A Developmental Model for Late-Onset Delinquency Gerald R. Patterson and Karen Yoerger (1997) Nebraska Symposium on Motivation Volume 44, 119-177.
1.	Current theories about delinquency have failed in two crucial aspects.
	a.	None of these theories can account for significant variance in societal rates of juvenile offending.
	b.	Although longitudinal studies identify many variables that predict later delinquency, the level of prediction is abysmally low. For example, a review of one major study showed false positive error of 85%.
2.	These failures are the outcomes of both inadequate theory and inadequate measurement. This chapter addresses both issues.
3.	Patterson and Yoerger believed that theory could be improved by examining the assumption that delinquents represent a homogeneous class. They took the position that there are two, or perhaps more, very different paths to delinquency. The general formulation has been cast as an early-onset and late-onset theory of juvenile delinquency.
4.	Patterson and Yoerger assumed that as compared to the late-onset group, the early-onset group was at significantly greater risk of adult criminal careers. They hypothesized that boys arrested early--before the age of 14 years--represent a very different kind of delinquent as compared to those who are arrested later in adolescence. They hypothesized that the two groups differ significantly in terms of
	a.	The context in which the family is embedded.
	b.	The level of parental skill in family management practices.
	c.	And the levels of child deviancy and social skills.
5.	The two groups can be differentiated by their growth patterns for both antisocial and delinquent behavior. The timing and the level achieved in growth clearly differentiate the two groups. Examination of the data relating to these hypotheses constitutes the primary focus of this chapter.
6.	All of the constructs in both the early- and late-onset models are defined by multiple (agent-method) indicators. This approach is thought to be a considerable improvement over the monoagent (adolescent or parent) or monomethod (self-report, interview) assessments that defined most of the child and family variables used in earlier longitudinal prediction studies.
7.	A social interactional perspective. The general formulation for the two paths to delinquency reflects a social interactional perspective. From this point of view, all social behaviors--including antisocial and delinquent behaviors--exist because they serve a recognizable function. The function is defined in part by the fact that social behaviors produce predictable reactions from the social environment. The predictability of some reactions by others to certain social behaviors is well documented in sequential lag analyses of both childhood and adult behaviors.
8.	Patterned reactions of others lead the child-adolescent to overselect some behaviors and to underselect others. One particularly interesting aspect of this process is that in certain situations the predictable reactions from the social environment actually lead the child to select deviant behaviors rather than prosocial ones. In some documented instances the selection process can become so extreme as to expose both the child and the reacting member of the social environment to extreme physical injury.
9.	Early-onset path. Patterson and Yoerger assumed that the training for boys arrested before the age of 14 actually begins very early, and it begins in the home. The proximal cause for antisocial behavior would be found in the immediate reactions provided by parents, siblings, and peers. The best predictors of antisocial behavior would, therefore, consist of measures of the reactions that lead simultaneously to an overselection of antisocial behaviors and an underselection of prosocial acts.
	Patterson and Yoerger assumed that much antisocial and delinquent behavior is overlearned and performed more or less automatically and is not the outcome of controlled cognitive processing. Driving a car is much like most coercive and antisocial behavior; neither activity requires much access to working memory or extensive attentional capacity. Much social interaction consists of automatic processing, where one may use all sorts of heuristics to compensate for the fact that each person has very limited channel capacity. Even under ideal conditions humans are capable of only a very limited rationality.
	The behavioral model begins with observations of families in which the reactions of both parents and siblings establish the child's coercive behaviors, such as whining, temper tantrums, and hitting, as having real functional value. In these families coercive behaviors are functional in the sense that they are the only effective means for terminating conflict.
	Later, the child learns that antisocial acts such as stealing and lying are also functional. With practice, the child learns when and in what settings these skills can be practiced. We suspect that the process begins as early as 18 months in some families.
