Research has documented the magnitude of youth violence and the trends in that violence over time. But what do we know about why young people become involved in violence? Why do some youths get caught up in violence while others do not? There is no simple answer to these questions, but scientists have identified a number of things that put children and adolescents at risk of violent behavior and some things that seem to protect them from the effects of risk.
The concepts of risk and protection are integral to public health. A risk factor is anything that increases the probability that a person will suffer harm. A protective factor is something that decreases the potential harmful effect of a risk factor. In the context of this report, risk factors increase the probability that a young person will become violent, while protective factors buffer the young person against those risks. The public health approach to youth violence involves identifying risk and protective factors, determining how they work, making the public aware of these findings, and designing programs to prevent or stop the violence.
Risk factors for violence are not static. Their predictive value changes depending on when they occur in a young person's development, in what social context, and under what circumstances. Risk factors may be found in the individual, the environment, or the individual's ability to respond to the demands or requirements of the environment. Some factors come into play during childhood or even earlier, whereas others do not appear until adolescence. Some involve the family, others the neighborhood, the school, or the peer group. Some become less important as a person matures, while others persist throughout the life span. To complicate the picture even further, some factors may constitute risks during one stage of development but not another. Finally, the factors that predict the onset of violence are not necessarily the same as those that predict the continuation or cessation of violence.
Violence prevention and intervention efforts hinge on identifying risk and protective factors and determining when in the course of development they emerge. To be effective, such efforts must be appropriate to a youth's stage of development. A program that is effective in childhood may be ineffective in adolescence and vice versa. Moreover, the risk and protective factors targeted by violence prevention programs may be different from those targeted by intervention programs, which are designed to prevent the reoccurrence of violence.
This report groups risk and protective factors into five domains: individual, family, peer group, school, and community, which includes both the neighborhood and the larger society (Box 4-1). Factors do not always fit neatly into these areas, however. Broken homes are classified as a family risk factor, but the presence of many such families in a community can contribute to social disorganization, an important community-level risk factor (Bursik & Grasmick, 1993; Elliott et al., 1996; Sampson & Lauritsen, 1994).
Risk factors are not necessarily causes. Researchers identify risk factors for youth violence by tracking the development of children and adolescents over the first two decades of life and measuring how frequently particular personal characteristics and social conditions at a given age are linked to violence at later stages of the life course. Evidence for these characteristics and social conditions must go beyond simple empirical relationships, however. To be considered risk factors, they must have both a theoretical rationale and a demonstrated ability to predict violence -- essential conditions for a causal relationship (Earls, 1994; Kraemer et al., 1997; Thornberry, 1998). The reason risk factors are not considered causes is that, in most cases, scientists lack experimental evidence that changing a risk factor produces changes in the onset or rate of violence.
As used in this report, risk factors are personal characteristics or environmental conditions that predict the onset, continuity, or escalation of violence.
The question of causality has practical implications for prevention efforts. Prevention depends largely on risk factors being true causes of violence. In practical terms, research has amassed enough strong, consistent evidence for the risk factors discussed in this report to provide a basis for prevention programs, even though a strict cause-and-effect relationship has been established for relatively few of them.
Most of the risk factors identified do not appear to have a strong biological basis. Instead, it is theorized, they result from social learning or the combination of social learning and biological processes. This means that violent youths who have violent parents are far more likely to have modeled their behavior on their parents' behavior -- to have learned violent behavior from them -- than simply to have inherited it from them. Likewise, society's differing expectations of boys and girls -- expecting boys to be more aggressive, for example -- can result in learned behaviors that increase or decrease the risk of violence.
The bulk of the research that has been done on risk factors identifies and measures their predictive value separately, without taking into account the influence of other risk factors. More important than any individual factor, however, is the accumulation of risk factors. Risk factors usually exist in clusters, not in isolation. Children who are abused or neglected, for example, tend to be in poor families with single parents living in disadvantaged neighborhoods beset with violence, drug use, and crime. Studies of multiple risk factors have found that they have independent, additive effects -- that is, the more risk factors a child is exposed to, the greater the likelihood that he or she will become violent. One study, for example, has found that a 10-year-old exposed to 6 or more risk factors is 10 times as likely to be violent by age 18 as a 10-year-old exposed to only one factor (Herrenkohl et al., 2000).
Researchers have theorized that risk factors also interact with each other, but to date they have found little evidence of interaction. What evidence does exist suggests that interactions between or among factors produce only small effects, but work in this area is continuing. To date, much more research has been done on risk factors than protective factors, but that picture, too, is changing.
Scientific theory and research take two different approaches to how youth violence develops -- one that focuses on the onset of violent behavior and its frequency, patterns, and continuity over the life course and one that focuses on the emergence of risk factors at different stages of the life course. Chapter 3 describes two developmental trajectories for the onset of violent behavior -- one in which violence begins in childhood (before puberty) and continues into adolescence, and one in which violence begins in adolescence.
In contrast, this chapter considers the timing of risk factors. It identifies the individual characteristics, experiences, and environmental conditions in childhood or adolescence that predict involvement in violent behavior in late adolescence -- that is, age 15 to 18, the peak years of offending. Research shows that different risk factors may emerge in these two developmental periods and that the same risk factors may have different effect sizes, or predictive power, in these periods.
The timing of risk factors and the onset of violence are connected. Only risk factors that emerge in early childhood can logically account for violence that begins before puberty. However, these early risk factors may or may not be implicated in violence that begins in adolescence. In fact, studies show that many youths with late-onset violence did not encounter the childhood risk factors responsible for early-onset violence. For these youths, risk factors for violence emerged in adolescence (Huizinga et al., 1995; Moffitt et al., 1996; Patterson & Yoerger, 1997; Simons et al., 1994).
Table 4-1 lists early and late risk factors and estimates their effect sizes for violence at age 15 to 18. It does not distinguish between youths who became violent before puberty and those who first became violent in adolescence; both groups are included among youths who were violent in late adolescence. However, the table does indicate that different risk factors emerge before puberty (age 6 to 11) and after puberty (age 12 to 14) and that the same risk factors have different effect sizes in these periods. Thus, for example, the table shows that substance use in childhood has a greater effect on violence at age 15 to 18 than parental abuse or neglect does and that substance use in childhood has a greater effect on violence than substance use in early adolescence. (The table is discussed at greater length below, in A Note on Sources.)
The distinction between early and late risk factors is important. To be effective, prevention programs must address the risk factors that appear at a particular stage of development. The observed clustering of risk factors in childhood and in adolescence provides clear targets for intervention during these stages of the life course.
Risk factors are powerful tools for identifying and locating populations and individuals with a high potential for becoming violent, and they provide valuable targets for programs aimed at preventing or reducing violence. But there are important limitations to our knowledge about and use of risk factors.
The following cautions are worth bearing in mind:
Nonetheless, most of the risk factors identified in this report do apply broadly to all young people. All children go through the same basic stages of human development -- and prevention of youth violence is based on understanding when and how risk factors come into play at various stages of development. Moreover, there is some evidence that most risk factors are equally valid predictors of delinquency and violence regardless of sex, race, or ethnicity (Rosay et al., 2000; Williams et al., 1999). Sophisticated studies that identify how cultural differences affect the interplay of the individual and his or her surroundings will make possible more effective prevention efforts.
There is some disagreement about exactly what protective factors are. They have been viewed both as the absence of risk and as something conceptually distinct from risk (Guerra, 1998; Jessor et al., 1995; Reiss & Roth, 1993; Wasserman & Miller, 1998). The former view typically places risk and protective factors on the opposite ends of a continuum. For example, good parent-child relations might be considered a protective factor because it is the opposite of poor parent-child relations, a known risk factor. But a simple linear relationship of this sort (where the risk of violence decreases as parent-child relations improve) blurs the distinction between risk and protection, making them essentially the same thing. 1
The view that protection is conceptually distinct from risk (the view used in this report) defines protective factors as characteristics or conditions that interact with risk factors to reduce their influence on violent behavior (Garmezy, 1985; Rutter, 1985; Stattin & Magnusson, 1996). For example, low family socioeconomic status is a risk factor for violence, and a warm, supportive relationship with a parent may be a protective factor. The warm relationship does not improve the child's economic status, but it does buffer the child from some of the adverse effects of poverty. Protective factors may or may not have a direct effect on violence (compare Jessor et al., 1995 and Stattin & Magnusson, 1996).
Interest in protective factors emerged from research in the field of developmental psychopathology. Investigators observed that children with exposure to multiple risk factors often escaped their impact. This led to a search for the characteristics or conditions that might confer resilience -- that is, factors that moderate or buffer the effects of risk (Davis, 1999; Garmezy, 1985; Rutter, 1987; Werner, 1989). Protective factors offer an explanation for why children and adolescents who face the same degree of risk may be affected differently.
The concept of protective factors is familiar in public health. Wearing seat belts, for example, reduces the risk of serious injury or death in a car crash. Identifying and measuring the effects of protective factors is a new area of violence research, and information about these factors is limited. Because they buffer the effect of risk factors, protective factors are an important tool in violence prevention.
