by Joseph A. D'Agostino

The habitually belligerent six-year-old boy who on February 29 shot and killed a six-year-old girl at the Theo Buell Elementary School in Mount Morris Township, Mich., liked to watch violent movies and television and "hated everybody," his incarcerated father told police.

The boy stole the murder weapon-a.32 caliber pistol that had previously been stolen by the person he stole it fromfrom the crackhouse where he lived with his 8-year-old brother and an uncle who was wanted on a criminal warrant.

Yet, on March 2, pushing for a federal law that would mandate trigger locks for all guns sold in America, President Clinton said of this first-grade murder: "That child would be alive today if that gun had had a child triggef lock on it that the other six-year-old child could not have fired."

Here is the real story as reported in a series of articles in the Detroit News:

The boy's mother, Tamarla Owens, had abandoned him at the house of an uncle, Sir Marcus Winfrey, together with his 8-year-old brother. She had just been evicted from her own broken-down home. In 1992, she had been found guilty of abusing yet another son. The uncle himself had an arrest warrant outstanding against him.

The boy's father, Dedric Darnell Owens, 28, is currently in jail with a prison record going back to 1989, when he broke into a car. In 1992, he was charged with attempted murder, but the case was dismissed. In 1995, he went to jail for burglary and cocaine possession, and then was released on parole on Dec. 21, 1999. On February 20, he was sent back to jail for violating parole.

"He was basically living in hell," Genesee County Sheriff Robert Pickell said of the boy. "His father said he was always watching violent movies. always interested in guns." The sheriff said that Owens told him that he asked the boy why he hit his classmates, and even once stabbed a girl with a pencil. "I hate them," the boy said.

Owens said publicly that his son "hated everybody. He didn't get along with anyone. He watched violent movies and TV" He admitted to taking the boy to see a Terminator movie which he later saw again on video. The boy, the father said., had been suspended from school three times before the shooting, once for the stabbing. "He looked up at me," said the imprisoned Owens. "He always wanted to go where I wanted to go."

Michigan's Family Independence Agency reported last year that Tamarla Owens was abusing an older son. And a social worker signed an affidavit the day after the shooting saying that the mother was "involved with drugs." The agency had been informed of his history of violence at school.

Six-year-old Kayla Rollands, the victim, had attracted the boy's ire by spurning his physical advances, the News reported. "He wanted to kiss her, she told him no and then she spit on him," fellow student Naomi Butler told the paper, saying she heard the story from Kayla and Kayla's sister.

The next day the boy arrived at school from the crackhouse carrying the stolen gun. The Davis Industries .32-caliber semi-automatic pistol had been stolen from the home of its legal owner in a nearby suburb, used to trade for crack, and then, just 84 days after its theft, used to kill Kayla. Police said that Owens told them that guns were exchanged for drugs at the house all the time.

Not One In 50,000

Police found another stolen gun, loaded, at the house.

Family friend Jamelle James, 19, who lived with the uncle, has been charged with involuntary manslaughter for bringing the gun used in the shooting into the home and "twirling" it in front of the boy and telling him where he kept it which police say James admitted he did.

Police said the boy claimed he planned ahead of time to bring the gun to school and "frighten" the girl with it, but not to kill her. As students were filing out of the first-grade room to head to a computer lab, the boy loaded the gun with three bullets, pointed it at one student, then turned on Kayla and fired a single shot that passed through her arm and chest. He then put the gun in his desk and ran to the principal's office. Kayla died 30 minutes later.

Michigan social workers took 50,000 children from their families from 1995 to 1997, but the shooter was not one of them.

by Linda Heath , Candace Kruttschnitt , David Ward

Over the past 20 years, numerous scholars have attempted to determine what, if any, relationship exists between exposure to media violence and aggressive behavior. Most of these studies find at least a modest relationship between media exposure and aggression (Andison, 1977). Two key questions, however, remain unanswered. First, do the effects of media violence exposure extend to actual violent, criminal behavior, or are they limited to minor aggressions and acts of juvenile delinquency? Second, what factors, if any, enhance or moderate the links between media violence and aggressive or violent behavior? We attempt to answer each of these questions, beginning our analysis with a brief (and by no means exhaustive) review of the research concerning media effects on aggression and crime. (Comprehensive reviews of this literature can be found in Andison, 1977; Comstock, Chaffee, Katzman, McCombs, & Roberts, 1978; Geen, 1976; Liebert & Baron, 1972; Liebert, Neale, & Davison, 1973; Murray & Kippax, 1979).

GENERALIZABILITY OF EFFECTS

The relationship between exposure to media violence and aggressive behavior has been demonstrated in a series of laboratory and field experiments and quasiexperiments. These studies have found that the media-aggression link is enhanced if (a) the media aggression is presented as being justified (e.g., Berkowitz, 1965); (b) salient cues are present during the retrieval period (e.g., Berkowitz & Frodi, 1977); (c) the respondents are predisposed to aggressive behavior (e.g., Friedrich & Stein, 1973; Parke, Berkowitz, Leyens, West, & Sebastian, 1977); and (d) the respondent identifies with the violent character (e.g., Huesmann, Eron, Klein, Brice, & Fischer, 1983; Turner & Berkowitz, 1972). These findings apply not only to young children (Bandura, Ross, & Ross, 1963; Liebert & Baron, 1972; Singer & Singer, 1980) but also to adolescents (Belson, 1978; Hartman, 1969) and to college students (Berkowitz & Rawlings, 1963; Berkowitz & Geen, 1966). Media effects are found on aggressive behaviors, including hitting a Bobo doll (e.g., Bandura et al., 1963), shocking confederates (e.g., Berkowitz & Geen, 1966), verbal and physical aggression (e.g., Eron & Huesmann, 1980; Feshbach & Singer, 1971), and minor acts of juvenile delinquency (e.g., Belson, 1978; Mclntyre, Teevan, & Hartnagel, 1972). Finally, controlling for factors such as socioeconomic status, intelligence, race, and mother's education does not eliminate the relationship between media exposure and aggression and in some cases strengthens it (e.g., Mclntyre et al., 1972).

While most of the findings from these studies point in the same direction (i.e., exposure to media violence is related to increased aggression), a number of researchers have questioned the practical significance of these findings. Specifically, causal ambiguities (Freedman, 1984), the modest size of the effects (e.g., Cook, Kendzierski, & Thomas, 1983), and the restricted generalizability of the dependent variables have led scholars to question whether, in fact, these findings have any bearing on the commission of a violent crime. Comstock, for example, concluded that the link between media violence and criminal activity

. . . rests on the willingness of the person who chooses to sit in judgment to extrapolate from the findings on interpersonal aggression to more serious, non-legal acts. Most important, the evidence does not tell us anything about the degree of social harm or criminal anti-social violence that may be attributable to television. It may be great, negligible, or nil. (Halloran, 1980, pp. 439-440)

In response to these concerns, some scholars have examined the relationship between mass media violence and actual criminal behavior. Milgram and Shodand (1973), for example, examined the rate of theft from a charity box immediately following exposure to one of three variations of a television program and found no relationship. However, the assumption that exposure to deviant behavior on television would, in fact, result in immediate imitation of criminal activity is questionable. Menzies (1971) compared the media habits of violent offenders and property offenders who were incarcerated and found no differences, although television habits while incarcerated might not be reflective of prior media exposure.

