Psych

Longitudinal Relations Between Children’s Exposure to TV Violence and Their Aggressive and Violent Behavior in Young Adulthood: 1977–1992 L. Rowell Huesmann, Jessica Moise-Titus, Cheryl-Lynn Podolski, and Leonard D. Eron University of Michigan Although the relation between TV-violence viewing and aggression in childhood has been clearly demonstrated, only a few studies have examined this relation from childhood to adulthood, and these studies of children growing up in the 1960s reported significant relations only for boys. The current study examines the longitudinal relations between TV-violence viewing at ages 6 to 10 and adult aggressive behavior about 15 years later for a sample growing up in the 1970s and 1980s. Follow-up archival data (N 450) and interview data (N 329) reveal that childhood exposure to media violence predicts young adult aggressive behavior for both males and females. Identification with aggressive TV characters and perceived realism of TV violence also predict later aggression. These relations persist even when the effects of socioeconomic status, intellectual ability, and a variety of parenting factors are controlled. Over the past 40 years, a body of literature has emerged that strongly supports the notion that media-violence viewing is one factor contributing to the development of aggression. The majority of empirical studies have focused on the effects of watching dramatic violence on TV and film. Numerous experimental stud- ies, many static observational studies, and a few longitudinal studies all indicate that exposure to dramatic violence on TV and in the movies is related to violent behavior (Huesmann & Miller, 1994; Huesmann, Moise, & Podolski, 1997). Furthermore, a sub- stantial body of psychological theory has developed explaining the processes through which exposure to violence in the mass media could cause both short- and long-term increases in a child’s ag- gressive and violent behavior (Bandura, 1977; Berkowitz, 1993; Eron, 1963; Huesmann, 1988, 1998; Zillmann, 1979). Long-term effects with children are now generally believed to be primarilydue to long-term observational learning of cognitions (schemas, beliefs, and biases) supporting aggression (Berkowitz, 1993; Hues- mann, 1988, 1998), whereas short-term effects with adults and children are recognized as also due to priming (Huesmann, 1998), excitation transfer (Zillmann, 1983), or imitation of specific be- haviors. Most researchers of aggression agree that severe aggres- sive and violent behavior seldom occurs unless there is a conver- gence of multiple predisposing and precipitating factors such as neurophysiological abnormalities, poor child rearing, socioeco- nomic deprivation, poor peer relations, attitudes and beliefs sup- porting aggression, drug and alcohol abuse, frustration and prov- ocation, and other factors. The evidence is already substantial that exposure to media violence is one such long-term predisposing and short-term precipitating factor. The current longitudinal study adds important additional empirical evidence that the effects of child- hood exposure to media violence last into young adulthood and increase aggressive behavior at that time for both males and females. Theoretical Background In discussing the alternative theoretical perspectives that have emerged to explain the obtained relations between exposure to violence (family, community, or mass media) and subsequent aggressive behavior in the observer, it is important to distinguish between short-term effects and longer term effects.

In recent theorizing, long-term relations have been ascribed mainly to acquisition through observational learning of three social-cognitive structures: schemas about a hostile world, scripts for social problem solving that focus on aggression, and normative beliefs that aggression is acceptable (Bushman & Huesmann, 2001; Huesmann, 1988, 1998). Building on the accumulating evidence that human and primate young have an innate tendency to imitate whomever they observe (Butterworth, 1999; Wyrwicka, 1996), these theories propose that very young children imitate almost any specific behaviors they see. Observation of specific aggressive behaviors around them increases children’s likelihood of behaving in exactly that way (Bandura, 1977; Bandura, Ross, & L. Rowell Huesmann, Jessica Moise-Titus, Cheryl-Lynn Podolski, and Leonard D. Eron, Research Center for Group Dynamics, Institute for Social Research, University of Michigan.

The research reported here was supported in part by Grants MH-28280, MH-31866, MH-38683, and MH-47474 from the National Institute of Mental Health (NIMH). We gratefully acknowledge the assistance of a large number of staff who have worked on this project over the course of 20 years, including Drew Battiger, Jennifer Blom, Pat Brice, Marla Commons, Vicki Crawshaw, Julie Crews, Eric Dubow, Paulette Fischer, Kathryn Foley, Janet Garcia, Sheree Hemphill, Gary Hudson, Rosemary Klein-Smith, Debbie Kopp, Darren Loomer, Cathy McGuire, Megan Ma- lecek, Rebecca Mermelstein, Laurie Miller, Patty Mullally, Richard Ro- manoff, Erica Rosenfeld, Pam Sama, Evelyn Seebauer, Jean Shin, Jane Swanson, Patti Timmons, David Tulsky, Linda White, Joe Wisler, Patty Yarmel, Arnaldo Zelli, and Lara Zuckert.

We also acknowledge the assistance of our colleagues who helped with the conception of the study, Riva Bachrach, Adam Fraczek, Nancy Guerra, Kirsti Lagerspetz, Simha Landau, Peter Sheehan, and Vappu Viemero, and the help of David Pearl of NIMH in funding the project.

Correspondence concerning this article should be addressed to L. Row- ell Huesmann, Research Center for Group Dynamics, Institute for Social Research, University of Michigan, 426 Thompson Street, Ann Arbor, Michigan 48106. E-mail: [email protected] Developmental PsychologyCopyright 2003 by the American Psychological Association, Inc.

2003, Vol. 39, No. 2, 201–2210012-1649/03/$12.00 DOI: 10.1037/0012-1649.39.2.201 201 Ross, 1963a). Proactive-instrumental aggressive behaviors in chil- dren 2 to 4 years old generally appear spontaneously (Tremblay, 2000), as may hostile temper tantrums. However, the observation of specific aggressive behaviors at that age leads to the acquisition of more coordinated aggressive scripts for social problem solving and counteracts environmental forces aimed at conditioning the child out of aggression. As the child grows older, the social scripts acquired through observation of family, peers, community, and the mass media become more complex, abstracted, and automatic in their invocation (Huesmann, 1988). In addition, children’s social- cognitive schemas about the world around them begin to be elaborated. In particular, extensive observation of violence around them biases children’s world schemas toward attributing hostility to others’actions (Comstock & Paik, 1991; Gerbner, Gross, Mor- gan, & Signorielli, 1994). Such attributions in turn increase chil- dren’s likelihood of behaving aggressively (Dodge, 1980; Dodge, Pettit, Bates, & Valente, 1995). As children mature further, nor- mative beliefs about what social behaviors are appropriate become crystallized and begin to act as filters to limit inappropriate social behaviors (Huesmann & Guerra, 1997). Children’s own behaviors influence the normative beliefs that develop, but so do the chil- dren’s observations of the behaviors of those around them, includ- ing those observed in the mass media (Guerra, Huesmann, Tolan, Van Acker, & Eron, 1995; Huesmann, 1999; Huesmann, Guerra, Zelli, & Miller, 1992). In summary, social-cognitive observational- learning theory postulates long-term effects of exposure to vio- lence through the influence of exposure on the development of aggressive problem-solving scripts, hostile attributional biases, and normative beliefs approving of aggression.

A major alternative or complementary theory explaining long- term effects is desensitization theory. This theory is based on the empirical fact that most humans seem to have an innate negative emotional response to observing blood, gore, and violence. In- creased heart rates, perspiration, and self-reports of discomfort often accompany such exposure (Cline, Croft, & Courier, 1973; Moise-Titus, 1999). However, with repeated exposure to violence, this negative emotional response habituates, and the observer becomes desensitized. The presumption is that lack of a negative emotional response to observing violence also indicates a flat response to planning violence or thinking about violence. Thus, proactive-instrumental aggressive acts become easier to commit.

There are two other quite different theoretical perspectives that have attempted to explain long-term relations between exposure to violence and aggression without hypothesizing any direct effect of violence viewing on aggression. One theory is that aggressive behavior or a correlate of aggressive behavior stimulates exposure to violence and thus engenders the observed relation between them. Observational studies of aggressive children do show that the aggressive child is likely to provoke others, who then respond aggressively to the child, creating a violent environment that the child“observes.”For media violence, however, the usual assump- tion is that the aggressive child simply“likes”watching media violence more than other children do (Atkin, Greenberg, Kor- zenny, & McDermott, 1979; Fowles, 1999; Goldstein, 1998).

Drawing on social comparison theory, Huesmann (1988, 1995, 1998) elaborated on this theme by suggesting that aggressive children feel happier and more justified if they believe they are not alone in their aggression, and viewing media violence makes them feel happier because it convinces them that they are not alone.The other alternative theory that has been widely discussed is best described as the“third variable”theory. A wide variety of demographic, family, and personal characteristics are known to be correlated both with TV viewing and with aggression, such as social class and IQ (Comstock & Paik, 1991). These theories suggest that the observed long-term positive relations between aggression and exposure to media violence are spurious and are derived from their joint association with one or more of these third variables.

These“third variable”explanations should not be confused with the developmental perspectives on the observational learning and desensitization theories, which also assign important roles to par- enting, intellectual ability, and social class as contributors to both exposure to violence and acceptability of aggressive behavior.

Children of lower socioeconomic status (SES) and lower IQ are known to watch more TV (Comstock & Paik, 1991), probably for multiple reasons including social norms, cost of alternative enter- tainments, and frustration with more intellectually demanding tasks. Parents’TV habits and child-rearing practices also influence the child’s TV habits (Comstock & Paik, 1991). Of course, early parenting factors such as harsh punishment, rejection of the child, and lack of discipline are also known to influence subsequent aggression by the child (Tremblay, 2000). The difference is that these factors are viewed not as explaining away the“effect”of exposure to violence on aggression but as explaining individual differences in exposure to violence and individual differences in the strength of the effect.

Any of these theoretical processes might also contribute to the shorter term relations between exposure to violence and aggressive behavior. However, two other processes have been widely dis- cussed as playing a role in short-term relations: priming and arousal processes. The observation of stimuli that have been paired in the past with observed violence or that inherently suggest violence (e.g., weapons) activates memory traces for aggressive scripts, schemas, and beliefs sufficiently to make their utilization more probable (Berkowitz, 1993; Josephson, 1987). A provocation that follows a priming stimulus is more likely to stimulate aggres- sion as a result of the priming. Although this effect is short-lived, the primed script, schema, or belief may have been acquired long ago and may have been acquired in a completely different context.

To the extent that observed violence (real world or media) arouses the observer, aggressive behavior may become more likely in the short run for two other possible reasons: excitation transfer (Zillmann, 1979, 1983) and general arousal (Berkowitz, 1993; Geen & O’Neal, 1969). First, a subsequent provocation may be perceived as more severe than it is because the emotional response stimulated by the observed violence is misattributed as being due to the provocation (Zillmann, 1979, 1983). Such excitation transfer could account for a more intense aggressive response in the short run. Alternatively, the increased general arousal stimulated by the observed violence may simply reach such a peak that the ability of inhibiting mechanisms such as normative beliefs to restrain ag- gression is reduced (Berkowitz, 1993).

It is important to recognize that these theoretical processes are not mutually exclusive. It is perfectly possible both that observa- tional learning, desensitization, priming, and excitation transfer all contribute to the stimulation of aggression by the observation of violence and that more aggressive children do like to watch more violence. However, there is one theory that is incompatible with all 202 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON of these processes. Catharsis theory (Feshbach & Singer, 1971; Fowles, 1999) would predict that violence viewing should be followed by reductions in aggression. Because the empirical evi- dence for any such negative relation is almost nonexistent (see Huesmann, Eron, Berkowitz, & Chaffee, 1991; Paik & Comstock, 1994), catharsis theory seems untenable at this time. Empirical Background In contrived experimental studies, children (both boys and girls) exposed to violent behavior on film or TV behave more aggres- sively immediately afterward (see reviews by Comstock, 1980; Geen, 1983, 1990; Geen & Thomas, 1986). The typical paradigm involves randomly selected children who are shown either a vio- lent or nonviolent short film and are observed afterward as they play with each other. The consistent finding is that children who see the violent film clip behave more aggressively toward each other (Bjorkqvist, 1985; Josephson, 1987) or toward surrogate objects (Bandura, Ross, & Ross, 1961, 1963a, 1963b). In these settings, exposure to violent film scenes clearly causes more ag- gressive behavior by children immediately afterward.

