A Life-Course View of the Development of Crime ROBERT J. SAMPSON and JOHN H. LAUB In this article, the authors present a life-course perspective on crime and a critique of the developmental criminology paradigm. Their fundamental argument is that persistent offending and desistance—or trajectories of crime—can be meaningfully understood within the same theoretical framework, namely, a revised age-graded theory of informal social control. The authors examine three major issues. First, they analyze data that undermine the idea that developmentally distinct groups of offenders can be explained by unique causal processes. Second, they revisit the concept of turning points from a time-varying view of key life events. Third, they stress the overlooked importance of human agency in the development of crime. The authors' life-course theory envisions development as the constant interaction between individuals and their environment, coupled with random developmental noise and a purposeful human agency that they distinguish from rational choice. Contrary to influential developmental theories in criminology, the authors thus conceptualize crime as an emergent process reducible neither to the individual nor the environment. Keywords: crime; development; trajectories; life course; typologies; prediction; desistance In this article, we argue for a life-course perspective on trajectories of crime, focusing on the question of whether (and why) adolescent Robert J. Sampson is chairman of the Department of Sociology and Henry Ford II Professor of the Social Sciences at Harvard University. His recent workfocuses on the limits of the prediction paradigm in criminology, durable forms of urban inequality, networks of community social organization, and theories of civil society. John H. Laub is a professor of criminology and criminal justice in the Department of Criminology and Criminal Justice at the University of Maryland at College Park. His areas of research include crime and deviance over the life course,juvenile delinquency and juvenile justice, and the history of criminology. He has published widely, including, most recently, Shared Beginnings, Divergent Lives: Delinquent Boys to Age 70, coauthored with Robert Sampson (Harvard University Press, 2003). This book is the 2004 recipient of the Michael J. Hindelang Book Award from the American Society of Criminology and the 2005 recipient of the Outstanding Book Award from the Academy of Criminal justice Sciences. DOL 10.1177/0002716205280075 12 ANNALS, AAPSS, 602, November 2005 Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 13 delinquents persist or desist from crime as they age across the adult life course. The growing tendency in developmental perspectives on crime, often called "developmental criminology," is to subdivide the offender population and assume different causal influences at different stages of the "criminal career." For example, it is now commonplace to assert that certain childhood factors uniquely explain persistent adult offenders, whereas another set of causal factors explain desistance in adolescence. A variation on this theme is that a small group of offenders continue to commit crimes at a persistently high rate as they grow older. In direct contrast, another view posits an "invariant" effect of age—that regardless of stable between-individual differences, all offenders will commit fewer crimes as they age. Although at first it may seem counterintuitive, our fundamental argument is that persistent offending and desistance—and hence trajectories of crime—can be meaningfully understood within the same theoretical framework. We do not argue that offender typologies are without merit, and in fact some of our analysis will estimate group-based trajectories of crime. Rather, our strategy is to start with the assumption of generality and see how far it takes us in understanding patterns of criminal offending across the full age range of the life course. We explore this logic in five sections, beginning with a summary of results from our prior research. This work serves as our point of departure for new analyses and theoretical reflection on key issues in life-course criminology. We specifically review, albeit in brief, the main results from Crime in the Making: Pathways and Turning Points through Life (Sampson and Laub 1993) and the more recent Shared Beginnings, Divergent Lives: Delinquent Boys to Age 70 (Laub and Sampson 2003) that bear most directly on this article. We then take on three major issues. 1. A life-course view of the idea of developmentally distinct groups that have unique causes. Here we revisit our position on typologies of crime, focusing on the dual taxonomy theory of Moffitt (1993) with the goal to identify points of agreement and disagreement. We present new analyses on the predictability of age at desistance and the life-course trajectory of crimes that are minor in nature. According to Moffitt (1993, 1994, forthcoming), to assess the validity of a life-course persistent versus adolescent-limited typology of offenders one must consider a sufficiently broad range of criminal and antisocial behaviors. We agree. We also concur that offender trajectory groups are of continuing analytic value and that there are men who offend at a high rate in adulthood. The main points of disagreement appear to be that (1) we find life-course desistance is the norm for all men and all crimes, including minor forms of deviance; and (2) we question the prospective or predictive power of offender groups and whether they are causally distinct with respect to later trajectories. NOTE: This article was presented at the Albany Symposium on Crime and Justice, "Developmental Criminology and Its Discontents: Offender Typologies and Trajectories of Crime," State University of New York at Albany, April 28-29, 2005. We thank Elaine Eggleston Doherty and Stacey J. Bosick for their assistance. Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 14 THE ANNALS OF THE AMERICAN ACADEMY 2. A revised life-course view of turning points. Unlike unusual events (e.g., Great Depression, war), many events are frequently recurring—people move in and out of various states over the life course in a repeated events fashion. We ask, How is this fact reconciled with the notion of long-term "development" or "growth"? We observe that developmental theory works well with many phenomena—the question here is what about crime and its time-varying predictors? We use marriage as a prime example and highlight results of work in progress that attempts to estimate the causal effects of marriage on crime. We find that conditioning on long-term histories of both outcome and treatment, the same man exhibits lower rates of crime in the state of marriage compared to not being in the state of marriage. We discuss how this finding fits a developmental theory of crime and the general idea of turning points. 3. A life-course view that takes human agency seriously. Developmental criminology, in practice if not in theory, tends to emphasize the notion that people get "locked" into certain trajectories. One of the lessons of prospective longitudinal research is that there is considerable heterogeneity in adult outcomes that cannot be predicted in advance. In this section, we highlight a life-course view that emphasizes human agency and choice over the life span, underscoring how people construct their lives within the context of ongoing constraints. From this view, trajectories are interpreted not from a lens of unfolding inevitability but rather continuous social reproduction. We want to ask the hard question of how men with a criminal past go about prospectively creating their own trajectories. The final section of the article considers the implications of these findings and theoretical reflections for the conception of development generally and life-course criminology in particular. Crime in the Making and the Origins of Life-Course Criminology Unravelingjuvenile Delinquency, along with subsequent follow-ups conducted by Sheldon and Eleanor Glueck of the Harvard Law School, is one of the most influential research projects in the history of criminological research (Glueck and Glueck 1950, 1968). The Gluecks' data were derived from a three-wave prospective study of juvenile and adult criminal behavior that began in 1940. The research design involved a sample of five hundred male delinquents aged ten to seventeen and five hundred male nondelinquents aged ten to seventeen matched case by case on age, race/ethnicity, IQ, and low-income residence in Boston. Extensive data were collected on the one thousand boys at three points in time—ages fourteen, twenty-five, and thirty-two. Over the period 1987 to 1993, we reconstructed, augmented, and analyzed these longitudinal data that, owing to the Gluecks' hard work over many years, are immensely rich and will likely never be repeated given modern institutional review board restrictions (e.g., wide-ranging interviews with Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 15 teachers, neighbors, and employers; detailed psychiatric and physical assessments; extensive searches of multiple agency records). In Crime in the Making, we developed a theoretical framework to explain childhood antisocial behavior, adolescent delinquency, and crime in early adulthood. The general organizing principle was that crime is more likely to occur when an individual's bond to society is attenuated. Our analysis of the causes of delinquency shared much in common with the focus in classical control theory (Hirschi 1969) on adolescence, but the reality of later life-course milestones required us to develop a modified theoretical perspective. After all, the transition to young adulthood brings with it new social control institutions and potential turning points that go well beyond adolescence. We thus developed an age-graded theory emphasizing informal social controls that are manifested in shifting and possibly transformative ways as individuals age (see Sampson and Laub 1993). For example, we focused on parenting styles (supervision, warmth, consistent discipline) and emotional attachment to parents in childhood; school attachment and peers in adolescence; and marital stability, military service, and employment in adulthood. Although these are manifestly distinct domains that are age graded, we argued that there are higher-order commonalities with respect to the concept of social connectivity through time. Stability and change in criminal behavior over the life course The delinquents and nondelinquents in the Gluecks' study displayed considerable between-individual stability in crime and many problematic behaviors well into adulthood. This stability held independent of age, IQ, ethnicity, and neighborhood SE S. Indeed, delinquency and other forms of antisocial conduct in childhood were strongly related to troublesome adult behavior across a variety of experiences (e.g., crime, military offenses, economic dependence, and marital discord). But why? One of the mechanisms of continuity that we emphasized was "cumulative disadvantage," whereby serious delinquency and its nearly inevitable correlates (such as incarceration) undermined later bonds of social control (such as employ-ability), which in turn enhanced the chances of continued offending (see also Sampson and Laub 1997). At the same time, we found that job stability and marital attachment in adulthood were significantly related to changes in adult crime—the stronger the adult ties to work and family, the less crime and deviance among both delinquents and nondelinquent controls. We even found that strong marital attachment inhibits crime and deviance regardless of that spouse's own deviant behavior and that job instability fosters crime regardless of heavy drinking. Despite differences in early childhood experiences, adult social