Other researchers have used aggregate rather than individual-level data to examine the relationship between crime rates and media events. Phillips (1983) documented an increase in the homicide rate following major televised prizefights, particularly among males of the same race as the loser of the fight. Using a similar approach, Hennigan and her colleagues (1982) documented an increase in the theft rate in cities immediately following the introduction of television into those cities. (The observed increase in theft rather than burglary indicates that the effect is not attributable to television sets being stolen.)

Although studies based on aggregate data greatly increase the external validity of the media-violence relationship, they do not easily lend themselves to testing the effect mediating variables might have on this relationship. A concern over the effect of such mediating variables has led to a reexamination of the processes by which media messages translate into action, resulting in expansion and reinterpretations of the traditional formulations.

MEDIATING VARIABLES AND PROCESSES

In 1972 the Surgeon General's Scientific Advisory Committee on Television and Social Behavior concluded that television can, under some circumstances, for some children, lead to increased aggressiveness. Since then researchers have been trying to pinpoint under what circumstances and for which people the relationship between viewing and aggression is strongest. To understand why and for whom these effects hold, researchers have ventured into the realm of cognitive psychology.

For example, Huesmann (1982) posited that social modeling effects rely on the principle of encoding specificity. The recall of an event (media or real life) depends, in part, on the similarity of the recall situation to the situation in which the encoding occurred. For example, a violent altercation might trigger memories of violent television episodes more easily if the television was viewed (or encoded) amid a setting of family violence. Further, Huesmann posited that characteristics that attract attention (such as familiarity) contribute to the possibility that the event will be encoded and stored in memory. Geen (1983) has suggested that one characteristic of television violence that might make scenes more salient and memorable is the perceived reality of the violence. Again, violence in the home could increase the perceived reality and ultimately the memorability of television violence.

Berkowitz (1984) suggested that cognitive neoassociationism can help researchers understand how the interpretations people give to media messages, and the thoughts that are activated by these messages, might influence the behavioral consequences of such messages. Basically, cognitive neoassociationism posits the memory as a collection of networks, composed of individual nodes or units that are connected by pathways. Factors such as contiguity, similarity, and semantic relatedness influence the strength of these pathways. The activation of any one node or unit (e.g., a thought, feeling, scene) will spread down the pathways to other related nodes, resulting in a priming effect. Nodes that have been recently primed will be brought into consciousness more easily than unprimed nodes.

By extension, although not explicitly tested by previous research, violent media depictions might trigger aggressive means of conflict resolution if those conflict resolution means are located on the same memory networks as the media depictions. For example, the coding of family violence and television violence on the same memory networks could result in more violent means of dealing with future interpersonal conflicts. Berkowitz (1984) further emphasized the importance of the meaning attached to the media depiction by the viewer: "Aggression is in the mind of the beholder, and a movie will not activate aggression-associated thoughts unless the viewer regards what is seen as aggression" (p. 419).

Another possibility, however, is that aggressive media acts may trigger behaviors that outside observers would label as aggressive or violent, even if the viewer him/ herself does not apply such a label. That is, media depictions of violence may lead some viewers to consider violence as a normal act. In essence, then, violent media actions may be miscoded onto the wrong network and associated with units representing "appropriate conflict resolution means," for example. Such coding could entirely sidestep the aggressiveness label, both at the encoding state and at retrieval stage, but could still result in behavior that would be labeled "aggressive" by the average observer.

This miscataloging of violent media messages might be especially prevalent when viewers are exposed to what Gerbner, Gross, Morgan, and Signorielli (1980) called the double-dose effect. That is, when media messages match real-world family experiences, the media depictions might not be cataloged as "fictional" or "deviant" but, instead, associated with "real life" memory units (such as "how Mom and Dad resolve disputes"). This formulation is congruent with Geen's (1983) and Huesmann's (1982) suggestions that events that are perceived as real rather than fictional might be more salient and therefore more likely to be encoded, leading to greater probability of later retrieval and possible behavioral effects.

In this research we examine the hypothesis that exposure to television violence during critical adolescent years-identified by Eron (1982) as between 8 and 12-is related to the commission of a violent crime as a young adult, particularly for those people whose home environments contained violence.

METHOD

Respondents

We defined violent offenders as individuals who had been convicted of an offense that involved the use or threat offeree on a victim. As a result, most of our offender population were persons incarcerated for murder, rape, robbery, and/or aggravated assault. Because there were some cases in which other forms of criminal sexual conduct (e.g., incest) and kidnapping met our criteria of "threat or use offeree on a victim," they were also included in our inmate population. Similarly, we also included several "burglars" for whom the recorded description of the crime indicated that sexual or physical assault or threat of assault was involved in the criminal act.

The second criterion we used to identify the violent offender respondents was age. Because the validity of our data depends in part on accurate recall of events that occurred during childhood and adolescence, we selected individuals for whom these life stages were fairly recent, that is, offenders who were, at the time of our interviews, between 18 and 25 years of age.

Our third criterion was conviction in the county where the study took place. The study design called for a match of our inmate population with a comparison group who grew up in the same neighborhood between the ages of 10 and 14, and we could access comparison group respondents only from the immediate geographic area. We therefore restricted our inmate population to those committed to the institutions from Hennepin County, Minnesota, which includes the city of Minneapolis and its suburbs. Finally, because there were so few women incarcerated in Minnesota prisons for crimes of violence, we limited our study to men.

We completed interviews with 48 inmates who met the above four criteria.1 To increase the internal validity of this study, we matched our violent offender population as closely as possible with a noncriminal population on the basis of age, race, sex, and neighborhood of residence during adolescence. We began this matching procedure by plotting on a map of Hennepin County the address of each inmate when he was approximately 10 to 14 years of age and his race. We then went to the relevant neighborhoods and placed posters in supermarkets, playgrounds, youth centers, and other recreational facilities. The posters offered $25 to males between the ages of 17 and 26, willing to participate in a university research project on "life styles." The term "life styles" was purposely chosen to satisfy the human subjects committee's informed consent requirement without indicating that we were particularly interested in crime, violence, mass media use, and family relations. We chose such a vague term to reduce possible self-selection and rejection problems. The posters contained a phone number for interested parties to call in response. We screened each caller to determine his residence between the ages of 10 and 14, his race, his age, and his criminal history.2 Questions about criminal history were included to ensure that our control group did not contain individuals who had been arrested for or convicted of violent crimes. These matching procedures ultimately produced completed interviews on a comparison group of 45 individuals of the same sex, age, socioeconomic background, and neighborhood during adolescence as our offender population.3 Although we were unable to obtain a one-for-one match on the racial dimension, the relative racial proportions between the control and offender populations are quite similar.4

The Instrument

Our survey instrument was composed of open- and closed-ended questions and paper-and-pencil items. Interviews were conducted by trained personnel (including the authors) and ranged in length from 3 to 8 hours. Each interview was tape recorded, transcribed, and then coded into quantitative form. The same interview format was used for both inmate and comparison groups, with only slight modifications in the criminal history questions for the comparison sample. Items were designed to document not only the extent of our subjects' exposure to television but also their exposure to violence in the home, their degree of family cohesiveness, their school and peer relationships, and their prior experiences with the criminal justice system.

