors and power • Exp lain differences between statistical and practical importance 5.2 Introduction Ideally, for any analysis we would have data about everyone we wished to study (i.e. , the whole population). In practice, we rarely have information about all members of our populat ion and instead collect information from a representative sample of the population. However, our goal is to make generalizations about various characteristics of the population based on the known facts about the sample. In this lesson we discuss the requir ements of making inferences to populations from samples. Business Context Understanding how to make inferences from a sample to a population is the basis of inferential statistics. This allows us to reach conclusions about the population without the need t o study every single individual. Hypothesis testing allows researchers to develop hypotheses which are then assessed to determine the probability or likelihood of the findings. Supporting Materials None 5.3 Basics of Making Inferences about Populations from Samples We choose a sample with the intention of using the data from that sample to make inferences about the “true” values in the population. These population measures are referred to as parameters while the equivalent measures from samples are known as statistics. It is unlikely that we will know the population parameters; therefore we use the sample statistics to infer what these population values will be. An important distinction between parameters and statistics is that parameters are fixed (although often not known) while statistics vary from one sample to another. Due to the effects of random variability, it is unlikely that any two samples drawn from the same population will produce the same statistics. By plotting the values of a particular statis tic (e.g., the mean) from a large number of samples, it is possible to obtain a sampling distribution of the statistic. For small numbers of samples, the mean of the sampling distribution may not closely resemble that of the population.
However, as the num ber of samples taken increases, the closer the mean of the sampling distribution INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -2 (the mean of all the means, if you like) gets to the population mean. For an infinitely large number of samples, the mean will be exactly the same as the population mean. Addi tionally, as sample size increases, the amount of variability in the distribution of sample means decreases. If you think of variability in terms of the error made in estimating the mean, it should be clear that the more evidence you have (i.e. , the more c ases in your sample), the smaller will be the error in estimating the mean. If repeated random samples of size N are drawn from any population, then as N becomes large, the sampling distribution of sample means approaches normality —a phenomenon known as the Central Limit Theorem . This is an extremely useful statistical concept as it does not require that the original population distribution is normal. In the next section, we'll take a closer look at the influence of sample size on the precision of the statistics. 5.4 Influence of Sample Size In statistical analysis sample size plays an important role, but one that can easily be overlooked since a minimum sample size is not required for the most commonly used statistical tests. Here we will demonstrate the effect of sample size in two common data analysis situations: crosstabulation tables and mean summaries. Precision of Percentages Precision is strongly influenced by the sample size. In the figures below we present a series of crosstabulation tables containing identical percentages, but with varying sample sizes. We will observe how the test statistics change with sample size and relate this result to the precision of the measurement. The Chi -square test of independence will be presented for each table as par t of the presentation of the effect of changing sample size. (See the lesson “ Combining Categorical Variables, Crosstabulations and the Chi -Square Statistic for a detailed discussion of the chi -square test. ) Sample Size of 100 The table below displays responses of men and women to a question asking for which candidate they would vote. MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -3 Figure 5.1 Crosstab Table with Sample of 100 • 46 % of the men and 54 % of the women choose candidate A —an 8% difference • The Chi-square test assesses whether men differ from women in the population • Significance value of .424 indicates probability that men and women share the same view (do not differ significantly) concerning candidate choice • Conclude there is no gender difference in a sample of 100 people Sample Size of 400 Now we view a table with percentages identical to the previous one, but based on a sample of 400 people, four times as large as before. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -4 Figure 5.2 Crosstabulation Table with Sample of 400 • 46 % of the men and 54 % of the women choose candidate A —an 8% difference • Significance value of .110 for Chi -Square test indicates probability that men and women share the same view (do not differ significantly) concerning candidate choice • Conclude there is no gender difference in a sample of 400 people Sample Size of 1,600 Finally we present the same table of percentages, but increase the sample size to 1,600; the increase is once again by a factor of four. MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -5 Figure 5.3 Crosstabulation Table with Sample of 1,600 • 46 % of the men and 54 % of the women choose candidate A -- an 8% difference • Significance value of .001 for Chi-Square test indicates probability that men and women share the same view (do differ significantly) concerning candidate choice • Conclude there is a gender difference in a sample of 1600 people • We have more precise estimates as our sample size increases and the 8% sampl e difference can more accurately be estimated Sample Size and Precision In the series of crosstabulation tables above we saw that as the sample size increased we were more likely to conclude there was a statistically significant difference between two groups when the magnitude of the sample difference was constant (8%). This is because the precision with which we estimate the population percentage increases with increasing sample size. This relation can be approximated (see note for the exact relationship) by a simple equation: the precision of a sample proportion is approximately equal to one divided by the square root of the sample size. The table below displays the precision for the sample sizes used in our examples. Table 5.1 Sample Size and Precision for Different Sample Sizes Sample Size Precision 100 1/sqrt(100) = 1/10 .10 or 10% 400 1/sqrt (400) = 1/20 .05 or 5% 1600 1/sqrt(1600) = 1/40 .025 or 2.5% And to obtain a precision of 1%, we would need a sample of 10,000 (1/sqrt(10,000) = 1/100). INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -6 Since precision increases as the square root of the sample size, in order to double the precision we must increase the sample size by a factor of four. This is an unfortunate and expensive fact of research. In practice, samples between 500 and 1,500 are often selected for national studies. Precision of Mean s The same basic relation—that precision increases with the square root of the sample size— applies to sample means as well. To illustrate this we display histograms based on different samples from a nor mally distributed population with mean 70 and standard deviation 10. Formally, for a binomial or multinomial distribution (a variable measured on a nominal or ordinal scale), the standard error of the sample proportion (P) is equal to N P P P StdErr ) 1 ( * ) ( − = Thus the standard error is a maximum when P = .5 and reaches a minimum of 0 when P = 0 or 1. A 95% confidence band is usually determined by taking the sample estimate plus or minus twi ce the standard error.
Precision ( pre) here is simply two times the standard error. Thus precision ( pre) is N P P P pre ) 1 ( * * 2 ) ( − = . If we substitute for P the value .5 which maximizes the expression (and is therefore conservative) we have ()( ) () N N NN pre 15 .
0 * 2 5 .
0 * 5 .
0 * 2 ) 5 .
0 1 ( * 5 .
0 * 2 ) 5 .
0 ( = = = − = This validates the rule of thumb used in the chapter. Since the rule of thumb employs the value of P=.5, which maximizes the standard deviation and thus the standard error, in practice, greater precision would be obtained when P departs from .5. Further Info MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -7 A Large Sample of Individuals Below is a histogram of 10,000 observations drawn from a normal distribution of mean 70 and standard deviation 10. This is for one sample. Figure 5.4 Histogram of 10,000 Observations • Sample of this size closely matches its population • Sample mean is very close to 70 • Sample standard deviation is near 10 • Shape of the distribution is normal Means Based o n Samples of 10 The second histogram displays 1,000 sample means drawn from the same population (mean 70, standard deviation 10). Here each observation is a mean based on only 10 data points. In other words we select 1,000 samples of ten observations each and plot their means in the histogram. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -8 Figure 5.5 Histogram of Means Based on Samples of 10 • Sample mean is very close to 70 • Sample standard deviation is reduced to 3.11 • Less variation among means based on groups of observations then among the observations themselves • Shape of the distribution is normal Means Based on Samples of 100 The next histogram is based on a sample of 100 means where each mean represents 100 observations. Here each observation is a m ean based on 100 data points. In other words we select 100 samples of 100 observations each and plot their means in the histogram. MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -9 Figure 5-6 Histogram of Means Based on Samples of 100 • Sample mean is very close to 70 • Sample standard deviation is reduced to 1.00 • Less variation among means based on groups of observations then among the observations themselves • Shape of the distribution is normal Thus with means as well as percents, precision increases with the square root of the sample size. Apply Your Knowledge 1. Consider the population distribution of the variable income, depicted below. Note, this distribution is highly skewed. If we are sampling repeatedly from this distribution, each time with sample size = 1000, then what is the di stribution of the sample means?
a. The sample means show the same distribution as income b. The sample means show a normal distribution c. The distribution of the sample means is unknown INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -10 2. Which of these statements is correct?
a. The bigger the sample, the higher b. The the precision with which we estimate a population percentage bigger the sample, the lower c. A sample size of 2000 will give a precision for a population percentage that is twice as precise as the precision for this percentage with sample size 1000 the precision with which we estimate a population mean d. If a population percentage is 0 or if it is 100, the standard error for the sample percentage will be 0 3. True or false? A population parameter varies from sample to sample? 5.5 Hypothesis Testing Whenever we wish to make an inference about a population from our sample, we must specify a hypothesis to test. It is common practice to state two hypotheses: the null hypothesis (also known as H0) and the alternative hypothesis (H1). The null hypothesis being tested is conventionally the one in which no effect is present. For example, we might be looking for differences in mean income between males and females , but the (null) hypothesis we are testing is that there is no difference between the groups. If the evidence is such that this null hypothesis is unlikely to be true, the alternative hypothesis should be accepted. Another way of thinking about the problem is to make a comparison with the criminal justice system. Here, a def endant is treated as innocent (i.e.. the null hypothesis is accepted) until there is enough evidence to suggest that they perpetrated the crime beyond any reasonable doubt (i.e. , the null hypothesis is rejected). The alternative hypothesis is generally (a lthough not exclusively) the one we are really interested in and can take any form. In the abo ve example, we might hypothesis that males will have a higher mean income than females . When the alternative hypothesis has a “ direction” (we expect a specific re sult), the test is referred to as one -tailed . Often, you do not know in which direction to expect a difference and may simply wish to leave the alternative hypothesis open- ended. This is a two -tailed test and the alternative hypothesis would simply be that the mean incomes of males and females are different. Whichever option you choose will have implications when interpreting the probability levels. In general, the probability of the occurrence of a particular statistic for a one- tailed test will be half that of a two -tailed test as only one extreme of the distribution is being considered in the former type of test. MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -11 5.6 The Nature of Probability Descriptive statistics describe the data in our sample through the use of a number of summary procedures and statistics. Inferential statistics allow us to infer the results from the sample on which we have data to the population which the sample represents. To do this , we use procedures that involve the calculation of probabilities . The fundamental issue with inferential statistical tests concerns whether any “ effects” (relationships or differences between groups) we have found are genuine or are a result of sampling variabili ty (in other words, mere chance). So we have two hypotheses and we want to know which hypothesis is true. The way hypotheses are assessed is by calculating the probability or the likelihood of finding our result. A probability value, which can range from 0 to 1 (corresponding to 0% to 100% in terms of percentages) , can be defined as “ the mathematical likelihood of a given event occurring ,” and as such we can use such values to assess whether the likelihood that any differences we have found are the result of random chance. Now in statistics, we want to be sure of our conclusions , so having formally stated your hypotheses, you must then select a criterion for acceptance or rejection of the null hypothesis. With probability tests such as the chi -square test or the t -test, you are testing the likelihood that a statistic of the magnitude obtained (or greater) would have occurred by chance assuming that the null hypothes is is true. You always assess the null hypothesis, which is the hypothesis that states there is no difference or relationship. In other words, we only wish to reject the null hypothesis when we can say that the result would have been extremely unlikely under the conditions set by the null hypothesis. In this case, the alternative hypothesis should be accepted. It is worth noting that this does not “ prove” the alternative hypothesis beyond doubt, it merely tells us that the null hypothesis is unlikely to be true. But what criterion (or alpha level , as it is often known) should we use? Unfortunatel y, there is no easy answer! Traditionally, a 5% level is chosen, indicating that a statistic of the size obtained would only be likely to occur on 5% of occasions (or once- in-twenty) should the null hypothesis be true. This also means that, by choosing a 5% criterion, you are accepting that you will make a mistake in rejecting the null hypothesis 5% of the time. 5.7 Types of Statistical Errors Recall that when performing statistical tests we are generally attempting to draw conclusions about the larger population based on information collected in the sample. There are two major types of errors in this process. False positives, or Type I errors, occur when no difference (or relation) exists in the population, but the sample tests indicate there are significant differences (or relations). Thus the researcher falsely concludes a positive result. This type of error is explicitly taken into account when performing statistical tests. When testing for statistical significance using a .05 criterion (alpha level), we acknowledge that if there is no effect in the population then the sample statistic will exceed the criterion on average 5 times in 1 00 (.05).
Type II errors, or false negatives, are mistakes in which there is a true effect in the population (difference or relation) but the sample test statistic is not significant, leading to a false conclusion of no effect. To put it briefly, a true effect remains undiscovered. The probability of making this type of error is often referred to as the beta level . Whereas you can select your own alpha levels, beta levels are dependent upon things such as the alpha level and the size of the sample. It is helpful to note that statistical power, the probability of detecting a true effect, equals 1 minus the Type II error and the higher the power the better. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -12 Table 5.2 Types of Statistical Errors in Hypothesi s Testing Statistical Test Outcome Not Significant Significant Population No Difference (Ho is True) Correct Type I error (α) False positive True Differenc e (H1 is True) Type II error (β) False negative Correct (Power) Statistical Power Analysis With increasing precision we are better able to detect small differences that exist between groups and small relationships between variables. Power analysis was developed to aid researchers in determining the minimum sample size required in order to have a specified chance of detecting a true difference or relationship of a given size. To put it more simply, power is used to quantify your ability to reject the null hypothesis when it is false. For example, suppose a researcher hopes to find a mean difference of .8 standard deviation units between two populations. A power calculation can determine the sample size necessary to have a 90% chance that a significant difference will be found between the sample means when performing a statistical test at a specified significance level. Thus a researcher can evaluate whether the sample is large enough for the purpose of the study. Books by Cohen (1988) and Kraemer & Thiemann (1987) discuss power analysis and present tables used to perform the calculation for common statistical tests. In addition specialty software is available for such analyses, such as SamplePower ® 5.8 Statistical Significance and Practical Importance . Power analysis can be very useful when planning a study, but does require such information as the magnitude of the hypothesized effect and an es timate of the variance. A related issue involves drawing a distinction between statistical significance and practical importance. W hen an effect is found to be statistically significant we conclude that the population effect (difference or relation) is not zero. However, this allows for a statistically significant effect that is not quite zero, yet so small as to be insignificant from a practical or policy perspective. This notion of practical or real wor ld importance is also called ecological significance. Recalling our discussion of precision and sample size, very large samples yield increased precision, and in such samples very small effects may be found to be statistically significant. In such situatio ns, the question arises as to whether the effects make any practical difference. For example, suppose a company is interested in customer ratings of one of its products and obtains rating scores from several thousand customers.
Furthermore, suppose mean ratings on a 1 to 5 satisfaction scale are 3.25 for male and 3.15 for female customers, and this difference is found to be significant. Would such a small difference be of any practical interest or use?
When sample sizes are small (say under 30), precision tends to be poor and so only relatively large (and ecologically significant) effects are found to be statistically significant. W ith moderate samples (say 50 to one or two hundred), small effects tend to show modest significance while large effects are hig hly significant. For very large samples, several hundreds or thousands, small effects can be highly significant; thus an important aspect of the analysis is to examine the effect size and determine if it is important from a practical, policy or ecological perspective. Apply Your Knowledge 1. Which of these statements is correct? MAKING INFERENCES ABOUT POPULATIONS FROM SAMPLES 5 -13 a. The nature of the null hypothesis is: there is no effect/difference b. The nature of the null hypothesis is: there is an effect/difference c. The nature of the alternative hypothesis is: there is no effect/difference 2. Which of these statements is correct?
a. If we reject the null hypothesis while the null hypothesis is true in reality, then we make a Type II error. b. If we reject the null hypothesis while the null hypothesis is true in reality, then we make a Type I error. c. If we reject the null hypothesis, while the alternative hypothesis is true in reality, then we make a Type I error. d. If we reject the null hypothesis, while the alternative hypothesis is true in reality, then we make a Type II error.
3. Which of these statements is false? a. The probability of detecting a true effect is known as “power .” b. Probability of making a Type II error + the probability of detecting a true effect=1 c. With an alpha level of .01, the probability of a Type II error will always be lower than that. d. With Alpha=.000001 the probability to commit a Type I error is lower than with Alpha=0.1. 5.9 Lesson Summary In this lesson, we explored how to make inferences from a sample to a population. This allows us to reach conclusions about the population without the need to study every single individual. Hypothesis testing allows researchers to develop hypotheses which are then assessed to determine the probability or likelihood of the findings. Lesson Objectives Review Students who have completed this lesson should now be able to: • Explain how to make inferences about populations from samples To support the achievement of the primary objective, students should now also be able to:
• Explain the influence of sample size • Explain the nature of probability • Explain hypothesis testing • Explain different types of statistical errors and power • Explain differences between statistical and practical importance 5.10 Learning Activity In this set of learning activities you won’t need any supporting mat erial. State the null and alternative hypotheses when assessing each of the following scenarios . W ould you do a one- or two -tailed test for each?
1. The relationship between gender and belief in the afterlife. 2. The relationship between number of years of education and income. 3. The difference in mean income between men and women. 4. The difference between happiness in marriage between married persons in America, Europe, Asia, Australia and Africa. INTRODUCTION TO STATISTICAL ANALYSIS WITH IBM SPSS STATISTICS 5 -14 RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -1 Lesson 6: Relationships Between Categorical Variables 6.1 Objectives After completing this lesson students will be able to: • Perform crosstab analysis on categorical variables • Perform a statistical test to determine whether there is a statistically significant relationship between categorical variables To support the achievement of this primary objective, students will also be able to: • Use the options in the Crosstabs procedur e • Request appropriate statistics for a crosstabulation • Interpret cell counts and percents in a crosstabulation • Use the Chi-Square test, interpret its results, and check its assumptions • Use the Chart Builder to visualize a crosstabulation 6.2 Introduction Many data analysts consider crosstabulations the core of data analysis. Crosstabulations display the joint distribution of two or more categorical variables. In this lesson we will provide examples and advice on how best to construct and interpret crosstabulati ons. W ith PASW Statistics, statistical tests are used to determine whether a relationship between two, or more, variables is statistically significant in a crosstabulation. Another way to state this is that a statistical test is done to determine whether a relationship observed in the sample is caused by chance sampling variation or instead is likely to exis t in the population of interest. To support the analysis, we also show examples of graphical displays of crosstabulations. Business Context When analyzing data, the focus is on both descriptive and causal relationships. When we examine a table with categorical variables, we would like to know whether a relationship we observe is likely to exist in our target population or instead is caused by random sampling variation. We might want to know whether:
• Satisfaction with the instructor in a training workshop was related to satisfaction with the course material. • Eating more often at fast -food restaurants was related to more frequent shopping at convenience stores. • Certain types of people are more likely to buy laptop versus desktop computers. Statistical testing tells us whether two categorical variables are related. W ithout that, we might make decisions based on observed category percentage dif ferences that are not likely to exist in a population of customers. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -2 6.3 Crosstabs Crosstabulations are commonly used to explore how demographic characteristics are related to attitudes and behaviors, but they are also used to see how one attitude is related to another. The key point is that crosstabulations are used to study the relationships between two, or more, categorical variables. Crosstabs Illustrated To provide context, consider the table depicted below. This table shows counts and percentages in the cells . For example, there are 206 fe male clerical employees . The percentages are calculated within Gender ( the column variable). Thus, 60.9% (157/ 258) of the men are clerical. To assess whether the two variables are related, there is a standard procedure to follow, depending on which percentages we are using. 1) If using percentages based on the row variable, compare percentages within each column.