The demonstration of a relation between the observation of dramatic TV or film violence and the commission of aggressive behavior has not been limited to the laboratory. Evidence from field studies has clearly shown that the amount of TV and film violence a child is regularly watching is positively related to the child’s aggressiveness. Children who watch more violence on TV and in the movies behave more violently and express beliefs more accepting of aggressive behavior (see reviews by Andison, 1977; Chaffee, 1972; Comstock, 1980; Hearold, 1979; Huesmann, 1982; Huesmann & Miller, 1994; Paik & Comstock, 1994; Wood, Wong, & Chachere, 1991). Although the correlations are modest by the standards used in the measurement of intellectual abilities (aver- age .41 for experiments and .19 for field studies [Paik & Comstock, 1994]), they are highly replicable and are substantial by public health standards (see Rosenthal, 1986). For example, as a comparison, the correlation between cigarette smoking and lung cancer was .34 in Wynder and Graham’s (1950) classic study.

Moreover, the correlation between childhood exposure to media violence and childhood aggression is highly replicable even across researchers who disagree about the reasons (e.g., Huesmann, La- gerspetz, & Eron, 1984; Milavsky, Kessler, Stipp, & Rubens, 1982) and across countries (Huesmann & Eron, 1986).

Although these one-shot field studies showing a correlation between media-violence viewing and aggression suggest that the causal conclusions of the experimental studies may well generalize to the real world, longitudinal studies with children can test the plausibility of long-term predisposing effects more directly. In perhaps the first longitudinal study on this topic, initiated in 1960 on 856 youth in New York State, Eron, Huesmann, Lefkowitz, and Walder (1972) found that boys’early childhood viewing of vio- lence on TV was statistically related to their aggressive and anti- social behavior 10 years later (after graduating from high school) even after initial aggressiveness, social class, education, and other relevant variables were controlled (Lefkowitz, Eron, Walder, & Huesmann, 1977). A 22-year follow-up of these same boys re- vealed that their early aggression predicted later criminality at age 30 and that early violence viewing also was independently but weakly related to their adult criminality (Huesmann, 1986, 1995).A more representative longitudinal study was initiated by Hues- mann and his colleagues in 1977 (Huesmann & Eron, 1986; Huesmann, Lagerspetz, & Eron, 1984). This 3-year longitudinal study of children in five countries also revealed that the TV habits of children as young as first graders also predicted subsequent childhood aggression even after initial levels of aggression were controlled. In contrast to earlier longitudinal studies, this effect was obtained for both boys and girls even in countries without large amounts of violent programming, such as Israel, Finland, and Poland (Huesmann & Eron, 1986). In most countries, the more aggressive children also watched more TV, preferred more violent programs, identified more with aggressive characters, and per- ceived TV violence as more like real life than did the less aggres- sive children. The combination of extensive exposure to violence coupled with identification with aggressive characters was a par- ticularly potent predictor of subsequent aggression for many chil- dren. Still, there were differences among the countries. Although the synchronous correlations were positive in all countries, the longitudinal effect of violence viewing on aggression was not significant for girls in Finland or for all children in Australia. In Israel, there were significant effects for children living in a city but not for children raised on a kibbutz.

A few longitudinal studies have seemed to produce results at odds with the thesis that media-violence viewing causes aggres- sion, but closer inspection of most of these studies reveals that their results are not discrepant but simply not strongly supportive of the thesis. (For a review, see Huesmann & Miller, 1994.) For example, although the National Broadcasting Company’s longitu- dinal study of middle-childhood youth conducted in the 1970s (Milavsky et al., 1982) reported significant regression coefficients for only 2 out of the 15 critical tests of the causal theory for boys, an additional 10 were in the predicted direction. Furthermore, for girls, 3 out of the 15 critical tests were significant and an addi- tional 7 were in the predicted direction. The Current Study As the above review indicates, over the past several decades, the correlation between TV-violence viewing and childhood or ado- lescent aggression has been unambiguously demonstrated. It has also been clearly confirmed that in the short run, exposure to violence causes an increase in immediate aggressive behavior.

These effects have been obtained repeatedly for both boys and girls. The few completed longitudinal studies have also suggested that there is a long-term effect of early childhood exposure on aggression later in childhood, in the teen years, and, less strongly, into adulthood. However, these longer term effects have been found only for boys in the existing studies that were initiated in the 1960s.

The current study is a follow-up of the 1977 longitudinal study of 557 children growing up in the Chicago area that we described above and have reported on elsewhere (Huesmann & Eron, 1986; Huesmann, Lagerspetz, & Eron, 1984). Our aim was to investigate the long-term relations between viewing media violence in child- hood and young-adult aggressive behavior. The study was de- signed to provide data that could be used to compare the relative plausibility of the violence effects theories described earlier (ob- servational learning theory or desensitization theory) with the plausibility of the preference-for-violence and“third variable” 203 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR theories. However, the study was not designed to contrast obser- vational learning theory with desensitization theory. In addition, the examination of the“third variable”hypotheses in this study, as always, is limited by the actual third variables included in the study.

In the follow-up study, we tracked down as many of the original boys and girls in that U.S. study as we could find 15 years later when they were in their early 20s. We interviewed them, inter- viewed their spouses or friends, and collected data on them from state archives. In this article, we address four major questions with data from this follow-up: (a) To what extent does early childhood exposure to media violence predict young-adult aggression and violence? (b) Are there gender differences in the predictability? (c) Does the extent to which the child viewer identifies with the aggressive character or believes the plot is realistic affect the strength of the prediction? and (d) To what extent does any long-term relation seem to be due to more aggressive children simply liking to watch violence or seem to be due to some environmental, family, or personal“third variable”that stimulates both childhood violence viewing and childhood and adult aggression? Method Participants and Procedure The data we are analyzing in this article were collected on a sample of youth during two time periods in their lives—an original 2-year period when they were in the first and second or third and fourth grades and a 2-year follow-up period approximately 15 years later when they were in their early 20s. The longitudinal design is summarized in Table 1.

In the original childhood study, 557 first and third graders from public schools in Oak Park, Illinois, and two parochial schools in Chicago, Illinois, were tested and interviewed at least twice, in the spring of their first- and second-grade years (younger cohort) or in the spring of their third- and fourth-grade years (older cohort). These interviews occurred in 1977 and 1978. The children were interviewed in their classrooms, and peer-nomination measures were obtained about observed behaviors, in- cluding aggression. The children’s scores over the 2-year period were averaged to provide more accurate single estimates of their TV viewing habits and aggressive behaviors during that early period. In addition, most parents were interviewed once during this 2-year period, and data on theintellectual ability of the children were obtained from school records. The exact procedures for participant selection and testing are described in Huesmann and Eron (1986). Essentially all first and third graders in Oak Park and all first and third graders in the two Chicago schools formed the pool from which participants were recruited. We obtained active parental permission for 758 children, or about 76% of the eligible children, and collected the initial 2 years of data on 557, or about 73%, of the permission participants. These 557 children constitute the initial sample for the current 15-year follow-up study.

A subset of 505 of these children was reinterviewed in 1979 in the spring of the younger cohort’s third-grade year and the older cohort’s fifth-grade year. Their scores on peer-nominated aggression at that time served as a criterion measure for the longitudinal analyses of childhood effects re- ported in Huesmann and Eron (1986) and Huesmann, Lagerspetz, and Eron (1984). However, for the current 15-year longitudinal study, we searched for all 557 original participants.

In 1991, when the participants were between 20 and 22 years old, we began the follow-up process. We searched for public archival data on all of the original 557 participants, and we attempted to locate and recruit as many as possible for a follow-up interview. Through searches of public driver’s license records and criminal justice records we obtained archival data on 450 of the original 557 participants (81%). This is called the archival follow-up sample. Its gender composition and age composition are shown in Table 1.

The process of locating and recontacting the original participants for interviews was more complex. None of the participants had been contacted during the prior 15 years. We searched phone records, school records, and pursued social networks for over 2 years. Those participants who were living in the Chicago area were asked to come to the University of Illinois at Chicago to complete the interview. Participants sat at a computer alone in a small room and completed the questionnaire. In addition, they were asked to give the names of the three people who knew them best (excluding parents and siblings) and to rate those people on how well they knew them.

If the participant listed his or her spouse, we selected the spouse as the “other”person to be interviewed; otherwise, we selected the person with the highest rating. With the participant’s permission, we then interviewed this other person. Participants were paid $50 to complete the 3–4-hr interview, and the other persons were paid $30 for completing a 1–1.5-hr interview.

At the point when no additional participants could be located who were available to do the interview in person in Chicago, the questionnaire was converted into a combination phone and mail interview. In this version, participants were again contacted initially by phone and asked to partici- Table 1 Data Collection Summary Cohort and genderChildhood dataChildhood and adult archival dataChildhood, adult archival, and adult interview data Wave 1 2 Wave 1 23Wave 4 Wave 1 23Wave 4 1977–1978 1992–1995 1992–1995 Older cohort (ages in years) Ages 8–9 Age 23 Ages 8–93Age 23 Males (n) 125 105 71 Females (n) 133 109 83 Younger cohort (ages in years) Ages 6–7 Age 21 Ages 6–73Age 21 Males (n) 144 119 82 Females (n) 155 117 93 Total (N) 557 450 329 Resampling rate (%) 81 60 204 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON pate. If they agreed, a short 20–30-min interview was conducted on the phone, and the remainder of the interview was mailed to the person. During the phone interview, the participants were also asked to give the names of the three people who knew them best and who were not family members and to rate how well those persons knew them. As with the in-person participants, we contacted the“other”person the participant named, either a spouse or the person who had the highest rating if there was no spouse (if the participant gave us permission to contact the person), and asked if that person would be willing to participate in our study. If the person agreed, he or she was sent a mail interview. Participants and“other” persons who participated by phone and mail were paid the same amounts for their participation as those who completed the in-person interviews.

Using these procedures, between 1992 and 1995 we were able to reinterview a total of 398 participants—299 through personal interviews and 99 through phone and mail interviews. In addition, we obtained a total of 356“other”person interviews (121 of whom were spouses)—181 in person and 175 by mail. However, of the 398 reinterviewed participants, only 329 had provided complete data during the first two waves of the study (i.e., were part of the original sample of 557). Thus our final interview sample size, as shown in Table 1, was 329—153 men and 176 women—or about 60% of the original Wave 1 2 sample of 557 children.

Although 40% attrition over a 15-year lag between interviews is about typical for this type of longitudinal research, it requires that some attention be paid to the representativeness of the follow-up sample. Within the sample of 329, there were 299 participants for whom we had obtained a parent interview during the first two waves of the study (161 women and 138 men). There were also 235 who reported that they currently had spouses or“significant others,”and we interviewed 121 of these during our “other”person interviews (65 women and 56 men).

At the time of the follow-up, the reinterviewed participants ranged in age from 20 to 25 years, with a mean age of 22 years. The lag since their initial first- or third-grade interview ranged from 15 to 18 years. The reinter- viewed sample was split fairly evenly by gender, with 153 men and 176 women. Although the majority of the reinterviewed sample was Caucasian, about 7.5% belonged to minority groups (compared with 12.9% of the original sample). At the time of the interview, the participants’level of education ranged from having completed ninth grade to having completed a graduate or professional degree, with the average education level being “some college.”Finally, in terms of SES, the reinterviewed sample was somewhat skewed toward the higher end of the scale, as measured by the Hollingshead and Redlich (1958) status rating of the participant’s father’s most recent occupation, with 65.3% having high-status jobs, 26.9% having medium-status jobs, 5.8% having low-status jobs, and 2% having jobs with an unknown status.