Television exposure was measured by giving photocopies of TV Guide fall programming summaries to our respondents. Because the relevant literature on exposure to television violence suggests that ages 8 through 12 are particularly important (Eron, 1982), each respondent was given a copy of the television listing for 1 week for the years when he was 8, 10, and 12 years old. He was asked to circle the shows he remembered watching all the time (or "regularly") and to put an X through the shows he watched only sometimes. In addition, we also asked our respondents to indicate the television shows and televison characters that were their favorites when they were between 8 and 12 years of age.

This retrospective television viewing self-report method was subjected to testretest reliability analysis prior to its use in this study. Heath and Petraitis (1984) found that with college students the same TV Guide sheets (corresponding to ages 8, 10, and 12) given out 7 to 10 days apart showed an exact-match test-retest reliability of .88. That is, the average respondent replied to 88% of the items identically during both of the sessions. Respondents were similar in age to the inmate and comparison samples and were therefore reporting viewing over a similar time span. Respondents in the reliability study did not know they would be asked to report television viewing a second time.

Our instrument also included six items that assessed physical abuse of the respondent by both his mother and his father. Factor analysis of these items indicated that they comprised two separate factors, one relating to paternal abuse and the other to maternal abuse. The paternal abuse items included three variables: (a) frequency of Dad losing his temper, coded in times per month or year; (b) what Dad did when he lost his temper, coded for degree of violence5; and (c) the consequence of the time Dad was the maddest, again coded for degree of violence. Maternal abuse was assessed by a similar method, using a scale composed of the above three items referencing mothers rather than fathers. Standardized scores were computed for the variables within each factor, and two additive factors were created (see Kruttschnitt, Heath, & Ward, 1986).

Our interviews revealed that family disorganization was a frequent characteristic of our respondents' childhoods. As a result, we thought that residential instability, which often accompanied family disorganization, might interfere with televisionviewing habits. Using a chronological indicator of changes in family life, we coded all individuals who were removed from their homes, for whatever reason, before 13 years of age as not having resided consistendy in their nuclear home. Those who stayed with their families until at least 13 years of age were coded as having had a consistent residence.

Finally, measures of the respondent's involvement in any recreational clubs and team sports and his race were included in the analysis. Recreational clubs and participation in team sports were included in order to control for time available for television viewing. The subject's race was included because we thought that individuals might be more likely to identify with television characters of their own race than characters of differing race. Previous research has shown not only that white television characters are more likely to use justified violence than minority characters (Gerbner, 1972) but also that viewing justified violence produces the more aggressive response in subjects (Geen, 1981). A respondent's race, therefore, may be related to his television viewing patterns and thereby to his subsequent aggression.

RESULTS

This study uses the epidemiological approach, whereby respondents are identified by variation on the dependent variable (in this case, violent behavior), and relationships to possible causal variables are examined. This approach is frequently used in medical research and is useful for studying correlates of rare events (such as the commission of violent crime). Although some statistics that are used with such an approach (e.g., ANOVA) are most often used in conjunction with experimental research, in epidemiological applications the results are clearly correlational. Consequently, we must be particularly concerned about possible confounding variables. By matching our inmate and comparison groups on neighborhood of residence during adolescence, we also succeeded in matching the groups on family size, family constellation at birth, number of household moves, socioeconomic status, and neighborhood crime rate. These variables should not, therefore, produce spurious effects. Further, none of these variables has been hypothesized to interact with media exposure in regard to violent behavior.

We did, however, examine time-usage factors that could be confounded with television exposure. Because adolescents have a limited amount of time for leisure activities, time devoted to any one activity could detract from time available for other activities. Two types of activities that could limit adolescents' available time for television viewing are involvement in organized sports and attendance at clubs. Neither of these activities, however, correlates significandy with time spent viewing television (for clubs and television, r = .08, N.S.; for sports and television, r = .07,N.S.). We also examined the possible interactions between time spent on sports and clubs and exposure to television in relation to violent behavior. The analyses revealed no significant interactions between television viewing and sports involvement in predicting violent behavior. Similarly, time spent attending club activities did not interact with television viewing in any systematic way.6

Another factor that could reduce the time adolescents have for televison viewing is removal of the adolescent from his natural home. Residence in a foster home, group home, or reformatory might entail restrictions on television viewing that would not be present in the natural home environment. To examine whether this variable would possibly confound our results, we compared the viewing habits of respondents who had been removed from their homes prior to age 12 with those of respondents who remained at home through age 12. Again, we found no relationship between viewing patterns and residential stability (r = .06, N.S.) and no patterns of interaction between media exposure and residence in regard to violent behavior (all F's < 1.00, MS.).

Finally, we examined possible parental controls over television viewing for confounds with actual television viewing. Respondents were asked if their parents had any rules about television viewing when they were young. Among the inmate sample, 34% indicated that their parents had some sort of television rules; and among the comparison sample, 54% indicated their parents had such rules. Almost all of these rules pertained to the amount of television viewing. (Only one inmate and two comparison respondents indicated parental restrictions on their viewing violent programs.) However, for both inmate and control respondents, these time restrictions do not appear to have been enforced. Among inmates whose parents supposedly had time restrictions on television viewing, 64% fell in the "high television exposure" group, compared with 48% of inmates whose parents had no rules. Similarly, among the comparison respondents whose parents restricted their TV viewing time, 50% fell in the high television exposure group, compared with 39% of those respondents whose parents had no time restrictions on viewing. We conclude, therefore, that variation in television viewing time was not merely an artifact of parental rule setting and supervision.

Analysis of Television Effects

We used discriminant analysis to differentiate between violent offenders and nonoffenders and included television exposure, maternal abuse, and paternal abuse as predictor variables. Discriminant analyses were performed on both the total sample and racial subsamples.7 We were also concerned about possible interaction effects. The correlations between the discriminant analysis interaction terms and their main effects (r's = .80 or above) indicated serious identification problems or evidence of multicollinearity. Therefore, we also used analysis of variance techniques; race, paternal abuse, maternal abuse, and television exposure (dichotomized at the respective medians) were entered as independent variables and violent behavior as the dependent variable.

The television exposure variables included the total number of shows watched "all the time" and the total number watched "sometimes" at 8, 10, and 12 years of age for each respondent. Because these six measures showed similar patterns, we collapsed across the three ages and across the frequency of viewing. The degree to which television programs contained portrayals of intentional harm served as our basis for distinguishing between violent and nonviolent television programs.

Table 1 presents the summaries of the discriminant analyses. As can be seen from this table, the television exposure variables (total viewing, violence viewing, and nonviolence viewing) enter all of the discriminant equations, but none of the equations provides particularly strong predictions of violent behavior. Analysis of the simple main effects indicates that inmates reported viewing more violent television as adolescents than did the comparison sample, F( 1,75) = 5.00,p <.05. Inmates reported viewing an average of ten violent television programs a week, and the comparison sample of respondents reported viewing an average of eight violent shows a week. This same trend is evident in regard to total television viewing, F (1,75) = 2.68, p <.11. Inmates watched a mean of 33 programs per week, while the comparison group mean was 28 programs per week. The pattern of results in regard to nonviolent programming was also in this direction, with inmates reporting watching 23 programs per week, compared with 19 programs per week for the control group. This finding, however, did not reach the standard level of statistical significance, F (1,75) = 1.86,jb = .18. The high correlation between violent television viewing and nonviolent television viewing (r = .82,p <.001) precludes our attributing the effect conclusively to either type of programming.