Look to see if the percentages are the same or different across rows within each column. If percentages are the same, there is no relationship between the variables. 2) If using percentages based on the column variable, compare percentages within each row.
Look to see if the percentages are the same or different across columns within each row.
Again, identical percentages indicate no re lationship between the variables. Here, percentages are based on Gender , so to assess whether Employment Category and Gender are related, we should compare the percentages 95.4% with 60.9% (95.4% of women are clerical versus 60.9% of men), 0% with 10.5% a nd 4.6% with 28.7%. Here, percentages differ substantially. Figure 6.1 Crosstabs Illustrated The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics . Supporting Materials RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -3 6.4 Crosstabs Assumptions To use Crosstabs, one condition has to be met: 1) Variables used in Crosstabs must be categorical (nominal, ordinal) .
6.5 Requesting Crosstabs Requesting Crosstabs is accomplished by following these steps: 1) Select variables for the Crosstabs procedure, at least one for the row and one for the column dimension; more than one variable can be used in a dimension of the table 2) Select percentage options .
3) Review the procedure output to investigate the relationship between the variables including: a. Cell counts b. Cell percentages .
6.6 Crosstabs Output The Crosstabs table has at least two rows and two columns. The information contained in the cells of this table will depend on what you requested: • By default, the (cell) Count is the only statistic displayed • Normally, you will ask for either a row or column percent, or both • These percentages are based on the variable categories in the row or column, respectively. • The last row and column of the table display marginal totals for each row and column • The lower right hand cell contains table total statistics, including the number of valid cases for the table. • The percentages are used to determine whether one variable is statistically associated with another Figure 6.2 Example of Crosstabs Output INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -4 6.7 Procedure: Crosstabs The Crosstabs procedure is accessed from the Analyze…Descriptives… Crosstabs menu choice.
With the Crosstabs dialog box open: 1) Place one or more variables in the Row(s) box .
2) Place one or more variables in the Column(s) box .
3) Open the Cells dialog to specify percents and other cell statistics . 4) The order of categories can be changed in the Format dialog .
A separate table will be created for all combinations of variables in the Rows and Columns boxes. Figure 6.3 Crosstabs Dialog In the Cells Display dialog: 1) By default the cell count is displayed (the Observed check box in the Counts area) .
2) Typically one or both of row and column percents are selected with the appropriate check boxes in the Percentages area. 3) Other more non- standard statistics available include residuals which help to understand where there are deviations from the expected counts if there is no relationship. RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -5 Figure 6.4 Crosstabs Cell Display Dialog 6.8 Example: Crosstabs We will work with the Census.sav data file in this lesson. In this example we want to study how overall happiness ( happy) is related to marital status ( marital).
We want to know what the percentage of overall happiness for each of the marital status groups , so we will percentage based on marital . The variable m arital is nominal while happy is ordinal , so both variables are categorical and appropriate for Crosstabs. Detailed Steps for Crosstabs 1) Place happy in the Row(s) box .
2) Place the variable marital in the Column(s) box .
3) Select Column in the Cells Display dialog. The percentages are based on the categories of marital . This indicates that we expect marital status to affect general happiness . Results from Crosstabs As with all analyses, you should first look to see how many cases are missing from an analysis. The first table in the Viewer window provides this information (not shown). There is actually very little missing data here. To examine a crosstabulation table: • First focus on the marginal totals, which are equivalent to frequency distributions for each variable. We see that most people are either very happy (597) or pretty happy (1099) with their life overall, and most respondents are either married (969) or never married (528). INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -6 • Next turn to the counts in the cells of the table. Focus on the upper left -hand cell where 398 respondents indicated they were very happy with their life and married. Conversely, from the bottom right -hand cell, we see that 123 people are not too happy with their life and single. • In the upper left -hand cell the first percentage is 41.1% (calculated within marital status). This tells us that, of the people who are married (398), 41.1% (398/969) are very happy in their life. The other percentages are calculated in a similar manner. We observe that: • 41.1% of those who are married say they are very happy. This is much larger than any other category. The next largest is divorced, at 20.0%. • Married people also have the lo west percentage of people who say they are not too happy at 8.4%. Those who are separated have the highest percentage (25.7%). These differences would certainly lead us to the conclusion that marital status is related to general happiness , and that married people are happier than others. Figure 6.5 Crosstabulation of marital and happy Apply Your Knowledge 1. What level of measurement should variables have to be used in Crosstabs?
a. Any level of measurement b. Ordinal and Scale c. Nominal and Ordinal 2. Consider a Crosstabs table with the variable sex on the rows and variable happiness on the columns. If we wanted to see how sex might affect happiness , how should the table be percentaged? a. By happiness b. By sex c. By both 3. Consider the output depicted below. W hich statements are correct ?
a. 11.9% of all women are manager s b. 4.6% of all women are manager s c. 60. 9% of all men are clericals d. 2.1% of all cases are female managers RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -7 6.9 Chi -Square Test Comparing percentages i s only part one of the story. What we don’t know is whether differences in percentages are due to sampling variation, or instead are likely to be real and exist in the population.
For this we turn to the Chi -Square test.
Every statistical test has a null hypothesis. In most cases, the null hypothesis is that there is no relationship between two variables. This is also true for the null hypothesis for a crosstabulation, so we use the Chi -Square test to determine whether there is a relationship between marital status and general happiness. If the significance is small enough, the null hypothesis is rejected. In the context of a crosstabulation, the null hypothesis would be that percentages across categories of an independent variable are statistically equivalent. As with other statistical tests, sample size has an effect on the calculated significance. Larger sample sizes, everything else being equal, will reduce the significance value of a test statistic and thus make it easier to reject the null hypothesis. Chi -Square is particularly sensitive to the effect of increased sample size. I n large samples, say 1000 to 1500 cases and above, it is fairly easy to show that two variables are related using the .05 significance level. So in large samples it is probably best to require a lower significance level before rejecting the null hypothesis (this is appropriate for the data file Census.sav , which contains over 2000 cases). Substantive Versus Statistical Significance A critical distinction to make is whether a relationship is statistically significant versus being substantively, or practically, significant. In very large samples, small differences will be statistically significant. You should not let statistical significance be the overriding determinant in deci ding whether a relationship or pattern you have discovered is interesting or important . Note INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -8 Chi -Square Test Assumptions To correctly use a Chi -Square statistical test in Crosstabs, one condition has to be met: 1) At most , 20% of the expected values in the cells of the table should be below 5. Chi -Square is calculated based on the expected values in each cell. If too many expected values are below 5, the reported significance can be incorrect. Because of this, PASW Statistics adds a footnote to the Chi -Square table noting the number of cells w ith expected counts less than 5. If more than 20 percent of the cells are in this condition, you should consider grouping categories of one or both variables .
6.10 Requesting the Chi-Square Test The procedure to request a Chi -Square test is to :
1) Select one or more row variables .
2) Select one or more column variables .
3) Request appropriate percentages .
4) Request the Chi -Square stati stic .
5) Inspect the Chi -Square test output, if the test results in a significant result, conclude the null hypothesis of independence is rejected and report the differences in percentages. If the Chi - Square is not significant, then conclude there are sample differences, but equality of percentages in the population cannot be rejected. 6.11 Chi-Square Output The table titled Chi -Square Tests shows the Chi -Square test.
There are three Chi -Square values listed, the first two of which are used to test for a relationship. We concentrate on the Pearson Chi -Square statistic, which is adequate for almost all purposes. The actual value (here 153.16), is not important, nor is the number of degrees of freedom (df), which is related to the number of cells in the table. These values are used to calculate the significance for the Chi -Square statistic, labeled “Asymp. Sig. (2- sided).” We interpret the significance value as the chance, if the null hypothesis is true, of finding a Chi-Square value at least this large. It is this value that we use to test the null hypothesis. If the significance va lue is smaller than the preset alpha, then we reject the null hypothesis of independence. Another option to deal with sparse cells is to use exact statistical tests (available in the PASW Statistics Exact Tests option). Note RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -9 Figure 6.6 . Chi -Square Tests Output 6.12 Procedure: Chi -Square Test To include the Chi -Square test in a Crosstabs , with the Crosstabs dialog open:
1) Place one or more variables in the Row(s) box.
2) Place one or more variables in the Column(s) box .
3) Optionally, inc lude appropriate percentages in the Cells dialog. 4) Select Chi-square in the Statistics dialog .
Figure 6.7 Statis tics Dialog In terms of formal testing, we often use significance values of .05 or .01 to determine whether we should reject the null hypothesis. Best Practice INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -10 6.13 Example : Chi-Square Test We will work with the Census.sav data file in this lesson. In this example we continue to examine the relationship between marital status ( marital) and happiness with one’s life overall ( happy). W e would like to determine whether the percentage differences we observed above are likely t o be observed in the total population.
Detailed Steps for Crosstabs with Chi-Square Test 1) Place the variable happy in the Row(s) box .
2) Place the variable marital in the Column(s) box .
3) Select Column in the Cells Display dialog. 4) Select Chi-square in the Statis tics dialog. Results from Crosstabs with Chi-Square Test The Crosstabs table percentages are based on the columns, or on marital . We observe that the marital groups differ substantially in their happiness (e.g. the percentage not too happy is 8.4% for thos e married versus 25.7% for those married). Figure 6.8 Crosstabulation of Marital Status and General Happiness The table titled Chi -Square Tests shows that the signifi cance value is quite small and is rounded off to .000, which means that the actual value is less than .0005. W ith the significance being so low, it is rather unlikely that the null hypothesis is true. Therefore, it is reasonable to conclude that there is a relationship between marital stat us and general happiness. The exact form of the relationship is as described above, i.e., as depicted in the table. Figure 6.9 Chi -Square Tests RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -11 Apply Your Knowledge 1. A chi -s quare value from a test of two variables is 53.4. What should we conclude about the relationship of these two variables? a. There is no relationship b. There is a relationship c. We need to know the significance of this value to reach a conclusion d. Both A and C 2. True or false ? A Chi -Square test is done differently for nominal compared to ordinal variables ?
3. See the output below. True or false? There is a statistically significant relationship between Minority Classification and Employment C ategory (alpha=0.05). 4. See the table below. True or false? It is incorrect to do the Chi -Square test, because more than 3 out of the 6 cells have an observed count less than 5. 6.14 Clustered Bar Chart For presentations it is often useful to show a graph of the relationship between two categorical variables. Clustered bar charts are the most effective method of doing this for Crosstabs. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -12 Clustered Bar Chart Illustrated As an illustration of a clustered bar chart, see the graph below. The bars show the percentage in each of the job categories by gender. For example, 95% of the women are clerical versus 61% of the men.
Figure 6.10 Clustered Bar Chart Illustrated 6.15 Requesting a Clustered Bar Chart with Chart Builder The Chart Builder procedure allows for the creation of a variety of charts, including clustered bar charts. It provides for great flexibility in creating charts, including formatting options. To create a clustered bar chart: 1) In the Chart Builder, select the clustered bar chart type of graph. 2) Specify the appropriate variables .
3) Specify the appropriate percentage statistic .
4) Inspect the graph in the output and compare the percentages .
6.16 Clustered Bar Chart from Chart Builder Output The clustered bar chart from Chart Builder gives the user control of:
• Which variable is used for clustering, and which variable defines the X -axis • What variable is used for percentages • The chart also correctly lists that the percentage is displayed, and it has labels and a legend that is placed outsi de the chart area.
In the graph depicted below, RS HIGHEST DEGREE [degree] is used for clustering (in the legend), GENERAL HAPPINESS [happy] defines the X -axis. Percentages are based on degree, so that percentages for each degree sum to 100%. RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -13 Figure 6.11 Example of a Clustered Bar Chart from Chart Builder 6.17 Procedure: Clustered Bar Chart with Chart Builder The Chart Builder procedure is accessed from the Graphs …Chart Builder menu. With the Chart Builder dialog open: 1) Select a clustered bar chart in the Chart Builder and drag it into the Chart preview area 2) Select a variable for the X -axis .
3) Select a variable for the Cluster on X: set c olor box.
INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -14 Figure 6.12 Chart Builder Dialog to Create Clustered Bar Chart To specify the statistic: 1) In the Element Properties dialog, specify the Percentage() statistic .
2) Select the appropriate base for percentages in the Set Parameters dialog. RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -15 Figure 6.13 Setting Percentage in Element Properties Dialog 6.18 Example: Clustered Bar Chart with Chart Builder We will create a clustered bar chart of the crosstab table of marital status and general happiness. We want to see how general happiness varies across categories of marital status, so we use marital status as the clustering variable. This is equivalent to how we percentage a crosstab by marital status to study this relationship. Detailed Steps for Clustered Bar Chart 1) Select the clustered bar chart icon and put it in the Chart Preview pane. 2) Place happy in the X -axis box .
3) Place marital in the Set Color box .
4) Select the Percentage() statistic in the Element Properties Statistics drop down 5) Select Total for Each Legend Variable Category in the Set Parameters dialog . Results from the Clustered Bar Chart Created with Chart Builder A crosstab table with percentages based from marital status lets us compare percentages of that variable within categories of general happiness, and this is mirrored in the clustered bar chart. • There are 3 values on the X -axis for each separate value of general happiness • At each value, the five categories of marital status are displayed with the bar representing percentages • We can readily compare the categories of marital status in this arrangement , equivalent to comparing across rows in the crosstabulati on Married respondents are by far more likely to say they are very happy, and they are less likely to say they are not too happy. There isn’t much difference between the other categories. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -16 Figure 6.14 Cluster ed Bar Chart of Marital Status and General Happiness 6.19 Adding a Control Variable Tables can be made more complex by adding variables to the layer dimension. • Adding one or more layer variables to a Crosstabs table is known as adding control variables • Usin g a control variable means that a subtable will be formed for every value of the control variable; thus, the relationship between the original variables will be controlled by examining it with in values of the third variable • You can add more than one contro l variable to a table by nesting variables in the layer (use the Next button), but unless you have a large sample size, or the variables have only a few categories, you may quickly create tables with only a few cases, or even no cases, in several cells Control Variable Crosstabs Illustrated The table depicted below includes sex as the control variable. Percentages are calculated per subtable. RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -17 Figure 6.15 A Control Variable Table 6.20 Requesting a Control Variable Including a control variable in Crosstabs is accomplished with these steps:
1) Select row and column variables .
2) Add a control variable. 3) Request appropriate percentages .
4) Optionally, request the Chi -Square statistic .
5) In the output, compare the percentages within each category of the control variable to see if there is a relationship between the row variable and column variable, for that particular category .
6) In the output, i nspect the Chi -Square test output for each of the subtables and compare these results .
6.21 Control Variable Output The layer variable will be included as outer -left variable in the crosstabulation. Here the control variable is gender. Choosing the proper control variables is somewhat of an art, since there is no reason (nor is there time) to try every possible other variable as a control. Select control variables based on the goals of the study, any available theory, and also based on which variables are possibly related to one or both variables in the table. Demographic variables are often used as controls. Best Practice INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -18 Figure 6.16 Crosstabs with a Control variable If a Chi -Square test is requested, a Chi-Square test will be performed for each of the values of the control variable. The interpretation of the test is the same as in any table, bu t it applies only to each subtable (here, for the male and female subtables separately). Figure 6.17 Chi -Square Test with a Control Variable 6.22 Procedure: Adding a Control Variable With the Crosstabs dialog open, t he procedure to include a control variable is: 1) Place one or more variables in the Rows box. 2) Place one or more variables in the Columns box. 3) Place the control variable in the Layer(s) box .
4) In the Cells dialog, select appropriate percentages .
5) Optionally, select Chi -S quare in the Statistics dialog . RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -19 Figure 6.18 Crosstabs Dialog with C ontrol Variable 6.23 Example: Adding a Control Variable We’ll create an entirely new table with three new variables to investigate the relationship between sex, race, and do you own a business ( ownbiz ). W e’ll use sex as the control variable to further illuminate the relationship between race and ownbiz .
To illustrate the benefit of adding a control variable, first we create the table of race and ownbiz .
Detailed Steps for the Two -Way Crosstabs 1) Select Reset button 2) Place [ race in the Column(s) box 3) Place ownbiz in the Rows box 4) In the Cells Display dialog, select Column percents 5) Select Chi-Square in the Statistics dialog Results for the Two-Way Crosstabs We see that the percentages in the table suggests that whites (13.6%) are most likely to own a business, followed by those of other races (10.9%) and then blacks (6.8%). Figure 6.19 Crosstab of Race and Owni ng a Business INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -20 Figure 6.20 Chi -Square Tests Detailed Steps for Control Variable Crosstabs Now we’ll see what happens when we add the variable sex to the layer. 1) Place sex in the Layer(s) box .
Results from Adding a Control Variable The resulting table is much larger. Actually, there are two subtables, one for males and one for females. Does the relationship vary by gender? To a certain extent it does. • White males are much more likely than black males to o wn a business (almost a 10 percentage point difference) • Black females are closer to white females in the percentage owning a business (a 4 percentage point difference) Figure 6.21 Crosstab of Race, Sex, and Owning a Business RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -21 Chi -Square Tests for Control Variable Are these differences we observe statistically significant? We can answer this, as before, with the c hi- s quare test.
Males. The chi -s quare test is significant (.018). T his means that, for males, race is related to owning a business. Females. The chi -s quare test is not significant (.263). This means that, for females, there is no relationship between race and owning a business. Figure 6.22 Chi -Square Tests with a Control Variable We added a control variable and discovered that the initial relationship depends somewhat on the respondent’s gender. What does this mean about our analysis of the bivariate table? Should the res ults in this table be entirely discarded in favor of the three- way table?