	In elementary grades, the "problem" child selects deviant peers to form his or her support group and in so doing receives further training. Under the group's tutelage, the child selects new forms of antisocial behavior that are identifiable delinquent acts. The high frequency of these acts leads to an early arrest and then to chronic and violent offending.
10.	Late-onset path. Training for the second path to juvenile offending is thought to begin early to middle adolescence. This path describes 55% of all those in the Oregon Youth Study who were arrested by the age of 18. Patterson and Yoerger assumed that this path also begins with disruption in parenting practices, particularly in the monitoring process. In late-onset families, the relative payoffs for prosocial skills (e.g., school work, chores, good relationships, keeping your word) are much better than payoffs found in the early onset families but are not as high as in families that will produce nondelinquent boys. Both the early and late paths are characterized by rapid growth in antisocial behavior, but they differ markedly in when that growth occurs and in the kind of antisocial behavior that is growing.
	Early-onset families allow coercive and antisocial acts to become functional as early as preschool years. However, late-onset families do not permit these behaviors to pay off until early adolescence. The early-onset families provide low pay offs for prosocial skills and the child's antisocial behavior further exacerbates the problem of training for prosocial skills. The effect is to produce a child who is both deviant and socially unskilled. The late onset families tend to be relatively more supportive of prosocial skills. When the growth in antisocial behavior occurs in early adolescence, the youth already has an array of moderately developed prosocial skills.
	For many early-onset boys, deviancy training begins during the preschool years. It is provided first by family and then in later stages by deviant peers. As the process continues for the early-onset group, disruptions in the parenting process make indirect contributions even during late adolescence. For the late-onset group, the training is primarily by deviant peers and secondarily by family members.
	In either case, Patterson and Yoerger hypothesized that deviancy training--regardless of when it occurs--is associated with a breakdown in parenting practices, such as monitoring, discipline, family problem solving, and positive reinforcement. In experimental studies, the magnitude of improvement in parenting skills correlates significantly with the magnitude of reduction in delinquency. In effect, the level of parenting skill is a major determinant of whether a child's deviant or prosocial behavior becomes functional in a given family.
	Finally, Patterson and Yoerger hypothesized that both paths may reflect a mix of biological and environmental variables. For example, Patterson and Yoerger suspected that early-onset boys may emerge from the combination of difficult-to-train (temperament) toddler with parents who are relatively inexperienced. Reviews of results from twin design studies suggest a genetic presence for at least some of the variance in temperament measures. Data from these adoption study designs showed significant environment-gene interaction terms predicting adolescent antisocial behavior. However, the authors do not know whether the mechanism that explains the gene-environment interaction is temperament. There are at least a half dozen equally plausible mechanisms. It is noted that twin designs show no contribution of genetics to official records of delinquency. Given known flaws in twin design studies, it is difficult to know how to interpret these findings. The fact that twin and adoption designs produce different outcomes suggests that tracing these contributions will not be an easy task.
	In summary, a social interactional perspective makes the assumption that in some special environments, child-adolescent social behaviors--including antisocial and delinquent acts--become functional. In this instance, the term functional implies that parents, siblings, or peers provide payoffs that enhance the likelihood that these behaviors will be selected in the future. Whether future growth is higher for prosocial or for deviant behaviors depends upon what the social environment provides for these behaviors. To understand deviancy we must study not only the payoffs for deviancy but also the relative payoffs for prosocial behavior.
11.	Improved measurement. Classic measurement theory has considered unreliability as the primary source of measurement error. Contemporary interest in modeling, with its emphasis on the use of multiple indicators, has led to an expanded list of concerns. Modern measurement theory is focused additionally on the effects of errors of specification (the omission of relevant measures) and on the necessity for disentangling correlated residuals. Additionally, even in a well-specified model, each indicator may reflect a systematic bias. In studies of family process, systematic bias may be a major source of measurement error. The implicit assumption was that all agents (e.g., mother, father, teacher, peer, or observer) may be selectively biased in reports of their own behavior and the behavior of others. Our preliminary studies suggest that the sources of bias may be quite different for each agent. For example, our data showed that teachers tend to underestimate the magnitude of deviancy for children whom they do not know well. But mothers, as compared to fathers or teachers, seem to select higher rates of relatively trivial exemplars of deviant child behavior and consequently have lower predictive validity for their scores.