In summary, the discriminant analysis revealed three noteworthy findings. First, paternal abuse, maternal abuse, and television viewing all associate with violent adult behavior. Second, violent and nonviolent television viewing are highly correlated, preventing us from examining one effect separate from the other. Our respondents did not seem to discriminate much in their program selection, lending credence to the television industry view that "hooking" people early in the evening ensures high ratings all evening long. Finally, the multicollinearity problems prevented us from examining the interaction terms with the discriminant analysis and led us to the next set of analyses.

Because of the aforementioned problems with the discriminant analysis, we also used analyses of variance (ANOVAs) to examine our interaction terms. These analyses revealed that paternal abuse is marginally related to later criminal behavior- F (1,78) = 3.48, p <.10-with 56% of the respondents in the high paternal abuse category being incarcerated, compared with 47% of those in the low paternal abuse category. This effect, however, is subsumed under the significant interaction among paternal abuse, maternal abuse, and total television viewing, F (1,78) = 5.95, p <.05. As Table 2 indicates, respondents who were exposed to two forms of violent behavior (i.e., maternal abuse and TV, paternal abuse and TV, or both parental and maternal abuse) were more likely to engage in violent criminal behavior than were those respondents who were exposed to only one form of violence or to none. Exposure to all three forms of violence did not increase the likelihood of violent behavior over that of exposure to two forms. Importandy, no one form of exposure to violent behavior in isolation increased the likelihood of later criminal involvement. This same three-way interaction is apparent when television viewing is broken down into violence viewing and nonviolence viewing-both Fs (1,78) > 4.44, ps <.05.8 Again, however, because of the high correlation between violence and nonviolence viewing, we cannot attribute the effects conclusively to one type of programming or the other. Race did not enter significantly as a main effect or in any interactions.

We also wanted to know whether inmates were more likely than comparison respondents to list violent characters or shows as their favorites. A majority of both the inmate sample (58%) and the comparison sample (56%) listed violent programs as their favorites during adolescence. Inmates were slightly more likely to list comedies as their favorites than were comparison respondents (39% versus 30%, respectively), and comparison respondents were more likely to list nonviolent dramatic programs as their favorites thin were inmates (14% versus 0%, respectively). These differences are not, however, statistically significant. In regard to favorite characters, the comparison sample was, if anything, more likely to list violent male characters as favorites than were inmates (53% versus 44%). Again, inmates were more likely to identify comedians as their favorite characters (33%) than were comparison respondents (28%), and race was irrelevant to favorite character choice.

Finally, in regard to the judged reality of the television messages, both the inmate and the comparison samples appear to be fairly credulous. Among inmates, 70% indicated that the things that happen on television could happen in real life. In the comparison sample, 72% agreed with this judgment.

DISCUSSION

The basic findings of this study are straightforward: high exposure to television during childhood years was related to the commission of a violent crime during young adulthood if violence was also present in the home. Exposure to television without violence in the home was not associated with violent crime. While our findings cannot speak to the exact cognitive processes underlying this relationship among television exposure, violence in the home, and violent behavior, they are congruent with at least three formulations that have been proposed to explain these processes: cognitive neoassociationism, encoding specificity, and the double-dose formulation.

According to cognitive neoassociationism (Berkowitz, 1984), activation of one node or memory spreads down the associative pathways to other nodes that reside on the same network. Perhaps children who view television violence amid a setting of family violence form associations between the televised violence and such other concepts (or nodes) as "conflict resolution," "family relations," or "goal attainment." If such associations were formed, later experiences with family conflict or blocked goals could prime or trigger the violent television images, leading, perhaps, to violent behaviors. Peripheral support for this notion comes from Belson's (1978) finding that one type of television violence that is strongly related to adolescent aggression is violence that occurs in close relationships. Similarly, among individuals raised in a violent family setting, television violence might not be coded on networks that contain nodes such as "deviant," "unkind," or even "aggressive." Violent behaviors could thus come to be considered part of normal life and not be seen as aggressive or nonnormative. Anecdotal evidence for this possibility comes from observations by the interviewers in this study that many inmates did not consider sticking a gun in someone's face as an unkind or aggressive act. Instead, they viewed such behavior simply as a means to an end (generally the acquisition of material possessions or money).

The findings from this research are also consistent with the concepts of encoding specificity (Huesmann, 1982). If children encode the messages from television violence in a hostile and aggressive family situation, they might be more likely, if later presented with a hostile or aggressive situation, to pull out those images of television violence.

Our findings are also congruent with Gerber and his colleagues' (1980) notion of the double-dose effect. Getting the same violent message from two sources (i.e., TV and Mom, TV and Dad, or Mom and Dad) increased the effect of each source. The double-dose message seems to be that television violence is not fiction when either Mom or Dad, or both, are also violent. Any one source of exposure to violence, in isolation, was not associated with criminal behavior, nor did a third source increase the harmful effect of two.

This research was not intended to pit these various theoretical formulations against each other, nor are these formulations necessarily contradictory. In fact, the overall results of our study are consistent with all of the above-noted formulations. Extensive exposure to television viewing, when coupled with physical abuse by either parent, produced a stronger relationship with later criminal activity than was observed with either force working in isolation.

These findings, however, must be muted with two caveats. First, we cannot be certain, from this research, that the crucial element of television exposure is the violence. Because we were unable to distinguish the effects of viewing violent and nonviolent television, we cannot rule out the possibility that some message other than violence (e.g., depiction of material wealth, which might lead to feelings of relative deprivation) is responsible for the increased violence. Second, we purposely chose respondents who had been convicted of serious, violent crimes, and this selection criterion produces serious external validity limitations. Our respondents were young, urban males, predominantly members of minority racial groups, predominantly of the lower socioeconomic class. This is the prototypic high-risk group for criminal involvement. The effects of television and family violence might be different among older populations, rural populations, or higher socioeconomic groupings.

In summary, the findings from this research indicate that television exposure is not only related to hitting Bobo dolls, shocking other subjects, rough playground activity, and petty delinquent acts; it is also related to serious, violent crimes. However, what is most intriguing about these findings is the way in which television exposure is related to acts of criminal violence. We have found that mass media effects do not operate in a vacuum. As current integrations between traditional social learning models and information processing models posit (e.g., Berkowitz, 1984; Geen, 1983), the effect that television exposure has on violent behavior depends, at least in part, on other aspects of the observer's environment. Only when the media message is congruent with the perceptions and life experiences of the observer will the message be associated with later action.

-1-

by Beverly Rivera , Cathy Spatz Widom

Childhood victimization and violent offending represent two serious social problems confronting society today. Langan and Innes (1985) estimated that 3% of Americans, or approximately six million people, are victims of a violent crime during the course of a year. According to the American Humane Association, in 1986 about 2.1 million children were reported to authorities as suspected victims of child abuse or neglect. Approximately 1,100 children die each year under circumstances suggestive of parental maltreatment.