bonds to work and family thus had similar consequences for the life-course trajectories of the five hundred delinquents and five hundred nondelinquent controls. These results were consistent for a wide variety of crime outcome measures, control variables (e.g., childhood antisocial behavior and individual-difference constructs), and analytical techniques ranging from Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 16 THE ANNALS OF THE AMERICAN ACADEMY methods that accounted for persistent unobserved heterogeneity in criminal propensity to analyses of qualitative data. Taken as a whole, these findings suggested to us that social ties embedded in adult transitions (e.g., marital attachment, job stability) explain variations in crime unaccounted for by childhood propensities. This empirical regularity supports a dual concern with continuity and change in the life course. A fundamental thesis of our age-graded theory of informal social control was that whereas individual traits and childhood experiences are important for understanding behavioral stability, experiences in adolescence and adulthood can redirect criminal trajectories in either a more positive or more negative manner. In this sense, we argue that all stages of the life course matter and that "turning points" are crucial for understanding processes of adult change. Drawing on the life-course paradigm (Elder 1985), we conceptualized a turning point as an alteration or deflection in a long-term pathway or trajectory that was initiated at an earlier point in time (see also Rutter 1996). Shared Beginnings, Divergent Lives: An Overview Crime in the Making raised many unanswered questions, and in its concluding chapter we highlighted directions for future research and theoretical development that appeared fruitful. Two of these directions seemed especially relevant for developmental/life-course theories of crime, namely, the merging of quantitative and qualitative data and further understanding of age and crime (Sampson and Laub 1993, 251-53). For example, what about crime in middle age? Older age? Is there really such a thing as a lifelong career criminal—or what have been dubbed "life-course persisters" (LCPs)? If so, can this group be prospectively identified? Another set of questions turned on the use of qualitative narratives to delve deeper into a person-based exploration of the life course. Can narratives help us unpack mechanisms that connect salient life events across the life course, especially personal choice and situational context? In our view, life-history narratives combined with quantitative approaches can be used to develop a richer and more comprehensive picture of why some men persist in offending and others stop. We made moves toward a narrative-based inquiry in Crime in the Making but were forced to rely on the Gluecks' written records rather than our own original interviews. These motivations led us to follow up the Glueck men to the present. Our study involved three sources of new data collection—criminal record checks (local and national), death record checks (local and national), and personal interviews with a sample of fifty-two of the original Glueck delinquents. The sample of men to interview was strategically selected to ensure variability in trajectories of adult crime. More specifically, using criminal history records we classified eligible men into strata that reflected persistence in crime, desistance, and "zigzag" offending patterns, including late desistance and late onset of violence (see Laub and Sampson 2003, chap. 4, for more details). The combined data represent a roughly fifty-year window from which to update the Glueck mens lives at the close of the twentieth Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 17 century and connect them to life experiences all the way back to early childhood. We believe these data represent the longest longitudinal study to date in criminology of the same men. The following sections briefly summarize the key findings. Age and crime Our analyses showed that, on one hand, the aggregate age-crime curve is not the same as individual age-crime trajectories, lending apparent support to one of the major claims of the criminal career model. There is enormous variability in peak ages of offending, for example, and age at desistance varied markedly across the Glueck men (Laub and Sampson 2003, chap. 5). On the other hand, we found that crime declines with age even for active offenders and that trajectories of desistance cannot be prospectively identified based on typological accounts rooted in childhood and individual differences. That is, offenses eventually decline for all groups of offenders identified according to extant theory and a multitude of childhood and adolescent risk factors. Whether low IQ, aggressive temperament, or early onset of antisocial behavior, desistance processes are at work even for the highest-risk and predicted life-course persistent offenders. While prognoses from childhood factors such as these are modestly accurate in predicting stable differences in later offending, they did not yield distinct groupings that were valid prospectively for troubled kids. Not only was prediction poor at the individual level, our data raised questions about the sorts of categorically distinct groupings that dominate theoretical and policy discussions (e.g., "life-course persistent offender," "superpredator"). These groupings tended to wither when placed under the microscope of long-term observation (Laub and Sampson 2003, chap. 5; Sampson and Laub 2003). We thus concluded that a middle-ground position was necessary in the criminal careers debate—yes, there is enormous variability in individual age-crime curves such that it renders the aggregate curve descriptive of few people, and yes, age has a direct effect on offending such that life-course desistance is the more accurate label. We believe this compromise position, which