CRIME IN EMERGING ADULTHOOD*

CRIME IN EMERGING ADULTHOOD*

ALEX R. PIQUERO University of Florida National Consortium on Violence Research

ROBERT BRAME University of South Carolina National Consortium on Violence Research

PAUL MAZEROLLE University of Queensland

RUDY HAAPANEN California Youth Authority

The extent to which local life circumstances influence criminal activity has been the focus of much theoretical debate. Although empirical research has been initiated, it remains limited. Herein, we use data on 524 serious offenders from the California Youth Authority for a seven- year post-parole period to examine the relationship between changes in local life circumstances and criminal activity. We extend previous research by employing a statistical model that accounts for the joint distribution of violent and nonviolent crime during the late teens and twenties in order to present information on patterns of criminal activity during a newly recognized developmental period of the life course, “emerging adulthood. ’’

Understanding processes that generate stability and change in criminal activity across the life course represents a dominant theme in the field of criminology. Relevant research examining this issue has revealed a range of important findings (Paternoster, Dean et al., 1997; Simons et al., 1998; Wright et al., 1999), and two key themes have surfaced: first, that changes in local life circumstances related to employment, marriage, and illicit substance abuse can materially alter the direction of offending trajectories; and second, that some combination of persistent individual differences in criminal potential and dynamic social experiences is related to crime over the life course. It is also the case that these two themes raise several more

*We are thankful to the anonymous reviewers and the editor for constructive comments. Support for this research was made possible from a grant from the U.S. Department of Justice, National Institute of Justice, 199958-IJ-CX-0058. Points of view expressed in this document do not necessarily express the views of the National Institute of Justice, U.S. Department of Justice.

CRIMINOLOGY VOLUME 40 NUMBER 1 2002 137

138 PIQUERO ET AL.

questions about the nature of continuity and change in criminal activity across the life course. For example, it is unclear if processes of stability and change vary across different groups of offenders or types of offenses. More specifically, it is unclear if the effects of changes in local life circumstances on criminal trajectories hold for all offenders.

Herein, we build on and integrate, to some extent, these themes by examining whether effects of local life circumstances on crime remain after controls for enduring individual differences are taken into consideration, and whether effects of local life circumstances vary within and between offenders in predicting crime over the life course, generally, and across crime types, specifically. In the current study, these questions are examined using longitudinal data covering a newly recognized period of the life course, “emerging adulthood” (Arnett, 2000).

THEORETICAL CONTEXT

The extent to which a common explanation of crime applies to all mem- bers of the offending population is a contentious issue within criminological circles. On one side of this debate are scholars who favor general explanations of crime. Gottfredson and Hirschi (1990), for example, argue that all criminal, deviant, and analogous acts can be attributed to variations in self-control and available opportunities. Their theory advocates a generallstatic viewpoint of criminal behavior that presumes that there is a general cause of crime for all offenders and that once the causal process has played out, change is unlikely. Although agreeing that general theories are in order, Sampson and Laub (1993) offer an important variation to this “common explanation” theory. To these scholars, crime can be under- stood as the product of informal social controls, such as the family, school, marriage, employment, and so on. Variation in these sources of informal social control over the life course, and especially in adulthood, are believed to be related to variations in involvement in criminal activity. Sampson and Laub’s position is such that although there is a general cause of crime for all offenders, changes in life circumstances can still materially affect criminal activity. This approach, characterized as a generaYdynamic viewpoint, attributes importance to change.

A key point of contention between the generallstatic and generay dynamic theories is whether controls on behavior are subject to variations within individuals over time and whether these variations are related to variations in criminal activity over time (Paternoster, Dean et al., 1997). According to Gottfredson and Hirschi, within-individual variation is not problematic, whereas for Sampson and Laub, it is central to understanding why some people persist in-and others desist from-crime. To the extent that Gottfredson and Hirschi are correct, controls for persistent individual

CRIME IN EMERGING ADULTHOOD 139

differences should render the effect of life circumstances on criminal activity irrelevant because the correlation between life circumstances and criminal activity is due primarily to a selection process (Hirschi and Gottfredson, 1995137). On the other hand, Sampson and Laub (1995146) argue that “salient life events and social bonds in adulthood. . . explain variations in criminal behavior independent of prior differences in criminal propensity.”

Although both Gottfredson and Hirschi and Sampson and Laub agree on the generality assumption, developmentalists relax the assumption (of one trajectory for all criminal offenders), thereby adding further complex- ity to the theoretical picture. Developmental theories are friendly to the notion that both persistent individual differences and changing life circumstances are related to involvement in crime, and that these factors affect different groups of offenders in different ways. The developmental view grants that there are different kinds of offenders, each possessing a unique sequelae to criminal activity, as well as a different criminal repertoire. As such, developmentalists contend that treating all offenders as emanating from the same population would be inconsistent with research docu- menting heterogeneity within offending populations (Cohen and Vila, 1996; D’Unger et al., 1998; Nagin and Land, 1993; Piquero et al., 2001).