THE ROLE OF CHILD ABUSE AND NEGLECT IN VIOLENT OFFENDING

The literature is replete with reports of child abuse and neglect in the family backgrounds of violent offenders. Most of these studies document whether a relationship between child abuse and neglect and violent behavior exists and report the magnitude of the association. A number of these studies have involved delinquents (Alfaro, 1981; Pagan & Wexler, 1987; Farrington, 1978; Geller & Ford-Somma, 1984; Gutierres & Reich, 1981; Hartstone & Hansen, 1984; Kratcoski, 1982; Krattschnitt, Heath & Ward, 1986; Lewis, Shanok, Grant, & Ritvo, 1984; Lewis, Shanok, Pincus, & Glaser, 1979; McCord, 1983). Others have described a variety of patient groups (Blount & Chandler, 1979; Climent & Ervin, 1972; Montane, Leichter & Lewis, 1984; Tarter, Hegedus, Winston, & Alderman, 1984). Some of these studies provide support for the linkage between abuse and neglect and later violent behavior; others do not (Alfaro, 1981; Kratcoski, 1982). In some, abused delinquents were less likely to engage in later aggressive crimes (Gutierres & Reich, 1981) or had low incidence of child abuse and parental violence (Pagan, Hansen, & Jang, 1983). In most, the relationship has been found to be consistent but modest.

With few exceptions, these studies have relied on retrospective, unsubstantiated reports of abuse and neglect Most do not incorporate normal control groups into their designs. Since many of the same family and demographic characteristics found in abusive home environments are also related to delinquency and adult criminality, appropriate control groups are necessary to accurately assess the independent effects of childhood victimization. As Monahan (1981) argued, the most important piece of information we can have in the prediction of violence is the base rate of violent behavior in the population with which we are dealing. Without control groups to provide rough estimates of base rates, it is difficult to assess the magnitude of relationships. Widom (1989d) concluded, in a recent review of the cycle of violence literature broadly defined, that "methodological problems play a major role in restricting our knowledge of the long-term consequences" (p. 181).

In the fall of 1986, Widom began a study designed to overcome a number of the methodological problems associated with previous research. Using a prospective cohorts design, official criminal histories for a large sample of substantiated and validated cases of physical and sexual abuse and neglect from the years 1967 through 1971 were compared to those of a matched control group of individuals with no official record of abuse or neglect In analyses restricted to adult behavior, Widom (1989a, 1989b) reported that abused and neglected children demonstrate increased risk of becoming adult criminals and of becoming adult violent criminals, although this was true primarily for males.

CHILDHOOD VICTIMIZATION AND PATTERNS OF VIOLENT OFFENDING

Few published studies have extended the investigation of the relationship of childhood victimization to later violent criminal behavior beyond establishing the relationship and reporting its magnitude. However, there are a number of widely held assumptions regarding violent offenders that may or may not be effected by documented previous abuse and/or neglect in the backgrounds of these violent offenders. For example, violent offenders are primarily male (Duxbury, 1980; Piper, 1985; Strasburg,1984; Weiner & Wolfgang, 1985) and tend to be disproportionately represented by nonwhites (Geller & Ford-Somma, 1984; Piper, 1985; Strasburg, 1984; Wolfgang, Figlio, & Sellin, 1972). A small proportion of chronically violent offenders commit a large proportion of violent crime (Hamparian, Schuster, Dinitz, & Conrad, 1978; Wolfgang et al., 1972).

Furthermore, it is generally assumed that once becoming a violent offender as a juvenile, there is a high probability of continuing as a violent offender as an adult. Violent offenders are assumed to commit their first offense early in their delinquent careers (Hartstone & Hansen, 1984; Wolfgang et al., 1972) and to have an early age of onset of violent offending (Piper, 1985; Wolfgang et al., 1972).

PURPOSE

The purpose of the present article is to examine further the relationship between childhood victimization and violent offending from adolescence through early adulthood. First, the overall relationship between childhood victimization and violent offending is described, followed by an examination of the extent to which there are sex-specific and race-specific effects that mediate the association. This article then describes the extent to which childhood abuse and/or neglect effects other characteristics of violent offending including chronicity, age of onset, patterns of violent offending, and continuity from violence as an adolescent to violence as a young adult. Finally, this article addresses the cycle of violence hypothesis directly by examining whether violent (narrowly defined as physical abuse) experienced as a child leads to later violence as an adult.

METHOD

The data for the present article are part of a larger project, funded by the National Institute of Justice, to study the relationship between child abuse and neglect and violent criminal behavior (Widom, 1989a). Building on past work, this research was designed to incorporate a number of methodological improvements. These included a relatively unambiguous operational definition of abuse and neglect; a prospective design; separate abused and neglected groups; a large sample to allow for subgroup comparisons and to allow for conclusions with respect to violent criminal behavior, a control group matched as closely as posible for age, sex, race, and approximate social class background; and assessment of the long-term consequences of abuse and neglect beyond adolescence or juvenile court into adulthood.

This research is based on a standard design referred to as specialized cohorts (Schulsinger, Mednick,&Knop,1981,p. 12). Matched cohorts that are free of the "disease" in question (violent behavior) at the time they are chosen for the study are assumed to differ only in the attribute to be examined, that is, having experienced abuse or neglect as a child. At present, this research involves (and is limited to) the collection, tabulation, and analysis of existing official records. (For a complete description of the design and subject selection, see Widom, 1989a.)

In comparisons of delinquent or violent behavior it is often difficult to judge what portion of the differences is due to the experience or factors under study and what portion is due to being labeled a delinquent or violent offender. Our research does not totally avoid this problem, but by use of a prospective design, with data collection started at the point of abuse and/or neglect and before the onset of delinquency and violent behavior, the problem is minimized.

In studies of the relationship between child abuse and neglect and later delinquency or criminality, it is also important to avoid ambiguity in the direction of causality of the events. Specifically, cases occur where delinquency precedes abuse and/or neglect, or may cause the abuse or neglect Thus, to minimize this likelihood and to maximize the likelihood that th temporal direction is clear (that is, abuse or neglect -» delinquency or criminality), abuse and neglect cases were restricted to those in which children were less than 11 years of age at the time of the abuse or neglect.

Identification of Abuse and Neglect Cases

Abuse and neglect cases were those brought to the attention of the authorities, and only those validated and substantiated by the court are included. Specifically, all cases of physical and sexual abuse and neglect processed during the years 1967 through 1971 in the county juvenile court in a metropolitan area in the midwest were initially included. Abuse and neglect cases from the adult criminal courts were also included. In these cases, a criminal charge was filed against the adult defendant. During 1967 through 1971, there were 140 cases (physical and sexual abuse and neglect) processed in adult criminal court in which the victim was 11 years of age or less. After examining 2623 abuse and neglect petitions, a total of 908 cases were retained for this study. Reasons for exclusion include (1) adoption of the child as an infant; (2) what we called "involuntary" neglect only1; (3) placement only; and (4) the large number of cases of "failure to pay child support."

Physical abuse, also referred to as "cruelty to children," refers to cases in which an individual had "knowingly and wilfully inflicted unnecessarily severe corporal punishment" or "unnecessary physical suffering" upon a child or children. These cases include injuries such as bruises/welts, burns, abrasions/lacerations, wounds/cuts, bone /skull fractures, and other evidence of physical injury to the child.

Sexual abuse refers to a variety of charges, ranging from relatively nonspecific charges of "assault and battery with intent to gratify sexual desires" to more specific and detailed charges of "fondling/touching in an obscene manner," sodomy, incest, and so forth.

Neglect refers to cases in which the court found a child (or children) to have no proper parent care or guardianship, to be destitute or homeless, or to be living in a physically dangerous environment. The neglect petition reflects the judgment that the behavior represents a serious omission(s) by the parents-beyond acceptable community and professional standards at the time. Neglect cases represent extreme failure to provide adequate food, clothing, shelter, and medical attention to children.