One additional question of importance is how the resampled participants differed from those participants who could not be found after 15 years. We compared the two groups on their initial childhood aggression. The par- ticipants for whom we succeeded in locating archival data did not differ significantly from the other participants on childhood aggressiveness.

However, the participants whom we succeeded in reinterviewing were less aggressive than those who were lost. This is a typical pattern for longitu- dinal studies of antisocial and aggressive behavior. The originally more aggressive and antisocial participants are underrepresented in the follow-up interview. Nevertheless, the reinterview process did not reduce the range of aggression scores (from 0 to .82 in the reinterview sample; from 0 to .65 in the dropped sample). The reinterviewed sample also scored higher than those not reinterviewed on childhood intellectual ability (p .01) and on parents’education (p .001), which again suggests that the reinterviewed sample is a less“at risk”part of the original sample. Given these kinds of resampling differences, we must be somewhat judicious in not underesti- mating effects because of the loss of more aggressive participants. How- ever, the distribution is not likely to bias us toward detecting effects that are not there. Child Measures (1970s) The key original child measures are listed in Table 2 with their reliability coefficients. Details of these measures and the procedures for administer- ing them have been published in Huesmann, Lagerspetz, and Eron (1984) and Huesmann and Eron (1986) and are only summarized below. Test– retest reliabilities were based on the analysis of a subset of 93 children who were retested after 1 month.

Childhood TV-violence viewing.Each year children were presented with 8 lists of 10 TV programs and asked to mark their favorite program on each list and how often they watched it—“every time it’s on,”“a lot, but not always,”or“once in a while.”As described in Huesmann and Eron (1986), the 80 programs used were the most popular that year according to the Nielsen ratings for 6- to 11-year-old children. The programs were divided so that each list of 10 had several violent and several nonviolent programs and was balanced for Nielsen popularity and time shown. The amount of on-screen physical violence portrayed on each program was coded by two raters on a 5-point scale fromnot violenttovery violent. The raters were clinical psychology graduate students who were trained to follow a written criterion for what constituted violence (i.e.,“visually depicted, physical, interpersonal acts that were intended to harm”; see Huesmann & Eron, 1986). The interrater reliability was .75 (Huesmann & Eron, 1986). Some examples of shows rated asvery violentareStarsky and Hutch,The Six Million Dollar Man, andRoadrunnercartoons. A child’s overall violence score was computed by summing the violence scores for the favorite programs and weighting them by how often they were watched.

On the basis of regression analyses predicting concurrent aggression (see Huesmann & Eron, 1986), shows that were watched“every time it is on” were weighted by 10, and shows that were watched“a lot, but not always” were weighted by 1. Shows that were watched only“once in a while”were weighted 0 and did not contribute to the violence score.

Childhood identification with aggressive TV characters.As described in Huesmann and Eron (1986), children were asked how much they acted like or did things like variousadultaggressive characters, such as the Bionic Woman and the Six Million Dollar Man (2 a lot,1 a little,0 not at all). In each year, two aggressive male characters and two aggressive female characters were presented. The children answered questions only Table 2 Reliabilities of Key Measures MeasureCoefficient alpha for self-reportCoefficient alpha for peer/other-reportOne-month test–retest reliability Childhood measures Peer-nominated aggression—.97 .91 TV-violence viewing——.75 Identification with aggressive TV characters .71—.60 Perceived realism of TV violence .86—.74 Adult and parent measures Indirect aggression .66 .75— Verbal aggression .75 .85— Mild physical aggression .69 .78— General aggression .82 .84— Severe physical aggression .59 .53— Aggression toward spouse .78 .72— Aggressive personality .78—— Aggressive fantasy .80—— Rejection of child .62—— Orientation toward mobility .63—— 205 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR about characters they had seen. The average identification withadult aggressive TV characters was then calculated separately for the male and female characters. We used this“believed similarity”identification mea- sure rather than a“wishful”identification measure because, theoretically, believed similarity should produce stronger observational learning from a model (Bandura, 1969).

Childhood judgments of realism of TV violence.As described in Hues- mann and Eron (1986), children were asked to rate how realistic they judged various violent programs to be. They were given a list of 10 violent TV shows, including cartoons, and were asked,“How true do you think this program is in telling what life is really like?”The rating scale was as follows: 3 just like it is in real life,2 a little like it is in real life, and 1 not at all like it is in real life. The average“realism”score for the programs they had seen was then calculated. Again, our hypothesis was that observational learning would be greater if the viewer perceived the observed behavior as“like it is in real life.” Childhood aggressive behavior.Aggressive behavior in the first two waves of the study was measured with a modified version of the Peer Nominated Index of Aggression (Eron, Walder, & Lefkowitz, 1971; Hues- mann & Eron, 1986; Walder, Abelson, Eron, Banta & Laulicht, 1961).

Each child was asked to report which children in the class engaged in 10 different aggressive behaviors, such as“starts a fight over nothing”and “pushes and shoves children.”One indirect aggression item was included.

A child’s aggression score was then computed by adding up the number of times the child was named by his or her peers on all 10 items and dividing the sum by the number of students in the class doing the ratings. As Table 2 shows, this scale is extremely reliable, with an internal consistency of .97 and a 1-month test–retest reliability of .91. This scale has been used in a wide variety of studies in different countries with participants of different ages and has shown construct, concurrent, and predictive validity in those settings (Huesmann & Eron, 1986; Huesmann, Guerra, Eron, & Crawshaw, 1994).

Childhood intellectual ability.Intellectual ability was assessed for all children in the third grade with the California Achievement Test (Tiegs & Clark, 1970). The test was given as part of the regular testing program in the schools during the spring of the children’s third-grade year. For the younger cohort, a local test, the Oak Park Reading Readiness Test, was used. Because these scores were on different scales, they were each converted to standard scores within grade to represent one common mea- sure of intellectual ability relative to grade.

Parent Measures (1970s) Theparents’ educational levelswere recorded from self-reports, and an average of the mother’s and father’s levels was computed to be used as a measure of SES. Similarly, thefather’s occupationwas recorded and coded according to Warner, Meeker, and Eells’s (1960) scale for SES. The scale scores were then reversed so that a higher score indicated higher SES.

Parents’educational level was chosen to be the primary measure of SES because it took into account both the mother’s and the father’s status.

Father’s occupation was used as a cross-validation measure.

Three measures of self-reported parents’aggressiveness were obtained.

Aggressive personalitywas assessed by parents’average score on Scales 4 and 9 of the Minnesota Multiphasic Personality Inventory (Huesmann, Lefkowitz, & Eron, 1978; .78).Severe physical aggressionwas measured by three items about how many times in the last year the parents had“choked, punched, or beaten another adult,”had“slapped or kicked another adult,”or had“threatened or actually cut someone with a knife or threatened or shot at someone with a gun”(Huesmann, Eron, Lefkowitz, & Walder, 1984; .60).Fantasizing about aggressionwas assessed with an adult version of the Aggression scale of the Children’s Fantasy Inven- tory (Rosenfeld, Huesmann, Eron, & Torney-Purta, 1982; .80).

Four measures of parenting practices and attitudes during the past year were assessed:rejection of the child( .62),nurturance of the child(not appropriate for alpha),harsh punishment of the child(not appropriate foralpha), andmobility orientation(i. e., willingness to sacrifice to get ahead; .63). Most of these parent measures have been described in more detail elsewhere (Eron et al., 1971; Huesmann & Eron, 1986).

Finally, two measures of parent TV usage were collected in the parent interviews:parents’ TV-viewing frequency, as estimated by the parents, andparents’ TV-violence viewing,based on their self-reports of programs watched (see Huesmann & Eron, 1986, and description in the following section).

Adult Measures (1990s) Only a subset of all the adult measures administered were used in this article. They included a self-report measure of TV-violence viewing and multiple measures of aggressive and antisocial behavior assessed from different sources.

Adult TV-violence viewing.During the personal or phone interviews, participants were asked to report their three favorite regularly scheduled TV programs during the current year and how frequently they watched them. All three programs were then coded for their level of violence on a scale from 0 (no visible or off-camera violence)to4(high visible violence).

Again, the raters were psychology graduate students who were trained to rate shows on the basis of the same written criterion used in the first two waves of the study. Raters were instructed to rate the shows on the basis of the frequency of violence that was visually depicted, physical, intentional, and interpersonal and to rate only those programs that they had viewed themselves. They were instructed to ignore verbal violence, accidental violence, and violence by animals in nature programs. However, violence by cartoon characters was counted. Twenty-seven different raters evaluated the 1,272 programs and videos that participants listed. Raters were in- structed to rate only those programs that they had viewed enough to be familiar with the content. The mean rating from all the raters who had viewed the show (two being the minimum acceptable) was then used. Two approaches to intercoder reliability were used: interrater correlations and the average discrepancy from the mean rating. The interrater correlations ranged from .39 to .96 with a mean of .78 (using Fisher’sz). The average absolute discrepancy from the mean rating was .34, and no discrepancy was greater than .61. 1A participant’s adult TV-violence viewing score was computed as the average violence rating for the participant’s three favorite regularly scheduled TV programs, weighted by the frequency of viewing each show. 2This approach has been used in a variety of other studies both for parents’reports of a child’s TV-violence viewing (Eron et al., 1971, 1972) and for adult reports of their own TV-violence viewing (Eron et al., 1972; Lefkowitz, Eron, Walder, & Huesmann, 1977). TV-violence viewing measured in this way has been shown to correlate moderately with other TV-viewing and behavior variables in expected directions in several sam- ples from different countries (Eron et al., 1972; Huesmann & Eron, 1986; Huesmann, Lagerspetz, & Eron, 1984; Lefkowitz et al., 1977), which lends it construct validity.

Adult aggressive behavior.Data about the participants’current aggres- sive behavior were obtained from three sources: self-reports,“other”per- 1The mean rating was computed for thenraters who evaluated the program. Then the average absolute discrepancy between that mean and each individual rating was computed.

2The weights for the frequencies were similar to those used with the children’s data (see above and Huesmann & Eron, 1986), but they were modified to fit the 4-point scale used for adult responses. Thus, as for children, a show watched“almost every time it is on”was weighted 10, and a show watched“hardly ever”was weighted 0.“Watching sometimes”was treated as equivalent to“watching once in a while”for children and was weighted 1. The response“usually,”which was not available for children, was weighted at 5, the midpoint between“sometimes”and“almost every time it is on.” 206 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON son reports, and archival state data. The measures used in the participant interview and the“other”person interview are listed in the bottom part of Table 2 along with their internal consistency reliability coefficients. The scaling schemes used for most of these adult measures and the means and standard deviations for men and women are shown in Table 3.