The first point to make in answering these questions is that this example confirms the importance of multivariate analysis. That said, we can add these comments: 1) The bivariate tabl e is still “correct,” although that word is better replaced by “valuable.” The bivariate table is our estimate of the relationship between owing a business and race. 2) However, if we generalize this relationship and believe it is the same for males and femal es, we would be incorrect. 3) Adding the control variable has given us additional insight into this relationship. It may help us understand more about the two- way relationship.
Apply Your Knowledge 1. See the output below. Which statements are correct ?
a. The tabl e shows that there is a statistically significant relationship between gender and minority classification (alpha=0.05) b. The relationship between current salary (binned) and minority classification is not significant for women (alpha=0.05) c. The relationship b etween current salary (binned) and minority classification is not significant for men (alpha=0.05) INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -22 d. The relationship between current salary (binned) and minority classification is not the same for men and women Brainstorming Exercise In regard to your own data, what are some potentially important control variables for your analyses? 6.24 Extensions: Beyond Crosstabs Decision Tree analysis is often used by data analysts who need to predict to which group an individual can be classified, based on potentially many nominal or ordinal background variables. For example, an insurance company is interested in the combination of demographics that best predict whether a client is likely to make a claim. Or a direct mail analyst is interested in the combinations of bac kground characteristics that yield the highest return rates. Here the emphasis is less on testing a hypothesis and more on a heuristic method of finding the optimal set of characteristics for prediction purposes. CHAID ( Chi-Square automatic interaction det ection), a commonly used type of decision- tree technique, along with other decision- tree methods are available in the PASW Decision Trees add- on module. A technique called loglinear modeling can also be used to analyze multi -way tables. This method requir es statistical sophistication and is well beyond the domain of this course. PASW Statistics has several procedures (Genlog, Loglinear and Hiloglinear) to perform such analyses. They provide a way of determining which variables relate to which others in the context of a multi-way crosstab (also called contingency) table. These procedures could be used to explicitly test for the three- way interaction suggested above. For an introduction to this methodology see Fienberg (1977). Academic researchers often use s uch models to test hypotheses based on many types of data. RELATIONSHIPS BETWEEN CATEGORICAL VARIABLES 6 -23 6.25 Association Measures Measures of association for categorical variables have been developed to summarize the strength of a relationship in a single statistic. This allows you to compare different tables (groups) concisely. • They are typically normed to range between 0 and 1 for variables on a nominal scale, or – 1 and 1 for variables on an ordinal scale. • The specific measure used thus depends on the level of measurement of the variables, among other t hings • In general, measures of association are bivariate rather than designed for multi -way tables The Crosstabs procedure provides many measures of association. Several are available because different aspects of the association are emphasized by particular measures. There are four tests each in the Nominal and Ordinal areas. The variable at the lowest level of measurement determines which test can be used, e.g., in a table with a nominal and ordinal variable, you must use one of the nominal tests. For a discussion of these tests, see Gibbons (2005). Figure 6.23 Crosstabs Statistics Dialog with Measures of Association 6.26 Lesson Summary We explored the use of the Crosstabs procedure to analyze relationships between categorical variables. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 6 -24 Lesson Objectives Review Students who have completed this lesson should now be able to: • Perform crosstab analysis on categorical variables To support the achievement of the primary obj ective, students should now also be able to:
• Use the options in the Crosstabs procedure • Request appropriate statistics for a crosstabulation • Interpret cell counts and percents in a crosstabulation • Use the Chi -Square test, interpret its results, and check its assumptions • Use the Chart Builder to visualize a crosstabulation 6.27 Learning Activity The overall goal of this learning activity is to create two and three- way crosstab ulations to explore the relationship between several variables and to use the Chart Builder to visualize the relationship. You’ll use the file Census.sav .
Supporting Materials 1. Investigate the relationship between the variables race and self-rated health ( health), can people be trusted ( cantrust), and support for spending on scientific research ( natsci). Request appropriate percentages and a Chi -Square test.
2. What is the significance of the Chi -Square test? Is there a statistically significant relationship between race and these variables ? Describe the relationships that you observe in the tables. 3. Create a clustered bar chart to display the r elationship of one of the tables with a significant relationship. 4. Now add sex as a control variable. Are there differences in the relationship by gender? Are the relationships in the subtables significant or not? Are they different or not? 5. Now remove the control variable sex and substitute the variable born (was the respondent born in the U.S. or not) . Does the relationship between race and self-rated health vary depending on place of birth?
6. For those with extra time: Select some variables that you find interesting and explore their relationship with Crosstabs. Begin with two- way tables before investigating more complex relationships. The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials THE INDEPENDENT - SAMPLES T TEST 7 -1 Lesson 7: The Independent - Samples T Test 7.1 Objectives After completing this lesson students will be able to: • Perform a statistical test to determine whether there is a statistically significant difference between two groups on a scale variable To support the achievement of this primary objective, students will also be able to: • Check the assumptions of the Inde pendent-Samples T Test • Use the Independent- Samples T Test to test the difference in means • Know how to interpret the results of a Independent- Samples T Test • Use the Chart Builder to create an error bar graph to display mean differences 7.2 Introduction When our purpose is to examine group differences on scale variables, we turn to the mean as the summary statistic since it provides a single measure of central tendency. In this lesson we outline the logic involved when testing for mean differences between groups , state the assumptions, and then perform an analysis comparing two groups. Business Context When analyzing data, we are concerned with whether groups differ from each other. W ithout statistical testing, we might make decisions based on perceptions that are not likely to exist in a population of customers . An Independent-Samples T Test allow s us to determine if two groups differ significantly on a scale variable. For example, we might want to know whether: • One customer group purchases more items , on average, than a second • Drug A reduces depression levels better than drug B • Student test scores in one class are higher than in a second class 7.3 The Independent -Samples T Test For historical reasons related to the development of statistics and lack of computing power, separate methods were developed to test for mean differences when the predictor or grouping var iable had two categories versus three or more. For dichotomous predictors we use the Independent -Samples T Test ; for predictors with more categories we use analysis of variance (ANOVA). So, the Independent- The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -2 Samples T Test applies when there are two As with any statistical test, the goal is to draw conclusions about the population, based on s ample data. For the Independent -Samples T Test the nul l hypothesis states that population means are the same , and we use sample data to evaluate this hypothesis. separate populations to compare on a scale dependent variable (for example, males and females on salar y). So, Ho ( the null h ypothesis) assumes that the population means are identical. W e then determine if the differences in sample means are consistent with this assumption. If the probability of obtaining sample means as far (or further) apart as we find in our sample is very small (less than 5 chances in 100 or .05), assuming no population differences, we reject our null hypothesis and conclude the popu lations are different. 7.4 Independent-Samples T Test Assumptions To correctly use the Independent -Samples T Test requires a number of assumptions to be made. 1) Only two population subgroups are compared. 2) The dependent variable has a scale measurement level .
3) The distribution of the dependent variable within each population subgroup follows the normal distribution (normality). 4) The variation is the same within each population subgroup. This is the so- called “homogeneity of variance” assumption. Violation of this assumption is more critical than violation of the normality assumption. When this assumption is violated, the significance or probability value reported by PASW Statistics is incorrect and the test statistics must be adjusted. 7.5 Requesting the Independent-Samples T Test Requesting an Independent- Samples T Test is accomplished with these steps: 1) Select one or more variables to be tested that are scale in measurement level .
2) Select a grouping or test variable and define the two groups .
3) Optionally, set the confidence interval if you prefer something other than 95% .
4) Review the procedure output to investigate the relationship between the variables including: a. Group Statistics Table The Independent- Samples T Test is robust to moderate violations of the homogeneity of variance assumption. If the ratio (greatest variance / smallest variance) is less than 2, with similar sample sizes of the groups, then violation is not serious. The Independent- Samples T Test provides a statistical test for the assumption of homogeneity of variance and an alternative test for testing equality of means, taking into account unequal variances. Note The Independent-Samples T Test is robust to moderate violations of the normality assumption when sample sizes are moderate to larg e (over 50 cases per group) and the dependent measure has the same distribution (for example, skewed to the right) within each group Note THE INDEPENDENT - SAMPLES T TEST 7 -3 b. Check the assumptions of the Independent- Samples T Test.
5) Examin e the t test statistics to determine whether there is a significant difference in the means . 7.6 Independe nt-Samples T Test Output The Group Statistics table provides sample sizes, means, standard deviations , and standard errors for the two groups. In the table below, t here is a 1.4 year difference in group means on highest year of school completed. The sample standard deviations (and so the variances) are quite different, indicating potentially unequal popul ation variances. Figure 7.1 Example of Group Statistics To understand how to work with the second table, the Independent Samples Test, we need to review the assumptions for conducting an Independent-Samples T Test . The most serious violation is that of the assumption of equal variances , so this assumption must be tested. Levene´s T est for Equality of Variances does exactly this. Levene´s homogeneity of variances test evaluates the null hypothesis that the dependent variable’s variance is the same in the two populations .
Since homogeneity of variance is assumed when performing the Independent -Samples T Test , the analyst hopes to find this test to be nonsignificant . In the firs t, left half, section of the Independent Samples Test table, Levene´s test for equality of variances is displayed. • The null hypothesis of Levene´s test is that the variances are equal • The F statistic is a technical detail to calculate the significance (Sig.) • The significance is the likelihood that the variances have the observed difference, or a greater difference, in the target population • We use a standard criterion value of .05 or .01 to reject the null hypothesis Figure 7.2 . Example of Levene´s Test for Equality of Variances Output As sample size increases, it is better to use the .01 level for the Levene’s test . Best Practice INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -4 Here, in the example output, we observe that the hypothesis of equal variances must be rejected because the significance value is low .
The second, right half, section of the Independent Samples Test table provides the test statistics for the null hypothesis of equal group means. The row labeled “Equal variances assumed” contains results of the standard t test. The second row labeled “Equal variances not assumed” contains an adjusted t test that corrects for lack of homogeneity of variances in the data. You would choose one or the other based on your evaluation of the homogeneity of variance question. Summarizing: t he result of the Levene test tells us which of the two r ows of t test statistics to use:
• If the null hypothesis of equal variances is not rejected • If the null hypothesis of equal variances , then use the row labeled “Equal v ariances assumed” is rejected , then use the row “ Equal Variances not assumed” To test equality of means: • The null hypothesis of the test is that the means are equal • The t and df statistics are t echnical details to calculate the s ignificance (Sig. 2-tailed) • The significance is the likelihood that the means have the observed difference, or a greater difference, in the target population • We use a standard criterion value of .05 or .01 to reject the null hypothesis Figure 7.3 Example of T T est for Equality of Means Output Here, we observe that the hypothesis of equal population means must be rejected ( reading the row Equal variances not assumed, since the null hypothesis of equal variances was rejected). The Independent Samples Test table provides an additional bit of useful information: the 95% confidence band for the population mean difference. T he 95% confidence band for the difference provides a measure of the precision with w hich we have estimated the true population difference. In the output shown below, the 95% confidence band for the mean difference between groups is .816 years to 1.914 years ( again using the Equal variances not assumed row). Note that the difference values does not include zero, because there is a difference between groups. So, the 95% confidence band indicates the likely range within which we expect the population mean difference to fall.
Speaking in a technically correct fashion, if we were to continually repeat this study, we would expect the true population difference to fall within the confidence bands 95% of the time. While the technical definition is not illuminating, the 95% confidence band provides a useful precision indicator of our estimate of the group difference. THE INDEPENDENT - SAMPLES T TEST 7 -5 7.7 Procedure: Independe nt-Samples T Test The Independent Samples T Test procedure is accessed from the Analyze…Compare Means…I ndependen t-Samples T Test me nu choice. With the Independent- Samples T Test dialog box open: 1) Place one or more scale dependent variables in the Test Variable box .
2) Place one categorical independent variable in the Grouping Variable box .
3) Click the Define Groups button to specify which two groups are being compared. Figure 7.4 Independent- Samples T Test Dialog In the Define Groups dialog :
4) Specify the two group values Figure 7.5 Independent- Samples T-Test --- Define Groups Dialog INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -6 The Options dialog can be used to change the confidence interval percentage and change the handling of missing values to listwise. Figure 7.6 T Test Options Dialog 7.8 Demonstration: Independe nt-Samples T Test We will work with the Census.sav data file in this lesson. In this example we examine the relationship between the respondent’s gender (sex) and number of children ( childs). You would expect that the means would be equal —it normally takes two to procreate— but let’s investigate the question. Before doing the actual test, we should explore the data to compare the distributions of number of children by gender. We’ll use the Explore procedure to do so. Detailed Steps for Explore Procedure for Number of Children by Gender The Explore dialog box is accessed from the Analyze…Descriptive Statistics…Explore menu.
With the dialog box open: We will not repeat details here on the Explore dialog box or the options available with the procedure. If you need a review, see the Lesson on Understanding Data Distributions for Scale Variables . Note If the independent variable is a numeric variable, specify a single cut point value to define the two groups. Those cases less than or equal to the cut point go into the first group, and those greater than the cut point fall into the second group. If the independent variable is categori cal but has more than two categories, it can be still used by using only two categories in an analysis. Tip THE INDEPENDENT - SAMPLES T TEST 7 -7 1) Place childs in the Dependent Lis t: box 2) Place sex in the Factor List: box 3) In the Plots dialog, select Histogram check box We’ll concentrate first on the boxplot. The highest value of childs is 8 (values any higher are coded to 8). Still, there is enough variation in the data to see that the male and female distributions are not identical. The inter -quartile range for males is wider than for females, and it extends all the way to 0.
The median for both genders is the same (2), and there are fewer outliers for males as a consequence of the wider IQR. Figure 7.7 Boxplot of Number of Children by Gender Looking next at the histograms, we observe that the distributions are not normal. However, given the sample size, this shouldn’t be a deterrent to doing a t test. We might be more concerned as to whether the mean is an suitable measure of central tendency of these distributions, especially for male s. But given the narrow range of childs , from 0 to 8, it seems reasonable to proceed. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -8 Figure 7.8 Histogram of Number of Children for Males Figure 7.9 Histog ram of Number of Children for Females D etailed Steps for Independent Samples T -Test 1) Place the variable childs in the Test Variable(s) box .
2) Place the variable sex in the Grouping Variable box .
3) Select the Define Groups button. Notice the question marks following sex . The Independent-Samples T Test dialog requires that you indicate which groups are to be compared, which is usually done by providing the data values for the two groups. 4) Specify that groups 1 and 2 are being compar ed THE INDEPENDENT - SAMPLES T TEST 7 -9 Results from Independent Samples T -Test As with all analyses, you should first look to see how many cases are in each group, along with the means and standard deviations. The Group Statistics table provides this information. We have fairly large samples for each group. Intriguingly, the means are not very close for males and females. The mean number of children for females is about .30 higher than for males (The actual sample mean difference is displayed in the Independent Samples Test table) . The standard deviations are similar for each group , which is a bit unexpected since the boxplots looked somewhat different. Figure 7.10 Group Statistics Table Reading an Independent Samples Test Table Looking at Levene´s test, the null hypothesis assuming homogeneity is not rejected at the .01 level.
Given the large sample size, it makes more sense to use a more stringent alpha level. So, we may assume homogeneity of variances , and we take the result of the t test fr om the Equal variances assumed row (actually, the two rows gives very similar results in this example). F igure 7.11 Independent Samples Levene’s Test To test equality of means, move to the column labeled “Sig. (2- tailed).” This is the probability of our obtaining sample means as far or further apart, by chance alone, if the two populations (males and females) actually have the same number of children. T hus the probability of obtaining such a large difference by chance alone is quite small (.000), so we conclude there is a significant difference in mean number of children between men and women. Can you suggest how that could be true? The 95% confidence b and for the mean difference between groups is from – .438 to –.142 years. Figure 7.12 Independent Samples T Test Results INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -10 Apply Your Knowledge 1. In which of the following situations can an Independent -Samples T Test be applied? a. Difference between men and women with respect to political preferences (liberal/ conservative/independent) b. Difference in mean age between liberals and conservatives c. Difference in mean income between three groups of po litical affiliation (liberals, conservatives and independents) d. Difference in mean expenditure between those shopping at Harrods and those not shopping at Harrods 2. True or False? Suppose we want to test whether boys and girls differ in mean hours (a week) d oing sports. Below we have the distribution of this variable, for both genders . Now we draw a random sample of 50 boys and 50 girls. The Independent -Samples T Test canno t be done because the distribution of hours sport is not normal within each of the groups ? 3. See the output belo w. Which statements are correct? a. To test equality of means, we use the “Equal variances assumed” row and disregard the row “Equal variances not assumed. ” b. The null hypothesis of equal group means is reject ed (alpha=0.05) c. The 95% confidence interval for the difference contains the value 0, which indicates that the null hypothesis of equal group means cannot be rejected (alpha=0.05) .
7.9 Error Bar Chart Although the Independent- Samples T Test procedure display s the appropriate statistical test information, a summary chart is often preferred as a way to present significant results. Bar charts displaying the group sample means can be produced using the Chart Builder procedure. However, many people prefer an error bar chart instead. It is a chart that focuses more on the precision of the estimated mean for each group than the mean itself. THE INDEPENDENT - SAMPLES T TEST 7 -11 Error Bar Chart I llustrated The figure below provides an example of an the error bar chart. The graph shows mean salary and its associated 95% confidence interval, for each of two groups. Mean salary for men is higher than for women and the intervals do not overlap, indicating differences between men and women in the population in mean salary. Figure 7.13 Error Bar Chart Illustrated 7.10 Requesting an Error Bar Chart with Chart Builder Requesting an Error bar chart is accomplished with these steps, using the Chart Builder procedure. 1) Place an Error Bar chart icon in the Chart Preview area 2) Select the scale variable for which you want a mean and confidence intervals .
3) Select the categorical variable defining the groups .
4) In the resulting graph, check if confidence intervals overlap. 7.11 Error Bar Chart Output The error bar chart will generate a graph depicting the relationship between a scale and categorical variable. It provides a visual sense of how far the groups are separated. • Note the means of each group • Note if the 95% confidence intervals of the groups overlap This method of comparing confidence intervals is not as precise as statistical testing. For example, an error bar chart does not take into account whether the homogeneity of variance assumption is met. Still, used carefully, they can be very useful Note INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -12 • If the error bar and statistical test lead to the same conclusion, support the statistical test with the error bar chart Procedure: Error Bar Chart with Chart Builder The Chart Builder procedure is accessed from the Graphs …Chart Builder menu. W ith the Chart Builder dialog open: 1) Select the simple error bar chart icon in the Bar Choose from: group and place it in the Chart preview area 2) Specify a variable for the X -axis .