Identification of Matched Control Group

One of the critical elements of this research design is the establishment of a control group matched as closely as possible on the basis of sex, age, race, and approximate family socioeconomic status during the time period under study (1967 through 1971). To accomplish this matching, the sample of abused and neglected cases were divided into two groups on the basis of their age at the time of the abuse or neglect incident* under and over school age.

Two control children were initially selected as matches for the abused or neglected child. This was done because one of the important elements of the design involves the assumption that the major difference between the abused/neglected group and the controls is in the abuse or neglect experience. Since this study is based on official records, the first attempt to gain control was to check the official records to determine if the proposed control subjects had been abused or neglected. If there was evidence that a control subject had been abused, then he or she was excluded from our control group. This occurred in 11 cases.

Children who were under school age at the time of the abuse or neglect were matched with children of the same sex, race, date of birth (+/-1 week), and hospital of birth through the use of county birth record information. Of the 319 abused and neglected children under school age, we have matches for 229 (72%).

For children of school age, elementary school records for the same time period were used to find matches with children of the same sex, race, date of birth, (+/- 6 months), and same class in elementary school during the years 1967 through 1971. Busing had not been implemented and the elementary schools were composed of students from small, socioeconomically homogeneous neighborhoods. Out of 589 school age children, we have matches for 438, representing about 74% of the group. Overall, we have 667 matches (73.7%) for the abused and neglected children.2

Demographic Characteristics

Abuse and Neglect Group. Among the abuse and neglect group, there are about equal numbers of males and females (49% versus 51%) and more whites than blacks (67% versus 31 %). The mean current age for the abused and neglected subjects is 25.69 (SD = 3.53). The majority of the sample are currently between the ages of 20 and 30 (85%), with about 10% under age 20 (the youngest is 16) and 5% older than 30 (the oldest is 32).

Control Group. There are also equal numbers of males and females in the control group, and approximately the same percentage of whites (65%) and blacks (35%). The mean age for the control group is 25.76 (SD = 3.53). As with the abuse and neglect group, most of the controls are between the ages of 20 to 30, with the youngest being 16 and oldest 33.

The current age distribution of both groups in the sample indicates that our design has allowed sufficient time for most of the subjects to come to the attention of authorities for delinquent, adult criminal, and violent behavior (Hartstone & Hansen, 1984; Rojek & Erikson, 1982; Strasburg, 1978; Wolfgang et al., 1972).

Data Collection

Abuse and Neglect Incident. Juvenile court and juvenile probation department files (which processed cases of delinquency, neglect, or dependent children) were examined in order to collect detailed information on the original abuse and/or neglect incident. Incident information included age of victim at the time of the incident type and extent of physical injuries, the characteristics of the perpetrator, and the disposition of the case or placement information.

Juvenile Delinquency. Juvenile probation department files were utilized to collect information on delinquent activity for both the abuse and neglect group as well as the controls. Records of the Bureau of Motor Vehicles were examined to find social security numbers to assist in further tracing of the subjects. Marriage license records were also searched for the females in both groups to find married names.

Adult Criminal History Information. Local, state, and federal agencies were searched for adult criminal histories on both the abuse and neglect group and the control group.

Dependent Variables

Definitions of the dependent variable violent criminal behavior and three age-related terms are as follows: Violent offenses include arrests for murder/attempted murder, manslaughter/ involuntary manslaughter/reckless homicide, rape/sodomy, robbery/robbery with injury, assault/assault and battery/aggravated assault, and battery/battery with injury3. Juvenile violent refers to an arrest for a violent offense committed while the person was less than 18 years of age. Adult violent refers to an arrest for a violent offense committed by a person 18 years of age or older. Any violent refers to any arrest for a violent offense (juvenile or adult).

RESULTS

Table 1 presents the percentage of violent offenders by group (abuse and neglect versus controls) for the three dependent variables-juvenile violent, adult violent, and any violent overall - and by sex and race separately. Interestingly, while the abuse/neglect group had significantly higher frequencies of any violent offending than the nonabused group (X2 = 4.68,4f= 1,p <.05), differences between the groups in the extent of violent offending as a juvenile (juvenile violent) and as an adult (adult violent) did not reach customary levels of significance. In all cases, however, the abuse/neglect group had higher frequencies of violent offending.4

Sex

The percentage of violent offenders by group and sex was also examined (see Table 1). Not unexpectedly, males have higher rates of violent offending than females for all levels of offending. However, in contrast to the existing literature on violent juvenile offenders, we did not find significant differences between abused and neglected and control males as juveniles. Specifically, males in our sample who were abused and neglected were not more likely to commit a violent offense as a juvenile than control males. Abused/neglected males were, however, at greater risk of committing a violent offense as an adult (x2 = 4.82, df = 1, p <.05) and any violent offense (x2 = 4.79, df = l,p <. 05) than control males.

Females who were abused/neglected were not significantly more lilely to commit a violent offense than nonabused females, although there was a tendency for abused/neglected females as juveniles to have a higher frequency of arrests for violent offenses than control females. This pattern contrasts strongly with the pattern for males described earlier. Abuse and neglect did not increase the likelihood of juvenile violence for males, whereas for females there was a trend toward increased violence as juveniles among those abused and neglected.

Race

Table 1 also presents the percentage of violent offenders by group and race. Surprisingly, there were no differences in violent offending between whites who were abused/neglected and those in the control group. For whites, then, being abused or neglected as a child does not appear to increase one's likelihood of being arrested for a violent offense. This contrasts dramatically with our findings for blacks. Blacks who were abused/neglected had higher rates of arrests for violent crimes for the three dependent variables in this analysis than blacks in the control group, although differences were significant only for arrests under the adult violent (x2 = 4.56, df = \,p <.05) and any violent (x2 = 7.22, df = 1,p <.01) categories.

In sum, abused/neglected males had a higher percentage of any violent offender than nonabused males. Blacks who wereabused/neglectedhadhigherpercentagesof any violentoffender than did blacks who were not abused. Abused and neglected females and whites did not show increased propensities for "any violent offending" in comparison to those not abused.

To go beyond simple bivariate tests, three separate logistic regression equations were estimated, predicted to dichotomous outcome variables (see Table 2). Using iterative maximum likelihood methods, three equations were estimated for violent arrests as a juvenile, as an adult, and any violent arrest as the dependent variables, with race, sex, and abused/ neglected versus control group status as explanatory variables.

Coefficients from logit models can be directly related to odds ratio measures. When "dummy" coding is used, exponentiating the coefficients of the association parameters gives the ratio of the odds of a positive response for the included versus the excluded group. As you can see from Table 2, males and blacks had greater odds for a violent arrest as a juvenile than females and whites. For adult arrests for violence and for any arrest for violence, all four predictor variables (sex, age, race, and abuse/neglect status) contributed significantly. The odds for an arrest for a violent crime were highest for the oldest black male who had been abused and/or neglected.

Age of Onset

Age of first officially recorded arrest was used to indicate the age of onset of criminal behavior in the abuse/neglect and control groups (see Table 3). A 2 x 2 analysis of variance was done comparing abused/neglected versus controls and violent versus nonviolent offenders with age at any first arrest (juvenile or adult) as the dependent variable. Both main effects were significant (for abuse/neglect versus controls: F = 5.64, df = 1,563, p <.02; for violent versus nonviolent F =6.l5,df = l,563,p <.02).Theinteraction was not significant. Overall, abused and neglected children began their delinquent careers earlier than controls (16.48 versus 17.30 years) and violent offenders were arrested earlier than nonviolent offenders (16.11 versus 17.01 years).