	Typically, studies show that maternal reports have both internal consistency and adequate test-retest reliabilities. There is growing evidence, however, that maternal ratings may vary from sample to sample. For example, ratings by socially disadvantaged mothers may have lower validity. Correlation's between parents' and teachers' ratings have been found to be appreciably lower for socially disadvantaged families than for families that were not disadvantaged.
	Patterson and his colleagues report evidence that maternal depression may be another source of maternal bias in ratings of both child adjustment and family process variables. The findings for the contribution of maternal depression to maternal ratings of family process and child adjustment exemplify the need to use multiple indicators to define key concepts. For this reason, each component of the early- and late-onset model was defined by multiple indicators.
12.	Early-onset model. The relation of age to crime has been well understood for years. Several scholars have taken the position that early onset was a key variable in predicting long-term negative outcomes. The earlier the onset, the worse the outcomes. Reanalyses of the data sets from this work strongly supported this position. However, there was little or no understanding of how these differences in outcome came about.
	This chapter by Patterson provides a detailed examination of differences between early and late onset in contextual, family process, and child adjustment variables. This discussion is theory driven in that a single model specifies how context, family process, and adjustment outcomes fit together.
	A basic premise in social interactional theory is that given repeated transactions, both members of a dyad are altered in some measurable fashion. The pattern describing how the two persons change over time might be thought of as a dyadic trait; it is quite stable over time. In formulating such a pattern for parent and child, Patterson and colleagues chose to proceed simultaneously at two different levels: they tried to explain why the child's deviant and social behaviors changed over time, and because parents were also changing over time, we wanted to explore whether some of these changes were brought about by what the child or adolescent was doing. However, it seemed to them that some changes in the dyad also reflected intrusions by forces from outside the dyad or even from contextual forces outside the family itself, such as changes in economic conditions, parental depression, or successful efforts to intervene. Patterson and colleagues assumed that the effect of contextual variables on child adjustment was mediated by changes in parenting practices, such as discipline, monitoring, or family problem solving. For example, in recently divorced families, the impact of massive changes in context on child adjustment would be minimal in those families where the now single parent maintained effective discipline, monitoring, and problem solving practices. The data collected in the longitudinal study of divorced families support this assumption.
	Patterson and colleagues hypothesized that parenting practices controlled the payoffs provided by family members for coercive and prosocial behaviors. Parenting practices serve as mediators between contextual variables and the contingencies that control child-adolescent behaviors. It became apparent, according to Patterson, the they needed to develop two parallel theories. One is concerned with why the child does what he or she does, and the other theory explains why parents do what they do.
13.	Why are some boys antisocial? Several decades of studies demonstrate that children's coercive antisocial behaviors are more functional in some environments than in others. It is only in the last decade that we have been able to resolve some of the difficulties in studying the nature of these functions. To date, most of the relevant studies of function are based on observation data and are focused on conflict episodes. We have known for some time now that conflict episodes occur two to three times more frequently in families with antisocial boys than with nonantisocial boys. These conflicts could produce negative emotion, such as anger, that in turn would cause the antisocial behavior. However, the Patterson modeling studies suggest that negative emotion does not play a direct role in causing antisocial behavior. The modeling studies do strongly suggest that emotions plays an indirect rather than direct role.