For indirect, verbal, and mild physical aggression, participants were asked to report the frequency of engaging in various behaviors when they “had problems with or got very angry at another person”(Bjorkqvist, Osterman, & Kaukiainen, 1992). Theindirect aggressionscale included six items such as how often the participant responded by“taking the person’s things”and“trying to get others to dislike the person.”Theverbal aggres- sionscale included four items such as how often the participant responded by“calling the person names”and“belittling the person’s physical abilities or looks.”Themild physical aggressionscale included three items asking how often the participant responded by hitting, kicking, or shoving the person. Thegeneral aggressionscale (Huesmann & Eron, 1986) in- cluded 12 items based on the original peer-nomination scale for children but that were rewritten for adults. It included such questions as“How often do you give the finger to others?”“How often do you start a fight overnothing?”and“How often do you take other people’s things without asking?”The response scale for all of these frequency measures ranged from 0 (never)to4(very often). Thesevere physical aggressionscale (Huesmann, Eron, et al., 1984) included three items about how many times in the last year the participant had“choked, punched, or beaten another adult,”had“slapped or kicked another adult,”or had“threatened or actually cut someone with a knife or threatened or shot at someone with a gun.” The participants were also asked to report how frequently theyag- gressed against their spouse(or significant other) using Straus’s Conflict Tactics Scale (Straus, Gelles, & Steinmetz, 1980), and they were assessed onantisocial/aggressive personalitywith Scales F, 4, and 9 from the MMPI (Huesmann et al., 1978). The reliabilities for all of these rating scales are respectable, as is shown in Table 2. We also asked the partici- pants to report their frequency of traffic violations and of committing different kinds of specific crimes and whether they had ever been arrested or convicted. For this purpose we used questions from the National Youth Survey (Elliott, Dunford, & Huizinga, 1987). Table 3 Mean Scores on Major Aggression Variables for Each Gender VariableM SD p Women (n 176)Men (n 153) Self-reported Frequency of verbal aggression (from 0 never to 4 very often) 0.95 1.11 0.77ns Frequency of indirect aggression (from 0 never to 4 very often) 0.86 0.72 0.60 .04 Frequency of mild physical aggression (from 0 neverto 4 very often) 0.37 0.41 0.52ns Severe physical aggression (no. of acts in past year; from 0 to 27) 0.57 1.15 2.15 .02 Frequency of general aggression (from 0 never to 4 very often) 0.94 0.94 0.46ns Aggressive personality (sum oftscores on MMPI scales F, 4, and 9) 178.13 185.19 26 .02 Aggression toward spouse (no. of acts in past year; from 0 to 72) 1.92 1.20 4.28ns Number of“arrests”in past 5 years (range 0–30) 0.10 0.97 2.04 .001 Mean violence rating of“arrests”(from 0 nonviolentto 1.92 homicide) 0.01 0.06 0.21 .03 Number of criminal acts in past 5 years (range 0–111) 4.57 15.88 15.7 .001 Other-reported Frequency of verbal aggression (from 0 never to 4 very often) 1.05 1.04 0.88ns Frequency of indirect aggression (from 0 never to 4 very often) 0.88 0.67 0.61 .01 Frequency of mild physical aggression (from 0 neverto 4 very often) 0.20 0.32 0.46 .04 Severe physical aggression (no. of acts in past year; from 0 to 27) 0.25 0.61 1.35 .04 Frequency of general aggression (from 0 never to 4 very often) 0.83 0.82 0.48ns Aggression toward spouse (no. of acts in past year; from 0 to 72) 1.32 1.02 2.59ns Composite aggression .14 .06 .69 .01 Archival measures (n 226) (n 224) Proportion ever convicted of a crime .00 .05 .16 .001 Number of moving traffic violations 0.36 1.09 1.52 .001 Note.MMPI Minnesota Multiphasic Personality Inventory. 207 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR As described earlier, a close friend, spouse, or significant other of the participant was also interviewed and was asked to rate the participant’s frequency of engaging in aggressive behavior on the first five rating scales described above as well as to report on the participant’s specific criminal and antisocial acts. If the“other”person was the participant’s spouse or “significant other,”he or she was also asked to complete Straus’s Conflict Tactics Scale (Straus et al., 1980). This procedure has been used success- fully in past studies (Huesmann, Eron, et al., 1984) and is advantageous because it provides an alternative source of information in addition to self-report. The reliabilities for these“other”person rating-scale aggression measures are also shown in Table 2 and ranged from .53 to .85.

Archival Measures of Aggression Last, archival data, including criminal conviction records and moving traffic violation records were obtained for each participant from state records. We used the existence of an Illinois driver’s license record as the mechanism for defining the sample on whom the state might have criminal records. In other words, if someone with a driver’s license did not appear in the criminal conviction registry, he or she was coded as having no “official”convictions. For the archival, self-report, and other-report data on specific arrests or convictions, we also computed a“violence”score based on a system by Rossi, Bose, and Berk (1974). Rossi et al.’s system yields scores that range from 0 for a crime involving no violence (e.g., embez- zlement) to 1.92 for a highly violent crime (e.g., homicide). To get driving violation data, we searched for traffic records in Illinois and in other states that would give us access. We assigned participants a score of 0 for moving traffic violations if we obtained access to their records in at least one state and they had no violations. Results Composite Aggression Score One of the first questions to be resolved was whether the multiple adult interview measures of adult aggression could be combined into a single measure of aggressive behavior that could be used in the major analyses. A key question in constructing such a composite was whether to measure aggression differently for men and women. Over the past decade, substantial evidence has accumulated suggesting that women are more likely to engage in indirect forms of aggression, men are more likely to engage in direct physical aggression, and both genders are about equally likely to engage in verbal aggression (Bjorkqvist, Lagerspetz, & Kaukiainen, 1992; Lagerspetz & Bjorkqvist, 1992; Lagerspetz, Bjorkqvist, & Peltonen, 1988). 3Because we wished to compare relations in men and women, we did not want to construct different measures of aggression for the two genders, but we also did not want to construct one measure that was biased toward assessing aggression in men. Our solution was to construct a composite measure from multiple indicators of aggression that assessed both direct and indirect forms of aggression.

We constructed a structural model, displayed in Figure 1, for such a composite from our 11 self-report and other-report rating scale measures of aggression and estimated its parameters from the correlations in the sample of our follow-up participants who had complete data on these measures (N 325). As shown in Figure 1, the structural measurement model fits the adult aggression data well, 2(26,N 325) 28.9,p .32, RMSE [root mean square error] .026, and gives weight to both typically male and typi- cally female behaviors. From this model, we derived a factor score regression equation that was used to estimate the latent aggressionscore for each participant with complete data. For participants who had only self-reports or only“other”person reports, we derived a regression prediction of the composite from only those elements they did have.

The composite aggression score within the sample of 329 par- ticipants has a mean of 0 with a standard deviation of 0.7. How- ever, the distribution is substantially positively skewed (1.9), as is typical for scales of aggression, with scores ranging from a min- imum of 1 to a maximum of 4. Thus, the median score is 0.25, and the 25th and 75th percentiles are 0.5 and 0.23, respectively.

In Table 3, the mean scores on the composite, its major com- ponents, and some other measures of aggressive behavior are displayed for male and female participants along with the scale for each measure. One can see that men scored higher on most questions about serious physical aggression and criminality, but women scored higher on indirect aggression. Men and women scored about equally on verbal aggression, general aggression, and aggression toward spouses no matter who reported it. These results are consistent with previous research on gender differences in aggression and add construct validity to the composite measure.

Overall, on composite aggression, men scored slightly but signif- icantly higher than women.

The composite measure of aggression correlated significantly with independent archival indices of aggression, confirming its validity. Archival state records indicated that both men and women who scored higher on our composite measure of aggression com- mitted more moving traffic violations (r .29,N 152,p .001 andr .14,N 169,p .07, respectively), whereas men who scored higher had also been convicted of more crimes (r .18, N 150,p .03). The composite measure also correlated with other-reported and self-reported indices of specific antisocial and violent acts. Both men and women who scored higher on the composite were more likely to have assaulted their spouses ac- cording to the reports of their spouses (r .71,N 56,p .001 andr .65,N 65,p .001, respectively). In addition, those men and women who scored higher on our composite measure of aggression were more likely themselves to report having engaged in criminal behavior (rs .33,Ns 150 and 174,ps .001, respectively). The more aggressive men also reported having been arrested more times (r .49,N 153,p .001).

In summary, the composite measure seems to be a valid measure of aggressive adult behavior for both men and women. It encom- passes a wide variety of different kinds of aggressive behavior, and although highly aggressive men and women may score high on different dimensions, both will score high on the composite. Con- sequently, the composite measure was the adult criterion measure of aggression used in all subsequent analyses. Correlations Between Childhood TV-Violence Viewing and Adult Composite Aggression In Table 4, the correlations between adult aggression and the early TV-viewing variables are shown. One can see that for both 3Crick and Grotpeter (1995), with minor changes, renamed the indirect aggression construct developed by Lagerspetz and Bjorkqvist (1992) as “relational aggression.”However, we prefer to use the original label, which seems to reflect the kind of aggression more accurately. 208 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON male and female participants, childhood TV-violence viewing correlates significantly with the composite measure of adult ag- gression 15 years later. In addition, childhood perceptions that TV violence reflects real life and childhood identification with same-sex aggressive TV characters significantly correlate with adult aggression 15 years later. The table also shows the correlations between the childhood TV measures and the subscales (see Figure 1) measuring physical aggression and indirect aggression. As Table 4 Correlations Between Childhood TV Violence Measures and Adult Aggression 15 Years Later Child TV measuresAdult composite aggressionAdult physical aggressionAdult indirect aggression Men Women Men Women Men Women TV-violence viewing .21** .19** .17* .15* .03 .20** Perceived realism of TV violence .22** .25*** .14†.14†.05 .28*** Identification with aggressive female characters .15†.23** .05 .09 .01 .19* Identification with aggressive male characters .29*** .22** .14†.12 .05 .22** Note.For men,n 153; for women,n 176. Physical aggression is defined in this table as the average of the self- and other-rated severe and mild physical aggression scales, whereas indirect aggression is the average of self- and other-rated indirect aggression scales. Also, all of the correlations with composite aggression remain significant even if the Minnesota Multiphasic Personality Inventory aggressive personality scale is removed from the composite score and only behavioral measures are used.

†p .10. *p .05. **p .01. ***p .001. Figure 1.A structural model showing the measurement model for the composite aggression score derived from 11 measures of adult aggressive behavior. Nonsignificant path coefficients are not shown. MMPI Minnesota Multiphasic Personality Inventory; RMSE root mean square error. 209 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR expected, early TV-violence viewing correlates with physical ag- gression for both men and women but with indirect aggression only for women. Furthermore, identifying with aggressive female TV characters as a child correlates with a woman’s indirect ag- gression but not with her physical aggression. Given that female characters on TV in the 1970s tended to display more indirect aggression than direct aggression, this finding is also not surprising.

In Figure 2, the meaning of these correlations with composite aggression is illustrated with a series of bar graphs. For each childhood TV variable, the participants were partitioned into three categories: those scoring in the upper 20% on the TV variable, those scoring in the middle 60%, and those scoring in the lower 20%. 4Then the mean aggression score was plotted for the partic- ipants in each category. The figures reveal that in each case the upper 20% of participants on the childhood TV measure scored substantially higher on aggression 15 years later. Female partici- pants in the high childhood TV-violence group, the high childhood TV-realism group, and the high childhood identification with same-sex aggressive character group all scored significantly higher on adult aggression 15 years later. The same was true for male participants in the high groups. One can conclude from these figures that the correlations between childhood TV-violence view- ing and adult aggression to a great extent result from the higher aggressive behavior of the adults who were the highest violence viewers as children. Relations to Specific Aggressive Behaviors How do these longitudinal relations between early TV-violence viewing and adult composite aggression translate into relations with specific aggressive behaviors? Given the overall significance of the effects on the composite aggression score, examining effects on specific items indicating the aggressive behaviors of most social concern seems justified. Therefore, we compared the par- ticipants who were in the high TV-violence viewing group during childhood (the upper 20%) with all other participants on a number of very specific serious aggressive behaviors in the categories of spouse abuse, physical aggression, criminal activities, and driving violations. Because most of these specific antisocial or aggressive behaviors occur with very low frequency, we classified the par- ticipants into those who reported engaging in the behavior at least once during the past year and those who said they had not done so during the past year. In Table 5, the results for eight specific behaviors are displayed.

One can see that men who were high TV-violence viewers as children were significantly more likely to have pushed, grabbed, or shoved their spouses, to have responded to an insult by shoving a person, to have been convicted of a crime (according to state records), and to have committed a moving traffic violation (ac- cording to state records). For example, men who were high TV- violence viewers in childhood were convicted of crimes at over three times the rate of other men. Women who were high TV- violence viewers as children were more likely to have thrown something at their spouses, to have responded to someone who made them mad by shoving, punching, beating, or choking the person, to have committed some type of criminal act, and to have committed a moving traffic violation. For example, women who were high TV-violence viewers as children reported havingpunched, beaten, or choked another adult at over four times the rate of other women. Of course, there are many other specific behaviors on which the high and low TV-violence viewers did not differ significantly. However, there was not a single aggressive behavior on which high TV-violence viewers scored significantly lower than medium and low TV-violence viewers. Predicting Adult Aggression From Childhood TV-Violence Viewing These correlations and frequencies demonstrate clearly that childhood exposure to media violence is related to adult aggres- sion. To gain a better understanding of the meaning of these longitudinal relations, we computed multiple regressions in which the relation of early TV-violence viewing to later aggression was examined after controlling for the effects of early aggressive behavior. Essentially, these regressions examined the contribution of early TV-violence viewing to“change”in aggression. 5The regressions for girls are shown in Table 6, and the regressions for boys in Table 7.