Like risk factors, proposed protective factors are grouped into individual, family, school, peer group, and community categories. They may differ at various stages of development, they may interact, and they may exert cumulative effects (Catalano et al., 1998; Furstenberg et al., 1999; Garmezy, 1985; Jessor et al., 1995; Rutter, 1979; Sameroff et al., 1993; Thornberry et al., 1995). Just as risk factors do not necessarily cause an individual child or young person to become violent, protective factors do not guarantee that an individual child or young person will not become violent. They reduce the probability that groups of young people facing a risk factor or factors will become involved in violence.
This chapter draws heavily on four important studies: Lipsey & Derzon's meta-analysis of 34 longitudinal studies on risk factors for violence (1998); Hawkins et al.'s study of malleable risk and protective factors drawn from 30 longitudinal studies, including some not included in the Lipsey & Derzon meta-analysis (Hawkins et al., 1998c); Paik and Comstock's meta-analysis of 217 studies of exposure to media violence and its effects on aggression and violence (1994); and the National Institute of Mental Health's Taking Stock report (Hann & Borek, in press), an extensive review of research on risk factors for aggression and other behavior problems.
Table 4-1 is adapted from the tables presented in the Lipsey and Derzon and Hawkins et al. meta-analyses. The risk factors in Table 4-1 predict felonies -- that is, violent and property crimes -- at ages 15 to 18, the peak years of involvement. 2 Entries in bold are effect sizes from the meta-analyses by Lipsey and Derzon, Hawkins et al., and Paik and Comstock for various classes of risk factors; other entries are effect sizes reported in two or more longitudinal studies. (Risk classes are described in Appendix 4-A and later sections of this chapter.) Some of the risk classes in Table 4-1 include several separate risk factors. For example, psychological condition includes hyperactivity, daring, and attention problems.
Additional risk factors and classes of risk factors have been added from other sources. For example, there is adequate evidence to establish harsh, lax, or inconsistent discipline as a separate risk factor, although Lipsey and Derzon include it in the poor parent-child relations class. Academic failure, family conflict, and belonging to a gang are additional examples of risk factors not included in any of the meta-analyses.
The measure of effect size used in these tables is a bivariate correlation (r), or simple correlation between two variables. All estimates of effect size are statistically significant and are based on multiple studies, with those for risk classes typically involving more studies than those for separate risk factors. The studies reviewed in Lipsey and Derzon, Paik and Comstock, and Hawkins et al. are not cited here; however, other studies that were used to establish a risk factor or that are included in estimates of effect size are cited. 3
There is a rich and extensive body of research on risks for antisocial behavior, externalizing behavior, conduct disorder, and aggression (Hann & Borek, in press). Each of these terms defines a pattern or set of behaviors that includes aggressive or violent behavior, but most of the behaviors included are either nonphysical, nonviolent acts or relatively minor forms of physical aggression. Risk factors for antisocial behavior may be quite different from those that predict violent behavior (robbery, aggravated assault, rape, and homicide). Since antisocial behavior does not present the potential for serious injury or death that violence does, this report relies on studies that identify risk factors for serious offenses generally and violent behavior specifically, bearing in mind that the vast majority of serious offenders report having been involved in violent offenses.
Risk and protective factors can be found in every area of a child or adolescent's life, they exert different effects at different stages of development, and they gain strength in numbers. The public health approach to the problem of youth violence seeks to identify risk and protective factors, determine when in the life course they typically occur and how they operate, and enable researchers to design preventive programs to be put in place at just the right time to be most effective.
This chapter describes what is known about individual, family, school, peer group, and community risk and protective factors that exert their effects in childhood and adolescence. It describes the power of early risk factors, which come into play before puberty, and late risk factors, which exert their influence after puberty, to predict the likelihood of youth violence.
The first decade of life encompasses a vast period of human development. Infants form attachments to parents or other loving adults and begin to become aware of themselves as separate beings. As toddlers, they begin to talk, to assert themselves, to explore the world around them, and to extend their emotional and social bonds to people other than their parents.
The start of school is a milestone in children's continuing social and intellectual development. Other children become more important in their lives, though still not as important as family members. They begin to empathize with others and hone their sense of right and wrong. As they progress through elementary school, children gain valuable reasoning and problem-solving skills as well as social skills.
Exposure to or involvement in violence can disrupt normal development of both children and adolescents, with profound effects on their mental, physical, and emotional health. 4 In addition, exposure to violence affects children and adolescents differently at different stages of development (Marans & Adelman, 1997).
Young children exposed to violence may have nightmares or be afraid to go to sleep, fear being left alone, or regress to earlier behavior, such as baby talk or bed-wetting. They may exhibit excessive irritability or excitability. Violence in the family, especially, may inhibit young children's ability to form trusting relationships and develop independence.
Elementary school children who live in violent neighborhoods may also experience sleep disturbances and be less likely to explore their environment. In addition, they can become frightened, anxious, depressed, and aggressive. They may have trouble concentrating in school. Because they understand that violence is intentional, they may worry about what they could have done to prevent or stop it (Osofsky, 1999).
Violence also affects parents. Adults living in violent households or neighborhoods may not be able to keep their children safe or to protect them from harmful influences. Some parents living in unsafe neighborhoods do not let their children play outside. While this solution may safeguard children temporarily, it can also impede healthy development. Parents in these situations understandably feel helpless and hopeless. Those who have been traumatized by violence themselves may, like their children, become anxious, withdrawn, or depressed. Under such circumstances, parents cannot respond spontaneously and joyously to their children, making it difficult for children to develop strong, secure attachments to their parents. Forming a bond with a loving, responsive parent or other adult caregiver is an essential factor in healthy development (Furstenberg et al., 1999; Osofsky, 1999; Patterson & Yoerger, 1997). Children and families exposed to or involved in violence may want to seek professional advice in addressing their mental, physical, and emotional health concerns.
A few risk factors for youth violence occur before birth. Others come into play as the child develops in response to his or her family and surroundings. Thus, most of the risk factors that exert an effect before puberty are found in the individual and family domains rather than in the larger world, a situation that changes dramatically in adolescence. Childhood risk factors are listed by domain in Box 4-1; effect sizes are listed in Table 4-1.
The most powerful early risk factors for violence at age 15 to 18 are involvement in general offenses and substance use before age 12. General offenses include serious, but not necessarily violent acts, such as burglary, grand theft, extortion, and conviction for a felony. Children engaging in such crimes often come to the attention of the police and juvenile justice system. Numerous studies have documented the overlap between serious nonviolent and violent offenses in adolescence, so early involvement in serious offenses carries a substantial risk for violence later.
Experimentation with drugs, alcohol, tobacco, or some combination of these substances is not particularly unusual by age 18, but use of these substances by children under the age of 12 is. Not only are these substances harmful to health, they are illegal. Thus, use of these substances signals antisocial attitudes and early involvement in a delinquent lifestyle that often comes to include violent behavior in adolescence (Fagan, 1993).
Two moderate risk factors emerge in childhood, being male and aggression. Boys (and young men) are far more likely than girls to be violent (see Chapter 2), yet some researchers have suggested that sex is a risk marker rather than a risk factor (Earls, 1994; Hawkins et al., 1998a; Kraemer et al., 1997). A risk marker is a characteristic or condition that is associated with known risk factors but exerts no causal influence of its own (Earls, 1994; Patterson & Yoerger, 1997). 5 For example, many more boys than girls are hyperactive, a risk factor with a small effect size, so some of the predictive power of being male may actually be the influence of hyperactivity. Moreover, boys have traditionally been exposed to more violence than girls, and socially approved male role models are more aggressive, suggesting that social learning plays a role in this risk factor. However, research indicates that being male confers risk even after accounting for other known risk factors. This suggests that being male is a risk factor rather than a risk marker, perhaps with some biological or biological-environmental interaction as the causal mechanism.
Many studies have found aggression -- characterized as aggressive and disruptive behavior, verbal aggression, and aggression toward objects -- to be a moderate risk factor among boys, although there is some evidence that physical aggressiveness is actually responsible for most of the observed effect (Nagin & Tremblay, 1999). Additional research is needed to sort out the unique influence of each of these types of aggression.
The remaining individual risk factors have relatively small effect sizes. Various psychological conditions, such as hyperactivity, impulsiveness, daring, and short attention span, pose a small risk for violence. A consistent individual predictor is hyperactivity/low attention, the central components of attention-deficit/hyperactivity disorder (ADHD), a cognitive disorder that may be genetically influenced in some way (Hawkins et al., 1998a). ADHD is characterized by restlessness, excessive activity, and difficulty paying attention, traits that may also contribute to low academic performance, a risk factor in school. Hyperactivity is often found in combination with physical aggression, another risk factor. Some researchers question the independent effect of hyperactivity on later violence, suggesting that the effect is actually physical aggression (and perhaps low academic performance) that was not controlled for in earlier studies of hyperactivity (Nagin & Tremblay, 1999). There is little agreement about the mechanism linking hyperactivity to violence.