	Although conflicts certainly generate negative emotion, their prime function in the Patterson model is to define the setting in which training for deviancy and prosocial skills can occur. Children learn the details about what works and does not work in their family, but what children learn in normal families is quite different from what they learn in distressed families. Given a conflict in a normal family, the child learns that certain prosocial behaviors, such as talking, laughing, or negotiating, are followed by a termination of the conflict (i.e., the child's behavior served a function; it "worked"). In such families, the child finds that coercive behaviors also work on occasion. The process for selecting what works and what does not work is analogous to escape conditioning. The individual is presented with an aversive stimulus (the conflict behavior of the family member) and tries out different behaviors until he finds one that works. This particular behavior is followed very quickly by a termination of a conflict (i.e., he escapes from the aversive stimulus). In escape conditioning parlance, the response that leads to termination of the aversive stimulus is strengthened or more likely to be selected on future occasions.
	The analysis was carried out at the intra-individual level and required examination of the usefulness (in terminating conflict) of every behavior the child tried. A statement about the usefulness of deviant behaviors for a given child must be compared with how useful nondeviant behaviors are in the same situation. Modern learning theory emphasizes the utility of such an application of the matching law to the study of social behavior. What this means is that in a given situation there should be a match between the relative frequencies of child behaviors and the relative payoffs for these behaviors. If the highest payoff is for whining or arguing, then that is the behavior that is most likely to occur in that setting. In laboratory situations where only two or three responses are available and all extraneous stimuli are controlled, correlations of .9 or above are expected. However, even in the highly complex give-and-take of family interaction, the match between relative payoffs and relative frequency of behavior is quite good. For example, in one study by Patterson and colleagues the average correlation was .65 for normal families and .81 for distressed families. It seems clear that the child matches his or her behavior to environmental payoffs.
	Further examination of dyadic exchanges in family settings reveals an extremely interesting finding. Patterson and colleagues found that in families with problem boys, prosocial behaviors were, by and large, simply not effective in terminating family conflict. In these families, the only thing that seemed to work was coercive behaviors. Given these findings, Patterson and colleagues were not surprised to find that antisocial boys are significantly retarded in a variety of prosocial skills. These families simply do not support the development of prosocial skills. It could be that these parents are put off by the density of aversive behaviors that accompany the problem child as suggested by corelational studies by Patterson and colleagues. However, Patterson believes that these parents tend to be noninvolved, even with their infants. Many of their reactions are not contingent upon what the infant is doing but directed more at their own mood. The fact that the parent does not reinforce the infant's prosocial overtures may create a situation where the infant must rely upon coercive behavior to have any impact on the social environment. Is this how coercion begins? The answer to this question is simply not known at this time.
	According to this formulation, an aggressive child is actively matching his or her behavior to family reactions. But if one just looks at the intra-individual scores for prosocial and deviant behaviors, they will not, by themselves, account for individual differences in aggression. For example, it is possible to have two boys with the same relative rates of payoffs for deviant behaviors. Let us say that in both cases child deviant behaviors are effective in terminating conflict about 40% of the time. What is missing here is information about the number of trials where this training occurs. Some families may have only 10 conflicts during the observation sessions whereas another family may have 100. Patterson and colleagues assume that if we know how often the training occurs and how high the relative payoffs are during training, then we should be able to account for individual differences in antisocial behavior. Assuming the model is correct, we should be able to observe mother-child interactions in the home and use the two variables (relative payoffs and frequency of conflict) to account for much of the variance in antisocial outcomes.
	As of the date of this chapter, Patterson has carried out two different studies that test this hypothesis. Observation data were collected in a laboratory setting with mothers and preschool children. Five hours of data served as the base for calculating the relative payoff and the conflict density scores. These data were used to predict child aggressiveness in interacting with the mother in the same setting a week later. The density and relative payoff variables accounted for 76% of the variance in the observed deviancy rates for children. The results were promising, but the samples were small. Confirmation has been reported in studies by other investigators.