The first regression in each table shows how adult aggression at the time of our follow-up can be predicted from childhood aggres- sion and cohort (i.e., age at time of the childhood testing). Each subsequent regression equation in these tables examines whether a childhood TV-viewing variable can improve the prediction of adult aggression over what was predicted from early aggression.

As Regression 2 in these tables shows, for both male and female participants, early exposure to TV violence significantly predicts adult aggression even when childhood aggression is partialed out of the relation. Moreover, as Regressions 3 and 4 illustrate, the same can be said for identification with aggressive TV characters and perception that TV violence is realistic. Even after the effects of early aggression in predicting later aggression are partialed out, each of these TV-viewing variables significantly predicts aggres- sion when the participants are in their early 20s. For both male and female participants, early TV-violence viewing, identification with same-sex aggressive TV characters, and the perception that TV violence is realistic all predict increases in aggressive behavior beyond what one would expect from participants’early aggres- siveness. However, for both genders the additive effect of all three predictors is not much greater than the individual contributions (R 2 .102 for females;R 2 .133 for males). 4Twenty percent was chosen for the size of the extreme groups because any smaller percentage would have produced ann 25 for some groups and reduced power too much. If the mean score for the variables to be plotted differed significantly between genders, we used different cut points for male and female participants; otherwise we used the same. The exact cut point selected was the most extreme score that placed at least 20% in the extreme group.

5Analysis of change scores can be viewed as a special case of“analysis of posttest with pretest as covariate”in which the regression coefficient for the pretest is fixed at 1. By subtracting out from the adult aggression score the maximal prediction from the pretest instead of the simple pretest score, one can interpret the coefficient for early TV-violence viewing as repre- senting the contribution of early TV-violence viewing to the change in adult aggression from what would have been expected on the basis of childhood aggression. 210 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON Interactive Effects of the TV-Viewing Scales on Later Aggression We next examined whether the identification with same-sex aggressive TV characters and the perception that TV violence is realistic not only had main effects on aggression but exacerbated the effects of viewing TV violence. We had found such an inter- active effect for boys during childhood in the early waves of this study (Huesmann & Eron, 1986; Huesmann, Lagerspetz, & Eron, 1984). To test for such effects, we expanded the regressions in Tables 6 and 7 that predicted adult aggression to include the interactive effect of childhood TV-violence viewing with identifi- cation with same-sex aggressive characters and with the perception that TV violence is realistic. For female participants, we found no significant interactive effects. For male partcipants, however, we found a significant effect for both interactions. As shown in Table 8, identifying with aggressive male TV characters and perceiving that TV violence is realistic each exacerbated the ef- fects of viewing TV violence on later aggression. Specifically, post hoc decomposition of the interaction showed that for a boy who scored 1SDabove the mean on identification with aggressive characters, the standardized partial slope of adult aggression on his childhood TV-violence viewing was .463. However, for a boy who scored at the mean on identification, the slope was only .185. The corresponding slopes for boys high and at the mean on perception of TV violence as realistic were .392 and .217, respectively. In other words, for boys, both identification with aggressive male characters and perception that TV violence is realistic significantly exacerbate the relation between childhood TV-violence viewing and adult aggression.

Testing Cross-Lagged Effects From Childhood to Adulthood The analyses presented thus far suggest that childhood TV- violence viewing is predictive of serious adult aggression and violence for both males and females even after controlling for childhood aggressiveness. The effect is exacerbated for boys who identify with aggressive males in the programs they watch. These results support the hypothesis that the causal effects of media violence that have been demonstrated in the laboratory extend into real life from childhood to adulthood. However, one might wonder if aggressiveness might be stimulating violence viewing as much as violence viewing is stimulating aggression.

To test this possibility, we constructed longitudinal structural models for males and females in which we simultaneously pre- Figure 2.Three bar graphs illustrating the relation between childhood TV habits and adult aggression for 153 males and 176 females. The y-axis variable in these graphs, composite aggression, is positively skewed with a mean of 0, a median of 0.25, and a standard deviation of 0.7. The high, medium (med), and low groups on the x-axis represent the upper 20%, the middle 60%, and the lower 20%, respectively, on the TV measure. The exact cut points were the highest and lowest scores that gave at least 20% in the extreme group. For TV-violence viewing, on which males and females scored significantly differently, the cut points were chosen sepa- rately for each gender. For the other two TV-viewing variables, the mean scores did not differ by gender and the same cut points were used for each gender. The women who were high TV-violence viewers in childhood scored significantly higher on adult aggression than did other women, t(173) 2.81,p .01, and the men who were high viewers of TV violence in childhood scored higher than did other men on adult aggression, t(150) 2.80,p .01. Similarly, both girls and boys who thought violent TV shows were“just like it is in real life”(Child Percep of Realism of TV Viol) grew up to behave more aggressively as adults:t(173) 3.43,p .001 for girls, andt(150) 3.62,p .001 for boys. Finally, both girls and boys who identified strongly with same-sex aggressive TV characters (Child Ident w Same Sex Agg TV Character) were significantly more aggressive as adults:t(162) 2.83,p .01 for girls, andt(144) 3.97, p .001 for boys. Percep perception; Viol violence; Ident identification; w with; Agg aggression. 211 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR dicted adult aggression from childhood TV-violence viewing and adult TV-violence viewing from childhood aggression. The pa- rameters of these structural models are based not just on the longitudinal correlations between aggression and TV-violence viewing but also on the contemporaneous correlations in childhood and adulthood. In Table 9 we present the complete set of observed correlations for males and females on which the models are based.

The significant correlations in Table 9 between childhood TV- violence viewing and childhood aggression for both male and female participants replicate those reported previously for this study with the complete Wave 1 2 sample (Huesmann & Eron, 1986). For both male and female participants, the correlations between childhood TV-violence viewing and adult aggression are also significant (as reported earlier in Table 4). However, only for women does adult TV-violence viewing correlate with adult ag- gression (r .23,p .01). Correspondingly, women’s adult TV-violence viewing is predicted significantly by their childhood TV-violence viewing (r .16,p .05), whereas men’s adultTV-violence viewing is not predicted from their childhood TV- violence viewing. For the developmental theoretical perspectives that emphasize early childhood as the important period for obser- vational learning, the lack of a correlation between adult TV- violence viewing and adult aggression is not unexpected and has been the frequent finding when the criterion measure of aggression is actual behavior (Bushman & Huesmann, 2001; Comstock & Paik, 1994; Eron et al., 1972). The more unusual finding here is that there is a significant relation for women. The other somewhat surprising finding is the lack of significant negative correlations between parents’social status, as measured by parents’education level, and the aggression and TV-violence viewing of their chil- dren. When father’s occupation was substituted as the indicator of SES, significant correlations did appear between SES and male participants’childhood aggression (r .19,p .05) and their later adult TV-violence viewing (r .19,p .05); however, for female participants, the previously significant correlations between SES and adult aggression and TV-violence viewing disappeared. Table 5 Differences in Frequency (%) of Spouse Abuse, Serious Physical Aggression, or Illegal Behavior“At Least Once”in Past 12 Months for High Childhood TV-Violence Viewers Compared With Other Viewers Adult aggressive behaviorMales Females High TV-violence viewersOther viewersChi-square significanceHigh violence viewersOther viewersChi-square significance Spouse abuse Pushed, grabbed, or shoved your spouse 41.7 22.2p .05 34.6 21.2ns Threw something at your spouse 20.8 14.8ns38.5 16.5p .02 Serious physical aggression Responded by shoving a person 68.8 50.4p .05 68.6 43.2p .01 Punched, beat, or choked another adult 21.9 16.9ns17.1 3.6p .01 Criminal behavior Self-reported any crime in last year 62.5 53.4ns48.6 25.9p .01 State-reported convictions 10.7 3.1p .03 00.0 00.0ns Driving behavior Self-reported moving traffic violations 87.5 76.3ns80.0 57.6p .01 State-reported moving traffic violations 60.0 39.4p .01 28.9 28.4ns Note.Chi-square significance test is Fisher’s exact one-sided test. Table 6 Predicting Adult Composite Aggression From Childhood TV Variables: Females (n 165) PredictorStandardized regression coefficients Regression 1 Regression 2 Regression 3 Regression 4 Cohort .156† .163* .093 .075 Childhood aggression .168* .125 .125 .133† TV-violence viewing .170* Identification with same-sex aggressive TV characters .188* Perception that TV violence is realistic.233** R 2increase over Regression 1—.027* .031* .048** R2 .039 .066 .070 .087 Note. R 2for five-variable model .102.

†p .10. *p .05. **p .01. 212 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON The structural models estimated from the TV-violence viewing and aggression data from Table 9 are shown in the upper panels of Figure 3 for male participants and of Figure 4 for female partici- pants. In these models, one-way paths are hypothesized only when the predictor is measured in time before the criterion. The model for male participants is an excellent fit, with a nonsignificant chi-square statistic and a low RMSE, 2(2,N 151) 0.018,p .99, RMSE .003. The model for female participants is adequate but not as good a fit, 2(2,N 174) 6.45,p .04, RMSE .047. However, for both genders, the path coefficients from child- hood TV-violence viewing to adult aggression are positive, statis- tically significant, and about twice the magnitude of the non- significant path coefficients from childhood aggression to adult TV-violence viewing. Although the positive, nontrivial path coef- ficients from early aggression to adult TV-violence viewing sug- gest that more aggressive children may have some tendency to turn to watching more TV violence, the much bigger effect implied bythese models is that early childhood exposure to TV violence stimulates increasesin young-adult aggressive behavior in both men and women.

An important issue in evaluating such structural models con- cerns the possibility that a construct is not measured reliably and therefore cannot predict or be predicted well by other constructs.

The lack of correlations for men between early TV-violence view- ing and adult TV-violence viewing and between adult TV-violence viewing and adult aggression generates some concern about the adult TV-violence measure along this line. However, adult TV- violence viewing was found to have substantial 1-month test–retest reliability in a pilot test (r .70), and it did correlate with a number of other variables that are not being investigated in this article (e.g.,r .16,p .01 with normative beliefs approving of aggression). In addition, we found the same pattern of correlations for adult TV-violence viewing by men in an earlier study (Eron et al., 1972). Therefore, these concerns can be somewhat alleviated. Controlling for SES and Intellectual Ability in the Longitudinal Models One might ask the extent to which the relations in these models might be explained by demographic characteristics such as the educational level or SES of the family. Lower income and lower educational status have generally been found to correlate with more exposure both to TV in general and to media violence in particular (Comstock & Paik, 1991; Eron et al., 1972; Huesmann & Eron, 1986). They also often correlate with greater risk for violent and aggressive behavior (Berkowitz, 1993). Similarly, a child’s intellectual ability might also be hypothesized to account for some of the longitudinal relation between childhood exposure to media violence and adult aggression. Scores on IQ and achieve- ment tests are known to be negatively correlated with both TV- violence viewing and aggression (Comstock & Paik, 1991; Hues- mann & Eron, 1986; Huesmann, Eron, & Yarmel, 1987). Some studies have suggested that observed longitudinal relations be- tween TV-violence viewing and aggression might be completely explained by these factors (e.g., Milavsky et al., 1982; Weigman, Kuttschreuter, & Baarda, 1986).