we subject to further testing in this article, has general implications for assessing key assumptions of developmental criminology and rethinking its conceptual meaning. Mechanisms of desistance A second goal of our book was to exploit life-history narratives to better understand patterns of stability and change in offending over the life course. In our narrative interviews, we asked the men to describe turning points in their life. We also had the men fill out life-history calendars so that we could more accurately determine the sequencing of major life events. Several turning points were implicated in the process of desistance from crime, including marriage/spouses, military service, reform school, work, and residential change. The mechanisms underlying the desistance process are consistent with the general idea of social control. Namely, what appears to be important about institutional or structural turning points is that they all involve, to varying degrees, (1) new situations that "knife off' the past from Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 18 THE ANNALS OF THE AMERICAN ACADEMY the present, (2) new situations that provide both supervision and monitoring as well as new opportunities of social support and growth, (3) new situations that change and structure routine activities, and (4) new situations that provide the opportunity for identity transformation (for details, see Laub and Sampson 2003, chaps. 6-8). The lesson we drew is that involvement in institutions such as marriage, work, and the military reorders short-term situational inducements to crime and, over time, redirects long-term commitments to conformity. In making the case for the importance of the adult life course, we have referred to involvement in these institutions as turning points because they can change trajectories over time (Laub and Sampson 1993; Sampson and Laub 1993). [Wjefind life-course desistance is the norm for all men and all crimes, including minor forms of deviance; and . . . we question the prospective or predictive power of offender groups and whether they are causally distinct with respect to later trajectories. A potential objection, however, is that turning points are a result of selection bias or, put differently, the unobserved characteristics of the person (e.g., Gottfredson and Hirschi 1990). To shed further light on life events, we exploited the longitudinal nature of the long-term data to examine within-individual change, where the unit of variation is across time. As such, stable characteristics of the person are held constant and we can exploit changes in social location, such as marriage, in terms of deviations from a person s expected trajectory. Holding age constant and allowing individual heterogeneity in age effects, we found that when in a state of marriage, the propensity to crime was lower for the same person than when not in marriage. Similar results were found for military service and steady employment. Quantitative models of within-individual change thus give statistical evidence of the probabilistic enhancement of desistance associated with life-course events like marriage, military service, and employment (Laub and Sampson 2003, chap. 9). With this brief summary as a backdrop, we can now turn to the heart of the current article's concern with group-based typologies, turning points, and human agency. Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 19 Group-Based Typologies A small group of persons is shown engaging in antisocial behavior of one sort or another at every stage of life. I have labeled these persons life-course-persistent to reflect the continuous course of their antisocial behavior. (Moffitt 1993, 676, italics added) Thus, in defiance of "regression to the mean," extremely antisocial persons remain extreme on measures taken at later ages and in different situations. (Moffitt 1994, 10) The "group" question is one of the most salient in modem developmental criminology. Here we revisit issues relating to Moffitt s (1993) dual taxonomy theory and group-based theories more generally (see Patterson and Yoerger 1993; Loeber and Hay 1997). One of the major strengths of our data is that they allow us to examine within-individual variability in crime over nearly the entire life course. Moreover, the original design in Unravelingjuvenile Delinquency targeted serious, persistent delinquents in adolescence, providing an important opportunity to assess patterns of continuity and change in crime for a population of high interest and concern to both criminal career theory and policy efforts that target high-risk children. The question we address here is whether our tests to date set up a "straw man" argument. In response to our research, Moffitt (forthcoming) claimed as much, arguing first and foremost that nowhere does her theory predict a "flat rate" of criminal offending for LCPs. Bypersistent, in otherwords, she simply means "high rate" over time in the between-individual mode of comparison (i.e., stability). We acknowledge that we did (and still do) read in the original theory an insistence that adult crime should be relatively flat for the distinct group of LCPs, for that is how persistence is typically defined (e.g., "continuous," "degrading only slowly"). Moreover, if the differences of note are really on a continuum with respect to levels of offending, then the idea of a distinct group is weakened. In defense of our interpretation, we would note that many others apparently read the theory in a similar way. Consider the following independent assessment (other similar interpretations are found in Cullen and Agnew2003,450; Thomberry 1997,2; Benson 2001,86): The second group of offenders in Moffitts taxonomy, "life-course-persistent," is hypothesized to engage in antisocial activities and criminal acts throughout the life span.... Unlike their adolescence-limited counterparts, life-course-persistent offenders continue their