	This seems to be a reasonable beginning. The variables seem to explain at a microlevel what families do that causes coercive and antisocial behavior. The combination of conflict density and relative payoffs consistently accounts for sufficient variance in a range of criterion variables to suggest that the procedures are appropriate to the model. Patterson and colleagues are only beginning to develop a complete account of this process. The studies need to be expanded to include the analyses of siblings and peers and to include relative payoffs for prosocial and deviant behaviors in nonconflict episodes. These studies together with an examination of genetic variables constitute the agenda for the next decade of work in this area.
14.	Early-onset parenting Models. A diagram of the early-onset parenting model (Figure 1) was distributed in class on 3-26-02. As shown in the figure, changes in discipline, monitoring, and family problem solving will produce changes in the contingencies provided by family members and peers for both prosocial and deviant behaviors. For example, an increase in parental monitoring might result in the child spending less time with deviant peers and more time in activities supervised by adults. Similarly, the use of more effective family problem solving could result in more clearly defined rules and the use of more effective negative sanctions for deviant behavior.
	Can a parenting model be replicated across samples, and how much variance in criterion measures of antisocial behavior are accounted for by such models. The findings (Figure 2) from three different studies were distributed in class on 3-26-02. Each of the studies used similar indicators to define three major constructs in the parenting model. The models accounted for a range of 30% to 52% of the variance in antisocial behavior. Such findings attest to the robustness of the model. The model has been replicated with a large sample of normal adolescents. It has also been replicated with families of normal preschool boys and girls.
	Antisocial children are doubly handicapped. Not only have they been trained to be deviant, but their families have failed to train them in the prosocial skills required for survival with normal peers and in the classroom. When these antisocial children enter a school, the effect of the dual nature of their problems is to produce a cascade of new problems. The antisocial children's deviant behaviors and omitted social-skill training lead to rejection by normal peers, academic failure, and depressed moods. Once children have acquired high rates of antisocial behavior, it becomes even more difficult to teach prosocial skills.
	The key hypothesis for the early-onset parenting model was that a composite of antisocial behavior measured in childhood could predict early first arrest (prior to 14 years of age). Data were collected from an at-risk sample living in the higher crime areas of a medium-sized metropolitan community. Police arrest data were collected on a yearly basis on boys from 10 to 14 years of age. A regression analysis showed that poor parenting skills and lower socioeconomic status contributed to increased risk for early arrest.
	Patterson and his colleagues hypothesized that it would be possible to predict early onset juvenile offending more effectively than to predict juvenile offending in general. The reason for this is that the predictors that are effective for early-onset delinquency are thought to be different from those that are most effective in predicting late-onset delinquency.
	Data for the antisocial indicators were assessed for children at 10 years of age. Police arrest data were collected each year through age 18. A median split analysis of this data (Figure 3) was distributed in class on 3-26-02. It successfully identified 47 of the 53 boys arrested by age 14. The very low false negative error of 5.8% means that the predictor missed only 6 of the 53 boys arrested early. The hit rate was 69.9%. Probably the best means for estimating the overall efficiency of the 2 X 2 table is to calculate the relative improvement over chance (RIOC). With this data, the RIOC value is 77.4% and is considerably higher than comparable values for other longitudinal prediction studies.
15.	Late-onset Model. The late-onset families live in contexts that place them at moderate risk, as do their somewhat limited parenting skills. As children, these boys tend to show about average levels of antisocial behavior; they also tend to be at least marginally socially skilled. Patterson and his colleagues hypothesized that members of the deviant peer group play a major role in training the late-onset boys in their movement from childhood forms of coercive and antisocial acts into juvenile offenses. An early-onset child may end up unsupervised and on the street as early as age 6 or 7. In the late-onset boys, the disruption of adult efforts to supervise takes place much later--during early to middle adolescence. Notice that the timing for the deviant peer-based metamorphosis is different for the two groups. What the boys bring to the process is also very different. The early-onset boys are both extremely deviant and extremely unskilled, whereas the late-onset boys are both moderately deviant and moderately skilled. The difference in starting points almost guarantees differences in growth patterns.