The effects of children's exposure to television and film violence have been studied extensively in regard to aggression, but there is relatively little research regarding the effects on more serious forms of violent behavior (for an extended discussion, see Appendix 4-B). Experimental studies have found that exposure to media violence has a small average effect size (.13) on serious forms of violence (Paik & Comstock, 1994); the average effect size in cross-sectional survey studies was very small (.06). Two frequently cited longitudinal studies have examined the effects that exposure to television violence in childhood produces on violent behavior during adolescence or early adulthood. One, in which participants reported having punched, beaten, or choked someone as young adults, found a significant predictive effect for women (.22) but no significant effect for men (Huesmann et al., (submitted)). The other study, in which teenage males reported being involved in a knife fight, car theft, mugging, gang fight, or similar delinquent behavior, found a statistically significant predictive effect in only one of nine tests (Milavsky et al., 1982). Exposure to violence appears to have a weak predictive effect on relatively immediate violence in experimental studies, but there is little consistent evidence to date for a long-term predictive effect.
Little research has been done on violence in other media -- video games, music videos, and the Internet. A recent meta-analysis by Anderson and Bushman (in press) reports that video game violence has a small average effect size (.19) on physical aggression in experimental and cross-sectional studies. Theoretically, the influence of these interactive media might well be greater than that of television and films, which present a passive form of exposure, but there are no studies to date of the effects of exposure to these types of media violence and violent behavior.
Problem behavior, another risk factor with a small effect size, refers to relatively minor problem behaviors such as stealing, truancy, disobedience, and temper tantrums. While not serious in themselves, antisocial behaviors may set the stage for more serious nonviolent or violent behavior later.
The medical or physical risk factor includes a number of conditions that as a group are somewhat predictive of violence. Prenatal and early postnatal complications, a more specific set of medical conditions, have been found to have inconsistent effects across a number of studies (Hawkins et al., 1998c). These complications encompass a broad group of genetic conditions or physical injuries to the brain and nervous system that interfere with normal development, including low birth weight, oxygen deprivation, and exposure to toxins such as lead, alcohol, or drugs (Hawkins et al., 1998b). Low resting heart rate, a condition that has been studied primarily in boys, is associated with fearlessness or stimulation seeking, both characteristics that may predispose them to aggression and violence (Raine et al., 1997; Hawkins et al., 1998c), but there is not enough evidence to establish this condition as a risk factor for violence. Some studies have even questioned its effects on aggression (Van Hulle et al., 2000; Wadsworth, 1976; Kindlon et al., 1995). There is also no evidence that internalizing disorders -- nervousness and withdrawal, anxiety, and worrying -- are related to later violence (Hawkins et al., 1998c).
Low IQ, or low intelligence, includes learning problems and poor language ability. This risk factor has a small effect size and is often accompanied by other risk factors with small effect sizes, such as hyperactivity/low attention and poor performance in school.
Antisocial beliefs and attitudes, including dishonesty, rule-breaking, hostility to police, and a generally favorable attitude toward violence, usually constitute a risk factor in adolescence, not childhood (Hawkins et al., 1998c). Only dishonesty in childhood is predictive of later violence or delinquency, and its effect is small.
There are no known strong risk factors for youth violence in the family domain, but low socioeconomic status/poverty and having antisocial parents are moderate factors. Socioeconomic status generally refers to parents' education and occupation as well as their income. Poorly educated parents may be unable to help their children with schoolwork, for example, and children living in poor neighborhoods generally have less access to recreational and cultural opportunities. In addition, many poor families live in violent neighborhoods, and exposure to violence can adversely affect both parents and children, as described above. Limited social and economic resources contribute to parental stress, child abuse and neglect, damaged parent-child relations, and family breakup -- all risk factors with small effects in childhood.
Studies suggest that antisocial parents -- that is, violent, criminal parents -- represent an environmental rather than a genetic risk factor (Moffitt, 1987). In other words, children learn violent behavior by observing their parents rather than by inheriting a propensity for violence. In fact, attachment to parents, a possible protective factor, can have the opposite effect if the parents are violent (Hawkins et al., 1998c).
Among the early risk factors with small effect sizes on youth violence is poor parent-child relations. One specific risk factor in this class -- harsh, lax, or inconsistent discipline -- is also somewhat predictive of later violence (Hawkins et al., 1998c). Children need reasonable, consistent discipline to establish the boundaries of acceptable and unacceptable behavior. Children who are treated harshly may view rough treatment as acceptable, those who are given no guidance may engage in whatever behavior gets them what they want, and children who receive mixed signals are completely at sea regarding appropriate behavior. Other family conditions, such as high stress, large size, and marital discord, also exert a small effect on later violence.
Another childhood predictor with a small effect size is broken homes, a category that includes divorced, separated, or never-married parents and a child's separation from parents before age 16. Separation from parents also operates as a distinct risk factor, again with a small effect size.
Abusive parenting in general and neglect in particular are predictors of later violence, but they have very small effect sizes. Neglect operates as a distinct risk factor, possibly because neglected children are less likely to be supervised or taught appropriate behavior. This is not to imply that child abuse and neglect do not cause serious problems in adolescence: Indeed, they have large effects on mental health problems, substance abuse, and poor school performance (Belsky & Vondra, 1987; Cicchetti & Toth, 1995; Dembo et al., 1992; Esbensen & Huizinga, 1991; Silverman, et al., 1996; Smith & Thornberry, 1995). This finding is discussed in more detail below, in the section on unexpected findings and effects.
The only early risk factor in the school domain is poor attitude toward and performance in school, and its effects are small. Numerous individual and family factors may contribute to poor performance, making it a fairly broad measure. For example, a child who is physically aggressive and is rejected by peers or who has difficulty concentrating or sitting still in class may understandably have difficulty performing academic tasks. Children who have been exposed to violence, as noted earlier, may also have trouble concentrating in school.
Young children do not socialize extensively with other children and are not strongly influenced by peers. Peers become more important as children progress through elementary school, although school-age children still look primarily to parents for cues on how to behave. Nonetheless, weak social ties to conventional peers and associating with antisocial peers both exert small effects in childhood.
Children with weak social ties are those who attend few social activities and have low popularity with conventional peers. School-age children often reject physically aggressive children because of their inappropriate behavior (Hann & Borek, in press; Reiss & Roth, 1993). The combination of rejection and aggressiveness exacerbates behavior problems, making it more difficult for aggressive children to form positive relationships with other children. Indeed, recent research indicates that children who are both aggressive and rejected show poorer adjustment in elementary school than children who are aggressive, rejected, or neither (Hann & Borek, in press).
Being drawn to antisocial peers may introduce or reinforce antisocial attitudes and behavior in children. Indeed, aggressive children tend to seek each other out (Hann & Borek, in press).
Community risk factors, such as living in socially disorganized neighborhoods or neighborhoods with high rates of crime, violence, and drugs, are not powerful individual-level predictors in childhood because these external influences have less direct impact on children than on adolescents. They may well exert indirect influences through poor parenting practices, lack of family resources, and parent criminality or antisocial behavior.
The most powerful early predictors of violence at age 15 to 18 are involvement in general offenses (serious, but not necessarily violent, criminal acts) and substance use. Moderate factors are being male, aggressiveness, low family socioeconomic status/poverty, and antisocial parents.
Violence increases dramatically in the second decade of life, peaking during late adolescence at 12 to 20 percent of all young people and dropping off again sharply by the early twenties. Some of these youths followed the childhood-onset trajectory, becoming violent before puberty and escalating their rate of offending during adolescence. But over half of all violent youths begin their violent behavior in mid- to late adolescence. These youths gave little indication of problem behavior in childhood and did not have poor relations with their parents.
There are numerous theories about why violence begins in adolescence, but a few themes run through most of them (Elliott & Tolan, 1999; Pepler & Slaby, 1994). Developmentally, puberty is accompanied by major physical and emotional changes that alter a young person's relationships and patterns of interaction with others. The transition into adolescence begins the move toward independence from parents and the need to establish one's own values, personal and sexual identity, and the skills and competencies needed to compete in adult society. Independence requires young people to renegotiate family rules and degree of supervision by parents, a process that can generate conflict and withdrawal from parents. At the same time, social networks expand, and relationships with peers and adults in new social contexts equal or exceed in importance the relationships with parents. The criteria for success and acceptance among peers and adults change.
Adapting to all of these changes in relationships, social contexts, status, and performance criteria can generate great stress, feelings of rejection, and anger at perceived or real failure. Young people may be attracted to violent behavior as a way of asserting their independence of the adult world and its rules, as a way of gaining the attention and respect of peers, as a way of compensating for limited personal competencies, or as a response to restricted opportunities for success at school or in the community. Good relationships with parents during childhood will help in a successful transition to adolescence, but they do not guarantee it.
Adolescents exposed to violence at home may experience some of the same emotions and difficulties as younger school-age children -- for example, fear, guilt, anxiety, depression, and trouble concentrating in school. In addition, adolescents may feel more vulnerable to violence from peers at school or gangs in their neighborhood and hopeless about their lives and their odds of surviving to adulthood. These young people may not experience the growing feelings of competence that are important at their stage of development. Ultimately, their exposure to violence may lead them to become violent themselves. Studies have shown that adolescents exposed to violence are more likely to engage in violent acts, often as preemptive strikes in the face of a perceived threat (Fagan & Wilkinson, 1998; Loeber et al., 1998; Singer et al., 1994, 1995).