3) Specify a variabl e for the Y-axis . This variable should be scale in measurement. Figure 7.14 Chart Builder Dialog to Create Error Bar Chart 7.12 Demonstration : Error Bar Chart with Chart Builder We will create an error bar chart corresponding to the Independent-Samples T-Test of gender and number of children. W e want to see how number of children varies across categories of gender, so we use childs as the Y -axis variable. This is equivalent to how we used the Independent- Samples T Test procedure to study this relationship. THE INDEPENDENT - SAMPLES T TEST 7 -13 Detailed Steps for Error Bar Chart Before beginning this example, to insure that childs is displayed properly in the error bar chart, do the following: 1) In the Data Editor, change the Measure level for childs to Scale 2) Change the value of Width to 3 ; change the number of Decimals to 1 Then, with the Chart Builder dialog open: 1) Select a simple error bar chart icon and put it in the Chart Preview pane .
2) Place childs in the Y -axis box .
3) Place the variable sex in the X-axis box .
Results from the Error Bar Chart We can observe the following details from the graph. • The mean number of children for each gender along with 95% confidence intervals is represented in th is chart • The confidence intervals for the two genders don’t quite overlap, which is consistent with the result from the T Test • The error bars have a small range compared to the range of childs , which indicates we are fairly precisely measuring number of children (because of large sample sizes) Figure 7.15 Error Bar Chart of Education and Gender INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 7 -14 7.13 Lesson Summary We explored the use of the Independent-Samples T-Test procedure and error bar c harts to test whether there are mean differences in a scale variable between two groups. Lesson Objectives Review Students who have completed this lesson should now be able to: • Perform a statistical test to determine whether there is a statistically significant difference between two groups on a scale dependent variable To support the achievement of the primary objective, students should now also be able to:
• Check the assumptions of the Independent- Samples T Test • Use the Independent- Samples T Test to test the difference in means • Know how to interpret the results of a Independent- Samples T Test • Use the Chart Builder to create an error bar graph to display mean differences 7.14 Learning Activity In these activities you will use the file Census.sav . The overall goal is to run the Independent- Samples T Test , to interpret the output and visualize the results with an error bar chart. 1. We want to see whether men and women differ in the ir mean socioeconomic index ( sei ) and their age when their first child was born ( agekdbrn). First, use the Explore procedure to view the distributions of these two variables by gender. Are they similar or different? Do you see any problems with doing a t test? 2. Now do a t test for each variable, by gender. Is the homogeneity of variance assumption met, or not? W hat do you conclude about mean differences by gender? 3. Create an error bar chart for each variable by gender. Is the graph consistent with the result from the t test? 4. Now do the same analysis with the variable race , testing whether there are differences in sei and agekdbrn comparing whites to blacks. Although race has three categories, you can use only two categories in the t test. As before, first use the Explore procedure to view the distributions of these two variables by race? Are they similar or different? Do you see any problems with doing a t test?
5. Now do a t test for each variable, by white versus black. Is the homogeneity of variance assumption met, or not? What do you conclude about mean differences between whites and blacks?
6. Create an error bar chart for each variable by race. Is the graph consistent with the result from the t test? 7. For those with more timed: How could you display an error bar chart with only the categories of white and black, not other? There are at least two methods. The file Census.sav, a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials THE PAIRED- SAMPLES T TEST 8 -1 Lesson 8: T he Paired-Samples T Test 8.1 Objectives After completing this lesson students will be able to: • Perform a statistical test to determine whether there is a statistically significant difference between the means of two scale variables To support the achievement of this primary objective, students will also be able to: • Use the Paired-Samples T Test p rocedure • Interpret the results of a Paired-Samples T Test 8.2 Introduction In this lesson we outline the logic involved when testing for a difference between two scale variables, and then perform an analysis comparing two variables . As with the Independent Sam ples T Test, we use the mean to compare the two variables, as the mean is an excellent measure of central tendency.
As an example, we might want to compare a student´s test score before and after participating in a particular program and see if the program had an effect by calculating the mean difference between the two test scores. Business Context When analyzing data, we are concerned with whether there is a difference between two scale variables. W ithout statistical testing, we might make decisions bas ed on perceptions that are not likely to exist in a population of customers. The Paired -Samples T Test allows us to determine if the difference between two variables in the sample reflects a difference in the population. For example: • In medical research a Paired-Samples T Test would be used to compare means on a measure administered both before and after some type of treatment. • In market research, if a subject were to rate the product they usually purchase and a competing product on some attribute, a Paired-Samples T Test would be needed to compare the mean ratings. • In customer satisfaction studies, if a special customer care program is implemented, we can test whether satisfaction beforehand is lower, or higher, than satisfaction after the program is in place. 8.3 The Paired -Samples T Test The Paired -Samples T Test is used to test for statistical significance between two population means when each observation (respondent) contributes to both means. With a Paired -Samples T Test each person serves as his own control. To the extent that an individual’s outcomes across the two T he file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 8 -2 conditions are related, the Paired-Samples T Test provides a more powerful statistical analysis (greater probability of finding true effects) than the Independent- Samples T Test. Moreover, due to the assumption of independent for the i ndependent-s amples test, a paired- samples test must be used when the same subject/respondent provides both scores. To clearly state the difference between the Independent-Samples T Test and the Paired -Samples T Test , it is instructive to compare the data structure needed for each of the two tests .
Figure 8.1 Data structure for Independent- Samples T Test (left) and Paired -Sam ples T Test (right) 8.4 Assumptions for the Paired -Samples T Test There are three assumptions to apply the Paired-Samples T T est:
1) Variables have a scale measurement level. 2) Variables should be in the same unit of measurement. 3) The difference between the two variables is normally distribut ed (because the mean difference is what is being tested) The Paired -Samples T Test is robust to violations of the normality assumption when sample sizes are moderate to large (over 50 cases per group). Note The homogeneity of variance assumption that holds for the Independent- Samples T Test does not apply to the Paired -Samples T Test since we are dealing w ith only one group. Note THE PAIRED- SAMPLES T TEST 8 -3 8.5 Requesting a Paired -Samples T Test Requesting a Paired -Samples T Test is accomplished by following these steps: 1) Choose pairs of variables for the Paired-Samples T Test .
2) Review the procedure output to investigate th e relationship between the variables including: a. Paired Samples Statistics Table b. Paired Samples Test Table. 8.6 Paired-Samples T Test Output The Paired Samples Statistics table provides summary information including sample sizes, means, standard deviations, and standard err ors for the pair of variables. The Paired Samples Correlations table displays the correlation between the pair of variables. The higher the correlations, the more statistical power there is to detect a mean difference. Figure 8.2 Example of Paired Samples Statistics Output Figure 8.3 Example of Paired Samples Correlations Output The null hypothesis is that the two means are equal. The mean difference in socioeconomic index is 2.42, reported along with the sample standard deviation and standard error in the Paired Samples Test table. Although this is not a large difference, the significance value (. 000) indicates that the two means are significantly different (at the .01 level). A 95% confidence interval for the mean difference is also reported, which can be quite useful. Figure 8-4 Paired Samples Test Table INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 8 -4 8.7 Procedure: Paired -Samples T Test The Paired -Samples T Test procedure is accessed from the Analyze…Compare Means…Paired - Samples T Test menu choice. W ith the Paired -Samples T Test dialog box open: 1) Select the pair of variables to compare and place them in the Paired Variables box (more than one pair of variables can be in the Paired Variable(s) box). Hint: Use Ctrl+Click to select the pair. Figure 8.5 Paired -Samples T Test Dialog 8.8 Demonstration: Paired-Samples T Test We will work with the Census.sav data file in this lesson. To demonstrate a P aired-Samples T Test , we will compare mean education levels of the respondent ( educ ) and his or her spouse (speduc ). The Paired -Samples T Test is appropriate because we will obtain data from a single respondent regarding his/her education and that of the spouse ’s education.
We are interested in testing whether there is a significant difference in education between spouses in the population. Although two people are involved—husband and wife— information is being obtained from only one person, so the paired- sample test is appropriate.
Detailed Steps for Paired -Samples T Test 1) Select the variables educ and speduc and place them in the Paired Variables box Results from Paired-Samples T Test The first table displays the mean, standard deviation and standard error for each of the variables. W e see that the means for respondent and spouse’s education are close, but the education for the spouse is a bit lower. This might i ndicate very close educational matching of people who marry. THE PAIRED- SAMPLES T TEST 8 -5 Figure 8.6 Paired Samples Statistics Table In the next table, the sample size (number of pairs) appears along with the correlation between the two variables . The correlation (. 594) is positive, high, and statistically significant (differs from zero in the population). This suggests that the power to detect a difference between the two means is substantial. Figure 8-7 Paired Samples Correlations T able The mean education difference, about .20 years, is reported along with the sample standard deviation of the difference. The significance of the difference is .024, so if we are using a cr iterion of .05, we would reject the null hypothesis and conclude the means are different (but we would reach a different conclusion if using an alpha value of .01). Whether or not the means are substantively different is a separate question. Figure 8.8 Paired Samples Test Table Apply Your Knowledge 1. True or false? Before doing the Paired- Samples T Test, a test of equality of variances should be done ?
2. See the output below. Which statements are correct ?
a. The sample mean difference is .6. b. The null hypothesis that the mean difference is 0 must be rejected (alpha=0.05) c. Normality of the distribution of mean differences is not a concern here. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 8 -6 8.9 Lesson Summary We explored the use of the Paired-Samples T Test to test the mean difference between two variables. Lesson Objectives Review After completing this lesson students will be able to: • Perform a statistical test to determine whether there is a statistically significant difference between the means of two scale variables To support the achievement of this primary objective, students will also be able to: • Use the Paired-Samples T Test procedure • Interpret the results of a Paired-Samples T Test 8.10 Learning Activity The overall goal of this learning activity is to use the Paired-Samples T Test .
1. One variable in the customer survey asked about agreement that SPSS products are a good value ( gdvalue). A second question asked about agreement that SPSS offers high quality products ( hiqualty). Use a paired- samples t test to see whether the means of these two questions differ (they are measured on a five- point scale). What do you conclude?
2. Then test whether there is a mean difference between agreement that SPSS products are easy to learn ( easylrn) and SPSS products ar e easy to use (easyuse). What do you conclude? 3. Could we use a paired- sample t test to compare how long a customer has used SPSS products ( usespss) and how frequently they use SPSS ( freqspss)? W hy or why not? The SPSS customer satisfaction data file SPSS_CUST.SAV . This data file was collected from a random sample of SPSS customers asking about their satisfaction with the software, service, and other features, and some background information on the customer and their company. Supporting Materials ONE-WAY ANOVA 9 -1 Lesson 9: One -Way AN O VA 9.1 Objectives After completing this lesson students will be able to: • Perform a statistical test to determine whether there is a statistically significant difference among three or more groups on a scale dependent variable To support the achievement of this primary objective, students will also be able to: • Use the options in the One -Way ANOVA procedure • Check the assumptions for One -Way ANOVA • Interpret the results of a One -Way ANOVA analysis • Use the Chart Builder to create an error bar to graph mean differences 9.2 Introduction Analysis of variance (ANOVA) is a general method of drawing conclusions regarding differences in population means when three or more comparison groups are involved. The I ndependent-Sam ples T Test applies only to the simplest instance (two groups), while the One-Way ANOVA procedure can accommodate more complex situations (three or more groups) . In this lesson we will provide information on the assumptions of using the One -Way ANOVA procedure and then provide examples of its use. Business Context When analyzing data, we are often concerned with whether groups differ from each other. W ithout statistical testing, we might make decisions based on perceptions that are not likely to exist in a population of customers. One -Way ANOVA allows us to determine if three or more groups significantly differ on scale var iables; thus we can determine which groups score higher or lower than the others. For example, we might want to know whether: • Customer groups differ on attitude toward a product or service. • Different drugs better reduce depression levels. 9.3 One -Way Anova The basic logic of significance testing for comparing group means on more than two groups is the same as that for comparing two group means (i. e., the Independent- Samples T Test). To summarize: • The null h ypothesis is that the population groups have the same means. The file Census.sav, a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-2 • We d etermine the probability of obtaining a sample with group mean differences as large (or larger) as what we find in our data. To make this assessment the amount of variation among group means (between- group variation) is compared to the amount of variation among observations within each group (within- group variation). Assuming in the population that the group means are identical (null hypothesis), the only source of variation among sample means would be the fact that the groups are composed of different individual observations. Thus a ratio of the two sources of variation (between group/within group) should be about 1 when there are no population differences. W hen the distribution of individua l observations within each group follows the normal curve, the statistical distribution of this ratio is known (F distribution) and we can make a probability statement about the consistency of our data with the null hypothesis. The final result is the probability of obtaining sample differences as large (or larger) as what we found if there were no population differences. If this probability is sufficiently small (usually less than 5 chances in 100, or .05) we conclude the population groups differ. Once we find a difference, we have to determine which groups differ f rom each other. (W hen the null hypothesis is rejected, it does not follow that all group means differ significantly ; the only thing that can be said is that not all group means are the same.) 9.4 Assumptions of One-Way ANOVA To correctly use the One-Way ANOVA procedure requires an understanding of additional issues. 1) The dependent variable must have a scale measurement level .
2) The i ndependent variable (named a “factor” in Anova analyses) must have a categorical measurement level . 3) The distribution of the dependent variable within each population subgroup follows the normal distribution (normality). One -Way ANOVA is robust to moderate violations when sample sizes are moderate to large (over 25 cases) and the dependent measure has the same distribution (for example, skewed to the right) within each comparison group. 4) The variation is the same within each population subgroup (homogeneity of variance) . One - Way ANOVA is robust to moderate violations when sam ple sizes of the groups are similar.
Similar to the Independent -Samples T Test , violation of the assumption of homogeneity of variances is more serious than violation of the assumption of normality. And like the Independent- Samples T Test , One -Way ANOVA applies a two -step strategy for testing: 1) Test the homogene ity of variance assumption.
2) If the assumption holds, proceed with the standard test (the ANOVA F Test ) to test equality of means ; if the null hypothesis of equal variances is rejected, use an adjust ed F test to test equality of means. 9.5 Requesting One -Way ANOVA Running a One -Way ANOVA is accomplished by following these steps: 1) Select the scale variable on which to test equality of group means 2) Select a factor variable. 3) Request the Levene homogeneity of variance test 4) Review the procedure output to: a. Check on the t est for homogeneity of variances b. Review the t est on equality of means, either using the standard ANOVA F test or an adjusted F test taking into account unequal variances 5) If the null hypothesis of equal population means is rejected, extend the analysis by adding a post hoc test within the One -Way ANOVA procedure. ONE-WAY ANOVA 9 -3 9.6 One -Way ANOVA Output The first table of output is the test for homogeneity of variance. The null hypothesis is that the variances are equal, so if the significance level is low enough (as it is in the table below), we reject the null hypothesis and conclude the variances are not equal. As with the independent -samples t test, this isn’t a problem, but it does mean that we should use the tests that adjust for unequal variance. Figure 9.1 Levene Test of Homgeneity of Variances Most of the information in the ANOVA table is technical in nature and is not directly interpreted.
Rather the summaries are used to obtain the F statistic and, more importantly, the probability value we use in evaluating the population differences. The standard ANOVA table will provide the following information: • The first column has a row for the between- groups and a row for within-groups variation. • Sums of squares are intermediate summary numbers used in calculating t he between- (deviations of individual group means around the total sample mean) and within- (deviations of individual observations around their respective sample group mean) group variances. • The “ df” column contains information about degrees of freedom, related to the number of groups and the number of individual observations within each group. • Mean Squares are measures of the between- group and within-group variation (Sum of Squares divided by their respective degrees of freedom). • The F statistic is the ratio of between to within group variation and will be about 1 if the null hypothesis is true. • The column labeled “Sig.” provides the probability of obtaining the sample F ratio (taking into account the number of groups and sample size), if the null hypothesis is true. In practice, most researchers move directly to the significance value since the columns containing the sums of squares, degrees of freedom, mean squares and F statistic are all necessary for the probability calculation but are rarely interpreted in their own right. In the table below, the low significance value leads us to reject the null hypothesis of equal means. You should also examine the actual standard deviations or variances for each group. In large samples, it is relatively easy to reject the null hypothesis of equal variances even when the variances are within a factor of 2 of each other. Tip INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-4 Figure 9.2 ANOV A Table Output When the condition of equal variances is not met, an adjusted F test has to be used. PASW Statistics provides two such tests, Welch and Brown- Forsythe. The table Robust Tests of Equality of Means provides the details. Again, the columns containing test statistic and degrees of freedom are technical details to compute the significance. Figure 9.3 Robust Tests of Equality of Means Output 9.7 Procedure: One-Way ANOVA The One-Way ANOVA procedure is accessed from the Analyze…Compare Means…One -Way ANOVA menu choice. With the One -Way ANOVA dialog box open: 1) Place one or more scale variables in the Dependent List box .
2) Place one categorical variable in the Factor box .
3) Open the Options dialog to request descriptive information and the test for homogeneity of variance. ONE-WAY ANOVA 9 -5 Figure 9.4 One -Way ANOVA Dialog In the Options dialog: 1) Ask for Descriptive statistics so that group means and standard deviations are displayed. 2) The homogeneity of variance test allows one to assess the assumption of homogeneity of variance. 3) Brown -Forsythe and W elch are robust tests that do not assume homogeneity of variance and thus can be used when this assumption is not met. Figure 9.5 One -Way ANOVA Options Dialog INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-6 9.8 Demo nstration : One-Way ANOVA We will work with the Census.sav data file in this lesson. In this example we investigate the relationship between marital status ( marital ) and education in years ( educ ). W e would like to determine whether there are educational differences among marital status groups. Detailed Steps for One- Way ANOVA 1) Place the variable educ in the Dependent List box .
2) Place the variable marital in the Factor box .
3) In the Options dialog, s elect Descriptive, Homogeneity of variance test , Brown -Forsythe , and Welch check boxes . Results from One -Way ANOVA As with all analyses, you should first look to see how many cases are in each group, along with the means and standard deviations. The first table in the Viewer window provides this information. The size of the groups ranges from 70 to 971 people. The means vary from 11.76 to 13.73 years (the One -Way ANOVA procedure will assess if these means differ), while the standard deviations vary from 2.89 to 3.45 (the test of homogeneity of variance will assess if thes e standard deviations differ).