The models in the bottom panels of Figures 3 and 4 test this hypothesis. The parents’education and the child’s intellectual Table 7 Predicting Adult Composite Aggression From Childhood TV Variables: Males (n 147) PredictorStandardized regression coefficients Regression 1 Regression 2 Regression 3 Regression 4 Cohort .097 .096 .055 .054 Childhood aggression .200* .161†.147†.164† TV-violence viewing .226** Identification with same-sex aggressive TV characters .251** Perception that TV violence is realistic.190* R 2increase over Regression 1—.050** .059** .033* R2 .043 .093 .102 .076 Note. R 2for five-variable model .133.

†p .10. *p .05. **p .01. Table 8 Predicting Adult Composite Aggression From Childhood TV Variables: Males (n 147) PredictorStandardized regression coefficients Regression 5 Regression 6 Initial cohort .122 .072 Childhood aggression .104 .126 TV-violence viewing (TV) .185* .217** Identification with same-sex aggressive TV characters (Identification) .113 TV Identification .278*** Perception that TV violence is realistic (Perception) .175* TV Perception .263*** R 2increase over Regression 2 from Table 7 .102*** .087*** R 2 .195 .180 Note.Variables in interactions were standardized (centered) within gen- der before combining.

*p .05. **p .01. ***p .001. 213 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR ability as assessed during Waves 1 and 2 of the study were introduced into the longitudinal structural models as control vari- ables with paths to every other variable. The resulting models for male and female participants both provide a good fit, as indicated by the nonsignificant chi-square statistics and the low RMSEs. 6 Furthermore, for both male and female participants, the paths from childhood exposure to TV violence to adult aggression remain significant, whereas the paths from early aggression to adult TV- violence viewing are not significant. For males, the difference between the two paths is larger with the control variables intro- duced; for females, the difference is smaller. In both cases, the most plausible conclusion is that childhood exposure to TV vio- lence is stimulating an increase in adult aggression regardless of the initial aggressiveness of the child, the intellectual ability of the child, or the educational background of the parents. For females, there may well also be a stimulating effect of childhood aggression on later TV-violence viewing, but it is a weaker effect. For males, there is little indication of such an effect. Finally, when father’s occupation was again substituted for parents’education as a mea- sure of SES, the key paths in the structural models did not change. 7 Controlling for Parent Aggression, Parent TV Habits, and Parenting Practices and Attitudes As discussed in the introduction, it has been established in the past that a variety of parent behaviors and practices are related both to a child’s TV-viewing habits and to a child’s aggression (Comstock & Paik, 1991; Huesmann & Eron, 1986). The question is whether these parent factors can account for the longitudinal relation between childhood TV-violence viewing and young-adult aggression. Three measures of parent aggression, four measures of parent child-rearing practices, and two measures of parents’TV habits were selected for investigation in this role because of their theoretical significance and likelihood of influence. Each of the nine measures was introduced into the structural equation models shown at the bottom of Figures 3 and 4, with paths of influenceleading to the childhood and adult TV-violence variables and the childhood and adult aggression measures. Table 10 summarizes the key path coefficients for the resulting nine new models for males and females.

In every case the path from early TV-violence viewing to adult aggression remained significant. It was never reduced substantially by the introduction of a parent variable. The parent factors did correlate with the child’s aggression and TV habits in some cases, but this did not account for the longitudinal relations. For example, for female participants, parent rejection of the child in the 1970s correlated .23 (p .003) with childhood aggression and .20 (p .009) with childhood TV-violence viewing; however, it did not reduce the path from childhood TV-violence viewing to adult aggression at all. For male participants, parental rejection of the child in the 1970s correlated .28 (p .001) with childhood aggression, .22 (p .009) with childhood TV-violence viewing, and .14 (p .108) with adult aggression; however, its introduction reduced the path coefficient only from .19 to .18. The implication is not that parent factors are unimportant. They certainly play a role in influencing both aggression and TV habits. However, these results imply that the parent factors that we measured probably do not account by themselves for the longitudinal relations between 6Parents’education and the childhood intellectual ability score were also introduced into the regression equations shown in Tables 7, 8, and 9.

For male participants, this introduction did not change any of the regression parameters in any table more than 4%. For female participants, the three positive regression coefficients for early TV viewing on later aggression were reduced slightly (17% on average) when parents’education was introduced, but all three increased when intellectual ability was introduced.

7For male participants, the path from early TV-violence viewing to later aggression became .196 (p .05), whereas the path from early aggression to later TV-violence viewing remained nonsignificant at .024. For female participants, the path from early TV-violence viewing to later aggression became .176 (p .05), whereas the path from early aggression to later TV-violence viewing remained nonsignificant at .146. Table 9 Longitudinal and Contemporaneous Correlations for Estimating Structural Models Relating TV-Violence Viewing and Aggression Measure 1 2 3 4 5 6 Males (n 151) 1. Childhood peer-nominated aggression—.18* .18* .08 .05 .33*** 2. Childhood TV-violence viewing—.21** .02 .12 .25** 3. Adult composite aggression—.00 .06 .07 4. Adult TV-violence viewing— .08 .13 5. Parents’educational status—.32** 6. Child’s intellectual ability— Females (n 174) 1. Childhood peer-nominated aggression—.28*** .13†.10 .00 .31*** 2. Childhood TV-violence viewing—.19** .16* .11 .19* 3. Adult composite aggression—.23** .17* .02 4. Adult TV-violence viewing— .23** .07 5. Parents’educational status—.32*** 6. Child’s intellectual ability— †p .10. *p .05. **p .01. ***p .001. 214 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON exposure to media violence inchildhood and young-adult ag- gression. Of course, there may be other unmeasured parent factors that do play a bigger role in the relation. Also, although the reliability of the measures of some parenting variables was high (e.g., .7 to .8) for most parental aggression measures and TV viewing, it was more modest for others (e.g., .6 to .7 for rejection and mobility orientation), and for some nonscale mea- sures it was uncertain (e.g., punishment and nurturance). More reliable measures of these constructs might have accounted for more of the effects. Discussion In this 15-year longitudinal study of 329 youth, we found that children’s TV-violence viewing between ages 6 and 9, children’s identification with aggressive same-sex TV characters, and chil- dren’s perceptions that TV violence is realistic were significantly correlated with their adult aggression. This was true for both male and female participants. It was true for physical aggression for both genders and for indirect aggression for women. Regression analyses that partialed out the effects of early aggression showed Figure 3.Two structural models showing the effects for males of TV-violence viewing and aggression in childhood on TV-violence viewing and aggression in adulthood. The models are the same except that parents’ education and children’s intellectual ability (Early Achievement) are included in the model in the lower panel.

Dashed lines indicate nonsignificant paths. 215 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR that these childhood TV habits were not just correlated with aggression but predicted increases or decreases in aggressive be- havior. For both male and female participants, more childhood exposure to TV violence, greater childhood identification with same-sex aggressive TV characters, and a stronger childhood belief that violent shows tell about life“just like it is”predicted more adult aggression regardless of how aggressive participants were as children.

The longitudinal relations primarily reflected the adult behavior of the highest TV-violence viewing children. The upper 20% of boys and girls on any of the three child TV-viewing variablesscored significantly higher on aggression as adults than did the rest of the participants. Furthermore, as adults they displayed a higher frequency of very serious antisocial and violent behaviors.

A longitudinal structural modeling analysis of the directionality of the effects suggested that it is more plausible that exposure to TV violence increases aggression than that aggression increases TV-violence viewing. These structural modeling analyses also demonstrated that the effects were not simply a consequence of lower SES children or less intellectually able children both watch- ing more violence and being more at risk for aggressive and violent behavior. The structural models show that for both boys Figure 4.Two structural models showing the effects for females of TV-violence viewing and aggression in childhood on TV-violence viewing and aggression in adulthood. The models are the same except that parents’ education and children’s intellectual ability (Early Achievement) are included in the model in the lower panel.

Dashed lines indicate nonsignificant paths. 216 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON and girls, habitual early exposure to TV violence is predictive of more aggression by them later in life independent of their own initial childhood aggression, their own intellectual capabilities, their social status as measured by their parents’education or their fathers’occupations, their parents’aggressiveness, their parents’ mobility orientation, their parents’TV viewing habits (including violence viewing), and their parents’rejection, nurturance, and punishment of them in childhood. Furthermore, the structural models suggest that being aggressive in early childhood has no effect on increasing males’exposure to media violence as adults and only a small effect for females.

Nevertheless, these results should not lead one to conclude that children’s aggressiveness plays no role in their TV and film preferences. The paths from childhood aggression to adult TV- violence viewing were not significant, but they were all positive, and adult TV-violence viewing is not predicted much better by almost any childhood variable. Furthermore, in the childhood waves of this study, the comparable paths from aggression in one year to TV-violence viewing in the next were positive and signif- icant (Huesmann & Eron, 1986, p. 61). These results certainly are consistent with“justification theory”—that more aggressive chil- dren are more likely to watch media violence because it makes their own behavior seem normal. Their subsequent viewing of violence then increases their aggressive scripts, schemas, and beliefs through observational learning and makes subsequent ag- gression even more likely.

It is particularly interesting that we found longitudinal results that were of about the same magnitude for female as for male participants. In our 1960–1970 (Eron et al., 1972) and 1960–1982 (Huesmann, 1986, 1995) studies of New York children, longitu- dinal effects were found only for boys. One possibility is that the change in social norms for appropriate female behavior that oc- curred with the feminist movement of the late 1960s and 1970s has disinhibited female aggression. In addition, the increase in aggres- sive female models in movies and TV might have engendered a stronger observational learning effect. The combination of these two factors may have led to an increase in the size of the effect for female participants, making detection easier. It is not that girls were not subject to the observational effect in earlier years. Indeed,they were, as the laboratory experiments showed (Paik & Com- stock, 1994). Rather, it is that their use of the learned aggressive behaviors or aggressive scripts was inhibited by their existing normative beliefs about appropriate female roles. This explanation is consistent with the information-processing perspectives on learning aggressive behavior that Huesmann (1986, 1988, 1998) and Dodge (1980) have offered. According to Huesmann’s model, learned scripts for aggressive behavior are not followed if they violate individuals’normative beliefs about what is appropriate for them.

Three notable gender differences in the results were found. First, early TV-violence viewing correlated with adult physical aggres- sion for both male and female participants but correlated with adult indirect aggression only for female participants. Lagerspetz et al.

(1988) pointed out that indirect aggression is more characteristic of females and more acceptable for them in most societies. The social-cognitive observational learning model suggests that nor- mative beliefs about aggression, hostile biases about the world, and aggressive social scripts are all learned from observing vio- lence. Female participants did not need to have observed indirect aggression to acquire it from observing violence. They only needed to have acquired beliefs more accepting of aggression. This reasoning also suggests that the lack of a finding of a relation between exposure to media violence and female aggression in earlier longitudinal studies may have been due to the failure to measure indirect aggression sufficiently in those studies.

Second, although identification with same-sex aggressive TV characters and the perception that violent TV shows tell about life “like it is”predicted adult aggression for both genders, these factors exacerbated the effect of TV-violence viewing only for male participants. Boys who viewed TV violenceandidentified with male aggressive TV characters or perceived TV violence as true to life were most at risk for adult aggression. The same gender difference had been found earlier for the childhood data relating TV viewing to subsequent aggression (Huesmann & Eron, 1986)—identification was a moderator for boys but not for girls.