Not surprisingly, different risk factors for violence assume importance in adolescence. Family factors lose predictive value relative to peer-oriented risk factors such as weak social ties to conventional peers, antisocial or delinquent friends, and membership in a gang (Table 4-1). Even involvement in general offenses, which had the largest effect size in childhood, has only a moderate effect size in adolescence.
In early adolescence, involvement in general offenses -- that is, illegal but not necessarily violent acts, including felonies -- becomes a moderate risk factor for violence between the ages of 15 and 18. Its predictive power lessens from childhood, largely because teenagers are somewhat more likely than children to engage in illegal behavior.
Psychological conditions, notably restlessness, difficulty concentrating, and risk taking, have small effect sizes in adolescence. Restlessness and difficulty concentrating can affect performance in school, a risk factor whose importance increases slightly in adolescence. Risk taking gains predictive power in early adolescence, particularly in combination with other factors. A reckless youth who sees violence as an acceptable means of expression, for example, is more likely to engage in violent behavior.
Aggressiveness exerts a small effect on later violence among adolescent males, as does simply being male. While aggressiveness is unusual in children between the ages of about 6 and 10, it is not terribly unusual in adolescence. Similarly, physical violence and crimes against persons in early adolescence have a small effect on the likelihood of violence at ages 15 to 18.
Antisocial attitudes and beliefs, including hostility toward police and a positive attitude toward violence, are more important predictors among adolescent boys than they are among children, but their effect sizes remain small. Antisocial behavior and low IQ continue to have small effect sizes in adolescence.
Substance use, which was a strong predictor of later violence for children, poses a small risk of later violence for adolescents. The question as to whether drug use causes young people to become violent is complex and has been widely studied (see Miczek et al., 1994 for a review), but there is little compelling pharmacological evidence linking illicit drug use and violence. In one large study, youths reported that over 80 percent of the violent incidents they initiated had not been preceded by drug use, including alcohol use (Huizinga et al., 1995). Thus, the risk may lie more in the characteristics of the social settings in which drug use and violence are likely to occur than in any effect of drugs on behavior (Parker & Auerhahn, 1998; Reiss & Roth, 1993).
The majority of violent adolescent offenders use alcohol and illicit drugs (see Chapter 3). Illicit drug use tends to begin after the onset of violence and to be associated with more frequent violent behavior and a longer criminal career (Elliott et al., 1989). This finding suggests that drug use may contribute to continued violence rather than to the onset of violence, but it is far from conclusive. Evidence shows that some violent behavior stems from robberies or other attempts to get money to support a drug habit but also that this link is relatively rare. If any substance can be said to cause youth violence, that substance is alcohol (APA, 1993; Parker & Auerhahn, 1998); however, this causal link is inconclusive because adolescent drinking is dependent to a large degree on the situation and social context in which it takes place (for reviews, see Parker & Auerhahn, 1998; Pernanen, 1991; Reiss & Roth, 1993; Roizen, 1993).
Parents' direct influence on behavior is largely eclipsed by peer influence during adolescence. Not surprisingly, therefore, most family risk factors diminish in importance, including the influence of antisocial parents and low socioeconomic status, the most powerful early risk factors. There are no large or even moderate risk factors in the family domain in adolescence.
Poor parent-child relations continue to have a small effect size, but for adolescents this category includes inadequate supervision and monitoring of young people's activities and low parental involvement, in addition to inappropriate discipline (Elliott et al., 1985; Hawkins et al., 1998a; Patterson & Yoerger, 1997; Roitberg & Menard, 1995). Broken homes and parental abuse also exert small effects. Other adverse family conditions present a risk factor; for example, some studies have found that family conflict is a risk factor for violence among adolescent males.
Although parents can and do influence their adolescents' behavior, they do so largely indirectly. The kind of peers chosen by young people, for example, is related to the relationship they have with their parents (Elliott et al., 1989; Hill et al., 1999; Patterson & Yoerger, 1997; Simons et al., 1994).
There are no large or moderate risk factors for violence in the school domain, but poor attitude toward or performance in school -- particularly if it leads to academic failure -- is a slightly larger risk factor in early adolescence than in childhood.
Research on school violence indicates that a culture of violence has arisen in some schools, adversely affecting not just students but teachers and administrators as well (Gottfredson et al., in press; Lorion, 1998). Students exposed to violence at school may react by staying home to avoid the threat or by taking weapons to school in order to defend themselves (Brener et al., 1999). For their part, teachers may burn out after years of dealing with discipline problems and threats of violence.
Schools located in socially disorganized neighborhoods are more likely to have a high rate of violence than schools in other neighborhoods (Laub & Lauritsen, 1998). At the same time, however, researchers emphasize that most of the violence to which young people are exposed takes place in their home neighborhood or the neighborhood surrounding the school, not in the school itself (Laub & Lauritsen, 1998). Individual schools, like individual students, do not necessarily reflect the characteristics of the surrounding neighborhood. A stable, well-administered school in a violent neighborhood may function as a safe haven for students.
Some gang activity takes place in schools, but school gangs are generally younger and less violent than street gangs, which form in neighborhoods (Laub & Lauritsen, 1998). Gangs in schools increased dramatically (by 87 percent) between 1989 and 1995 but have recently declined (see Chapter 3). The chances of becoming a victim of violence are more than two and one-half times as great in schools where gangs are reported, and these schools are disproportionately located in disadvantaged, disorganized neighborhoods (Met Life, 1993; Snyder & Sickmund, 1999).
Peer groups complicate the picture further. They operate both in neighborhoods and in schools, but the concentration of young people in schools may intensify the influence of these groups. One large study of adolescent males found that some schools have dominant peer groups that value academic achievement and disapprove of violence, while others have groups that approve of the use of violence (Felson et al., 1994). This study found that the risk of becoming involved in violence varied depending on the dominant peer culture in their school, regardless of their own views about the use of violence.
Peer groups are all-important in adolescence. Adolescents who have weak social ties -- that is, who are not involved in conventional social activities and are unpopular at school -- are at high risk of becoming violent, as are adolescents with antisocial, delinquent peers. These two types of peer relationships often go together, since adolescents who are rejected by or unpopular with conventional peers may find acceptance only in antisocial or delinquent peer groups. Social isolation -- having neither conventional nor antisocial friends -- is not a risk factor for violence, however (Cairns & Cairns, 1991; Elliott & Menard, 1996; Fergusson & Lynskey, 1996; Patterson & Yoerger, 1997). A third risk factor with a large effect size on violence is belonging to a gang. Gang membership increases the risk of violence above and beyond the risk posed by having delinquent peers (Thornberry, 1998). These three peer group factors appear to have independent effects, they sometimes cluster together, and they are all powerful late predictors of violence in adolescence.
Researchers who have studied what causes young people to join gangs have found that the risk factors for gang membership are virtually the same as those for violence generally (Hill et al., 1999). The notion that gangs act as surrogate families for children who do not have close ties to their own families is not borne out by recent data (Hill et al., 1999), but gangs do strengthen young people's sense of belonging, their independence from parents, and their self-esteem. Estimates from law enforcement agencies indicate that gang members are overwhelmingly male and the great majority (almost 80 percent) are African American or Hispanic (Snyder & Sickmund, 1999). But surveys in which young people identify themselves as gang members suggest that there are substantially larger proportions of white and female gang members. In a survey of nearly 6,000 8th graders in 1995, 25 percent of white students and 38 percent of female students reported they were gang members (Esbensen & Osgood, 1997). Lacking comparisons within ethnic groups, it is difficult to tell whether ethnicity per se is a risk factor in gang membership.
Increasing involvement in the community is a healthy part of adolescent development, unless the community itself poses a threat to health and safety. Social disorganization and the presence of crime and drugs in the neighborhood pose a small risk of violence when measured on an individual level, as they are in Table 4-1. As noted in the table, however, both of these risk factors have a substantially greater effect on the neighborhood level, where they measure the average rate of violent offending by youths living in the neighborhood or community.
Socially disorganized communities are characterized in part by economic and social flux, high turnover of residents, and a large proportion of disrupted or single-parent families, all of which lessen the likelihood that adults will be involved in informal networks of social control. As a result, there is generally little adult knowledge or supervision of the activities of teenagers and a high rate of crime. Moreover, in areas experiencing economic decline, there are likely to be few neighborhood businesses. In such an environment, it is hard for young people to avoid being drawn into violence. Not only are they on their own after school, they are exposed to violent adults and youth gangs, they have few part-time job opportunities, and their neighborhood is not likely to offer many after-school activities such as sports or youth groups (Bursik & Grasmick, 1993; Sampson et al., 1997; Wilson, 1987).
Social disorganization is also a risk factor for violence in rural areas. One study of rural communities found that poverty plays a less important role in predicting violence than residential instability, broken homes, and other indicators of social disorganization (Osgood & Chambers, 2000). In fact, very poor areas were not characterized by high residential instability or a large proportion of broken homes. In cities, however, the combination of poverty with instability and family disruption is predictive of violence (Bursik & Grasmick, 1993; Elliott et al., 1996).
Adolescents who are exposed to violence in their neighborhood feel vulnerable and unable to control their lives. These feelings can lead to helplessness and hopelessness. Such young people may turn to violence as a way of asserting control over their surroundings. They may arm themselves or even join a gang for protection. Studies have shown that adolescents exposed to violence are more likely to engage in violent acts, often as preemptive strikes in the face of a perceived threat (Singer et al., 1994, 1995).