Given the emphasis here on violent offending and the importance in the criminological literature on age at first violent offense as a predictor of later serious outcomes, we also examined age at first violent offense for abuse/neglected subjects versus controls. A t -test was first done to determine whether, in addition to an earlier age of onset generally, abused/ neglected subjects had an earlier age of onset of violent offending. Surprisingly, abused and neglected subjects and controls overall did not differ significantly on age of first violent arrest (19.05 versus 19.68).

The assumption is often made that the earlier age of onset of delinquent behavior, the greater the probability of continuation and escalation. This is based on the notion that delinquent behavior "untreated" will continue to spread. Or, in learning theory terms, children socialized into violence who manifest this behavior early retain a commitment to such behavior through the life span. For example, Hamparian et al. (1978) found that those first convicted at the earliest ages tended to be the most persistent offenders. While our results indicate that abused/neglected children and violent offenders began their delinquent careers earlier than those nonabused/neglected and nonviolent offenders, our findings also support the general principle that earlier age of onset predicts to more offending. In separate multiple regression analyses controlling for current age of the subjects, age at first offense was significantly negatively correlated with total number of any offenses (ft - ~.32,p <.001) and for total number of any violent offenses ()3 = -.27, p <.001).

Patterns of Violent Offending

Based on a reading of criminal histories, it seemed possible that aggregating the sample of violent offenders and treating them as a homogeneous group might obscure important differences. Since there are a number of pathways through which one can become a violent offender, it also seemed possible that these different patterns may be useful in furthering our understanding of the antecedents and consequences of violent offending and that these patterns should be explored. Using the notion of criminal careers (Blumstein, Cohen, & Farrington, 1988), we examined violent offending over the course of the life span, dividing our data into two distinct time periods-adolescence (approximately ages lOto 17) and early adulthood (ages 18 to 33). Accordingly, there are five patterns of violent offending (illustrated in Table 4). Interestingly, abused/neglected and controls were distributed roughly in proportion to their numbers across the five violent offending patterns (%2 = 4.3, df= 4, ns).

Two separate ANOVAs were done for age at first any offense (Table 4) and for age at first violent offense (Table 5) for the five patterns. For each ANOVA, the pattern of violent offending had a significant impact on age at first offense [Any first (Table 4): F - 50.25 df = 4,145,p <.001; First violent (Table 5): F = 45.89, df = 4,144, p <.001]. In part, this is not terribly surprising because those people with arrests as juveniles obviously had an earlier age at first offense than those with adult arrests only. Among these violent offenders, however, the group effect (abuse/neglect versus controls) did not differentiate for age of onset for any first arrest (F = 1.08, df =1,145, p =.30) or for age of first violent arrest (F =.21, #=1,144, p =.65). The interaction for Any first (Table 4) was significant (F = 3.11, df =4,145, p <.05), but for First violent (Table 5), the interaction was not significant (F =1.72, df =4,144, p =.15).

Some striking findings should be noted. Based on past literature, one might have expected that those in the type I category (violent -> violent) would have begun offending at the earliest age. However, the mean age at Any first arrest is relatively the same across the patterns of violent offending, and the mean age at First violent arrest for this group is later. As a whole, the violent -»violentoffenderswerenotmorelikelytobeginviolentoffending at an earlier age.

Looking at Any first, it is clear that the first four patterns of violent offending (types I-IV: violent -> violent, violent -> nonviolent, violent -> no arrest, and not violent -*? violent) came in contact with the criminal justice system at about the same time, at approximately age 13 to 14. However, the last group (type V: no arrest -» violent) had no official contact as a juvenile (no arrests at all), then became an offender as an adult (mean age = 20.40), and shortly thereafter had an arrest for a violent crime (mean age = 21.90).

Looking at age at First violent arrest, a somewhat different pattern is evident Here, the first three groups (I-III) began violent offending at age 14 to 15 (about a year or so after they began offending in general). These early onset violent offenders represent 37% of the group of violent offenders and 10% of all offenders in our sample. The majority of violent offenders do not follow this pattern of early onset In our sample, 63% of the violent offenders became violent as adults. Looking backwards, of those with violent arrests as adults, only 17% had arrests for violent crimes as juveniles.

Of the entire group of violent offenders, there is one small group of abused/neglected children (within type I, n = 13) who began their criminal careers very early (mean age at Any first arrest = 11.58) and earlier than any of the other groups. With one exception, these are cases of neglect, not abuse. Interestingly, based on our information from the juvenile probation department records, only one of the 13 was described as having behavior problems or being ungovernable, and none were described as having problems with truancy. Looking at available information on other family and demographic characteristics, there appears to be no distinguishing feature of this small group. However, based on these patterns of violent offending, it would appear to be useful to examine more carefully the factors that influence entrance into violent offending at different life stages, and particularly the relevance of these patterns to antecedents.

Exploring the potential usefulness of this temporal typology of violent offending a little further, we examined the extent (average number) to which the five groups were arrested for violent offenses. In two separate ANOVAs, using current age of the subjects as the covariate, the total number of any offenses (see Table 6) and total number of violent offenses (see Table 7) differed significantly across the five violent off ending patterns. Within these violent offenders, two groups (one early onset and one later onset) account for a higher rate of offending, a finding consistent with previous literature. However, in terms of prevention of further violence, Group III (violent-» not violent) ought to be studied since they were violent as juveniles but appear to desist from violence as adults.

Chronicity

Along with the notion that violent offenders begin their criminal careers early, there is another assumption that violent offenders are more likely to be chronic offenders. The extent of chronicity of these violent offenders was examined, and these findings are presented in Table 8. Following Wolfgang et al. (1972), we divided our subjects into three groups as follows: none (no arrest), one (one arrest for a violent crime), and chronic or frequent (two or more arrests for a violent crime).

As juveniles, there were significantly more chronic violent offenders among the abused and neglected children than among the controls, by a factor of about 12 to 1 (1.2 versus .1). However, as adults, the groups did not differ in chronicity (3.5% versus 2.1%), and differences in chronicity over the life span (ever having an arrest for a violent crime-as a juvenile or as an adult), only approached significance (p <.10).

Continuity

A related assumption is that there is continuity in criminal careers-that once a person begins a delinquent career, the likelihood is high that it will continue. Given that abused and/ orneglected subjects are at higher risk for delinquency andadultcriminality(Widom,1989c), it might also be expected that they would be more likely to continue in crime, unable to break the cycle of violence. Table 9 presents our findings on the continuity between juvenile antisocial behavior and adult criminal behavior for any arrests and for violent arrests separately for abused and neglected subjects and for controls.

Of those with juvenile offenses, roughly the same proportion of abused and neglected children and controls go on to commit offenses as an adult (53% versus 50%). This compares with about 20% of the abused and neglected subjects, and 15% of the controls who did not have an arrest as a juvenile. Of those with violent offenses as juveniles, approximately the same proportion go on to commit violence as an adult in the abused and neglected group (34%) as do in the controls (37%). Despite major differences in the extent of involvement in criminal activity, non-abused and neglected subjects are just as likely as abused and neglected individuals to continue criminal activity once they have begun.