We observe that: • The married group has the most education (13.73) while those separated ha ve the least education (11.76). Figure 9.6 Table of Descriptive Statistics What we don’t know is whether these differences are due to sampling variation, or instead are likely to be real and exist in the population. For this we turn to the ANOVA. Levene Test of Homogeneity of Variance First, we must review Levene’s test of homogeneity of variance. The null hypothesis of homogeneity of within- group variance is not rejected (significance .694). This means we can use the standard ANOVA table. Figure 9.7 Test of Homogeneity of Variances Table ONE-WAY ANOVA 9 -7 ANOVA Table Every statistical test has a null hypothesis. In most cases, the null hypothesis is that there is no difference between groups. This is also true for the null hypothesis for the One-Way ANOVA procedure, so we test with One -Way ANOVA whether there is no difference in mean education among marital status groups. If the significance is small enough, we reject the null hypothesis and conclude that there are differences. Figure 9 .8 ANOVA Table for Educ ation by Marital Status We see that the probability of the null hypothesis being correct is extremely small, less than .05, therefore we reject the null hypothesis and conclude that there are differences in education among these groups. If we had not met the homogeneity of variances assumption, and given our disparate sample sizes, we would have to turn to the Brown- Forsythe and Welch tests, which test for equality of group means without assuming homogeneity of variance. These tests are s hown below, although we would not report them in this situation (where equal variances may be assumed). Figure 9.9 Robust Tests of Mean Differences Both of these measures mathematically attempt to adjust for the lack of homogeneity of variance. • When calculating the between- group to within-group variance ratio, the Brown -Forsythe test explicitly adjusts for heterogeneity of variance by adjusting each group's contribution to the between- group variation by a weight related to its within- group variation.
• The Welch test adjusts the denominator of the F ratio so it has the same expectation as the numerator, when the null hypothesis is true, despite the heterogeneity of within- group variance. Both tests indicate there are highly significant differences in average highest year of school completed between the marital status groups, which are consistent with the conclusions we drew from the standard ANOVA. Having concluded that there are differences in amount of education among different marital status groups, we probe to find specifically which groups differ from which others. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-8 Apply Your Knowledge 1. True or false? Suppose we have collected data for four groups of respondents on region (north/east/south/west) and their income categories (very low/low/moderate/high/very high).
Is One- W ay ANOVA the correct procedure to test whether there are differences between the regions wi th respect to income categories? 2. True or false? When the F -test is not significant, then we will not follow up this analysis with an analysis on which group means differ ?
3. In a dataset about employees and their salaries we tested whether there are differences in mean salary according to the job category of the employee. The output is depicted below.
Which statements are correct ?
a. Although the sample standard deviations are different, the null hypothesis of equal population variances cannot be rejected (alpha=0.05) b. The table titled ANOVA must be discarded, as the null hypothesis of equal population variances is rejected (alpha=0.05) c. The null hypothesis of equal population group means is rejected by both Welch and Brown -Forsythe tests (alpha=0.05) 9.9 Post Hoc Tests with a One -Way ANOVA Post hoc tests are ty pically performed only after the overall F test indicates that population differences exist, although for a broader view see Milliken and Johnson ( 2004). At this point there is usually interest in discovering just which group means differ from which others . In one aspect, the procedure is quite straightforward: every possible pair of group means is tested for population differences and a summary table produced. However, a problem exists in that as more tests are performed, the probability of obtaining at le ast one false-positive result increases. As an extreme example, if there are ten groups, then 45 pairwise group comparisons (n*(n- 1)/2) can be made. If we are testing at the .05 level, we would expect to obtain on average about 2 (.05 * 45) false- positive tests. In an attempt to reduce the false- positive rate when multiple tests of this type are done, statisticians have developed a number of methods. ONE-WAY ANOVA 9 -9 Often, more than one post hoc test is used and the results are compared to provide for more evidence about potential mean differences. 9.10 Requesting Post Hoc Tests with a One- Way ANOVA If the null hypothesis of equal population group means is rejected, a post hoc analysis is required, following these steps: 1) Ask for one or more appropriate post hoc test s in the Post Hoc dialog. 2) Inspect the output and report which groups differ significantly in their population mean.
9.11 Post Hoc Tests Output The table labeled Multiple Comparisons provides all pairwise comparisons. The rows are formed by every possible combination of groups. The column labeled “Mean Difference (I -J)” contains the sample mean difference between each pairing of groups. If this difference is statistically significant at the specified level after applying the post hoc adjustments, then an asterisk (*) appears beside the mean difference. Notice the actual significance value for the test appears in the column labeled “Sig.” . In addition, the standard errors and 95% confidence intervals for each mean difference appear. These provide information on the precision with which we have estimated the mean differences. Note that, as you would expect, if a mean difference is not significant, the confidence interval includes 0. Also notice that each pairwise comparison appears twice. For each such duplicate pair the significance value is the same, but the signs are reversed for the mean difference and confidence interval values. Figure 9.10 Multiple Comparisons Output INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-10 9.12 Procedure: Post Hoc Tests with a One- Way ANOVA Post hoc analyses are accessed from the One -Way ANOVA dialog box. With the One-Way ANOVA dialog box open: 1) Open the Post Hoc Multiple Comparisons dialog 2) Select the appropriate method of multiple comparisons , which will depend on whether the assumpti on of homogeneity of variance has been met .
Figure 9.11 Post Hoc Testing Dialog Why So Many Tests? The Post Hoc dialog lists over a dozen tests just in the Equal Variances Assumed area. We need to review why so many tests are available. The ideal post hoc test would demonstrate tight control of Type I (false- positive) error, have good statistical power (probability of detecting true population differences), and be robust over assumption violations (failure of homogeneity of variance, non- normal error distributions). Unfortunately, there are implicit tradeoffs involving some of these desired features (Type I error and power) and no current post hoc procedure is best in all these areas. Add to this the fact that pairwise tests can be based on different statistical distributions (t, F, studentized range, and others) and that Type I error can be controlled at different levels (per individual test, per family of tests, variations in between), and you have a large collection of post hoc tests. We will briefly compare post hoc tests from the perspective of being liberal or conservative regarding control of the false- positive rate (Type 1 error). The existence of numerous post hoc tests suggests that there is no si ngle approach that statisticians agree will be optimal in all situations. LSD The LSD or least significant difference method simply applies standard t tests to all possible pairs of group means. No adjustment is made based on the number of tests performed. The argument is that since an overall difference in group means has already been established at the selected criterion level (say .05), no additional control is necessary. This is the most liberal of the post hoc tests. ONE-WAY ANOVA 9 -11 SNK, REGWF, REGWQ & Duncan The SNK (Student -Newman -Keuls), REGWF (Ryan -Einot -Gabriel -Welsh F), REGWQ (Ryan- Einot- Gabriel -Welsh Q, based on the studentized range statistic) and Duncan methods involve sequential testing. After ordering the group means from lowest to highest, the two most ext reme means are tested for a significant difference using a critical value adjusted for the fact that these are the extremes from a larger set of means. If these means are found not to be significantly different, the testing stops; if they are different then the testing continues with the next most extreme set, and so on. All are more conservative than the LSD. REGWF and REGWQ improve on the traditionally used SNK in that they adjust for the slightly elevated false- positive rate (Type I error) that SNK has when the set of means tested is much smaller than the full set. Bonferroni & Sidak The Bonferroni (also called the Dunn procedure) and Sidak (also called Dunn- Sidak) perform each test at a stringent significance level to insure that the family -wise (applying to the set of tests) false- positive rate does not exceed the specified value. They are based on inequalities relating the probability of a false- positive result on each individual test to the probability of one or more false positives for a set of indepe ndent tests. For example, the Bonferroni is based on an additive inequality, so the criterion level for each pairwise test is obtained by dividing the original criterion level (say .05) by the number of pairwise comparisons made. Thus with five means, and therefore ten pairwise comparisons, each Bonferroni test will be performed at the .05/10 or .005 level. Tukey (b) The Tukey (b) test is a compromise test, combining the Tukey (see next test) and the SNK criterion producing a test result that falls between the t wo.
Tukey Tukey’s HSD (Honestly Significant Difference; also called Tukey HSD, WSD, or Tukey(a) test) controls the false- positive rate family -wise. This means if you are testing at the .05 level, that when performing all pairwise comparisons, the probabil ity of obtaining one or more false positives is .05. It is more conservative than the Duncan and SNK. If all pairwise comparisons are of interest, which is usually the case, Tukey’s test is more powerful than the Bonferroni and Sidak. Scheffe Scheffe’s met hod also controls the family -wise error rate. It adjusts not only for the pairwise comparisons, but also for any possible comparison the researcher might ask. As such it is the most conservative of the available methods (false- positive rate is least), but has less statistical power. Specialized Post Hoc Tests Hochberg’s GT2 & Gabriel: Unequal Ns Most post hoc procedures mentioned above (excepting LSD, Bonferroni & Sidak) were derived assuming equal group sample sizes in addition to homogeneity of variance and normality of error.
When the subgroup sizes are unequal, PASW Statistics substitutes a single value (the harmonic mean) for the sample size. Hochberg’s GT2 and Gabriel’s post hoc test explicitly allow for unequal sample sizes. Waller -Duncan The Waller -Duncan takes an approach (Bayesian) that adjusts the criterion value based on the size of the overall F statistic in order to be sensitive to the types of group differences associated with the F (for example, large or small). Also, you can specify the rat io of Type I (false positive) to Type II (false negative) error in the test. This feature allows for adjustments if there are differential costs to the two types of errors. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-12 Unequal Variances and Unequal Ns Tamhane T2, Dunnett’s T3, Games-Howell, Dunnett’s C Each of these post hoc tests adjusts for unequal variances and sample sizes in the groups.
Simulation studies (summarized in Toothaker, 1991) suggest that although Games -Howell can be too liberal when the group variances are equal and sample sizes are unequal, it is more powerful than the others. The bottom line is that your choice in post hoc tests should reflect your preference for the power/false- positive tradeoff and your evaluation of how well the data meet the assumptions of the analysis, and you li ve with the results of that choice. 9.13 Demonstration : Post Hoc Tests with a One -Way ANOVA We will work with the Census.sav data file in this example. Previous analysis showed that the null hypothesis of equal mean education for the marital status groups was rejected. Post hoc analysis will reveal which groups differ. Detailed Steps for a Post Hoc Test 1) Place the variable educ in the Dependent List box .
2) Place the variable marital in the Factor box .
3) In the Post hoc dialog, select Bonferroni .
Results for the Post Hoc Tests We will move directly to the post hoc test results. ONE-WAY ANOVA 9 -13 Figure 9.12 Bonferroni Post Hoc Results We see the Married and W idowed groups have a mean difference of 1.26 years of education. This difference is statistically significant at the specified level after applying the post hoc adjustments. The interval [.54, 1.99] contains the difference between these two population means with 95% confidence. Summarizing the entire table, we would say that the Married, Divorced, and Never Married groups differs in amount of education from the Separated and Widowed groups. We could not be sure of this just from examining the means. Apply Your Knowledge 1. In a dataset about employees and their salaries we tested whether there are differences in mean salary according to the political preference of the employee. The output is depicted below. Which statements are correct ?
1) The null hypothesis of equal variances is not rejected (alpha=0.05) 2) The null hypothesis of equal group means is rejected (alpha=0.05) 3) Post -hoc tests show significant differences between all three groups A, B, and C (alpha=.05). INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-14 9.14 Error Bar Chart with Chart Builder For presentations it is often useful to show a graph of the relationship between a scale and categorical variable. Error bar charts are the most effective method of doing this for ANOVAs. The Chart Builder allows for the creation of a variety of charts, including error bar charts. It provides for great flexibility in creating charts, including formatting options. 9.15 Requesting an Error Bar Chart with Chart Builder Follow these steps to produce an error bar using the Chart Builder :
1) Select an error bar chart type of graph. 2) Specify the scale variable for which means and confidence intervals are requested. 3) Specify the categorical variable that defines the groups. 4) Inspect the error bar in the output to see which groups do (not) overlap in their confidence intervals. 9.16 Error Bar Chart Output The standard error bar chart will generate a graph depicting the relationship between a scale and categorical variable. It provides a visual sense of how far the groups are separated. • Note the means of each group (the small circle) • Note if the 95% confidence intervals of the groups overlap ONE-WAY ANOVA 9 -15 Figure 9.13 Error Bar Chart of TV Hours by Highest Degree 9.17 Procedure: Error Bar Chart with Chart Builder The Chart Builder procedure is accessed from the Graphs …Chart Builder menu. W ith the Chart Builder dialog open: 1) Select an error bar chart icon and place it in the Chart preview area. 2) Specify a categorical variable for the X -axis .
3) Specify a scale variable for the Y -axis . The mean is calculated by default. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-16 Figure 9.14 Chart Builder Dialog to Create Error Bar Chart 9.18 Demonstration : Error Bar Chart with Chart Builder We will create an error bar chart for the ANOVA of marital status and education. We want to see visually how education varies across categories of marital status, so we use education as the Y -axis variable. This is equivalent to how we used ANOVA to study this relationship. Detailed Steps for Error Bar Chart 1) Select an error bar chart icon and put it in the Chart Preview pane. 2) Place the variable educ in the Y -axis box .
3) Place the variable marital in the X -axis box . Results from the Error Bar Chart Created with Chart Builder An ANOVA table with means based from education within categories of marital status lets us compare groups, and this is mirrored in the error bar chart. • The mean education of each marital status group along with 95% confidence intervals is represented in thi s chart • Confidence intervals that do not overlap indicate that those groups differ from each other • Confidence intervals that do overlap indicate that those groups do not differ from each other • We can readily compare the categories of marital status in this arrangement ONE-WAY ANOVA 9 -17 Married, divorced and never married respondents have more education than the separated and widowed categories. Figure 9.15 Error Bar Chart of Education and Marital Status Additional Resources For additional information on multiple comparison tests with ANOVA, see: Klockars, Alan J. and Sax, G. 1986. Multiple Comparisons . Newbury Park, CA: Sage. Further Info The confidence intervals on the error bar chart are determined for each group separately and no adjustment is made based on the number of groups that are being compared, or for unequal variance. So an error bar chart should never be used without doing statistical tests. Further Info INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 9-18 9.19 Lesson Summary We explored the use of the One-Way ANOVA procedure to analyze relationships between a categorical and a scale variable. Lesson Objectives Review Students who have completed this lesson should now be able to: • Perform a statistical test to determine whether there is a statistically significant difference among three or more groups on a scale dependent variable Lesson Objectives Review To support the achievement of the primary objective, students should now also be able to: • Use the options in the One -Way ANOVA procedure • Check the assumptions for One -Way ANOVA • Interpret the results of a One -Way ANOVA analysis • Use the Chart Builder to create an error bar to graph mean differences 9.20 Learning Activity The overall goal of this learning activity is to use One-W ay ANOVA with post hoc tests to explore the relationship between several variables. You will use the PASW Statistics data file Census.sav.
1. Investigate how the number of siblings ( sibs) varies by highest degree ( degree). Ask for appropriate statistics. 2. Is the assumption of homogeneity of variance met? Is the ANOVA test significant at the .01 level? 3. Do a post hoc analysis, if justified. Ask for both the Bonferroni and Scheffe tests? What do you conclude from these tests? Which education groups have different mean numbers of children? Are the Bonferroni and Scheffe tests consistent? 4. Create an error bar chart to display the mean differences for sibs by degree. Is the error bar chart a correct representation of which means are different? 5. Now do another analysis of political position ( polviews) by degree. Repeat the same steps from the analysis above. Which education groups differ in their political position? The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES 10-1 Lesson 10: Bivariate Plots and Correlations for Scale Variables 10.1 Objectives After completing this lesson students will be able to: • Perform a statistical test to determine whether two scale variables are correlated (related) To support the achievement of the primary objective, students will also be able to: • Visually assess the relationship between two scale variables with scatterplots , using the Chart Builder procedure • Explain the options of the Bivariate Correlations procedure • Explain the Pearson correlation coefficient and its assumptions • Interpret a Pearson correlation coefficient 10.2 Introduction In this lesson we examine and quantify the relationship between two scale variables. A scatterplot visually presents the relationship between two scale variables, while the Pearson correlation c oefficient is used to quantify the strength and direction of the relationship between scale variables.
The Pearson correlation coefficient (formally named the Pearson product -moment correlation coefficient) is a measure of the extent to which there is a linear (or straight line) relationship between two variables. Business Context When we examine the distributions of two scale variables, we would like to know whether a relationship we observe is likely to exist in our target population or instead is caused by random sampling variation. Statistical testing tells us whether two scale variables are related. Assessing correlation coefficients allows us to determine the direction and strength of this relationship. For example, we might want to know whether: • Higher SAT scores are associated with higher first year college GPAs • Eating more often at fast -food r estaurants was related to more frequent shopping at convenience stores • Lower levels of depression are associated with higher self -esteem scores 10.3 Scatterplots The scatterplot visually presents the relationship between two scale variables. A scatterplot displays individual observations in an area determined by a vertical and a horizontal axis, each of which The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 10-2 repres ent the variables of interest (note that the variables must be scale). In a scatterplot, look for a relationship between the two variables and note any patterns or extreme points. Scatterplot illustrated For an illustration, see the scatterplot below, showing the relationship between variables age and salary . T he relationship is linear (a straight line describes the relationship between age and salary ), and positive (if age increases, then so does salary ); there is one person having a salary which is “out of line .” Figure 10.1 Scatterplot of Age and Salary 10.4 Requesting a Scatterplot The scatterplot is available in the Chart Builder procedure. The steps to create a scatterplot are: 1) Select a chart (simple scatter) from the Gallery. 2) Place the dependent variable on the vertic al (y) axis and the independent variable on the horizontal (x). 3) Review the procedure output to investigate the relationship between the variables including: a. Linearity b. Directionality c. Outliers . Place the independent variable on the x -axis and the dependent variable on the y -axis. Here, salary depends on age and not vice versa, so salary is the dependent variable on the y -axis, age the independent variable on the x -axis. If the situation is such that no dependent or independent variable can be identified (say number of hours watching tv and number of hours of internet use), then the choice of which variable goes where is arbitrary. Best Practice BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES 10-3 10.5 Scatterplot Output The standard scatterplot will generate a graph depicting the relationship between the two variables. • Note if there is a linear relationship • Note the direction of the relationship • Note any outliers Figure 10.2 Scatterplot Showing the R elationship between Beginning Salary and E ducation L evel 10.6 Procedure : Scatterplot The scatterplot is created from the Graphs …Chart Builder menu choice. With the Chart Builder dialog box open: 1) Select a Simple Scatter graph on the canvas and drag and drop i t in the Chart Preview area. 2) Place the dependent variable in the vertical axis .