criminal involvement throughout most of their lives (i.e., they are unlikely to desist). (Piquero, Farrington, and Blumstein 2003, 398, italics added) Even more striking, consider Nagin's (2005, 183) recent book where he graphs a flat expectation trajectory derived from Moffitts theory and states, "The life-course persistent trajectory is flat and high, whereas the adolescent-limited trajectory rises and falls with age" (p. 182). Moffit s (forthcoming) clarification clears the air considerably for if the theory is that "high rate yet declining with age" equals life-course persistent, then we have little or no disagreement, and in fact our data (and that of many others) clearly sup- Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 20 THE ANNALS OF THE AMERICAN ACADEMY ports the assertion that there are high-rate offenders. We remain a bit puzzled, however, because this concept would then seem to revert back to the classic "chronic offenders" from the Wolfgang, Figlio, and Sellin (1972) birth cohort study. In other words, in the clarified or revised position, it is not clear to us what is new or different in the "life-course-persistent" concept versus "chronic" other than the label. A second line of critique is that we did not examine a population-based study and as a result cannot fully test the dual taxonomy theory. We fully agree in one important sense—we cannot assess the validity of the adolescence-limited hypothesis. Much of the testing of Moffitt's (1993) theory requires a population-based sample, and the limitations of our data conflict with the ideal testing conditions she prefers. So we are in agreement here as well. The apparent exception is that we remain convinced that our data are quite relevant to examining long-term trajectories of crime and thus the existence of life-course persistent offender groups. As we have argued elsewhere (Laub and Sampson 2003,113), it would be hard to write an analytic script that would be more conducive to finding troubled adult men than the one laid out in the behavioral story of the delinquent group in the Gluecks' Unraveling Juvenile Delinquency (Glueck and Glueck 1950). These five hundred men generated some ten thousand criminal and deviant offenses to age seventy. [W]e highlight a life-course view that emphasizes human agency and choice over the life span, underscoring how people construct their lives within the context of ongoing constraints. From this view, trajectories are interpreted not from a lens of unfolding inevitability but rather continuous social reproduction. Thus, it seems not at all a straw-man argument to say that if we cannot find convincing evidence that a life-course persistent group can be prospectively identified in these data based on theoretical risk factors at the individual level in childhood and adolescence, then that aspect of the theory is in trouble. Our finding confirms what some have called the Robins paradox, namely, antisocial behavior in children Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 21 is one of the best predictors of antisocial behavior in adults, yet most antisocial children do not grow up to be antisocial adults (Robins 1978). In retrospect, high-rate adult offenders will almost always be drawn from the pool of high-risk children, but looking forward from high-risk children, we cannot distinguish well who will persist or desist as adults. A third line of critique of our work was advanced by Blumstein (2003), who suggested that if one calculated rates of offending ("lambda") among "active offenders" as well as offenders distinguished by crime types and various combinations thereof, evidence of LCPs might be found. We are open to the possibility that if one decomposes the data into smaller and smaller subgroups, a subset of men may be found that in retrospect appear to look somewhat flat in their offending for some period of the adult life course. But to our mind, these exceptions may merely prove the rule. Here we are admittedly old-fashioned in our approach to data—if one has to search hard and long for patterns that one cannot otherwise see, we are skeptical about the replicability and generality of the results, especially in light of sample selection strategies based on the dependent variable. We further wonder about the overall import of the findings for theory and public policy (for an earlier exchange along similar lines, see Gottfredson and Hirschi 1986; Blumstein, Cohen, and Farrington 1988). For us, the key question remains: Is there a predictable group of offenders who commit crime at a high rate and maintain that high rate of offending over the full life course with some degree of persistence? Revisiting the predictability of persistence We now turn to critiques of our work that we address with further data analyses. Moffitt (forthcoming) contends that we did not consider offenses by LCPs that are minor in nature from a legal perspective but that nonetheless capture important dimensions of deviant or antisocial activities in adulthood. Crime, in the sense of serious predatory offending, for example, might be declining over time but "bad behavior" will not. We assess this important argument by measuring within-individual variations in relatively minor offenses that tap various types of deviant behavior. Specifically, we calculated person-year counts of "other" offenses recorded in arrest histories—disorderly conduct, vagrancy, gambling, speeding, conspiracy, lewdness, impersonation of apolice officer, resisting arrest, desertion, nonsupport, and hunting near a dwelling. Although these offenses are by definition violations in a legal sense, they reflect the type of antisocial tendencies that Moffitt has emphasized, especially family conflict (see Moffitt 1993, 680; Moffitt, forthcoming). It is further true that the information we analyze is by definition based on official record keeping, but we would emphasize that our comparisons are within