Neighborhood adults who are involved in crime pose a risk because young people may emulate them. Easily available drugs add to the risk of violence. As noted earlier, drug use is associated with both a higher rate of offending and a longer criminal career (Elliott et al., 1989). More important, ready availability of drugs indicates that considerable drug trafficking is taking place in the neighborhood -- and drug trafficking is dangerous for buyer and seller alike.
Violence peaks during the second decade of life. The youths who first became violent in childhood escalate their violence in adolescence, and a larger group of young people embarks on violence in adolescence. For some young people, violence represents a way of gaining the respect of peers, enhancing their sense of self-worth, or declaring their independence from adults. Violence drops off as adolescents enter adulthood and assume adult roles.
Parents' direct influence on behavior is largely supplanted in adolescence by peer influences. Thus the most powerful peer predictors of violence in adolescence are weak social ties to conventional peers, ties to antisocial, delinquent peers, and belonging to a gang.
This chapter does not identify a number of characteristics and conditions frequently thought of as risk factors. Furthermore, some of the risk factors that have been identified may exhibit smaller effect sizes than expected. There are two reasons for this. First, this report relies on longitudinal studies, which identify risk factors and their effect sizes on the basis of their ability to predict future behavior. Much of this research involves identifying risks for aggression, externalizing behavior, or antisocial behavior -- not risks for violence. While there is considerable overlap between the risk factors for aggression and those for violence, there are some important differences, particularly with respect to effect sizes (Hann & Borek, in press). Television violence, for example, has a very large effect on aggressive behavior but only a small effect on violence. Second, some studies that have been widely cited in the media involve cross-sectional and retrospective research designs, which are inappropriate for identifying factors that predict future violence.
Conduct disorder has been linked to youth violence in numerous studies, but the cluster of symptoms used to determine this disorder includes physical aggressiveness, nonphysical aggressiveness, and antisocial attitudes and beliefs. For purposes of predicting violence, the critical question is: What components of this disorder actually confer risk? There is some evidence that physical aggression accounts for most of the predictive power of conduct disorder (Nagin & Tremblay, 1999) and has a moderate to small effect size as a predictor of violence. Antisocial attitudes and beliefs also predict violence, but with an even smaller effect size. The three components of conduct disorder generally cluster together, which accounts for their having been combined into a single risk factor in earlier studies. Other childhood disorders such as attention-deficit/hyperactivity disorder, depressive and anxiety disorders, and their symptoms do not cause violent behavior, but their presence often signals serious behavioral and emotional problems that negatively affect family, social, and academic functioning, domains of risk for violent behavior.
Race has long been considered a risk factor for the onset of violence, and it is included as a risk factor in most studies using simple bivariate predictors of violence. The question is whether race predicts violence once other known risk factors are taken into account. Studies that have accounted for the effects of other known risk factors have typically found no significant effect of race on youth violence (Elliott et al., 1989; Reiss & Roth, 1993; Roitberg & Menard, 1995). Thus, race appears to be a risk marker rather than a risk factor. Race is a proxy for other known risk factors -- living in poor, single-parent families, doing poorly at school, and being exposed to neighborhood disadvantage, gangs, violence, and crime. The evidence suggests that the link between race and violence is based largely on social and political distinctions rather than biological differences.
Ethnicity has also been proposed as a risk factor, but it has not been studied extensively enough to include here. Young people from ethnic minorities may be subject to prejudice and thus to limited opportunity, and they may face unique stresses when their family culture conflicts with the dominant U.S. culture. At the same time, their ethnic culture may offer them strong support and guidance and thus function as a protective factor (APA, 1993).
Child abuse is widely considered to be a powerful risk factor for youth violence. This belief is based on a number of early studies that suffered from serious methodological problems (see Dodge et al., 1990; Garbarino & Plantz, 1986; Howing et al., 1990; and Widom, 1989 for reviews). In more sophisticated, controlled longitudinal studies, the effects are much smaller (see Table 4-1), a finding that holds for both self-report and official record studies. In addition, studies reporting on child abuse as a predictor of nonviolent delinquent behavior or less serious offenses find larger effect sizes than those cited here for violence or serious delinquency (Bolton et al., 1977; Smith & Thornberry, 1995; Widom, 1989, 1991; Zingraff et al., 1993, 1994). Neither sexual abuse nor physical abuse is a significant predictor of youth violence when considered alone (Hawkins et al., 1998c). Sexual abuse has been linked to criminal behavior in adulthood (Widom & Ames, 1994), but not to violence in adolescence.
Although the effect size of child abuse or neglect is small when a correlation measure is used (as in Table 4-1), the relative risk of violence among abused or neglected children can be substantial. Knowing that a child was abused does not help much in predicting future violence, however, since the vast majority of abused children do not become violent. For example, one longitudinal study showed that 5 percent of abused children were arrested for a violent crime by age 18, compared to 3 percent of nonabused children (Widom, 1991). The relative risk of arrest for violence is nearly twice as great in the abused group as in the nonabused group, yet the correlation for this relationship is .07, a small effect size. 6 In other words, even though the probability of later violence is substantially higher among abused than nonabused youths in this study, the correlation is small because the majority of all youths (95 percent of abused and 97 percent of nonabused youths) did not become violent.
When the proportion of youths who become violent is greater, the relative risks appear to be lower. Thus, when subjects in the 1991 Widom study were tracked to age 30, the relative risk of violence dropped to 1.3 (Widom, 2000). In the one longitudinal self-report study to date, which had relatively high proportions of abused children reporting violence, the relative risk of violence was 1.2 (Smith & Thornberry, 1995). In both of these cases, the correlation was less than .10.
Heredity does not seem to play a strong role in violence (see Cary, 1994 for a review). While there is some evidence supporting a genetic effect, the proposed mechanisms are very complex and nonspecific (Turbin, 2000). Neurotransmitters such as dopamine, serotonin, and GABA may play a role in aggression, but so far their mechanisms of action are unclear and there is insufficient evidence to consider them predictors of violence. In general, there are no known neurobiological patterns that are precise and specific enough to be considered reliable risk factors for violent behavior (Reiss & Roth, 1993).
Drug trafficking in early adolescence predicts later violence (Hawkins et al., 1998c; Herrenkohl et al., 2000; Huizinga et al., 1995; Menard et al., in press; Reiss & Roth, 1993). This risk factor is not included here because only one study presents correlations (or the data necessary to calculate them); therefore, average effect size could not be estimated. In the Menard et al. study, the correlation between selling marijuana and violence in adolescence was .33; for selling hard drugs, it was .27. In the Hawkins et al. study, the odds ratio for selling drugs at age 14 and violence at age 18 was 3.34; it was 4.55 for selling at age 16 and violence at age 18. Drug selling thus appears to have at least a moderate effect size.
Research on resilience and the public health approach to the problem of youth violence have brought a new awareness of, and research on, protective factors -- those aspects of the individual and his or her environment that buffer or moderate the effect of risk. Identifying and understanding how protective factors operate is potentially as important to violence prevention and intervention efforts as research on risk factors.
To date, the evidence regarding protective factors against violence has not met the standards established for risk factors. Therefore, this report does not refer to protective factors, only to proposed protective factors (Table 4-2). There are several reasons for this: Not all studies define protective factors as buffering the effects of risk; most studies have looked for an effect on antisocial behavior in general, not on violence specifically; and those that have found buffering effects on violence have not been adequately replicated. This does not mean that protective factors do not exist, just that more research is needed to identify them. 7
Most studies of protective factors do not specify when in the course of development these factors exert their buffering effects or how they change over the life course. Further study is needed to clarify these points; therefore, Table 4-2 does not show age of onset for the proposed protective factors listed.
The authors of a 1995 longitudinal study on protective factors and their buffering effects on the risk of problem behavior in adolescence (Jessor et al., 1995) recently reexamined their data to see whether they could find any buffering effect specifically on violence. They did find a buffering effect, but their results must be considered preliminary until they are replicated by others. Nonetheless, these findings are encouraging, since they indicate that several of the factors identified as protective against problem behavior also provided a buffering effect against violence. By implication, other studies that have demonstrated buffering effects on the risk of antisocial behavior or general delinquency (for example, Fergusson & Lynskey, 1996) may also contain evidence of potential protective factors against violence. The discussion of proposed protective factors in this report rests on the reanalysis of the 1995 study data (Turbin, 2000), as well as on results from other studies, bearing in mind the caveats noted above.
The 1995 Jessor study grouped possible protective factors together and found that students who scored high on this index of protection were buffered from the effects of risk, compared to students who scored low on the index. The index was composed of seven psychosocial protective factors: attitudinal intolerance of deviance, positive orientation to health, religiosity, positive relations with adults, perceived consequences for misbehavior, friends as models for conventional behaviors, and high involvement in conventional activities. In an analysis of specific factors, however, only two -- an intolerant attitude toward deviance and commitment to school -- had significant protective effects. The new findings show that the same two factors appear to exert a significant, though small, buffering effect on risk factors for violence.