3) Place the independent variable in the horizontal axis . INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 10-4 Figure 10.3 Chart Builder Dialog Box to Create a Scatterplot 10.7 Demonstr ation: Scatterplot We will work with the C ensus.sav data file in this example. In this demonstration we examine the relationship between mother’s education (maeduc ) and father’s education ( paeduc). This will allow us to investigate whether people marry those with a similar amount of education (we could also investigate this question by examining the scatterplot of the respondent’ education and his/her spouse). Detailed Steps for a Scatterplot 1) Place the Simple Scatter icon in the Char t Preview area 2) Place the variable maeduc in the Y-axis box .
3) Place the variable paeduc in the X -axis box . Results from the S catterplot The scatterplot visually presents the relationship between two variables by displaying individual observations. Observe that for the most part couples with low education tend to marry each. Also, couples with high education tend to marry each other, thus there is a positive, linear relationship. BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES 10-5 Figure 10.4 Scatterplo t of Mother’s and F ather’s Highest Year of School Completed In preparation for our discussion of the correlation coefficient, we will edit the scatterplot to superimpose a best fit straight line to the data. 10.8 Adding a Best Fit Straight Line to the Scatterplot All charts in PASW Statistics can be edited. The type of editing that can be done depends on the type of chart. For a scatterplot, one option is to add a best fit line, which can be done in one step. Detailed Steps to Edit Scatterplot We need to open the scatterplot in the Chart Editor. 1) Double- click on the chart to open it in the Chart Editor 2) Select Elements…Fit Line at Total 3) Close the Chart Editor The straight line tracks the positive relationship between father and mother’s educational attai nment.
How well do you think it describes or models the relationship? Does it match what you would have drawn by hand? We use scatterplots to get a sense of whether or not it is appropriate to use a correlation coefficient to describe this relationship wi th one number. As we will learn, the correlation coefficient assumes a linear relationship. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 10-6 Figure 10.5 Scatterplot with Fit Line Added Apply Your Knowledge 1. Consider the scatterplot between X and Y depicted below. Which statements are correct ? a. There is a linear relationship between X and Y b. There is a positive relationship between X and Y c. The point labeled A is far from the straight line describing the relationship between X and Y d. The point labeled B is far from the straight line describing the relationship between X and Y BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES 10-7 10.9 Pearson Correlation Coefficient The Pearson Correlation Coefficient is a measure of the extent to which there is a linear (or straight line) relationsh ip between two scale variables. It is normed so that a correlation of +1 indicates that the data fall on a perfect straight line sloping upwards (positive relationship), while a correlation of – 1 would represent data forming a straight line sloping downwards (ne gative relationship). A correlation of 0 indicates there is no straight -line relationship at all. Below are four scatterplots with their accompanying correlations, all based on simulated data following normal distributions. Four different correlations appear ( here, and in general, the letter “r” is used to denote the Pearson Correlation Coefficient ). All are positive, but represent the full range in strength of linear association (from 0 to 1). As an aid in interpretation, a best -fitting straight line is superimposed on each chart. Figure 10.6 Scatterplots Based on Various Correlations For the perfect correlation of 1.0, all points fall on the straight line trending upwards. In the scatterplot with a cor relation of .8 in the upper right, the strong positive relation is apparent, but there is some variation around the line. Looking at the plot of data with correlation of .4 in the lower left, the association is clearly less pronounced than with the data correlating .8 (note greater scatter of points around the line). The final chart displays a correlation of 0: there is no linear association present and the best -fitting straight line is a horizontal line. We use statistical tests to determine whether a rel ationship between two or more variables is statistically significant. That is, we want to test whether the correlation differs from zero (zero indicates no linear association) in the population, based on the sample results. In other words, the null hypothesis is the correlation coefficient is 0 in the population, and we use a statistical test to assess this hypothesis. Pearson Correlation Coefficient Assumptions To correctly use the Pearson c orrelation coefficient and apply statistical tests, three conditions must be met: 1) Variables must have a scale measurement level .
2) Variables must be linearly related. 3) Variables must be normally distributed. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 10-8 10.10 Requesting a Pearson Correlation Coefficient The Pearson Correlation Coefficient is available in the Bivariate Correlations procedure. Requesting a Pearson Correlation Coefficient is accomplished with these steps: 1) Choose variables for the correlation (We simply list the variables to be analyzed; there is no designation of dependent and independent variables. Correlations will be calculated on all pairs of variables listed.). 2) Optionally s elect the correlation statistic to calculate. Pearson is used for scale variables, while Spearman and Kendall’s tau- b (less common) are used for non-normal data or ordinal data, as relationships are evaluated after the original data have been transformed into ranks. 3) Optionally, s elect a one- or two- tailed significance test to perform. 4) The Flag significant correlations check box is checked by default. W hen checked, asterisks appearing beside the correlations will identify significant correlations. 5) Examine the output to see which correlations are significant. 10.11 Bivariate Correlation Output The standard correlation table will provide the following information: • The Pearson Correl ation, which will range from +1 to - 1, the further away from 0, the stronger the relationship • The 2- tailed significance level, the test of the null hypothesis that the correlation is 0 in the population; all correlations with a significance level less than .05 will be considered statistically significant and will have an asterisk next to the coefficient • N, which is the sample size • The correlations in the major diagonal will always be 1, because these are the correlations of each variable with itself • The cor relation matrix is symmetric, so that the same information is represented above and below the major diagonal Figure 10.7 Example of Bivariate Correlations Output BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES 10-9 10.12 Procedure: Pearson Correlation with Bivariate Correlations The Bivariate Correlations procedure is accessed from the Analyze…Correlate…Bivariate menu choice. With the Bivariate Correlations dialog box open: 1) Place two or more variables in the Variables box .
2) Optionally, s elect the correlation coefficient to calculate. 3) Optionally, select the type of significance test to perform .
4) Open the Options dialog box to display descriptive information and determine how to handle missing values .
Figure 10.8 Bivariate Correlations Dialog Box Dichotomous variables with only two categories can be used to calculate correlation coefficients. A scatterplot won’t be useful to study the relationship between a dichotomous variable and a scale variable, but the correlation coefficient still provides information on the strength and direction of the relationship. Further Info INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 10-10 10.13 Demonstration : Pearson Correlation with Bivariate Correlations We will work with the Census.sav data file in this lesson. In this demonstration we examine the relationship between mother’s education (maeduc ), father’s education (paeduc ), and the education of the respondent ( educ). W e would like to determine whether, for example, people marry people with a similar amount of education and if the children also have a similar amount of education as their parents. Detailed Steps for Bivariate Correlations 1) Place maeduc, p aeduc and educ in the Variables box . Results from Bivariate Correlations In the Correlations table we see that the three correlations are moderate to high, and positive .
C ouples with high education tended to marry each other, thus we have a positive, linear relationship (r = .68). There is a similar level of association between the respondent’s education and that of his father (r=.48) and that of his mother (r=.44). Figure 10.9 Correlations Table of Education Variables What we don’t know is whether these relationships are due to sampling variation, or instead are likely to be real and exist in the population of adults. For this we turn to the significance level of the Pearson Correlation Coefficient .
We see that the probability of the null hypothesis being true for all of these relationships is extremely small, less than .01; therefore we reject the null hypothesis and conclude that there is a positive, linear relationship between these variables. Apply Your Knowledge 1. True or false? If we remove point A from the data, the correlation between X and Y will be lower? BIVARIATE PLOTS AND CORRELATIONS FOR SCALE VARIABLES 10-11 2. True or false? In the two scatterplots below, t he correlation between age and income is higher in A than in B ?
3. What is the range of a Pearson correlation coefficient? a. From 0 to 1 b. Can take on any positive or negative value c. From –1 to 1 d. Depends on the standard deviation of the variables 10.14 Lesson Summary We explored the use of scatterplots and c orrelations to examine relationships between scale variables. Lesson Objectives Review Students who have completed this lesson should now be able to: • Perform a statistical test to determine whether two scale variables are correlated (related) To support the achievement of the primary objective, students should now also be able to: INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 10-12 • Visually assess the relationship between two scale variables with scat terplots, using the Chart Builder procedure • Explain the options of the Bivariate Correlations procedure • Explain the Pearson correlation coefficient and its assumptions • Interpret a Pearson correlation coefficient 10.15 Learning Activity The overall goal of this learning activity is to visualize the relationship between two scale variables creating scatterplots and to quantify this relationship with the correlation coefficient. In this set of learning activities you will use the data file Bank .sav. 1. Suppose you are interested in understanding how an employees demographic characteristics, beginning salary, and time at the bank and in the work force are related to current salary. Start by producing scatterplots of salbeg, sex, time, age, edlevel , and work with salnow . Add a fit line to each plot. Check on the variable labels for time and work so you understand what these variables are measuring. 2. Describe the relationships based on the scatterplots. Do they all appear to be linear? Are any relationships negative? W hat is the strongest relationship? 3. Now produce correlations with all these variables. W hich correlations with salnow are significant? What is the largest correlation in absolute value with salnow ? Did this match what you thought based on the scatterplots? 4. Examine the correlations between the other variables? W hich variables are most strongly related? Create scatterplots for these as well to check for linearity. 5. For those with more time: Go back and review the scatterplots with salnow . Are there any employees who are outliers —far from the fit line—in any of the scatterplots? How might they be affecting the relationship?
The file Bank.sav , a PASW Statistics data file that contains information on employees of a major bank.
Included is data on beginning and current salary position, time working, and demographic information. Supporting Materials REGRESSION ANALYSIS 11-1 Lesson 11: Regression Analysis 11.1 Objectives After completing this lesson students will be able to: • Perform linear regression to determine whether one or more variables can significantly predict or explain a dependent variable To support the achievement of the primary objective, students will also be able to: • Explain linear r egression and its assumptions • Explain the options of the Linear Regression procedure • Interpret the results of the Linear Regression p rocedure 11.2 Introduction Correlations allow one to determine if two scale variables are l inearly related to each other. So for example, beginning salary and education might be positively related for employees. Regression allows one to further quantify this relation by developing an equation predicting start ing salary based on education. Linear regression is a statistical method used to predict a variable (a scale dependent measure) from one or more independent (scale) variables. Commonly, straight lines are used, although other forms of regression allow nonli near functions. In this lesson we will focus on linear regression. Business Context When we examine the relationships between scale variables, we would like to know whether a relationship we observe is likely to exist in our target population or instead is caused by random sampling variation. Statistical testing tells us whether scale variables are related. T he results of linear regression allow us to determine if one or more scale variables predict an outcome variable and the impact each independent variable has on this variable. For example, we might want to know whether: • Higher SAT scores explain higher first year college GPAs • Eating more often at fast -food restaurants predicts more frequent shopping at convenience stores • Increasing income leads to more customer purchases 11.3 Simple Linear Regression Linear regression involving a single independent (scale) variable is the simplest case and is called simple linear regression. In other words, one scale variable, say X , is used to predict another scale variable, say Y. The file Bank.sav , a PASW Statistics data file that contains information on employees of a major bank.
Included is data on beginning and current salary position, time working, and demographic information. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-2 Simple Regression Illustrated When there is a single independent variable, the relationship between the independent variable and dependent variable can be visualized in a scatterplot, and the concept of l inear regression can be explained using the scatterplot. Figure 11.1 Scatterplot of Height and Weight The line superimposed on the scatterplot is the best straight line t hat describes the relationship. The line is represented in general form by the equation, Y = a + b*X where, b is the slope (the change in Y per unit change in X) and a is the intercept (the value of Y when X is zero). (Here, Y is weight in pounds and X is height in inches ).
The value of the equation is linked to how well it actually describes or fits the data, and so part of the regression output includes fit measures. T o quantify the extent to which the straight line fits the data, the fit measure, R Square , was developed. R 2 h as the dual advantages of falling on a standardized scale and having a practical interpretation. The R Square measure (which is simply the correlation squared, or r 2, when there is a single predictor variable, and thus its name) is on a scale from 0 (no linear association) to 1 (perfect linear prediction). Also, the R Square value can be interpreted as the proportion of variation in one variable that can be predicted from the other. Thus an R Square of .50 indicates that we can account for 50% of the variance in one variable if we know values of the other.
You can think of this value as a measure of the improvement in your ability to predict one variable from the other (or others if there are multiple independent variables). We use statistical tests to determine whether a relationship between the independent var iable and dependent variable is statistically significant. That is, we want to test whether the predictor can significantly explain the dependent variable. REGRESSION ANALYSIS 11-3 • Linear Regression is used to determine whet her the independent variable can significantly explain the dependent variable, in other words, test the null hypothesis that R Square is zero • Once we find a relationship, we have to assess the effect of the independent variable on the dependent variable Finally, referring to the scatterplot, we see that many points fall near the line, but some are quite a distance from it. For each point, the difference between the value of the dependent variable and the value predicted by the equation (value on the line) is called the residual (also known as the error).
Points above the line have positive residuals (they were under -predicted), those below the line have negative residuals (they were over -predicted), and a point falling on the line has a residual of zero (p erfect prediction). Points having relatively large residuals are of interest because they represent instances where t he prediction line did poorly. Outliers, or points far from the mass of the others, are of interest in regression because they can exert co nsiderable influence on the equation (especially if the sample size is small). 11.4 Simple Linear Regression Assumptions To correctly use simple linear regression and apply statistical tests, four conditions must be met: 1) Variables must have a scale measurement level .
2) Var iables must be linearly related. 3) Residuals must be normally distributed. 4) Residuals are assumed to be independent of the predicted values, implying that the variation of the residuals around the line is homogeneous .
A variable coded as a dichotomy (say 0 and 1) can technically be considered as a scale variable. A scale variable assumes that a one- unit change has the same meaning throughout the range of the scale. If a variable’s only possible codes are 0 and 1 (or 1 and 2, etc.), then a one- unit change does mean the same change thr oughout the scale. Thus dichotomous variables, for example sex, can be used as predictor variables in regression. It also permits the use of nominal predictor variables if they are converted into a series of dichotomous variables; this technique is called dummy coding and is considered in most regression texts (Draper and Smith, 1998; Cohen and Cohen, 2002). Note While not covered in this course, PASW Statistics can provide influence statistics to aid in judging whether the equation was strongly affected by a particular observation. Note INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-4 11.5 Requesting Simple Linear Regression Requesting a linear regression involves these steps: 1) Select a dependent variable and an independent variable 2) Review the procedure output to investigate the relationship between the variables including: a. R b. Adjusted R 2 2 3) Examine the regression test statistics to determine whether the observed relationship is statistically significant .
.
4) Determine the impact of the independent variable on the dependent variable. 11.6 Simple Linear Regression Output The standard linear regression will generate three tables depicting the relationship between the two variables.
• The Model Summary table provides several measures of how well the model fits the data • R, the multiple correlation coefficient , which can range from 0 to 1, is a generalization of the correlation coefficient. It is the correlation between the dependent measure and the combination of the independent variable(s), thus the closer the multiple R is to 1, the better the fit. If there is only one predictor, the multiple correlation coefficient is equivalent to the Pearson correlation coefficient. • R Square, which can range from 0 to 1, is the correlation coefficient squared. It can be interpreted as the proportion of variance of the depe ndent measure that can be predicted from the independent variable(s). • Adjusted R Square represents a technical improvement over R Square in that it explicitly adjusts for the number of predictor variables relative to the sample size. If Adjusted R Square a nd R Square differ dramatically, it is a sign that you have used too many predictor variables for your sample size. • Standard Error of the Estimate is a standard deviation type summary of the dependent variable that measures the deviation of observations ar ound the best fitting straight line. As such it provides, in the scale of the dependent variable, an estimate of how much variation remains to be accounted for after the line is fit. Figure 11.2 Model Summary and ANOVA Tables REGRESSION ANALYSIS 11-5 While the fit measures indicate how well we can expect to predict the dependent variable, they do not tell whether there is a statistically significant relationship between the dependent and independent variable( s). The analys is of variance table ( ANOVA in the Output Viewer) presents technical summaries (sums of squares and mean square statistics) of the variation accounted for by the prediction equation. Our main interest is in determining whether there is a statistically significant (non- zero) linear relation between the dependent variable and the independent variable(s) in the population. • The Sig. column provides the probability that the null hypothesis is true (i.e., no relationship between the independent and dependent variable). We use the significance, as in any other hypothesis test, to determine whether or not to reject the null hypothesis. If significant results are found, we turn to the next table, Coefficients , to view the regression coefficients. Figure 11.3 Regression Coefficients • The first column contains a list of the independent variables plus the intercept ( Constant ).
The intercept is the value of the dependent variable when the independent variable is 0. • The column labeled B contains the estimated regression coefficients we would use in a prediction equation. The coefficient for height indicates that on average, each additional inch in height was associated with an increase in weight of 5.33 pounds. • The Standard Error (of B) column contains standard errors of the regression coefficients. The standard errors can be used to create a 95% confidence interval around the B coefficients. • Betas are standardized regression coefficients and are used to judge the r elative importance of each of several independent variables. • The t statistics provide a significance test for each B coefficient, testing whether it differs from zero in the population. 11.7 Procedure: Simple Linear Regression Linear Regression is available i n the Regression …Linear menu choice. With the Linear Regression dialog box open: 1) Place the dependent variable in the Dependent: box .
2) Place the independent variable in the Independent(s): box .
INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-6 Figure 11.4 Linear Regression Dialog Box In the Linear Regression dialog: 1) The Independent(s) list box will permit more than one independent variable, and so this dialog box can be used for both simple and multiple regression. 2) The block controls permit an analy st to build a series of regression models with the variables entered at each stage (block), as specified by the user. 3) By default, the regression Method is Enter, which means that all independent variables in the block will be entered into the regression eq uation simultaneously. This method is selected to run one regression based on all variables you specify. If you wish the program to select, from a larger set of independent variables, those that in some statistical sense are the best predictors, you can re quest the Stepwise method. 4) The Selection Variable option permits cross -validation of regression results. Only cases whose values meet the rule specified for a selection variable will be used in the regression analysis ; then the resulting prediction equati on will be applied to the other cases. Thus you can evaluate the regression on cases not used in the analysis, or apply the equation derived from one subgroup of your data to other groups. 5) The Statistics dialog box presents many additional (and some of the m quite technical) statistics. 6) The Plots dialog box is used to generate various diagnostic plots used in regression, including residual plots. 7) The Save dialog box permits you to add new variables to the data file. These variables contain such statistics as the predicted values from the regression equation, various residuals and influence measures. 8) The Options dialog box controls the criteria when running stepwise regression and choices in handling missing data. By default, PASW Statistics excludes a case from regression if it has one or more values missing for the variables used in the analysis. REGRESSION ANALYSIS 11-7 11.8 Demonstration : Simple Linear Regression We will work with the Bank.sav data file in this example. In this example we examine the relationship between beginning salary at the bank ( salbeg) and highest year of school c ompleted (edlevel ). We would like to determine whether, for example, people with more education receive a higher initial salary, and then determine the impact of each additional year of education on salary.