individuals. It is hard to imagine why a fifty-five-year-old man, for example, compared to the same man at fifty, would be any more or less likely to be arrested for nonsupport of children or gambling. Within-individual trajectories do not compare different groups or cohorts of men with different characteristics often thought to influence processing (e.g., race and social class). And it turns out that the Glueck men as adults engaged in all sorts Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 22 THE ANNALS OF THE AMERICAN ACADEMY of deviant activities that the Boston police appeared only too happy to record— indeed, there were more than three thousand arrests for these minor offenses! We do not have data on things like being fired from jobs, but then again, that kind of behavior is not illegal or of a rule-breaking kind and thus remains outside the bounds of a theory of crime and deviance. To assess the predictability of trajectories of offending relating to miscellaneous minor offenses, we employ the validated child-risk predictor used in Sampson and Laub (2003) based on a summary of thirteen measures listed in Table 1. These measures are derived from multiple sources (parents, teachers, official records, and the boys themselves) that tap classic individual-difference risk factors and the observed propensity to offend of the boys in their early years and adolescence. Measures of individual differences include some of the most venerable and sturdy predictors of crime, especially cognitive abilities (Moffitt 1994,16), temperament (Moffitt 1993, 695), personality traits (Caspi et al. 1994; Hawkins et al. 2000), and childhood behaviors (Moffitt 1994,15). In addition, guided by the substantial body of research on criminal careers, we focused on early and frequent involvement in crime and delinquency (Blumstein et al. 1986, 72, 94). [CJrime is more likely to occur when an individuals bond to society is attenuated. Verbal intelligence (see Moffitt 1993) was assessed using the Wechsler-Bellevue IQ test and coded into eight categories ranging from one (120 and above) to eight (59 and below). The mean verbal IQ for the delinquent sample was 88.6. We also examine the full-scale IQ score that includes both math and verbal skills, unrecoded. From detailed psychiatric assessments of the boy, we use four dichotomous variables of personality traits: extroverted ("uninhibited in regard to motor responses to stimuli"), adventurous ("desirous of change, excitement, or risk"), egocentric ("self-centered"), and aggressive ("inclined to impose ones will on others"). To capture the early onset of childhood behavior, we used self-reported age of onset of misbehavior, a dichotomous indicator based on teacher and parent reports of the subject engaging in violent and habitual temper tantrums while growing up, and a report from the mother as to whether the subject was overly restless and irritable growing up (we labeled this "difficult child"). The level of delinquent conduct in adolescence was measured in several ways. We used an indicator of the average annual frequency of arrests in adolescence while not incarcerated and a composite scale (ranging from 1 to 26) based on unof- Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 23 TABLE 1 CHILD AND ADOLESCENT RISK FACTORS MEASURED BEFORE ADULTHOOD (YOUNGER THAN SEVENTEEN) Cognitive Measured intelligence (IQ, Full-Scale Wechsler-Bellevue) Verbal IQ (Wechsler-Bellevue) Psychiatric assessments Extroversion Adventurousness Egocentricity Aggressiveness Early onset/conduct disorder Age of onset of misbehavior (self-reported) Age at first arrest (police) Age at first incarceration (correctional) Violent temper tantrums (teacher and parent reports) Difficult child behavior (mother reports) Antisocial behavior Frequency of arrest per days free (up to age seventeen) "Unofficial" delinquency (self-, parent, and teacher reports) ficial self-, parent, and teacher reports of delinquent behavior (e.g., stealing, vandalism) and other misconduct (e.g., truancy, running away) not necessarily known to the police. Following the logic of the criminal career approach, we also included measures of the age at first arrest and age at first incarceration for each boy. Overall, the delinquency measures capture both the level and the developmental pattern of official and unofficial behavior up to an average of about fourteen years of age for each boy. To assess summary patterns, we followed the logic of risk factor theory by giving emphasis to the combination of individual-level risks within the person. We combined standardized indicators of all thirteen variables in a single child-risk indicator, with constituent items scored such that a high value indicated either the presence of antisocial behavior or an individual-level risk (e.g., low verbal IQ, engaging in tantrums, early age of onset of antisocial behavior, and so on). We then looked at the distribution across all boys and created a group at highest risk for what Moffitt would call life-course persistent offenders—namely, those boys in the upper 20 percent of the distribution. The bottom 80 percent group is defined as low risk. What is important to point out is that the groups were defined prospectively with respect to adult offending, as all the measurement was completed prior to age seventeen. Other than delinquency, which we separate out in a later analysis, the vast majority of measures refer to individual differences of the boys in childhood. The prospective ability of these measures to predict later involvement in crime was demonstrated in earlier work (Sampson and Laub 1993, 92). Thus, while retrospective reporting is a concern we fully acknowledge, the multimethod and multireporter approach, combined with the diversity of measures and their demonstrated validity in predicting stability of offending, speaks to the utility of Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 24 THE