	Patterson and his colleagues hypothesized that involvement with deviant peers leads to the rapid development of covert (substance use, stealing, truancy, etc.), but not overt antisocial acts for both groups. They also assume that training for overt acts, such as temper tantrums and physical hitting, takes place only in interactions with siblings and parents. If this is true, then both early and late groups should evidence growth in covert antisocial acts. However, the growth will occur at different times, starting in childhood for the early onset group and in middle adolescence for the late-onset group.
	Eventually the results of the training of late-onset boys in covert antisocial acts are noticed by parents and teachers. The increase in covert acts is followed by increasing risk of late-onset police arrest. This general model (Figure 4) was distributed in class on 3-26-02. Again, the contribution of contextual factors to the development of the marginal prosocial and antisocial skills is thought to be mediated by parenting practices.
	a.	Marginality Hypothesis. The marginality hypothesis is that late-onset boys are less deviant than early onset boys, but more deviant than nondelinquent boys.
	b.	The Social Skills Hypothesis. Three groups (early, late, and nondelinquent boys) were compared on mean values of four social skills measures (child social skill, poor peer relations, academic achievement, and self esteem). The hypothesis that late-onset boys are marginally deficient in social skills was confirmed. Table 1 Social Skills Tests for the Marginality Hypothesis was distributed in class on 3-28-02. Evidence from several studies suggested that as young adults, late onset boys who possessed even marginal levels of work- and relationship-enhancing skills would be more likely to desist in their offending. They would find payoffs for prosocial activities relatively higher than the payoffs for continued offending.
	c.	Contextual and Parent-Practice Variables. Patterson and his colleagues also hypothesized that the marginality hypothesis could be extended to include contextual and parenting-skill variables as well. Presumably the late-onset group would emerge from contexts that were midway between the nondelinquent and early-onset groups. Families of late-onset youth would be characterized by marginally skilled discipline practices. The findings from this analysis were reported in Table 2 which was distributed in class on 3-28-02. The data showed that, as compared to late-onset families, the early onset families were embedded in more negative contexts. They had significantly more transitions (from intact to single-parent families, etc.), higher parental antisocial behavior, lower likelihood of parent employment, and lower socioeconomic status. The two groups did not differ significantly on income.
		As compared to the families of nondelinquent boys, late-onset families showed significantly higher parental antisocial behavior and lower income. The two groups did not differ on transitions, parent employment, socioeconomic status, or discipline. Finally, the levels of childhood antisocial behaviors were also consistent with the marginality hypothesis. The early-onset boys were significantly more antisocial than were late onset boys, who in turn were significantly more antisocial than were uninvolved boys.
	c.	Contribution of Disrupted Monitoring. It was hypothesized that for the late-onset boys, the role played by parenting skills varies from one developmental phase to another. During childhood, parental discipline played important roles in determining the marginal levels of deviancy and prosocial skills. However, in early to middle adolescence, it was disruptions in parental monitoring that contributed indirectly to juvenile offending. The assumption was that parents can and do control the amount of time the adolescent invests in deviant peers.
		Studies from various sources show that during adolescence there is increasing conflict between adolescents and their parents. Patterson and colleagues hypothesized that these conflicts would be associated with disrupted parental monitoring. They also assumed that outside disrupters might also be associated with ineffective parental supervision. To test this idea, a disrupter risk score was formed where each variable was scored 0 or +1. The list of disrupters included unemployment or recent financial loss, a severe illness or death, family transition (divorce, etc.), change of residence, or pubescence. An event was scored only if it had occurred during the interval when the boys were 11 to 12 years of age.