One of the proposed protective factors shown to have a buffering effect on the risk of violence is an individual characteristic, and the other falls into the domain of school; both are classed as having a small effect. No other factors in the individual, family, school, or peer group domains have been shown to exert significant buffering effects on risk factors for violence, although they have been shown to moderate the risk of antisocial behavior or delinquency. No protective factors have been proposed yet in the community domain.
An intolerant attitude toward deviance, including violent behavior, is the strongest proposed protective factor. It reflects a commitment to traditional values and norms as well as disapproval of activities that violate these norms. Young people whose attitudes are antithetical to violence are unlikely to become involved in activities that could lead to violence or to associate with peers who are delinquent or violent.
The four remaining individual factors have not yet been shown to moderate violence, although they may buffer risks for antisocial behavior or general delinquency. High IQ has been cited as a possible protective factor (Fergusson & Lynskey, 1996; Garmezy, 1985; Rutter, 1985; Werner & Smith, 1982). Children with above-average IQs may exhibit qualities, such as curiosity and creativity, that help them make the most of early educational, artistic, and cultural experiences. Above-average IQ can also help a child excel in school. High IQ may increase an adolescent's chances of benefiting from educational, creative, and cultural opportunities. For youths facing multiple risk factors, exposure to the wider world may open a window on alternative values and lifestyles.
Being born female has also been cited (Garmezy, 1985; Rutter, 1985; Werner & Smith, 1982), but it is the opposite of being born male, a risk factor, and as yet there is no evidence of a buffering effect. Being a girl entails less exposure to violence, less impulsiveness and daring, and being expected to behave less aggressively than boys.
Some studies have proposed positive social orientation as a protective factor (Garmezy, 1985; Jessor et al., 1998; Rutter, 1985; Werner & Smith, 1982). Like commitment to school, a positive social orientation indicates that a young person has adopted traditional values and norms, a slightly different emphasis than intolerance of deviance. This proposed factor appears to be the opposite of antisocial attitudes and beliefs, a late-onset risk factor that has a small effect size.
Perceived sanctions for transgressions, a protective factor in the earlier Jessor study (1995), refers to perceived peer disapproval of deviant behavior. The reanalysis of those original data reveals that this proposed factor has no significant protective effect on risk of violence or problem behavior.
There is no doubt that an essential aspect of healthy child development is forming a secure attachment in infancy to a parent or other adult who senses and responds to a baby's needs (Bell & Fink, 2000). Likewise, researchers agree that having a loving adult who is interested in and supportive of a child or young person's ideas and activities helps that child or adolescent develop the confidence and competence needed to progress from one stage of development to the next. Good relations with an adult who supports conventional behavior and disapproves of delinquent behavior can provide invaluable guidance for young people. The question is whether these relationships moderate the effects of exposure to risk and thus fit the definition of a protective factor.
A warm, supportive relationship with parents or other adults has been shown to protect against antisocial behavior, but studies so far have not found a significant buffering effect on the risk of violence (Hawkins et al., 1998c; Klein & Forehand, 2000; Rutter, 1979; Turbin, 2000; Werner & Smith, 1992).
It is uncertain whether family protective factors, like family risk factors, become less influential as young people progress through adolescence. Parental support and encouragement remain important, but even parents who have had a good relationship with their children before puberty may affect their adolescents' behavior only indirectly -- for example, through choice of friends (Elliott et al., 1989). This indirect influence is not inconsequential, however; associating with peers who disapprove of violence may inhibit later violence in young people (Hawkins et al., 1998c), and parents' positive evaluation of peers has been found to reduce the risk of delinquency (Smith et al., 1995).
Several studies have pointed to monitoring or supervision of activities as a protective factor against delinquency and antisocial behavior, but this is essentially the opposite of failure to monitor, an adolescent-onset risk factor with a small effect size. To date, no evidence of moderating effects on the risk of violence has been presented (Baldwin et al., 1990; Klein & Forehand, 2000; Smith et al., 1995).
Commitment to school is the second proposed protective factor that has been found to buffer the risk of youth violence. Young people who are committed to school have embraced the goals and values of an influential social institution. Such young people are unlikely to engage in violence, both because it is incompatible with their orientation and because it would jeopardize their achievement in school and their standing with adults (Jessor et al., 1995; Turbin, 2000). This proposed factor is included because it appears to buffer the risk of violence, not because it is the opposite of poor attitude toward or performance in school, a risk factor with small effect sizes in both childhood and adolescence.
School can give adolescents who face multiple risk factors a place in which to excel socially and academically. Achievement in school and the approval of teachers provide the recognition so important to adolescent development -- recognition some adolescents do not receive from other sources. Encouragement from teachers can give young people the confidence to seek continued educational or job skills training. In addition, schools with peer groups that value academic achievement may lower students' risk of becoming involved in violence (Felson et al., 1994). Unfortunately, schools with a culture of violence may be unable to exert their very important protective function.
Extracurricular activities in art, music, drama, school publications, and the like give adolescents an opportunity to participate in constructive group activities and achieve recognition for their efforts. Studies have found that recognition for or involvement in conventional activities -- whether family, school, extracurricular, religious, or community -- is a protective factor against antisocial behavior (Jessor et al., 1995; Rae-Grant et al., 1989). The reanalysis of the Jessor data shows that involvement in family, volunteer, and school club activities other than sports has an insignificant effect on risk for violence (Turbin, 2000).
Having friends who behave conventionally is a proposed protective factor that seems to reduce the risk of delinquency, but there is no evidence of a true buffering effect on specific risk factors. Buffering effects on violence were not significant in the reanalysis of the Jessor data (Turbin, 2000; see also Smith et al., 1995). However, as noted earlier, researchers have found that associating with peers who disapprove of violence may inhibit violence in young people (Hawkins et al., 1998c; Jessor et al., 1995).
Although the body of research on protective factors is growing, very little work has been done specifically on protective factors that buffer the risk of violence. Some researchers have identified individual and environmental characteristics that can be considered candidates for protective factors. Lacking adequate scientific evidence of the nature, mechanism, size, and timing of these candidates' moderating effects, however, this report considers all of them proposed protective factors.
One recent reanalysis of earlier data has found two proposed protective factors that seem to buffer the risk of violence -- an intolerant attitude toward deviance and commitment to school. These two factors appear to exert a statistically significant, though small, buffering effect on the risk of violence, but until these findings are replicated, they must be considered preliminary.
Identifying and understanding how protective factors operate is as important to preventing and stopping violence as identifying and understanding risk factors. This area of the public health approach to youth violence cries out for more research.
Scientists have identified a number of personal characteristics and environmental conditions that put children and adolescents at risk of violent behavior and some that seem to protect them from the effects of risk. These risk and protective factors can be found in every area of life, they exert different effects at different stages of development, they tend to appear in clusters, and they appear to gain strength in numbers. The public health approach to youth violence involves identifying risk and protective factors, determining when in the life course they typically come into play, designing preventive programs that can be put in place at just the right time to be most effective, and making the public aware of these findings.
Many years of research have yielded valuable insights into the risk factors involved in the onset and developmental course of violence. Less work has been done on protective factors, but that situation is changing. Some basic principles have emerged from these studies:
Violence prevention and intervention efforts hinge on identifying risk and protective factors and determining when in the course of development they emerge. More research in these areas is needed, particularly concerning why violence stops or continues in childhood and adolescence. Nonetheless, the research carried out to date provides a solid foundation for programs aimed at reducing risk factors and promoting protective ones -- and thereby preventing violence, the subject of Chapter 5.
Americans have been concerned about the prevalence of violence in the media and its potential harm to children and adolescents for at least 40 years. The body of research on television violence has grown tremendously since the first major Federal reports on the subject in 1972 and 1982 (National Institute of Mental Health, 1982; U.S. Surgeon General's Scientific Advisory Committee on Television and Social Behavior, 1972). During this period, new media emerged -- video games, cable television, music videos, and the Internet. As they gained popularity, these media, along with television, prompted public concern and research attention.
Recent surveys depict the abundance of (primarily electronic) media in U.S. homes (Roberts et al., 1999; Woodard, 1998) and the extensive presence of violence within the media landscape (Wilson et al., 1997, 1998). They also show that the proliferation of new media has expanded the opportunities for children to be exposed to media violence at home. Current psychological theory suggests that the interactive nature of many of these new media may affect children's behavior more powerfully than passive media such as television. Research to test this assumption is not yet well developed, and accurate measurement is needed to determine how much violence children are actually exposed to through various media -- and how patterns of exposure vary among American youths.
In reading this discussion of research on the impact of media violence on America's youth, a few major points should be kept in mind:
American children and youths spend, on average, more than 4 hours a day with television, computers, videotaped movies, and video games (Roberts et al., 1999; Woodard, 2000). But their exposure to media varies considerably, depending on their age, parental viewing habits, and family socioeconomic status (SES). Most systematic research on children's exposure to violent media dates back to the 1970s, when most families did not have access to cable television, music videos, video games, or the Internet. As noted earlier, very few contemporary studies systematically document children's actual consumption of violent media; this is particularly true for the newer media.