ANNALS OF THE AMERICAN ACADEMY considering the link between childhood risk and trajectories of crime throughout life. When we examine the predictive power of childhood risk groups, the dominant pattern is parallel offender trajectories by age and no evidence of distinct developmental pathways. We demonstrate this pattern in several different ways, beginning in Figure 1 with the predicted probability of offending for the most general outcome—total crime (sum of violent, property, alcohol/drug, and other offenses). This replicates the main picture painted in Laub and Sampson (2003, chap. 5). Next we turn to the "raw data" trajectories for the predatory crimes (violent and property offenses) emphasized in traditional criminological theory (see Figure 2). We present the raw data because some critics have wondered whether our smoothing of the data through age-expected trajectories might have masked subgroups of offenders. Even without smoothing, however, one sees a sharp rise and then decline in predatory offense counts for both risk groups, with the main difference in the level of offending. In Figure 3, even the relatively messy trajectories for alcohol- and drug-related offenses reveal remarkably similar patterns for each child-risk group. As predicted by Moffitt (1993, 1994), there are definitely men who offend well into middle age, yet as is also evident in our data, the same pattern holds for both childhood risk groups, and a sharp decline in offending (desistance) is the eventual pattern for all men. In Figure 4, we turn to the key findings predicting the miscellaneous offenses described earlier from age seven to seventy by childhood risk. Once again the same pattern is displayed as found for other offense types—both child-risk groups show parallel patterns of offending with sharp declines in crime by age. Figure 5 replicates these results using smoothed age-crime trajectories, with strikingly similar patterns. In short, there is no prospective evidence of a flat-line offending trajectory when we examine raw or smoothed age-crime patterns for various crime types, including minor forms of illegal activity and deviance. In analyses not shown, these findings hold up when incarceration and active offender designations are taken into account and when we disaggregate childhood risk into constituent measures (see also Sampson and Laub 2003). Childhood risk in family adversity As another test, we conducted analyses that interacted individual-risk characteristics, both the overall scale and constituent measures, with criminogenic family environments during the turbulent years of child and adolescent development. A long history of research, including on the Gluecks' data, has shown that family structural conditions (e.g., poverty, large family size, and residential mobility) and family social processes (e.g., poor supervision, erratic/threatening discipline, and weak parental attachment) are strong predictors of adolescent delinquency (see Sampson and Laub 1993, chap. 4). Moffitt (1993) argued that when a child's vulnerability is compounded with such negative family conditions, life-course-persistent offending is most likely. Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 25 FIGURE 1 PREDICTED TOTAL OFFENSE TRAJECTORIES: AGE SMOOTHED, SEVEN TO SEVENTY, RY CHILDHOOD RISK Age Drawing on Sampson and Laub (1993), we conducted a principal components analysis that reduced the dimensionality of a set of theoretically and empirically salient items measuring family adversity. Two key dimensions emerged, the first defined by high residential mobility, parental emotional instability, low maternal supervision, and hostility between father and son. Poverty, large families, and erratic/harsh methods of discipline defined the second dimension. We then selected those boys who were in the upper half of the distribution of each orthogonal factor (hence approximately 25 percent of the boys) and who were in the upper 20 percent of the distribution of the individual-level childhood-risk score. In other words, we examined the interaction of the multiple indicators, with the end result that approximately 4 percent of the delinquent group members are defined as truly high risk. These boys experienced not only the extremes of criminogenic family environments; they were vulnerable from the start based on multiple childhood risks. In Figure 6, we present the raw plots of trajectories of "other" offending for the boys at the highest child and family risk compared with the rest of the delinquent group. Perhaps not surprisingly, the rate of offending for the high-risk group is Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 26 THE ANNALS OF THE AMERICAN ACADEMY FIGURE 2 RAW TRAJECTORY OF PREDATORY OFFENSES: AGES SEVEN TO SEVENTY, RY CHILDHOOD RISK 1.0 7.00 11.00 15.00 19.00 23.00 27.00 31.00 35.00 39.00 43.00 47.00 51.00 55.00 59.00 63.00 67.00 Age higher in the early years up to the point of the traditional peak age of offending— about age fifteen. Thereafter, the rate of offending drops off, and these boys desist just like all other boys in the study. Amazingly, in fact, the rates of offending are higher in later life for the group predicted to be at lower risk based on early child and family circumstances. But the big picture is clear—the age-crime curves look the same as in the earlier figures, where we see increasing and then declining involvement in crime for all risk groups. Our basic conclusion thus continues to hold, namely, that desistance and aging out of crime appear to reflect a general (almost fractal) process for all groups of offenders. Latent class models of desistance So far, we have restricted our analysis to prospectively defined groups of offenders based on childhood and adolescent risk factors. A quite different approach is to take the full life course as a given and ask whether there are distinct and latent offender groups based on ex post trajectories of