Before we begin, we should view a scatterplot of these two variables. Detailed Steps for Scatterplot In the Chart Builder dialog, accessed from Graphs…Chart Builder :
1) Place the Simple Scatter icon in the Chart Preview area 2) Place the variable salbeg in the Y -axis box. 3) Place the variable edlevel in the X -a xis box. We can observe that there is a positive relationship between education and beginning salary. This relationship, though, is not strong, as there is a fair amount of scatter in the data. The relationship does seem reasonably linear. The PASW Statistics Missing Values add- on module provides more sophisticated methods for handling missing values. This module includes procedures for displaying patterns of missing data and imputing (estimating) missing values using multiple variable imputation methods. Note INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-8 Figure 11.5 Scatterplot of Beginning Salary and Education Level Once we have viewed the scatterplot, we are read for the simple linear regression. Detailed Steps for Simple Linear Regression 1) Place salbeg in the Dependent box .
2) Place edlevel in the Independent(s) box . Results from Simple Linear Regression Here we can observe that the multiple R or correlation coefficient between highest year of school completed and beginning salary is .6 33—this is the correlatio n between these two variables. If we square the m ultiple R, we get R S quare, which tells us that about 40 % of the variance (.401) in the dependent variable can be predicted from the independent variable. Adjusted R Square is basically the same, as there is only one independent variable in the model. If you wish, you can edit the chart and request a regression fit line to determine whether linear regression seems appropriate for these data. Note REGRESSION ANALYSIS 11-9 Figure 11.6 Model Summary Table for Regression So now we know the correlation between the dependent and independent variables and the percent of variance accounted for by t he model. However, we don’t know whether this relationship is due to sampling variation, or instead is likely to be real and exist in the population. For this we turn to the ANOVA table. Every statistical test has a null hypothesis. In most cases, the null hypothesis is that there is no relationship between two v ariables. This is also true for the null hypothesis for Linear Regression (the null hypothesis is that we have no relationship between the dependent and the combination of independent variable(s)). If the significance is small enough, the null hypothesis h as to be rejected. T he probability of the null hypothesis being correct for this relationship is extremely small, less than .0 1, therefore the null hypothesis has to be rejected and the conclusion is that there is a linear relationship between these variables. Figure 11.7 ANOVA Table Because a significant relationship has been found between our dependent and independent variable, next step is to determine what the impact on the dependent variable is .
F igure 11.8 Coefficients Table Looking at the B column, we see that for each additional year of education completed, the expected increase in beginning salary is $691. 01. In fact, if we wanted to predict beginning salary based on education, we would use the following equation: salbeg = –2,516.387 + 691.011*( educ). INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-10 The column t is a technical detail and tells that the observed sample coefficient ( 691.011) is 17 .773 times the standard error ( 38.879) away from the value that we expect if the null hypothesis is true, i.e., zero. Such a t value leads to a significance value of .000 (column Sig .) for the B coefficient. Apply Your Knowledge 1. True or false? In the figure depicted below the point labeled A has a positive residual ?
2. True or false. Suppose we predict Salary with Age and find an R Square of .25. Then 75% of the variation in Salary cannot be accounted for by Age?
3. Which coefficient is used in the regression equation to make predictions? a. Beta b. B The column labeled Standardized Coefficients will be discussed in the Multiple Regression section. Note In case of simple regression the test on the null hypothesis R Square = 0 tests the same as the test on B = 0 . In both cases the test is whether X and Y are linearly related. This equivalence can be seen in the test statistics, as the F value in the ANOVA table is the square of the t value in the Coefficients table. Further Info REGRESSION ANALYSIS 11-11 11.9 Multiple Regression A regression involving more than one independent variable is called multiple regression and is a direct extension of simple linear regression. When we run multiple regression we will again be concerned with how well the equation fits the data, whether a linear model is the best fit to the data, whether any of the variables are sign ificant predictors, and estimating the coefficients for the best - fitting prediction equation. In addition, we are interested in the relative importance of the independent variables in predicting the dependent measure. 11.10 Multiple Linear Regression Assumptions To correctly use multiple regression and apply statistical tests, one extra condition has to be met, in addition to the assumptions stated for simple linear regression. We restate the four assumptions for simple linear regression and add the extra assumption:
1) Variables must have a scale measurement level .
2) Variables must be linearly related. 3) Residuals must be normally distributed. 4) Residuals are assumed to be independent of the predicted values, implying that the variation of the residuals around the line is homogeneous .
5) Absence of multi -collinearity —no exact or nearly exact linear relation between the independent variables .
11.11 Requesting Multiple Linear Regression Requesting multiple linear regression is accomplished with these steps: 1) Select a dependent var iable and two or more independent variables 2) Request a histogram of the residuals; this allows for a check of the normality of errors 3) Request a scatterplot of the standardized residuals and standardized predicted value; this allows for a check of the homogeneity of errors 4) Review the procedure output to investigate the relationship between the variables including: a. R b. Adjusted R 2 2 5) Examine the regression test statistics to determine whether the observed relationship is statistically significant .
.
6) Determine which independent variables are significantly related to the dependent variable. 7) Determine the impact of each independent variable on the dependent variable. 11.12 Multiple Linear Regression Output The standard linear regression will generate the same tables depicting the relationship between the variables that were discussed earlier.
The R square and Adjusted R square are interpreted the same, except now the amount of explained variance is from a group of predictors, not just one. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-12 Figure 11.9 Variables in the Model and Model Summary The ANOVA table has a similar interpretation, except that now it tests whether any variable has a significant effect on the dependent variable. Figure 11.10 ANOVA Table for Multiple Regression In the Coefficients table, there is an additional twist to the interpretation of the B coefficient. For example, the effect here of height on weight can be stated as every additional inch of height predicts an additional 4.261 pounds of weight. However, that estimate controls for the other variables in the model . Here, we must also examine the significance values for each coefficient, because a regression that is overall significant does not imply that each coefficient is statistically significant. REGRESSION ANALYSIS 11-13 Figure 11.11 Multiple Regression Coefficients Table To test the assumptions of regression, we turn to the histogram of residuals. The residuals should be approximately normally distributed, which is basically true for the histogram below. Figure 11.12 Histogram of Residuals Additionally, the scatterplot of the standardized error (residu al) and standardized predicted value (here of weight) should show no pattern if homogeneity of variance holds. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-14 Figure 11.13 Scatterplot of Residuals and Predicted Value 11.13 Procedure: Multiple Linear Regressio n Multiple linear r egression is available in the Regression…Linear menu choice. With the Linear Regression dialog box open: 1) Place the dependent variable in the Dependent: box .
2) Place the independent variables in the Independent (s): box.
3) Select the Plots button to open that dialog REGRESSION ANALYSIS 11-15 Figure 11.14 Linear Regression Dialog for Multiple Regression In the Plots dialog: 4) Select Histogram 5) Move *ZRESID into the Y: box in the Plots dialog and*ZPRED into the X: box in the Plots dialog Figure 11.15 Linear Regression Plots Dialog INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-16 11.14 Demonstration : Multiple Linear Regression We will work with the Bank.sav data file in this lesson. In this example we examine the relationship between beginning salary at the bank and several predictors, including education, years of previous work experience ( work), age, and gender ( sex).
Note that gender is a dichotomous variable coded 0 for males and 1 for females, but it can be included as an independent variable (see Message Box above). Detailed Steps for Multiple Linear Regression 1) Place salbeg in the Dependent : box .
2) Place edlevel , sex, age, and work in the Independent(s) : box .
While we can run mult iple regression at this point, we will request some diagnostic plots involving residuals and information about outliers. By default no residual plots will appear. 3) Select the Plots button 4) Select Histogram .
5) Move *ZRESID into the Y: box in the Plots dialog .
6) Move *ZPRED into the X: box in the Plots dialog. We requested a histogram of the standardized residuals because regression assumes that the residuals follow a normal distribution. Regression can produce summaries concerning various types of residuals. W e request a scatterplot of the standardized residuals ( *ZRESID) versus the standardized predicted values ( *ZPRED) because regression assumes that residuals are independent of predicted values, thus if we see any patterns (as opposed to a random blob) in this plot, it might suggest a way of adjusting and improving the analysis. Next we request casewise diagnostics in the Statistics dialog . The Casewise Diagnostics check box requests information about all cases whose standardized residuals are more than 3 standard deviations from the fit line. 7) Select the Statistics button 8) Select Casewise Diagnostics Results from Multiple Linear Regression Recall that Linear Regression uses listwise deletion of missing data so that if a case is missing data on any of the five variables used in the regression it will be dropped from the analysis. If this results in heavy data loss, other choices for handling missing values are available in th e Regression Options dialog box. Here we can observe that the multiple R or correlation coefficient between our combination of predictors and the dependent variable is .699. If we s quare the multiple R, we get R square, which is .489. The Adjusted R square is .485, which is about the same. When reporting on explained variance—R square— always report the adjusted R square. Further Info REGRESSION ANALYSIS 11-17 Therefore, about 48.5% of the variance in beginning salary can be predicted from the four independent variables. Figure 11.16 Model Summary Table for Multiple Regression We see that the probability of the null hypothesis being correct for this relationship is extremely small, less than .0 1, therefore we reject the null hypothesis and conclude that there is a linear relationship between thes e variables and beginning sal ary. Figure 11.17 ANOVA Table for Regression of Beginning Salary Figure 11.18 Regression Coefficients Table to Predict Beginning Salary In the Coeffic ients table, the independent variables appear in the order they were listed in the Linear Regression dialog box, not in order of importance. Although the B coefficients are important for prediction and interpretive purposes, analysts usually look first to the t test at the end of each line to determine which independent variables are significantly related to the outcome measure. Since four variables are in t he equation, we are testing if there is a linear relationship between each independent INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-18 variable and the dependent variable after adjusting for the effects of the three other independent variables. Looking at the significance values we see that edlevel , sex , and age significant at the .05 significance level, while the work experience is not (sig = .352 after controlling for the other predictors). The estimated regression ( B) coefficient for edlevel is about $651, similar but not identical to the coefficient ( 691) found in the simple regression using edlevel alone. In the simple regression we estimated the B coefficient for edlevel ignoring any other effects, since none were included in the model. Here we evaluate the effect of edlevel after controlling (statistically adjusting) for age, sex, and work . If the independent variables are correlated, the change in B coefficient from simple to multiple regression can be substantial. So, after controlling (holding constant) age, sex, and work , one year of formal education, on average, was worth another $651 in beginning salary .
Continuing on: • The variable sex has a B coefficient of about – $1,526. This means that a one- unit change in gender (which means moving from male status to female, or compar ing females to males), controlling for the other variables, is associated with a drop in beginning salary of –$1,526. • The variable age has a B coefficient of $33, so each additional year increases beginning salary by $33. • Since we found work experience’s coefficient to be not significantly different from zero, we treat it as 0. If we simply look at the estimated B coefficients we might think that sex is the most important variable. However, the magnitude of the B coefficient is influenced by the unit of m easurement (or standard deviation if you like) of the independent variable. The Beta coefficients explicitly adjust for such standard deviation differences in the independent variables. • They indicate what the regression coefficients would be if all variables were standardized to have means of 0 and standard deviations of 1. • A Beta coefficient thus indicates the expected change (in standard deviation units) of the dependent variable per one standard deviation unit increase in the independent variable (after adjusting for other predictors). This provides a means of assessing relative importance of the different predictor variables in multiple regression. • The Betas are normed so that the maximum should be less than or equal to one in absolute value (if any Betas are above 1 in absolute value, it suggests a problem with the data: multi - collinearity). Examining the Betas, we see that edlevel is the most important predictor, followed by sex , and then age. The Beta for work is near zero, as we would expect. If we needed to predict salbeg from these background variables (dropping work ) we would use the B coefficients. Rounding to whole numbers, we would say: salbeg = –2,667 + 651* edlevel – 1526 *sex + 33* age .
Diagnostic Statistics The request for casewise di agnostics produces two tables , the most important of which is shown below. The Casewise Diagnostics table lists those observations more than three standard deviations (in error) from the regression fit line. Assuming a normal distribution, this would happe n less than 1% of the time by chance alone. In this data file that would be about 5 outliers (.01* 474), so the six case s does not seem excessive. Residuals should normally be balanced between positive and negative values; when they are not, you should investigate the data further . In these data, all six residuals are positive, so this does indicate that some additional investigation is required. We could, for example s ee if these observations have anything in common ( very high education which may be out of line REGRESSION ANALYSIS 11-19 with others). Since we know their case numbers (an ID variable can be substituted), we could find them easily at them more closely. We also don’t want to discover very large prediction errors, but here residuals are very, very high, over 6 standard deviations above the fit line. Figure 11.19 Casewise Listing of Outlier s In the diagnostic plots involving residuals we see the distribution of the residuals with a normal bell - shaped curve superimposed, depicted in the figure below. Th e residuals are fairly normal , although they are a bit too concentrated in the center . They are also somewhat positive skewed. Given this pattern, a data analyst might try a data transformation on the dependent measure, whic h might improve the properties of the residual distribution, e.g., the log. However, just as with ANOVA, larger sample sizes protect against moderate departures from normality. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-20 Figure 11.20 Histogram of the Residuals In the scatterplot of residuals, we hope to see a horizontally oriented blob of points with the residuals showing the same spread acr oss different predicted values. Unfortunately, we see a hint of a curving pattern: the residuals seem to slowl y decrease, then swing up at higher salaries. This type of pattern can mean the relationship is curvilinear. Also, the spread of the residuals is much more pronounced at higher predicted salaries. This suggests lack of homogeneity of variance. REGRESSION ANALYSIS 11-21 Figure 11.21 Scatterplot of Residuals and Predicted Val ue for Beginning Salary Apply Your Knowledge 1. Consider the output below , for the regression where Miles per gallon for a vehicle is predicted from Engine Size, Horsepower, W eight and American car or not (coded as 1 for American cars and 0 for cars from other countries). Which statements are correct ?
a. We predict lower mpg for an American car than for a non-American car a. All predictors have an effect significantly di fferent from 0 b. The most important predictor for Miles per Gallon is Vehicle Weight c. The fact that there is a n Unstandardized coefficient greater than 1 (in absolute value) indicates a problem due to multi -collinearity. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-22 Additional Resources 11.15 Lesson Summary We explored the use of Linear Regression to test relationships between scale variables and develop prediction equations to prediction the dependent variable and determine the impact of each independent variable on the dependent variable. Lesson Objectives Review Students who have completed this lesson should be able to: • Perform linear regression to determine whether one or more variables can significantly predict or explain a dependent variable To support the achievement of the primary objective above, students should also be able to: • Explain linear regression and its assumptions • Explain the options of the Linear Regression procedure • Interpret the results of the Linear Regression procedure 11.16 Learning Activity The overall goal of this learning activity is to run linear regressions and to interpret the output . You will use the PASW Statistics data file Census.sav.
1. Run a linear regression to predict total family income ( income06) with highest year of education ( educ). First, do a scatterplot of these two variables and superimpose a fit line.
Does the relationship seem linear? How would you characterize the relationship?
2. Now run the linear regression. What is the Adjusted R square value? Is the regression significant? What is the B coefficient for educ ? Interpret it.
The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials For additional information on linear regression analysis, see:
Allison, Paul D. 1998. Multiple Regression: A Primer .
Thousand Oaks, CA: Pine Forge. Draper, Norman and Smith, Harry. 1998. Applied Regression Analysis. 3rd ed. New York: Wiley. Further Info REGRESSION ANALYSIS 11-23 3. Next add the variables born (born in the U.S. or overseas), age, sex , and number of brothers and sisters ( sibs). Check the coding on born so you can interpret its coefficient. First, do a scatterplot of age and sibs with income06. Superimpose a fit line. Does the relationship seem linear? How would you characterize the relationship? Why not do scatterplots of income06 with sex and born ?
4. Use all these variables to predict income06. Request residual statistics including the histogram of errors and the scatterplot of standardized values. Also request casewise diagnostics. What is the Adjusted R square? How much has it increased from above? 5. Which variables are significant predictors? What is the effect of each on income06? W hich variable is the strongest predictor? The weakest?
6. Examine the casewise diagnostics. Do you see any pattern? Are there more cases with large errors than we would expect? 7. Examine the histogram and scatterplot. Are the errors normally distributed? Do you see any pattern in the scatterplot? What might that mean? 8. What is the prediction equation for income06 ?
9. For those with more time: Add additional variables to the regression equation for income06 .
Examples are father and mother’s education, or number of children. Be careful to add variables that are at least on an interval scale of measurement. Repeat the exercise above.
Are the new variables significant predictors? Does adding variables change the effects of the variables already in the model from above? INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 11-24 NONPARAMETRIC TESTS 12-1 Lesson 12: Nonparametric Tests 12.1 Objectives After completing this lesson students will be able to: • Perform non-parametric tests on data that don’t meet the assumptions for standard statistical tests To support the achievement of this primary objective, students will also be able to: • Describe when non- parametric tests should and can be used • Describe the options in the Nonp arametric procedure dialog box and tabs • Interpret t he results of several types of nonparametric tests 12.2 Introduction Parametric tests, such as the t test, ANOVA, or the Pearson correlation coefficient , make several assumptions about the data. They typically assume normality of the variable(s) distribution for scale variables, and they often assume that the variance is equal within categories of a grouping or factor variable. If these assumptions are clearly violated, the results of these tests are in doubt. Fortunately there are alternatives. There are a whole family of various tests and methods that make fewer assumptions about the data.
T hese tests fall under the class of nonparametric statistics.
1) These methods generally don’t assume normality or variance equality .
2) They are generally less powerful than parametric tests, which means they have a lower chance of finding true differences .