Several content analyses over the last 30 years have systematically examined violence on television (Gerbner et al., 1980; Potter et al., 1995; Signorielli, 1990). The largest and most recent of these was the National Television Violence Survey (NTVS) 1 (Wilson et al., 1997, 1998), which examined the amount and content of violence 2 on American television for three consecutive years, as well as contextual variables that may make it more likely for aggression and violence to be accepted, learned, and imitated. Smith and Donnerstein (1998) report the following NTVS findings:
The NTVS report notes that many television programs fail to depict the harmful consequences of violence. Specifically, it finds that of all violent behavioral interactions on television, 58 percent depict no pain, 47 percent depict no harm, and 40 percent depict harm unrealistically. Of all violent scenes on television, 86 percent feature no blood or gore. Only 16 percent of violent programs feature the long-term, realistic consequences of violence.
Because an exhaustive description of the research literature is not possible within this brief discussion, findings from meta-analyses are reported 3 where available. In meta-analyses, the results of multiple studies are combined and compared systematically and an overall effect size computed. These analyses include findings from randomized experiments that look at aggression immediately after viewing violence, as well as cross-sectional surveys that provide a snapshot of the relationship between viewing violence and behavior at a fixed point in time. Also presented are findings from longitudinal studies that examine whether exposure to media violence affects violence and aggression over time.
Many anecdotal reports have described instances in which television and film violence led to immediate violent behavior in individual children, but scientific studies of this relationship draw a more complex and qualified picture. Most of the relevant research has focused on how watching dramatic violence on television and film affects aggressive thoughts and emotions, as well as aggressive behavior. Some important studies address violence as well.
A substantial number of laboratory and field experiments over the past half-century have examined whether children exposed to violent behavior on film or television behave more aggressively immediately afterwards (see reviews by Bushman & Huesmann, 2000; Comstock & Scharrer, 1999; Geen, 1990; Geen & Thomas, 1986; Huesmann et al., 1997). Many studies have also examined the immediate effect of media violence on aggressive thoughts or emotions (Rule & Ferguson, 1986), which have been shown to increase the risk of aggressive behavior (Dodge & Frame, 1982; Huesmann & Guerra, 1997).
The most recent and comprehensive meta-analysis of media violence was conducted by Paik and Comstock (1994), who examined effect sizes from 217 empirical studies on media violence and aggressive and violent behavior published between 1957 and 1990. The analysis indicates clearly that brief exposure to violent dramatic presentations on television or in films causes short-term increases in the aggressive behavior of youths, including physically aggressive behavior. Across all the randomized experiments, the unweighted average effect size was large (r = .37). 4 When only experiments examining physical aggression as the outcome were examined, the effect size was also large (r = .32).
Although the experimental methods used in these studies enable researchers to test causality more readily than other research methods as noted by Comstock and Paik (1991), the findings may not necessarily apply to all real-world settings. Because experiments are narrowly focused on testing specific causal hypotheses, they do not examine the effects of all factors that might be present in more realistic situations. This means that some real-world influences might actually lessen or even eliminate the aggressive reactions observed in experiments. For example, while television, film, and other media contain a variety of antisocial and other messages, most laboratory studies to date have exposed study participants primarily to violent materials. In addition, participants may react differently in the laboratory when they realize that their expressions of aggression will not be punished (Gunter, 1983). Any summary of these experimental results should also acknowledge the argument raised by some critics (such as Freedman, 1992) that many study participants provide the responses they believe the researcher wants. Despite these limitations, laboratory experiments are important because they allow researchers to isolate the unique effect of exposure to violence on subsequent behavior.
An important general finding from these experimental studies is that not all youths seem to be affected equally by media violence. Effects seem to be strongest on youths who are predisposed to be aggressive for some reason or who have been aroused or provoked (Berkowitz, 1993; Bushman, 1995; Geen & O'Neal, 1969).
Cross-sectional surveys over the past 40 years have generally focused on establishing a link between the current aggressiveness of children and the amount of television and film violence they watch regularly (see reviews by Bushman & Huesmann, 2000; Chaffee, 1972; Comstock & Scharrer, 1999; Eysenck & Nias, 1978; Huesmann & Miller, 1994).
Paik and Comstock's meta-analysis (1994) indicates that in cross-sectional surveys viewing media violence was positively correlated with various measures of aggression. They reported small to moderate effect sizes across all measures of aggression (r = .19) and for physical aggression alone (r = .20). For the outcome of most concern to this report -- criminal violence against a person -- the effect size was small (r = .06). These results suggest that the link between media violence and aggressive behavior found in laboratory studies may also hold for behaviors outside the laboratory. However, cross-sectional surveys do not by themselves indicate whether media violence is causing aggression, whether aggressive youths are attracted to media violence, or whether some other factor is predisposing some youths to watch more violence and behave more aggressively.
Long-term studies in which exposure to media violence in early childhood is related to later aggression and violence (such as aggravated assault, robbery, rape, and homicide) can identify the enduring effects of media violence. In most such studies to date, however, aggression, not violence, has been the primary outcome measured. In the absence of a meta-analysis, the findings of three frequently cited longitudinal studies on the effects of media violence are discussed briefly below. Studies examining effects over shorter time periods (Singer et al., 1984) or with international samples (Huesmann & Eron, 1986) are not included here.
In a study begun in 1960 on a sample of 875 youths in New York State, Eron and colleagues found that for boys, but not for girls, exposure to media violence at age 8 was significantly related to aggressive behavior a decade later (r = .31, N = 211, p < .01) (Eron et al., 1972; Lefkowitz et al., 1977). At both times, peers assessed physical and verbal aggression. The longitudinal correlation remained above .25, even in separate analyses statistically controlling for factors such as the child's initial aggressiveness, the child's intelligence, family SES, parents' aggressiveness, and parents' punishment and nurturance of the child.
Milavsky et al. (1982) examined the probability of initiating aggression after exposure to violence on television in 2,400 boys and girls age 7 to 12 from two midwestern cities who had been surveyed up to six times between 1970 and 1973. A sample of 800 teenage boys 5 was studied at five times to identify the effect of violent television on aggression and violence. For the elementary school sample, the average cross-sectional correlation between exposure to media violence and personal aggression was small for boys (r = .17) and large for girls (r = .30). The researchers then attempted to predict aggressive behavior at one point in time from the extent to which children viewed television violence at an earlier time, while controlling for earlier aggressive characteristics. They examined this prediction over 15 time intervals ranging from 5 months to 3 years apart. For elementary school boys, only 2 of the 15 predictions at different intervals were statistically significant. For girls, only three predictions were statistically significant. In the teenage male sample, only one of eight correlations was significant. In only one of nine analyses using measures of violence (for example, knife fight, car theft, mugging, gang fight) were boys with greater exposure to television violence more likely to initiate violence 2 years later than those with less exposure.
The third longitudinal study of media violence effects began in the late 1970s and spanned five countries (Huesmann et al., (submitted); Huesmann et al., 1984; Huesmann & Eron, 1986). In each locale, samples of middle-class youths were examined three times between age 6 to 8 or age 8 to 11. Both physical and verbal aggression were assessed by peers. The correlations between aggression and overall viewing of television violence at a single point in time were small to moderate and often significant. In the United States, the 3-year average correlation was moderate for boys and for girls (r = .25 and r = .29, respectively; p< .001). The predictive power of viewing television violence for childhood aggression a year later varied substantially. In the United States, girls' viewing of television violence had a significant effect (β = .17, N = 89, p < .05) on their later aggression, even after accounting for early levels of aggression, SES, and scholastic achievement. For boys, television violence alone did not predict later aggression. When the investigators took into account both exposure to television violence and identification with aggressive television characters, they found a positive relation with aggressiveness (β = .19, N = 84, p < .05).
A follow-up study of over 300 people in the U.S. sample 15 years later suggested that media violence has a delayed effect on aggression (Huesmann et al., (submitted)). There was a small to moderate longitudinal correlation between childhood television viewing and a composite measure of young adult aggression (physical, verbal, and indirect aggression) for both men (r = .21, N = 153, p < .01) and women (r = .19, N = 176, p < .01). When the outcome was limited to physical aggression, the correlations were smaller (r = .17 and r = .15, respectively). Furthermore, women who had watched relatively more television violence as girls committed significantly more specific acts of violence as adults, such as "punching, beating, or choking another adult," than did the other women (17 percent versus 4 percent). There were no significant differences among the men. Other analyses showed that effects remained significant even when researchers controlled for parent education and children's scholastic achievement (β = .19 for boys, β = .17 for girls, p < .05). In addition, aggressive behavior did not significantly increase boys' or girls' viewing of television violence (β = .08 for boys and β = .04 for girls; p = ns).
In summary, these longitudinal studies show a small, but often statistically significant, long-term relationship between viewing television violence in childhood and later aggression, especially in late adolescence and early adulthood. Some evidence suggests that more aggressive children watch more violence, but the evidence is stronger that watching media violence is a precursor of increased aggression.
Other studies have explored the behavioral impact of introducing television in several countries (Centerwall, 1989a, 1989b, 1992; Joy et al., 1986; Williams, 1986). These studies indicate that when television was introduced, aggression and violence increased. The findings must be viewed with caution, however, because they do not take into account a range of other factors that may influence national crime rates and the amount of violence watched on television.