offending. And if so, can the resulting trajectory groups be linked to preexisting or childhood differences? Despite its Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 27 FIGURE 3 RAW TRAJECTORY OF ALCOHOL OFFENDING: AGES SEVEN TO SEVENTY, RY CHILDHOOD RISK 7.00 11.00 15.00 19.00 23.00 27.00 31.00 36.00 39.00 43.00 47.00 51.00 55.00 59.00 63.00 67.00 Age prospective nature, in the analysis above we might have masked underlying trajectory groups, such as life-course persisters. Nagin's (2005) semiparametric group-based modeling approach offers an innovative way to satisfy our objective. In general, the mixed Poisson model assumes that the population is comprised of discrete Poisson distributions with respect to the rate of offending. Each trajectory assumes a polynomial relationship that links age and crime. Based on our earlier analysis, we use a cubic function of age for the seven to seventy models and estimate the equation, H(K)=K+Wage)u+v2(age2\+v3(age% where X{t is the predicted rate of offending for person i in groupj for time period t, AGE.t is the age of person i for time period t, AGEft is the squared age of person i for time period t, and AGE I is the cubed age of person i for time period t; and the coefficients (3^, (3/, (3^, and(3^ structure the shape of the trajectory for each groupj. Although every individual in each group is constrained to the same slope and intercept of that trajectory, these parameters, which determine the level and shape of the trajectory, are free to vary by group. Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 28 THE ANNALS OF THE AMERICAN ACADEMY FIGURE 4 RAW TRAJECTORY OF "OTHER" OFFENSES: AGES SEVEN TO SEVENTY, RY CHILDHOOD RISK 7.00 11.00 15.00 19.00 23.00 27.00 31.00 35.00 39.00 43.00 47.00 51.00 55.00 59.00 63.00 67.00 Age Figure 7 shows the results from semiparametric mixed Poisson models for "other" crimes. Again, heterogeneity in trajectories is present and the data firmly reject a simple typology of two offender groups. There are instead five groups of offending patterns by age for "other" offenses, similar to patterns for total offending as well as crime specific trajectories (see Laub and Sampson 2003,104-6). Most important to our discussion here is that the differences across groups seem to be age at desistance and rate of offending, but with all groups eventually declining with age. Furthermore, we see in Table 2 that the different subgroups in the data for "other" offenses are not systematically predicted by key constituent indicators of our child-risk measure (for similar results for total crime, see Laub and Sampson 2003, 108-9). For good measure, we also consider in Table 2 two measures of considerable interest in intergenerational studies of the transmission of crime risk—the criminality/deviance of parents and parental mental health status. Parental risk is not a consistent predictor. Interestingly, the group that peaks the latest in terms of other offenses (group 2, in their late thirties) has the lowest score on parental criminality and second lowest score on mental disturbance. Overall, the patterns in Table 2 are Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 29 FIGURE 5 PREDICTED "OTHER" OFFENSE TRAJECTORIES: AGE SMOOTHED, SEVEN TO SEVENTY, RY CHILDHOOD RISK 7.00 11.00 15.00 19.00 23.00 27.00 31.00 35.00 39.00 43.00 47.00 51.00 55.00 59.00 63.00 67.00 Age contradictory and do not add up to a consistent story about the past as prologue (c£ Glueck and Glueck 1968). Age at desistance Finally, we address in more detail the notion that it may not be the rate of individual offending that is at issue but the length of criminal career (Moffitt, forthcoming). More precisely, the question we pose is, Do those prospectively predicted to be high-rate offenders, or LCPs, offend to a later point in their lives than the low-risk group? Because we have not specifically addressed this argument before, we present in Table 3 the age at last offense for all crime types by the childhood risk factor. The bottom line is that we do not see consistent evidence of differential age at termination based on prospective childhood risk. Note that none of the mean ages at termination across five different crime types differ significantly by group. These results maintain when we examine the age at last offense taking into account both childhood risk and family risk factors together (see Table 4). Thus, there is no evidence in the Glueck data that prospectively defined life-course-persistent Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 30 THE ANNALS OF THE AMERICAN ACADEMY FIGURE 6 RAW TRAJECTORY OF "OTHER" OFFENDING: AGES SEVEN TO SEVENTY, BY CHILD/FAMILY RISK .8 .7 . 7 11 15 19 23 27 31 35 39 43 47 51 55 59 63 67 Age offenders display "unusually extended offending careers" (Moffitt, forthcoming) over the life course once conditioned on a troubled adolescence. Summary We believe our analyses on minor or miscellaneous offenses and age at termination, in conjunction with reanalyses of raw data trajectories from our previous work, again raise questions about what might be termed the causal theory of groups and the idea that offender groupings are prospectively valid. While our analyses focused on Moffitt s taxonomy, the most detailed and articulate statement of a group-based theory to date, our results have implications for other group-based theories of crime trajectories (e.g., Patterson andYoerger 1993; Loeber and Hay 1997). We would add that empirical research has by now firmly rejected the notion that there are only two groups of latent-class offenders. Setting aside the present study, Piquero (2005) has recently and independently provided an exten- Downloaded from ann.sagepub.com at Masarykova Univerzita on February 4, 2015 A LIFE-COURSE VIEW OF THE DEVELOPMENT OF CRIME 31 FIGURE 7 TRAJECTORIES OF OFFENDING FOR "OTHER" OFFENSES: AGES SEVEN TO SEVENTY 0.9 n A í ^ « * ^ í í ^ # í f * ŕ í í # ŕ ŕ ŕ ŕ ŕ í # # í