DISCUSSION

Early childhood victimization demonstrably increases the risk of violent offending through the life span; however, the effects are not distributed similarly across the sexes and races. Increased risk of ever becoming a violent offender was found primarily in abused and neglected males compared to control males. Interestingly, as juveniles abused and neglected males did not show higher rates of violent offending than controls.

In contrast to the effects of early childhood victimization on females in terms of increased risk for delinquency and adult criminality in general (Widom, 1989c), adult females who were abused and neglected were not more likely to commit a violent offense than nonabused females. As juveniles, though, abused and neglected females tended to be at greater risk of being arrested for a violent offense than control females. Because of the very low rates of violent offending in females, the lack of significant differences is not surprising. However, the pattern of this difference (in contrast to that found for males) is very provocative. Although this may be a spurious finding, related to an increased surveillance by the justice system of abused and neglected females, further exploration of this unexpected result is planned.

The long-term consequences of childhood victimization for violent offending differed dramatically by race. For whites, abused and neglected children did not have significantly higher rates of violent arrests than the controls. In contrast, blacks who were abused and neglected, compared to blacks who were not abused, had substantially higher arrest rates for violent offending (juvenile, adult, and any). This striking finding warrants discussion and further examination. Because these findings are based on official records and because official records overrepresent minority groups, the most obvious explanation for the higher rates of violent arrests among blacks is that these results reflect bias and discrimination within the criminal justice system. However, this explanation does not seem to explain the differences within blacks and the lack of differences within whites.

Other possible explanations for these race-specific findings are that parental violence is more severe among blacks than whites or that nonwhites are more physically abusive with their children and within their homes than whites. To the extent that we can address these issues.however.ourdataindicatethatthisisnof the case. Within whites, approximately 20% suffered physical abuse, compared to the percentage for blacks of less than 9%. Blacks suffered more neglect relative to whites in our sample.

Furthermore, based on data from the National Study on the Incidence and Severity of Child Abuse and Neglect (National Center on Child Abuse and Neglect, 1981) and the National Study on Child Neglect and Abuse Reporting (American Humane Association, 1984), the percentage of black and white children reported for child abuse and neglect appears to be representative of the United States population at large. If anything, black families were characterized by more neglect and slightly less physical and sexual abuse (as found in our study), whereas white families were characterized by more abuse or combined abuse/ neglect Thus racial bias in official reports showing more blacks being reported for abuse or neglect was not found in the NCCAN and AHA studies.

It is possible that another type of discriminatory treatment by the system might be at work here. Abused and neglected black children in our sample could have been subjected to more extreme victimization than white children before coming to the attention of officials at the time, if we postulate a justice system that expects and tolerates higher levels of violence among black families. Following this scenario, only the most extreme cases of physical abuse in black families would be reported to the authorities, and thus these cases would not be comparable to cases of physical abuse from white families. Unfortunately, our data do not permit examination of this question. Given our very dramatic findings about these racespecific effects and that other researchers have found background experiences associated with violent crimes to vary depending on the person's race (Kruttschnitt et al., 1986), future research should examine these issues.

Age of onset of offending is believed to have important implications for future criminality (Robin, 1964). We found that abused and neglected children and violent offenders began their delinquent careers earlier than nonabused and neglected children and nonviolent offenders. Within the violent offenders in our sample, we also found that earlier age of onset was negatively associated with more arrests in general and for violent crimes in particular.

In a survey of violent juveniles, Hamparian et al. (1978) found that the mean age at first arrest for violence was 14.4, in comparison to the average age of 13 for first arrests generally. In England, Wadsworth (1979) also found that the first conviction for violence occurred at a later age than the first conviction for other offenses. We also found that arrest for violence occurred later than more general arrests (about a year later on average). However, for the group of violent offenders as a whole, the mean age is 16 for any first arrest and 17 for first arrest for a violent crime. This is much later than age of onset of arrests reported in studies of violent/wvem'fe offenders, such as that from Hamparian etal. (1978)describedabove. Age of onset varies dramatically with the pattern of violent offending. Our findings suggest that a typology based on temporal patterns of violent offending, in line with the construct of criminal careers, may be useful in examining factors associated with violent careers and in the development and testing of theories of violent offending. Based on our findings, crosssectional glimpses into the careers of violent offenders miss these differences and may in fact be misleading.

Despite the fact that victims of early child abuse and neglect differ from nonabused and neglected children on a number of indices of violent offending, not all aspects of violent offending differentiate the groups. One example of such a similarity between the groups is provided by our finding of no differences between the groups in continuity between arrests for violent crimes as a juvenile and violence as an adult Despite significant differences in the extent of involvement in violence and the age of onset, nonabused and neglected children are just as likely as abused and neglected children to continue violent offending once they have begun.

These findings are interesting in connection with recent literature in criminology, particularly debates on criminal careers (Blumstein et al., 1988; Gottfredson & Hirschi, 1988). While our findings indicate that officially recorded abuse and neglect increase the likelihood of having an official criminal record and speed up entrance into officially recorded delinquent activities, early childhood victimization does not appear to place one at increased risk for continuing in a life of crime. These findings lend weight to the importance of distinguishing factors that may stimulate individuals to become involved in crime and violence from factors that affect the continuity or discontinuity of criminal careers. Furthermore, these findings suggest that there may be aspects of criminal careers that are relatively resistant and not affected by factors such as early childhood victimization.

This study suffers a number of limitations that should be kept in mind in interpreting these findings. Pagelow (1982) has suggested that labeling children involved in officially recorded cases of abuse and neglect, in many cases disrupting their residence with their families and stigmatizing their parents, may create a self-fulfilling prophecy that may be difficult to resist or overcome. Because this research is based on official records of both independent (abuse and neglect) and dependent variables (violent offending), generalizing to different types of abuse and neglect cases should be avoided (Widom, 1988). For example, one cannot generalize from these findings to unreported cases of abuse and neglect, and especially to cases handled unofficially by private medical doctors. In addition, because of our exclusions, these findings are not generalizable to cases of abused and neglected children who were adopted in early childhood. It is also important to determine the extent to which all subjects (abused and neglected and controls) report having experienced child abuse or neglect and contacts with criminal justice agencies that have not been disclosed through official records.

Our findings support some frequently held assumptions about violent offending, but they also seriously challenge some popular assumptions. While early child abuse and neglect increase the risk of violent offending, not all childhood victims become violent offenders. While there seems to be a small group of chronically violent offenders, most violent juveniles do not become adult violent offenders. Although abused and neglected children are at increased risk for violent offending, the relationship between early childhood victimization and later violent offending is not universal. Thus further research is needed to understand what accounts for the differences in children who succumb to violent behavior and those who appear relatively invulnerable to these long-term consequences.

One obvious place to begin intervention programs would be to focus on the early onset violent offender (type I) who commits a large number of general and violent offenses. However, before initiating such programs, it is important to understand more about this small group-their family backgrounds, personal characteristics, social supports, and so forth-factors that may have contributed to the maintenance of their delinquent and violent criminal behavior. At the same time, we need to examine the lives of the individuals in the other groups (II and HI) who were violent as juveniles but were not rearrested for violent crimes as adults. Unfortunately, designs such as this one that depend on archival records are weak in understanding process, since they are typically undertaken to record performance out-comes and not the processes mediating performance (Cook & Campbell, 1979). We are beginning a follow-up study of these violent offenders to remedy some of the limitations of the current study and to examine more carefully the roles of risk and protective factors in the lives of these abused and neglected and control children.

-1-