Despite anecdotal reports of a "contagion of violence," relatively little systematic research has examined whether seeing or hearing about violence in news coverage encourages violent or aggressive behavior. On the whole, the limited data available support the notion of a contagion effect. This evidence is derived from studies examining how reports of a well-known person's suicide affect the likelihood of imitative suicide (Phillips, 1979, 1982; Simon, 1979; Stack, 1989). Other studies of the contagion effect (Berkowitz & Macaulay, 1971; Phillips, 1983) have been questioned because of their research methods and the ambiguity of their results (Baron & Reiss, 1985; see Phillips & Bollen, 1985 for a response). This area merits additional research.
Theoretically, the effects of exposure to media violence extend to Internet media as well. To date, however, no studies have been published regarding the effects of Web-based media violence on youth aggression and violence.
The impact of video games containing violence has recently become a focus of research because children are theoretically more susceptible to behavioral influences when they are active participants than when they are observers. To date, violent video games have not been studied as extensively as violent television or movies. The number of studies investigating the impact of such games on youth aggression is small, there have been none on serious violence, and none has been longitudinal.
A recent meta-analysis of these studies found that the overall effect size for both randomized and correlational studies was small for physical aggression (r = .19) and moderate for aggressive thinking (r = .27) (Anderson & Bushman, in press). In separate analyses, the effect sizes for both randomized and cross-sectional studies was small (r = .18 and .19, respectively). The impact of video games on violent behavior remains to be determined.
Research suggests that not all youths are affected in the same way by viewing media violence. Factors that appear to influence the effects of media violence on aggressive or violent behavior include characteristics of the viewer (such as age, intelligence, aggressiveness, and whether the child perceives the media as realistic and identifies with aggressive characters) and his or her social environment (for example, parental influences), as well as aspects of media content (including characteristics of perpetrators, degree of realism and justification for violence, and depiction of consequences of violence).
Evidence that these factors moderate the influence of media violence is limited, and it is more relevant to aggression than to violence. For example, studies of responses to violent television and films and violent video games have found that people who were initially more aggressive than other subjects were more affected in behavior, thoughts, and emotions (Anderson & Dill, 2000; Bushman, 1995; Bushman & Geen, 1990; Friedrich & Stein, 1973; Josephson, 1987). Research in this area clearly suggests that the impact of violent television, film, and video games on aggression is moderated by viewers' aggressive characteristics.
Evidence that other individual, environmental, and content factors moderate the effects of exposure to media violence is less clear. Some studies suggest that these factors may buffer or enhance effects, but few have tested for such influences. Although limited in scope and depth, such studies provide clues to potential avenues for prevention efforts. For example, preliminary data point to the potentially vital role of parents in supervising their children's exposure to violent media and in helping them interpret it (Nathanson, 1999).
A substantial body of research now indicates that exposure to media violence increases children's physically and verbally aggressive behavior in the short term (within hours to days of exposure). Media violence also increases aggressive attitudes and emotions, which are theoretically linked to aggressive and violent behavior. Findings from a smaller body of longitudinal studies suggest a small but statistically significant impact on aggression over many years. The evidence for long-term effects on violence is inconsistent.
Based on the findings of studies reported here, the average effect sizes of exposure to media violence on various measures of aggression range from small (r = .15) to quite large (r = .64). The evidence that exposure to media violence is a risk factor for violent behavior is more limited, with small average effect sizes of r = .06 in cross-sectional surveys, r = .13 in experimental studies (Paik & Comstock, 1994), and r = .00 to .22 in longitudinal studies (Huesmann et al., (submitted); Milavsky et al., 1982). Taken together, findings to date suggest that media violence has a relatively small impact on violence. The effect on aggression is stronger, ranging from small to moderate.
Although there is clear scientific evidence of a correlation between exposure to media violence and some violent behaviors, randomized experiments -- the research methodology best suited to determining causality -- cannot ethically be used in studies of violent behavior. Thus, the causal links between media violence and behavior are more firmly established for aggressive behavior than for violent behavior. Longitudinal studies, which also provide some insights into this issue, have linked repeated exposure to media violence in the early years with an increased likelihood of aggressive behavior in the teen and adult years. However, few of these studies have reported on violence as an outcome. Moreover, the violent behaviors that are the focus of this report (homicide, forcible rape, aggravated assault, and robbery) occur infrequently and are subject to multiple influences. At present, it is extremely difficult to distinguish between the relatively small, long-term effects of media exposure and those other influences.
In sum, a diverse body of research provides strong evidence that exposure to violence in the media can increase children's aggressive behavior in the short term. Some studies suggest that long-term effects exist, and there are strong theoretical reasons why this is the case. But many questions remain regarding the short- and long-term effects of media violence, especially on violent behavior. Despite considerable advances in research, it is not yet possible to describe accurately how much exposure, of what types, for how long, at what ages, for what types of children, or in what types of settings will predict violent behavior in adolescents and adults.
Efforts to reduce the presumed harmful effects of media violence on youths have taken various forms, including:
From a public health perspective, this preventive domain is largely uncharted territory. Few preventive efforts have been studied systematically. Furthermore, not enough research has been done to form a basis for the design of many experimental interventions. As noted in other parts of this report, an extensive body of scientific research undergirds our emerging knowledge about effective ways of preventing youth violence. Although many violence prevention programs address a complex array of risk and protective factors in the lives of young people, they have not yet addressed the role of the media. This gap needs to be filled.
Research to date justifies sustained efforts to curb the adverse effects of media violence on youths. Although our knowledge is incomplete, it is sufficient to develop a coherent public health approach to violence prevention that builds upon what is known, even as more research is under way. Unlike earlier Federal research reports on media violence and youth (National Institute of Mental Health, 1982; U.S. Surgeon General's Scientific Advisory Committee on Television and Social Behavior, 1972), this discussion takes place within a broader examination of the causes and prevention of youth violence. This context is vital. It permits media violence to be regarded as one of many complex influences on the behavior of America's children and young people. It also suggests that multilayered solutions are needed to address aggressive and violent behavior.
A variety of media violence is present in the homes of young children, with considerable variation in the degree of parental supervision (Woodard, 1998). Regardless of government and other interested groups' attempts to limit the amount of violence reaching American families, families themselves play a critical role in guiding what reaches their children. Whether by adopting V-chip technology for home television programming, by using Internet violence screening, or simply by monitoring closely children's use of televisions, computers, and video games, parents can limit and shape their children's selection of, interaction with, and response to media violence. Community groups -- including schools, faith-based organizations, and Parent Teacher Associations -- can teach parents and children how to be more critical consumers of media. Federal agencies can be more active in encouraging needed research, in sharing research findings with the public, in encouraging increased interaction between violence prevention researchers and media researchers, and in creating networks for sharing solutions to social and public health problems.
If the relationship to violence is nonlinear, risk and protection may take on a different meaning. However, the conditions and characteristics identified as protective factors by those using the absence-of-risk conceptualization rarely, if ever, involve a nonlinear relationship to violence.2
As noted in Chapter 3, most violent offenders commit many serious property offenses (such as burglary, auto theft, and larceny), and most youths involved in serious property offenses (FBI index offenses) are also involved in violent offenses. The risk factors described here are based on longitudinal studies that use self-reports to predict violent offenses. Several of the studies also include official arrest data and thus predict self-reported offenses, arrests for serious or violent offenses, or both.34
Numerous studies of these effects have been done, notably those of Robert Pynoos and colleagues. See, for example, Pynoos, R. & Nader K. (1988). Psychological first aid for children who witness community violence. Journal of Traumatic Stress, 1, 445-473.5
This is a different use of the term "risk marker" than that proposed by Kraemer et al. (1997). They use risk marker to refer to a risk factor or cause (such as sex or race) that cannot, in practical terms, be changed by an intervention. This report focuses on its causal role rather than its amenability to change.6
In another study, the relative risk of later violence was two to three times as great among abused children as nonabused children (Zingraff et al., 1993).7
There is a fairly extensive body of research on protective factors in the field of psychopathology (Garmezy, 1985; Rae-Grant et al., 1989; Rolf et al., 1993; Rutter, 1979, 1985; Rutter et al., 1979; Stattin et al., 1997; Werner and Smith, 1982, 1992). There are also a number of studies focusing on delinquency that purport to identify protective factors (Brewer et al., 1995; Farrington and West, 1993; Hawkins et al., 1992; Resnick et al., 1997; Smith et al., 1995).1
The NTVS randomly sampled programs from 6:00 a.m. to 11:00 p.m. on 23 broadcast and cable channels over a 20-week period from October to June during the 1994 through 1997 viewing seasons. A sum of 119 hours per channel, or 2,500 hours of television programming, was assessed each year.2
The NTVS defined violence as "overt depiction of a credible threat of physical force, or the actual use of such force intended to physically harm an animate being or group of beings." Content analyses of television programs generally treat the program itself as the unit of analysis and exclude advertisements. "Violence also includes certain depictions of physically harmful consequences against an animate being or group that occur as a result of unseen violent means. Thus, there are three primary types of violent depictions: credible threats, behavioral acts, and harmful consequences" (Smith & Donnerstein, 1998, p. 170).3
In the text to follow, all reported results are statistically significant (p < .05).4
In this study, all effects are unweighted average effects.5
These predictions are based on subsamples from which many of the most aggressive children had been dropped by the research team, reportedly because they were not accurately describing their television viewing.