3) These tests are most useful with questions using a short response scale with 3, 4, or 5 points. These scales are not truly interval in measurement. 4) They are also useful when variables have very skewed distributions and so the normality assumption is violated. 5) These tests are also commonly used when sample size is small. Business Context Nonparametric tests allow us to determine if we have relationships in our data when we do not meet important distributional assumptions. They permit us to do standard analysis —looking at group differences, or assessing associations between variables —with all types of data, thus extending data analysis capabilities. The file Census.sav , a PASW Statistics data file from a survey done on the general adult population. Questions were included about various attitudes and demographic characteristics. Supporting Materials INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-2 12.3 Nonparametric Analyse s PASW Statistics includes a wizard to guide you through selecting the appropriate nonparametric test for a particular set of variables. You need to know whether the general situation is: • One Sample: A dependent variable with no grouping variable • Independent Samples: A dependent variable with a grouping (factor) variable • Related Samples: Two (or more) dependent variables whose association you wish to test (such as experiments with pre- and post -tes t measurements) Figure 12.1 Nonparametric Menu Choices 12.4 The Independent Samples Nonparametric Analysis When the dependent variable is scale and we want to test equality of population means for two , or three or more groups, the procedures that we h ave used were the Independent- Samples T Test and By default, the Nonparametric Tests procedures will use the declared scale of measurement of variables to determine how they are used. In parti cular, recall that nonparametric tests are an alternative to parametric methods such as One -Way ANOVA , and these parametric procedures assume that a dependent variable is scale. Therefore, to conduct the equivalent nonparametric test, the dependent variabl e must have scale level of measurement in PASW Statistics.. If you use the default settings in the Nonparametric Tests Wizard, it is critical that the level of measurement be set correctly for each variable in the analysis. Important NONPARAMETRIC TESTS 12-3 One-W ay ANOVA, respec tively. Certain conditions (normality, homogeneity of variances) had to be satisfied to use these procedures. If the assumptions are violated, or if the variable is ordinal in nature, t hese tests cannot be used and an alternative is needed. N onparametric Independent Samples test s provide this alternative. The wizard will select the appropriate test, depending on whether there are two groups or three or more groups. If there are more than two groups, a post hoc analysis can be run to determine which groups differ significantly, analogous to the post hoc pairwise comparisons in the One-Way ANOVA procedure. Independent Samples Nonparametric Assumptions The nonparametric tests for tw o or more independent samples only assume: 1) There is a categorical independent variable defining two or more groups 2) There is an ordinal or scale dependent variable on which group differences are tested.
12.5 Requesting an Independent Samples Nonparametric Analysis Requesting a nonparametric test for two or more independent samples is accomplished with these steps: 1) Select Independent Samples from the Nonparametric Tests menu entry.
2) Select whether you want to compare the shape of the distributions, the medians, or customize the analysis .
3) Specify the Test field and Group variables .
4) To see the equivalent of post hoc pairwise comparisons, you can request a specific test and the appropriate com parison.
5) Review the significance test and other output in the Model Viewer . 12.6 Independent Samples Nonparametric Tests Output The output from the Independent Samples test s, and for all procedures in the Nonparametric Tests menu entry, is produced in the Model Viewer. Initially, all that is displayed is the actual test result. • The null hypothesis being tested is described in plain language • The specific test used is noted, here the Kruskal -Wallis test • The significance level is listed • The decision about the null hypothesis is listed, using the .05 level of significance Figure 12.2 Nonparametric Independent Samples Hypothesis Test Summary INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-4 By double-clicking on the Model output, the Model V iewer window is opened, with additional output.
The Model Viewer contains two panes: 1) The Main View appears on the left side. It displays general information about the tests, and there is a dropdown list at the bottom of the pane that allows you to switch between views about the test/model. 2) The Auxiliary view appears on the right side. This view displays a more detailed visualization (including tables and graphs) of the model compared to the general visualization in the main view. Like the main view, the auxiliary view may have more than one element that can be displayed. Initially, in the main view, the overall test is listed, identical to what was displaye d in the Output Viewer. In the A uxiliary view, the distribution of the dependent variable is displayed via a boxplot, and details about the test are listed. Figure 12.3 Model Viewer Panes for Nonparametric Independent Samples Test In the A uxiliary view, we can open the Pairwise Comparisons view, as shown in the figure below, where all the possible pairwise comparisons are listed (with no redundancy). The tests are adjusted for the fact that multiple comparisons are being done (6 tests in this example). Here only one comp arison is significant at the .05 level, that between the middle and working class groups. They differ in their attitude toward spending on supporting scientific research. The distance network chart —the graph above the table—lists the average rank for eac h category of the grouping variable. The rank is used because this is a nonparametric test, and data can be ranked without making many assumptions about the dependent variable. It is a graphical representation of the comparisons table in which the distances between nodes in the network correspond to differences between samples. Yellow lines correspond to statistically significant differences; black lines correspond to non- significant differences. Hovering over a line in the network displays a tooltip with t he adjusted significance of the difference between the nodes connected by the line. NONPARAMETRIC TESTS 12-5 Figure 12.4 Pairwise Comparison Tests 12.7 Procedure: Independent Samples Nonparametric Tests The nonparametric Independent Samples tests are accessed from the Analyze …Nonparametric Tests…Independent Samples menu choice. W ith the dialog box open: In the Objectives tab: 1) Select Customize analysis INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-6 Figure 12.5 Nonparametric Tests Two or More Independent Samples Dialog In the Field s tab: 1) Specify the test variable(s) 2) Specify the Groups variable NONPARAMETRIC TESTS 12-7 Figure 12.6 Two or More Independent Samples Fields Tab Dialog In the Settings tab: 1) Select Customize tests and s pecify the desired test and, if applicable, any multiple comparisons option. INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-8 Figure 12.7 Two or More Independent Samples Settings Tab Dialog 12.8 Demonstration : Independent Samples Nonparametric Tests In this example we will use the file Census.sav. Our objective is to see how one’s political position (from liberal to conservative; the variable polviews ) is related to marital status. For example, are married people more conservative than others? The variable polviews is measured on a seven- point scale, and it is truly ordinal, not interval, in measurement scale. Therefore, testing for differences by marital status is best done with a nonparametric method. Detailed Steps for Independent Samples Nonparametric Test 1) In the Objectives tab, s elect Customize analysis. 2) In the Fields tab, specify polviews as the Test Fields variable.
3) Specify marital as the Groups variable .
4) In the Settings tab, select Customize tests The variable polviews may not be Scale in measurement level in the data. If not, change its measurement level before beginning this example. Note NONPARAMETRIC TESTS 12-9 5) Select the Kruskal -Wallis 1 -way ANOVA test, and also select the All pairwise option on the Multiple comparisons : dropdown Results from Independent Samples Nonparametric Test The model view output, initially condensed in the Viewer window, shows that we reject the null hypothesis, as the significance of the Kruskal -Wallis test is .000. This test uses the ranks of cases on the dependent variable to determine whether there are differences between categories, and we conclude that there are. Figure 12.8 Nonparametric Test of Political Position by Marital Status To decide which categories differ from others, we need to open the Model viewer and look in the Auxiliary pane at the Pairwise Comparisons view. There are many possible comparisons to make, and those significant at the .05 level are highlighted. We find that: • Those who are never married are significantly different than those who are divorced, widowed, or married • There are no other pairs that are significantly different (although the separated- married pair has an adjusted significance of .063). INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-10 Figure 12.9 Pairwise Comparison Tests for Marital Status To see the average rank on polviews for the pairs that differ, we can look at the distance network chart. The average rank for the never married category is 800.51, lower than any other category. And lower values on polviews indicate more liberal attitudes , so this implies that those who have never been married are more liberal than married, divorced, and widowed respondents. Figure 12.10 Distance Network Chart for Marital Status NONPARAMETRIC TESTS 12-11 Apply Your Knowledge 1. True or false? In order to assess if there is a relationship between region and gender (two nominal variables), a nonparametric Independent Samples test can be run ?
2. See the dataset shown below , with data collected on students, with grade for mathematics at two different points in time. What is the appropriate test to see whether grade differs at time 2 from time 1? a. Nonparametric test s: One Sample b. Nonparametric test s: Independent Samples c. Nonpar ametric test s: Related Samples d. Parametric test: Independent Samples T-Test 12.9 The Related Samples Non parametric Analysis As the parametric Independent- Samples T Test and One -Way ANOVA have their analog in the nonparametric Independent Samples tests, so has the parametric Paired-Samples T Test its equivalent in the nonparametric Related Samples test. (Actually the parametric Paired-Samples T Test is done with two paired variables, while the nonparametric Related Samples procedure allows for two or more paired variables.) The nonparametric tests for two or more paired samples only assume that:
A One -Way ANOVA would find similar results as the Kruskal -Wallis test, but the nonparametric test is more appropriate for the data. However, sometimes an analyst will perform both tests, and if the nonparametric test is consistent with the parametric test results, report only on the latter. Further Info INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-12 1) The measurement level of the variables is ordinal or scale (depending on the specific test chosen) .
12.10 Requesting a Related Samples Nonparametric Analysis Requesting a nonparametric analysis is accomplished with these steps: 1) Select the Related Samples menu selection.
2) Select whether you want to compare observed data to hypothesized or customize the analysis .
3) Specify the Test fields. 4) Review the significance test and oth er output in the Model Viewer.
12.11 Related Samples Nonparametric Tests Output The output from the nonparametric Related Samples procedure is produced in the Model Viewer.
Initially, all that is displayed is the actual test result. • The null hypothesis being tested is described in plain language • The specific test used is noted, here the Wilcoxon Signed Ranks test • The significance level is listed • The decision about the null hypothesis is listed, using the .05 level of significance Figure 12.11 N onparametric Related Samples Test Summary By double- clicking on the Model output, the Model Viewer window is opened, with additional output.
The Model Viewer contains two panes: 1) The Main View appears on the left side. It displays general information about the tests, and there is a dropdown list at the bottom of the pane that allows you to switch between views about the test/model. 2) The Auxiliary view appears on the right side. This view displays a more detailed visualization (including tables and graphs) of the model compared to the general visualization in the main view. Like the main view, the auxiliary view may have more than one element that can be displayed. Initially, in the main view, the overall test is listed, identical to what was displayed in the Output Viewer. In the auxiliary view, the distribution of the variables is displayed via a histogram, and details about the W ilcoxon Signed Ranks test are listed. NONPARAMETRIC TESTS 12-13 Figure 12.12 Model Viewer Panes Related Samples Test 12.12 Procedure: Related Samples Nonparametric Tests The nonparametric Related Samples procedure is accessed from the Analyze …Nonparametric Tests…Related Samples menu choice.
In the Objectives tab: 1) Select Customize analysis . INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-14 Figure 12.13 Nonparametric Tests Two or More Related Samples Dialog In the Fields tab: 1) Specify the Test Fields variables NONPARAMETRIC TESTS 12-15 Figure 12.14 Field Tab Dialog Related Samples In the Settings tab: 1) Select Customize tests and specify the desired test INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-16 Figure 12.15 Settings Tab Dialog Related Samples 12.13 Demonstration : Related Samples Nonparametric Tests In this example we will continue to use the data file Census.sav . Several questions asked about the respondent’s interest in various subjects, on a three- point scale from Very Interested to Not at all interested. W e wo uld like to see whether the respondents have more interest in medical discoveries ( intmed) or in scientific discoveries ( intsci). To understand what we are testing, below are the two frequency tables for these two variables. These variables are measured on an ordinal scale, so a paired- sample t test not justified. We can see that the percentage of people saying they are Very Interested in medical discoveries is higher than for scientific discoveries, but we want to see if this difference is statistically significant. We will use a test based on the median of the distribution. NONPARAMETRIC TESTS 12-17 Figure 12.16 Frequencies for Interest in Medicine and Interest in Science Detailed Steps for Related Samples Nonparametric Test 1) Chan ge the level of measurement to Scale for intmed and intsci 2) In the Objectives tab of the Nonparametric tests: Two or More Related Samples dialog , select Customize analysis.
3) In the Fields tab, specify intmed and intsci as the Test Field variables 4) In the Sett ings tab, select the Wilcoxon matched -pair signed -rank test. Results from Related Samples Nonparametric Test The model view output, initially condensed in the Viewer window, shows that we reject the null hypothesis, as the significance of the Wilcoxon Signed Ranks test is . 000. This test uses the ranks of cases on the variables to determine whether there are differences between them .
INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-18 Figure 12.17 Nonparametric Test of Interest in Medicine and Science To see the dist ributions of the differences between these two variables, we need to open the Model viewer and look in the A uxiliary pane. The bar chart of differences varies from – 2 to 2 because the variables are coded from 1 to 3, so the difference can vary from (3- 1) to (1-3). It is clear that there are more differences in one direction than another (the actual direction depends on how the variables are coded and the order of variables in the dialog box). It is the number of positive and negative differ ences that is used to calculate the test statistic. Figure 12.18 Distribution of Differences between Interest in Medicine and Science The actual output doesn’t tell us which variable has a higher level of interest, which is why we viewed the frequency tables first. We now know that there is more interest in medical discoveries than scientific discoveries. NONPARAMETRIC TESTS 12-19 The distribution of the each variable can be viewed within the Model Viewer, though. 1) Select Continuous Field Information from the View dropdown 2) Select INTERESTED IN MEDICAL DISCOVERIES from the Field(s): dropdown Figure 12.19 Distributi on of Interest in Medical Discoveries Apply Your Knowledge 1. Would you use a related - samples nonparametric test in the following situations ? Select all that apply. a. When related variables are not truly interval/scale in measurement.
b. When we want to compar e two groups of respondents c. When we want to compare three ordinal variables measured on the same response scale Additional Resources 12.14 Lesson Summary We demonstrated the use of the Nonparametric Tests procedure in this lesson for data that don’t meet the assumptions for parametric tests. For additional information on nonparametric tests, see:
Daniel, W ayne W. 2000. Applied Nonparametric Statistics. 2nd ed. Boston: Duxbury Press. Gibbons, Jean D. 2005. Nonparametric Measures of Association. Newbury Park, CA: Sage. Further Info INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 12-20 Lesson Objectives Review Students who have completed this lesson should now be able to:
• Perform non-parametric tests on data that don’t meet the assumptions for standard statistical tests To support the achievement of this primary objective, students should now also be able to: • Describe when non- parametric tests should and can be used • Describe the options in the Nonp arametric procedure dialog box and tabs • Interpret the results of several types of nonparametric tests 12.15 Learning Activity The overall goal of this learning activity is to use nonparametric tests to explore the relationship between several variables, using the data file SPSS_CUST.SAV .
1. Most of the questions asking for customer evaluation of the software and service are measured on a five- point scale from Strongly Agree to Strongly Disagree (lower values are more agreement, equivalent to higher satisfaction). Review the data file. 2. Test whether overall customer satisfaction ( satcust) is different by highest degree earned ( degree). Be sure to code satcust as Scale beforehand. Try both the Kruskal-Wallis test for independent samples and the Median test. What do you conclude? Are the test results consistent? 3. Look at the pairwise results for each test. Which degree groups are more satisfied overall?
Are the pairwise results consistent? If not, how do they differ? Is there something odd about the pairwise results for the Kruskal -Wallis test? 4. To help think further about these results, use Crosstabs to request a table of satcust by degree. Request appropriate percentages and a chi -square t est. Is this analysis consistent with the nonparametric analysis? 5. For those with more time: Temporarily remove the five respondents with only some high school education and rerun the nonparametric tests. Does this change any of the results? 6. Two questions on the survey ask whether SPSS products are easy to learn ( easylrn) and easy to use ( easyuse). Test whether customers think that products are easier to learn than use, or vice- versa with a related samples test. Use both the related- samples W ilcoxon signed rank test and the related- samples Friedman test. Are the results consistent for both tests?
What do you conclude? The SPSS customer satisfaction data file SPSS_CUST.SAV . This data file was collected from a random sample of SPSS customers asking about their satisfaction with the software, service, and other features, and some background information on the customer and their company. Supporting Materials COURSE SUMMARY 13-1 Lesson 13: Course Summary 13.1 Course Objectives Review Now that you have completed the course, you should be able to: • Perform basic statistical analysis using selected statistical techniques with PASW Statistics And you should be able to: • Explain the basic elements of quantitative research and issues that should be considered in data analysis • Determine the level of measurement of variables and obtain appropriate summar y statistics based on the level of measurement • Run the Frequencies procedure to obtain appropriate summary statistics for categorical variables • Request and interpret appropriate summary statistics for scale variables • Explain how to make inferences about p opulations from samples • Perform crosstab analysis on categorical variables • Perform a statistical test to determine whether there is a statistically significant relationship between categorical variables • Perform a statistical test to determine whether there is a statistically significant difference between two groups on a scale variable • Perform a statistical test to determine whether there is a statistically significant difference between the means of two scale variables • Perform a statistical test to determ ine whether there is a statistically significant difference among three or more groups on a scale dependent variable • Perform a statistical test to determine whether two scale variables are correlated (related) • Perform linear regression to determine whether one or more variables can significantly predict or explain a dependent variable • Perform non- parametric tests on data that don’t meet the assumptions for standard statistical tests 13.2 Course Review: Discussion Questions 1. Is there a “correct” order to the steps for a statistical analysis? 2. What factors would help you to decide whether to use a table or a chart to report on an analysis? Or to use both? 3. Which procedure do you prefer to analyze scale variables, Frequencies or Descriptives? 4. When should you use nonparametric methods of analysis?
5. What should you do when a relationship that is substantively significant is not statistically significant? INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS 13-2 13.3 Next Steps Thought Starters How might you use Paired T Tests in analyzing data for your organization? How might y ou use ANOVA in data analysis? How might you use Regression in data analysis? Next Courses This course discussed many statistical techniques . In this section we provide direction for what courses you can attend to broaden your knowledge in specific areas .
If you want to learn more about this: Take this course: Regression and related methods Advanced Techniques: Regression ANOVA and related methods Advanced Techniques: ANOVA A variety of advanced statistical methods Advanced Statistical Analysis Using PASW Statistics INTORDUCTION TO STATISTICAL ANALYSIS REFERENCES A-1 Appendix A: Introduction to Statistical Analysis References 1.1 Introduction This appendix lists only references that cover several of the techniques and statistical methods discussed in this course. Some of these references, such as the book by Field, also cover advanced statistics. More detailed references for specific techniques are included in some of the lessons.
1.2 References INTRODUCTORY AND INTERMEDIATE STATISTICS TEXTS Berenson, Mark L., Timothy C. Krehbiel, David M. Levine. 2005. B asic Business Statistics: Concepts and Applications. 10th ed. New York: Prentice Hall. Burns, Robert P and Burns, Richar d. 2008 . Business Research Methods and Statistics Using SPSS .
London: Sage Publications Ltd. Field, Andy. 2009. Discovering Statistics Using SPSS . 3rd ed. London: Sage Publications Ltd. Knoke, David, Bohrnstedt, George W. and Mee, Alisa Potter. 2002. Statistics for Social Data Analysis . 4th ed. W adsworth Publishing. Norusis, Marija J. 2009. SPSS 17.0 Guide to Data Analysis . New York: Prentice-Hall.
Norusis, Marija J. 2010. PASW Statistics 18.0 Statistical Procedures Companion. (forthcoming) New York: Prentice- Hall.
INTRODUCTION TO STATISTICAL ANALYSIS USING IBM SPSS STATISTICS A - 2