		The present analyses included all boys scoring below the median on the antisocial construct at grade 4. Figure 5 was distributed in class on 3-28-02. All path coefficients were significant. As expected, the disrupter variables are related to less effective parental monitoring; parental monitoring in turn is related (path .61) to increasing involvement with deviant peers. In Figure 5, increasing disruption was correlated with increasing conflict between parent and adolescent. This conflict in turn contributes to further involvement with deviant peers, which offers nice support for the hypothesis that parent-adolescent conflict fuels a "flight to the peer group." The findings are consistent with the idea that training parents to be more vigorous in their efforts to supervise and monitor their adolescent may be an important deterrent for juvenile offending.
	d.	Involvement with Deviant Peers. As noted earlier, involvement with deviant peers plays a primary role in the early-onset model's metamorphosis of childhood antisocial acts to delinquent activities. Deviant peers play a comparable role in the late-onset model, where they both model and shape the performance of antisocial attitudes and covert antisocial acts, such as substance abuse and truancy. Eventually the modeling and support include behaviors that would be classified as juvenile offenses, and this in turn is related to increasing risk of police arrest.
	e.	Availability of Deviant Peers. Societies and individual families differ in their approaches to granting autonomy to adolescents. Much of the uncertainty revolves around the issue of unsupervised time (i.e., when, where, with whom, and doing what?). The question is when and in what ways should parental supervision be relaxed. If the controls are relaxed too soon, the child may be at risk of hanging out with deviant peers. Alternatively, if the controls remain in place too long, the adolescent is at risk of never becoming autonomous.
		The ambiguity of the parental monitoring role increases with the age of the adolescent; this is further exacerbated by the increasing availability of deviant peers. Our public schools may maximize contacts with deviant peers. In an earlier Patterson study, over half the boys that the target boys hang out with were met in the school setting. It is the ideal setting for contacts with members of the deviant peer group. During early and middle adolescence, most of the peer group reports to this setting 5 days a week. During that time there are substantial periods of time when the interaction cannot be closely supervised by adults, and there are some settings where there is not adult supervision at all. In many ways, this setting is ideal for shopping among members of the peer group for a wide range of partners, from crime to marriage.
		Several studies speak to the fact that, by and large, antisocial children and adolescents are not social isolates. They belong to extensive and rapidly changing social networks. It seems that deviant peers may be ubiquitous presence in many of these social groups. An extensive self-report study provides a detailed portrayal of just how commonplace contacts with deviant peers for youth between the ages of 11 and 24 years. Figure 6 was distributed in class on 3-28-02.
		During childhood, less than a quarter of the boys reported contacts with groups that contain moderate to high densities of deviant peers. From early to middle adolescence, there is rapid growth in such contacts, which peaks between the ages of 16 and 18 years. For example, by age 14, about 40% of the boys report contact with groups that contain moderate to high densities of deviant members. By age 17, this applies to slightly over half of the boys.
		The school setting is rich in possibilities for contacts with deviant peers. We assumed that the parents can and do have an impact on the frequency and duration of these contacts that occur in the evenings and on the weekends. We assumed that it is at such times that the intensive training for antisocial attitudes and for covert acts actually take place.
	f.	Selective Seekers. Patterson and colleagues hypothesized that each individual actively shops among settings, activities, and peers in such a manner as to maximize the relative payoffs available. There is a cluster of three hypotheses required for the selective seeker formulation. The first hypothesis is the well-known "likes tend to select likes." Presumably this is a general principle that applies to the selection of friends, lovers, marriage partners, and perhaps business associates as well. Eight studies have shown that bright people tend to self-select bright people to intermarry. Patterson and his colleagues found a correlation of .39 (p=.001) between latent constructs for antisocial fathers and antisocial mothers. Another study showed that antisocial boys tend to select one another to "hang out with." A corollary hypothesis is that the more antisocial the child, the earlier he or she will become a member of a deviant peer group.
		The second assumption of the selective seeker model is that children select friends with whom to hang out who maximize the individual's payoffs. An observational study of the playground interactions of preschool children, computed cost-benefit ratios for children dyads. Maximizing payoffs was defined as the ratio of positive to negative exchanges. The data in fact showed correlations in the .6 to .7 range between friendship preferences and observed cost-benefit ratios. Children preferred as friends those peers who provided a higher ratio of higher to negative exchanges.