Why would this exacerbating effect not occur for girls? Theoret- ically, it is difficult to believe that identification and perception of realism do not enhance observational learning in girls as well as Table 10 Structural Models’Cross-Lagged Longitudinal Path Coefficients Between Childhood Aggression and Adult TV-Violence Viewing and Between Childhood TV-Violence Viewing and Adult Aggression While Controlling for Parent Factors During Childhood Parent variable controlledMales Females Child TV violence3adult aggressionzChild aggression3adult TV violencezChild TV violence3adult aggressionzChild aggression3adult TV violencez Parents’physical aggression .20 2.41* .06 0.62 .17 2.22* .09 1.13 Parents’aggressive personality .19 2.36* .07 0.83 .17 2.17* .12 1.48 Parents’aggressive fantasy .21 2.60** .04 0.46 .16 2.14* .11 1.33 Parents’nurturance of child .19 2.40* .05 0.56 .17 2.17* .10 1.24 Parents’punishment of child .20 2.41* .06 0.60 .17 2.29* .14 1.68† Parents’rejection of child .18 2.12* .05 0.56 .17 2.26* .12 1.40 Parents’mobility orientation .19 2.32* .04 0.50 .19 2.43* .12 1.45 Parents’TV-viewing frequency .20 2.41* .05 0.57 .15 2.03* .12 1.46 Parents’TV-violence viewing .19 2.37* .05 0.62 .16 2.16* .12 1.48 †p .10. *p .05. **p .01. 217 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR boys. One possibility is that for girls, the relation between early exposure to violence and subsequent aggression is due more to cognitive and emotional desensitization to violence than to obser- vational learning. Desensitization should not depend as much as observational learning on identification or perceptions of realism.

Such a hypothesis cannot be tested with the current data but should be examined in future research.

The third notable gender difference was apparent in the struc- tural models. For both male and female participants, there was no significant statistical effect of childhood aggression on adult TV- violence viewing. However, whereas for males the path coefficient from aggression to TV-violence viewing was virtually zero, for females it was a nontrivial positive value. This finding suggests that aggressive females may be more prone than aggressive males to use violent media to make themselves feel better and more justified about their own behavior (Huesmann, 1988). In a culture in which female aggressiveness is still less accepted than male aggressiveness, feeling justified about one’s aggressive behavior could well be more important for females than for males.

Although longitudinal nonexperimental data do not provide a strong test of causation, they can be used to compare the relative plausibility of alternative causal perspectives. These results are certainly consistent with the observational learning and desensiti- zation theories, which predict long-term statistical effects of early TV-violence viewing on later aggression after the effects of early aggression are statistically controlled. The results are much less consistent with the theoretical perspective that more aggressive children turn to watching more violence because they like it or because it serves as justification for their own aggression (Hues- mann, 1982). Only for female participants was there a suggestion of such a long-term effect, and the effect estimate was not signif- icant. The hypotheses that these longitudinal effects could be completely explained by“third”variables such as social class, intellectual ability, parent aggression, or parenting differences also did not receive much support. The effects remained even when the variance due to these factors was partialed out in our regression equations and structural models. However, intellectual ability and parents’education did seem to account for some of the effect in female participants. It may be that social norms for female behav- ior that inhibit the modeling of media violence are related to educational level or intelligence whereas for males this is not true.

Of course, a variety of parenting factors have been shown in the past to be related to both a child’s exposure to media violence and that child’s later aggression—for example, parents’intellectual ability and social class, parents’viewing habits, and parents’ aggression (Comstock & Paik, 1991; Potter, 1999). In the current study, many of these correlations were again found for at least one gender (e.g., for intellectual ability, parental rejection of the child, and parents’frequency of TV viewing). It seems plausible that these factors are indeed influencing both the child’s aggressiveness and the child’s exposure to media violence. Which children are placed most at risk of being exposed to media violence and of experiencing other learning conditions that reinforce the lessons taught by media violence is undoubtedly influenced by parent factors. However, given the pattern of results obtained, it is not very likely that the relations between early exposure to media violence and subsequent aggression are completely due to these “third”variables.The effect sizes for media violence on aggression revealed in this longitudinal study are modest; however, there are few other factors that have been shown to have larger effects. That is not surprising considering the large number of factors that must con- verge before serious adult aggression occurs. Furthermore, as Rosenthal (1986) has pointed out, a correlation of .20 can represent a change in the probability of violence from 50/50 to 60/40, which is large enough to generate social concern.

One might also wonder whether the attrition in the sample made it unrepresentative and biased the results. This seems unlikely.

Archival data were obtained on 80% of the original sample and interview data on 60%. An analysis of the attrition data showed that those who were not reinterviewed tended to be slightly more aggressive as children. Thus, it seems more plausible that, if anything, the attrition weakened the relations between TV- violence viewing and aggression. Implications for Prevention of Violence Overall, these results suggest that both males and females from all social strata and all levels of initial aggressiveness are placed at increased risk for the development of adult aggressive and violent behavior when they view a high and steady diet of violent TV shows in early childhood. The obvious follow-up question is whether society can do anything to prevent or at least moderate this effect.

Several points provide us with guidelines. First, we do not need to be as concerned about adults’or even teenagers’exposure to media violence as much as we do with childrens’exposure. Media violence may have short-term effects on adults, but the real long- term effects seem to occur only with children. This makes some societal controls more palatable in a society that places a high premium on the rights of adults to watch whatever they want.

Second, we need to be aware that media violence can affect any child from any family. The psychological laws of observational learning, habituation/desensitization, priming, and excitation transfer are immutable and universal. It is not, as some have suggested, only the already violence-prone child who is likely to be affected. True, media violence is not going to turn an otherwise fine child into a violent criminal. But just as every cigarette one smokes increases a little bit the likelihood of a lung tumor some day, the theory supported by this research suggests that every violent TV show increases a little bit the likelihood of a child growing up to behave more aggressively in some situation.

Third, the violent films and TV programs that probably have the most deleterious effects on children are not always the ones that adults and critics believe are the most violent. What type of violent scene is the child most likely to use as a model for violent behavior? It is one in which the child identifies with the perpetrator of the violence, the child perceives the scene as telling about life like it is, and the perpetrator is rewarded for the violence. Thus, a violent act by someone like Dirty Harry that results in a criminal being eliminated and brings glory to Harry is of more concern than a bloodier murder by a despicable criminal who is brought to justice. Parents need to be educated about these facts.

Finally, we must recognize the economic realities of media violence. Violence sells. Both children and adults are attracted to violent scenes by the action and intense emotions. Many of the most popular shows and popular films for children have contained 218 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON violence. Violent TV shows appear to be a little cheaper to produce on the average. Hamilton (1998) reported that from 1991 to 1993, the average production fee per hour for network prime time TV programming was about $1,094,000 for nonviolent shows and $998,000 for violent shows—about 10% cheaper in other words. A more telling statistic may be the finding that among shows with some violence, those with more violence actually cost less. Each additional violent act seems to reduce the cost by about $1,500. Of course, these are only averages for production costs. What really counts are the ability of a TV show to attract enough sponsors to cover its cost or to attract enough syndicated buyers and the ability of a video game to attract enough buyers to cover its cost. Here, a variety of marketing issues become important. For example, for- eign markets become very important in these calculations, and generally violent shows and games are easier to sell in foreign markets than are other kinds of games or shows. More specifically, the probability of a TV show being exported successfully increases about 16% if it is violent (Hamilton, 1998).

The easiest way to reduce the effects of media violence on children, of course, is to reduce children’s exposure to such vio- lence. Prevention programs aimed at reducing exposure could obviously be targeted either at the production sources of the violence or at the child viewing the violence. In a society with strong protections for free speech, it is probably always going to be easier to target prevention efforts at the viewing child than at the producer. However, a more informed legal debate is needed on this subject. Broadcasters and film and program makers cannot avoid all responsibility for what children are exposed to. The argument that“people watch it so we give it to them”is not valid in a modern socially conscious society, and it is unrealistic to expect parents to control completely what children watch in a society with multiple TVs in each household, VCRs everywhere, and both parents working. Furthermore, it is the exposure of the 2- to 14-year-old child that is of the greatest concern here, as described in this article. The social value of reductions in the exposure of adults and even older teenagers is probably small compared to the social value of reducing younger children’s exposure.

The ongoing V-chip social experiment is one such attempt to reduce children’s exposure by giving parents a mechanism to control what the TV will allow to be broadcast through it. The problem is that the possibilities for this technology were greatly reduced from the start by the producers of violent shows, who managed to scuttle any idea of a content-based rating system that would actually allow parents to make judgments on the basis of violent content. Instead, only age guideline ratings are broadcast for most programs. Why did the producers do this? One can only speculate, but it is certainly likely that income from violent shows would be substantially reduced if violent labels were added—as much because sponsors would withdraw as because parents would actually program them out with the V-chip.

With regard to interventions aimed at the viewing child, there are a number of possibilities. Again, of course, simply reducing children’s exposure through parental intervention is an obvious approach. However, the theory and results described in this article suggest a number of approaches aimed at changing the effect of any observed violence on the child as well. Nathanson (1999) recently found that parental co-viewing of and commenting on the programs seems to reduce the effects of TV violence on the child, probably because it reduces the child’s identification with theperpetrator, reduces the child’s perception of the violence as real, and reduces the likelihood that the child will rehearse the observed violent script in fantasy or play immediately after observation.

Huesmann, Eron, Klein, Brice, and Fischer (1983) showed that the effects of violence on second graders could be reduced by a targeted school-based attitude change intervention that inculcates them with the beliefs that violence on TV does not tell about the world as it is and should not be imitated. A number of interven- tions based on teaching critical viewing skills in schools are being promoted, though few have yet to undergo rigorous evaluation.

One of the problems with many of these interventions may be that they do not focus on those moderating variables that have been shown to be theoretically relevant, such as identification by the child with perpetrators, perception of violent acts as justified, and perception of violent scripts as realistic.

Although some questions remain to be resolved about the exact extent of the effect of observed violence on aggressive and violent behavior and its importance relative to other causal factors, the current study provides compelling additional evidence that habit- ual exposure of children to violence in the media (or in the real world around them) does have lasting effects on their propensity to behave aggressively and violently. Future research should proba- bly be directed much more at elaborating and testing the kinds of interventions that parents, schools, producers, and the government can promote that will mitigate these long-term effects. References Andison, F. S. (1977). TV violence and viewer aggression: A cumulation of study results 1956–1976.Public Opinion Quarterly, 41,314–331.

Atkin, C., Greenberg, B., Korzenny, F., & McDermott, S. (1979). Selective exposure to televised violence.Journal of Broadcasting, 23,5–13.

Bandura, A. (1969). Social learning theory of identificatory processes. In D. Goslin (Ed.),Handbook of socialization theory and research. Chi- cago: Rand McNally.

Bandura, A. (1977).Social learning theory. Englewood Cliffs, NJ: Prentice Hall.

Bandura, A., Ross, D., & Ross, S. A. (1961). Transmission of aggression through imitation of aggressive models.Journal of Abnormal Social Psychology, 63,575–582.

Bandura, A., Ross, D., & Ross, S. A. (1963a). Imitation of film-mediated aggressive models.Journal of Abnormal and Social Psychology, 66, 3–11.

Bandura, A., Ross, D., & Ross, S. A. (1963b). Vicarious reinforcement and imitative learning.Journal of Abnormal and Social Psychology, 67, 601–607.

Berkowitz, L. (1993).Aggression: Its causes, consequences, and control.

New York: McGraw-Hill.

Bjorkqvist, K. (1985).Violent films, anxiety and aggression. Helsinki, Finland: Finnish Society of Sciences and Letters.

Bjorkqvist, K., Lagerspetz, K., & Kaukiainen, A. (1992). Do girls manip- ulate and boys fight? Developmental trends in regard to direct and indirect aggression.Aggressive Behavior, 18,117–127.

Bjorkqvist, K., Osterman, K., & Kaukiainen, A. (1992). The development of direct and indirect aggressive strategies in males and females. In K.

Bjorkqvist & P. Niemela (Eds.),Of mice and women: Aspects of female aggression(pp. 51–64). New York: Academic Press.

Bushman, B. J., & Huesmann, L. R. (2001). Effects of televised violence on aggression. In D. Singer & J. Singer (Eds.),Handbook of children and the media(pp. 223–254). Thousand Oaks, CA: Sage.

Butterworth, G. (1999). Neonatal imitation: Existence, mechanisms, and 219 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR motives. In J. Nadel & G. Butterworth (Eds.),Imitation in infancy(pp.

63–88). Cambridge, England: Cambridge University Press.