		The third assumption of the selective seeker model is that deviant individuals are more likely to reinforce each other for deviancy. If that is true, for what do deviant individuals reinforce each other? Clinical work with groups of deviant adolescents suggests that they use rich schedules of positive reinforcement for antisocial talk and behaviors. The clinical impressions were strongly supported by an observation study in a corrections institution. Both deviant peers and staff were more likely to provide positive reinforcement for deviant behavior than for prosocial behavior.
		These findings were replicated and extended in a series of studies by Dishion. Analyses of videotaped interactions between dyads of boys who spent a good deal of time together hanging out showed a significant match between the relative rate of rule-breaking talk for the interchange and the relative rate of payoff by the peer for rule-breaking talk. In the normal dyads the same reinforcement was available, but it was contingent upon prosocial talk. Deviant dyads mutually reinforce each other for deviant talk. Even a casual viewing of these tapes leaves one with the strong impression that one is observing a process that leads to the formation of antisocial attitudes. And it can serve as an analogue for what would occur in a natural setting where rule-breaking talk can quickly change to rule breaking behavior.
16.	Implications
	a.	A case was made for early and late onset as being two different paths to juvenile offending. Early onset boys come from more disadvantaged homes and are significantly more deviant and less socially skilled than are late-onset boys. The term marginally adjusted seems to be an apt description for the late-onset group.
	b.	It has been known for some time that early-onset boys were at greater risk for long-term negative outcomes, such as chronic and violent offending and perhaps adult offending as well. Earlier theories have done little to explicate the process that produces these outcomes. Patterson assumes that the proximal variable that accounts for risk of arrest is the overall frequency of antisocial acts. Presumably the individual is first trained in a variety of overt and covert antisocial acts under the tutelage of the deviant peer group. The antisocial acts are then followed by the emergence of a wide spectrum of trivial and nontrivial delinquent acts. In the predictive sense, one precedes the other. The implication is that the process can be interrupted by reducing the level of antisocial behavior. This in turn would be followed by a corresponding decrease in risk for offending as shown in research by Patterson and his colleagues.
	c.	The other facet of the early-onset and late-onset models that is crucial to the understanding of long-term negative outcomes concerns the role of antisocial behavior as a determinant for the timing (as well as frequency) of juvenile offending. Extremely high levels of antisocial behavior in childhood are associated with the very early emergence of juvenile offending and accompanying risk of earliest police risk. Presumably this comes about because the extremely deviant child has simply overwhelmed any parental efforts to supervise or control. The young, extremely antisocial child starts earliest and is also the most social unskilled.
	d.	Parental skills play a key role in both the early-onset and late-onset models. Presumably these parenting practices in turn control the reinforcing and punishing contingencies for the child's behavior. Three studies have been designed to test this prediction. One study has been completed. The findings showed that Parent Training Therapy produced significant improvements in parenting scores and the magnitude of improvements in parenting correlated significantly with the magnitude of improvements in teachers' ratings of antisocial behavior.
	e.	The late-onset boys not only start later but also perform juvenile offenses at lower rates. But the key set of findings has to do with the fact that by the time late-onset boys start offending, they are already moderately skilled in work and relational areas. Patterson predicts that, in the long run, the presence of these social skills will generate the relative payoffs that pull them out of the deviancy.
	f.	The early-onset and late-onset models bear a close resemblance to factor analysis studies of delinquents carried out in the 1940s. These studies produced two groups labeled "socialized delinquents" and "unsocialized delinquents." The unsocialized delinquent sounds like the prototype of Patterson's early-onset boys. The socialized delinquent hangs out with deviant peers, has some modicum of social skills, and tends to have only a transitory career in crime. The Patterson research, however, goes a long way toward pinpointing the process that leads to early-onset and late-onset delinquency.