Chaffee, S. H. (1972). Television and adolescent aggressiveness (over- view). In G. A. Comstock & E. A. Rubinstein (Eds.),Television and social behavior: Vol. 3. Television and adolescent aggressiveness(pp.

1–34). Washington, DC: U.S. Government Printing Office.

Cline, V. B., Croft, R. G., & Courier, S. (1973). Desensitization of children to television violence.Journal of Personality and Social Psychology, 27, 360–365.

Comstock, G. A. (1980). New emphases in research on the effects of television and film violence. In E. L. Paler & A. Dorr (Eds.),Children and the faces of television: Teaching, violence, selling. New York:

Academic Press.

Comstock, G. A., & Paik, H. (1991).Television and the American child.

San Diego, CA: Academic Press.

Comstock, G. A., & Paik, H. (1994). The effects of television violence on aggressive behavior: A meta-analysis.Communication Research, 21, 516–546.

Crick, N. R., & Grotpeter, J. K. (1995). Relational aggression, gender, and social-psychological adjustment.Child Development, 66,710–722.

Dodge, K. A. (1980). Social cognition and children’s aggressive behavior.

Child Development, 51(1), 162–170.

Dodge, K. A., Pettit, G. S., Bates, J. E., & Valente, E. (1995). Social information processing patterns partially mediate the effect of early physical abuse on later conduct problems.Journal of Abnormal Psy- chology, 104,632–643.

Elliott, D. S., Dunford, F. W., & Huizinga, D. H. (1987). The identification and prediction of career offenders utilizing self-reported and official data. In J. D. Burchard & S. N. Burchard (Eds.),Prevention of delin- quent behavior(pp. 90–121). Thousand Oaks, CA: Sage.

Eron, L. D. (1963). Relationship of TV viewing habits and aggressive behavior in children.Journal of Abnormal and Social Psychology, 67(2), 193–196.

Eron, L. D., Huesmann, L. R., Lefkowitz, M. M., & Walder, L. O. (1972).

Does television violence cause aggression?American Psychologist, 27, 253–263.

Eron, L. D., Walder, L. O., & Lefkowitz, M. M. (1971).Learning of aggression in children. Boston: Little, Brown.

Feshbach, S., & Singer, R. D. (1971).Television and aggression: An experimental field study. San Francisco: Jossey-Bass.

Fowles, J. (1999).The case for television violence. Thousand Oaks, CA:

Sage.

Geen, R. G. (1983). Aggression and television violence. In R. G. Geen & E. I. Donnerstein (Eds.),Aggression: Theoretical and empirical reviews:

Vol. 2. Issues and research(pp. 103–125). New York: Academic Press.

Geen, R. G. (1990).Human aggression. Pacific Grove, CA: Brooks/Cole.

Geen, R. G., & O’Neal, E. C. (1969). Activation of cue-elicited aggression by general arousal.Journal of Personality and Social Psychology, 11, 289–292.

Geen, R. G., & Thomas, S. L. (1986). The immediate effects of media violence on behavior.Journal of Social Issues, 42,7–28.

Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. (1994). Growing up with television: The cultivation perspective. In J. Bryant & D. Zillmann (Eds.),Media effects(pp. 17–41). Hillsdale, NJ: Erlbaum.

Goldstein, J. (1998).Why we watch: The attractions of violent entertain- ment. Oxford, England: Oxford University Press.

Guerra, N. G., Huesmann, L. R., Tolan, P., Van Acker, R., & Eron, L. D.

(1995). Stressful events and individual beliefs as correlates of economic disadvantage and aggression among urban children.Journal of Consult- ing and Clinical Psychology, 63,518–528.

Hamilton, J. T. (1998).Channeling violence: The economic market for violent television programming,Princeton, NJ: Princeton University Press.Hearold, S. L. (1979).Meta-analysis of the effects of television on social behavior. Unpublished doctoral dissertation, University of Colorado.

Hollingshead, A. B., & Redlich, F. C. (1958).Social class and mental illness. New York: Wiley.

Huesmann, L. R. (1982). Television violence and aggressive behavior. In D. Pearl, L. Bouthilet, & J. Lazar (Eds.),Television and behavior: Ten years of scientific progress and implications for the eighties: Vol. 2.

Technical reviews(pp. 126–137). Washington, DC: U.S. Government Printing Office.

Huesmann, L. R. (1986). Psychological processes promoting the relation between exposure to media violence and aggressive behavior by the viewer.Journal of Social Issues, 42,125–139.

Huesmann, L. R. (1988). An information processing model for the devel- opment of aggression.Aggressive Behavior, 14,13–24.

Huesmann, L. R. (1995).Screen violence and real violence: Understand- ing the link[Brochure]. Auckland, New Zealand: Media Aware.

Huesmann, L. R. (1998). The role of social information processing and cognitive schemas in the acquisition and maintenance of habitual ag- gressive behavior. In R. G. Geen & E. Donnerstein (Eds.),Human aggression: Theories, research, and implications for policy(pp. 73– 109). New York: Academic Press.

Huesmann, L. R. (1999). The effects of childhood aggression and exposure to media violence on adult behaviors, attitudes, and mood: Evidence from a 15-year cross-national longitudinal study.Aggressive Behav- ior, 25,18–29.

Huesmann, L. R., & Eron, L. D. (1986).Television and the aggressive child: A cross-national comparison. Hillsdale, NJ: Erlbaum.

Huesmann, L. R., Eron, L. D., Berkowitz, L., & Chaffee, S. (1991). The effects of television violence on aggression: A reply to a skeptic. In P.

Suedfeld & P. Tetlock (Eds.),Psychology and social policy(pp. 191– 200). New York: Hemisphere.

Huesmann, L. R., Eron, L. D., Klein, R., Brice, P., & Fischer, P. (1983).

Mitigating the imitation of aggressive behaviors by children’s attitudes about media violence.Journal of Personality and Social Psychology, 44, 899–910.

Huesmann, L. R., Eron, L. D., Lefkowitz, M. M., & Walder, L. O. (1984).

Stability of aggression over time and generations.Developmental Psy- chology, 20,1120–1134.

Huesmann, L. R., Eron, L. D., & Yarmel, P. W. (1987). Intellectual functioning and aggression.Journal of Personality and Social Psychol- ogy, 52,232–240.

Huesmann, L. R., & Guerra, N. G. (1997). Children’s normative beliefs about aggression and aggressive behavior.Journal of Personality and Social Psychology, 72,408–419.

Huesmann, L. R., Guerra, N. G., Eron, L. D., & Crawshaw, V. B. (1994).

Measuring children’s aggression with teachers’predictions of peer nom- inations.Psychological Assessment, 6,329–336.

Huesmann, L. R., Guerra, N. G., Zelli, A., & Miller, L. (1992). Differing cognitions relating to TV viewing and aggression among boys and girls.

In K. Bjorkqvist & P. Niemela (Eds.),Of mice and women(pp. 77–87).

New York: Academic Press.

Huesmann, L. R., Lagerspetz, K., & Eron, L. D. (1984). Intervening variables in the TV violence-aggression relation: Evidence from two countries.Developmental Psychology, 20,746–775.

Huesmann, L. R., Lefkowitz, M. M., & Eron, L. D. (1978). Sum of MMPI scales F, 4, and 9 as a measure of aggression.Journal of Consulting and Clinical Psychology, 46,1071–1078.

Huesmann, L. R., & Miller, L. S. (1994). Long-term effects of repeated exposure to media violence in childhood. In L. R. Huesmann (Ed.), Aggressive behavior: Current perspectives(pp. 153–186). New York:

Plenum Press.

Huesmann, L. R., Moise, J. F., & Podolski, C. P. (1997). The effects of media violence on the development of antisocial behavior. In D. Stoff, 220 HUESMANN, MOISE-TITUS, PODOLSKI, AND ERON J. Breiling, & J. Maser (Eds.),Handbook of antisocial behavior(pp.

181–193). New York: Wiley.

Josephson, W. L. (1987). Television violence and children’s aggression:

Testing the priming, social script, and disinhibition predictions.Journal of Personality and Social Psychology, 53,882–890.

Lagerspetz, K. M., & Bjorkqvist, K. (1992). Indirect aggression in girls and boys. In L. R. Huesmann (Ed.),Aggressive behavior: Current perspec- tives(pp. 131–150 ). New York: Plenum Press.

Lagerspetz, K. M., Bjorkqvist, K., & Peltonen, T. (1988). Is indirect aggression typical of females? Gender differences in aggressiveness in 11- to 12-year-old children.Aggressive Behavior, 14,403–414.

Lefkowitz, M. M., Eron, L. D., Walder, L. O., & Huesmann, L. R. (1977).

Growing up to be violent: A longitudinal study of the development of aggression. New York: Pergamon Press.

Milavsky, J. R., Kessler, R., Stipp, H., & Rubens, W. S. (1982). Television and aggression: Results of a panel study. In D. Pearl, L. Bouthilet, & J.

Lazar (Eds.),Television and behavior: Ten years of scientific progress and implications for the eighties: Vol. 2. Technical reviews(pp. 138– 157). Washington, DC: U.S. Government Printing Office.

Moise-Titus, J. (1999).The role of negative emotions in the media violence–aggression relation. Unpublished doctoral dissertation, Uni- versity of Michigan, Ann Arbor.

Nathanson, A. I. (1999). Identifying and explaining the relationship be- tween parental mediation and children’s aggression.Communication Research, 26(2), 124–143.

Paik, H., & Comstock, G. A. (1994). The effects of television violence on antisocial behavior: A meta-analysis.Communication Research, 21, 516–546.

Potter, W. J. (1999).On media violence. Thousand Oaks, CA: Sage.

Rosenfeld, E., Huesmann, L. R., Eron, L. D. & Torney-Purta, J. V. (1982).

Measuring patterns of fantasy behaviors in children.Journal of Person- ality and Social Psychology, 42,347–366.

Rosenthal, R. (1986). Media violence, antisocial behavior, and the social consequences of small effects.Journal of Social Issues, 42,141–154.

Rossi, P., Bose, C., & Berk, R. (1974). The seriousness of crime, normativestructure and individual differences.American Sociological Review, 39, 224–237.

Straus, M. A., Gelles, R. J., & Steinmetz, S. K. (1980).Behind closed doors: Violence in the American family(Appendix B). Garden City, NY:

Anchor Books.

Tiegs, E. W., & Clark, W. W. (1970).California Achievement Test. New York: McGraw-Hill.

Tremblay, R. E. (2000). The development of aggressive behavior during childhood: What have we learned in the past century?International Journal of Behavioral Development, 24,129–141.

Walder, L. O., Abelson, R., Eron, L. D., Banta, T. J., & Laulicht, J. H.

(1961). Development of a peer rating measure of aggression.Psycho- logical Reports, 9,497–556 (Monograph Supplement 4–19).

Warner, W. L., Meeker, M., & Eells, K. (1960).Social class in America.

New York: Harcourt.

Weigman, O., Kuttschreuter, M., & Baarda, B. (1986).Television viewing related to aggressive and prosocial behavior. The Hague: Stitching voor Onderzoek van het Onderwijs.

Wood, W., Wong, F. Y., & Chachere, G. (1991). Effects of media violence on viewers’aggression in unconstrained social interaction.Psychologi- cal Bulletin, 109,371–383.

Wynder, E. L., & Graham, E. A. (1950). Tobacco smoking as a possible etiological factor in bronchiogenic carcinoma.Journal of the American Medical Association, 143,329–336.

Wyrwicka, W. (1996).Imitation in human and animal behavior. New Brunswick, NJ: Transaction Publishers.

Zillmann, D. (1979).Hostility and aggression. Hillsdale, NJ: Erlbaum.

Zillmann, D. (1983). Transfer of excitation in emotional behavior. In J. T.

Cacioppo & R. E. Petty (Eds.),Social psychophysiology: A sourcebook (pp. 215–240). New York: Guilford Press. Received February 17, 2000 Revision received December 19, 2001 Accepted December 19, 2001 221 SPECIAL ISSUE: TV AND VIOLENT BEHAVIOR