Although several developmental theories exist (Loeber and Stouthamer-Loeber, 1998; Patterson and Yoerger, 1993), one prominent example is Moffitt’s (1993) developmental taxonomy. She rejects the assumption that there is a general theory of crime and argues for the existence of two distinct groups of offenders. One of these offender groups, life-course-persistent, is characterized by continuity in offending, and the other group, adolescence-limited, is characterized by change in offending. According to her perspective, life-course-persistent offenders originate as a result of the interaction between neuropsychological deficits and defi- cient familial and neighborhood environments. These individuals are typically born into families who are ill-prepared to perform effective socialization. As a result of ineffective socialization, life-course persisters fail in their family life, their school work, and all sorts of interpersonal relationships. Because they never learn to control their antisocial proclivi- ties, they act impulsively as children, adolescents, and adults. For this group of offenders, continuity is the modal behavior, and change is highly unlikely. The criminal repertoire of life-course persisters is believed to include all sorts of criminal acts, including violence. On the other hand, adolescence-limited offenders begin offending in adolescence as a function of the perceived maturity gap and the peer social context of adolescence. Because adolescence-limiteds do not suffer from life-course-persistent- type risk factors, as adulthood ensues, they are likely to embrace their prosocial tendencies and skills and desist. For this group of offenders,

140 PIQUERO ET AL.

change is the modal behavior. The criminal repertoire of adolescence-limiteds is believed to include mainly status and nonviolent delinquent acts.

As can be seen, the debate between developmental and general theories hinges on differences, if any, between individuals in their patterns of offending and in covariate measures that reflect possible underlying causal forces (McDermott and Nagin, 2001). Related to this debate is the extent to which offenders specialize in offending. Interestingly, the theoretical models discussed above make markedly different predictions regarding specialization. For both Gottfredson and Hirschi and Sampson and Laub, versatility in offending is the norm: in other words, both of these general theories claim that offenders rarely (if ever) specialize. For example, Gottfredson and Hirschi (1990:91) contend that “within the domain of crime. . .there will be much versatility among offenders in the criminal acts in which they engage.” Similarly, Sampson and Laub (199356) contend that “[because of] the low level of specialization in specific crimes committed by the Glueck men. . .[our] theoretical framework does not make crime-specific predictions.” Recently, Laub and Sampson (2001:63) noted that their “life history narratives [suggest] no major differences in the process of desistance for non-violent and violent juvenile offenders.”

However, developmentalists anticipate both specialized and generalized patterns of criminal activity. In Moffitt’s scheme, for example, life-course persisters engage in both nonviolent and violent crimes, whereas adolescence-limited offenders concentrate their crime in the nonviolent domain. Even further specialized patterns appear in Loeber’s developmental model as he argues for three groups of offenders, each of whom follow an ordered progression of antisocial behavior from least to more serious behaviors as age ensues. Loeber’s model claims that there is a different offending pattern within offenders, across groups, over time.

“EMERGING ADULTHOOD” AND LOCAL LIFE CIRCUMSTANCES

Two key questions from the theoretical discussion above arise: (1) Do the effects of local life circumstances on crime remain after controls for enduring individual differences are taken into consideration, and (2) do the effects of local life circumstances vary within and between offenders in predicting crime over the life course generally, and across crime types specifically? Empirical evidence regarding these two questions bears relevance for a recently acknowledged developmental period of the life course, “emerging adulthood” (Arnett, 2000).

Arnett (2000:469) suggests that the period of the life course between ages 18 and 25 reflects a distinctive developmental phase characterized by change and a process of exploration of possible life directions:


Having left the dependency of childhood and adolescence, and having not yet entered the enduring responsibilities that are normative in adulthood, emerging adults often explore a variety of possible life directions in love, work, and worldviews. Emerging adulthood is a time of life when many different directions remain possible, when little about the future has been decided for certain, when the scope of independent exploration of life’s possibilities is greater for most people than it will be at any other period of the life course.

Although the examination of emerging adulthood as a distinctive developmental period is relatively new, past research highlights, to some extent, how changing social experiences during this time period can influence crime.

For example, Uggen (2000) found that compared with younger offenders, older offenders in their late 20s who were given marginal employment opportunities were less likely to reoffend. Similarly, Farrington et al. (1986) found that boys had higher crime rates during periods of unemployment than they did during periods of employment. Graham and Bowling (1995) studied patterns of persistence/desistance among individuals aged 14 to 27 and found that for women, desistance was primarily a function of leaving school, leaving home, and having partnerships, whereas for men, it was the disengagement from deviant peers. Mischkowitz’s (1994) study found that desistance between ages 20 and 30 resulted from changes toward a more conventional lifestyle across a variety of domains, including work and family. Labouvie (1996) reported that the benefits of marriage and parenthood in deterring substance use was strongest between ages 28 and 31, which suggests that the timing of events is important. Rand (1987) reported discrepant findings across transitional life events such that by age 30, marriage was associated with reduced crime, but the results varied across offender and crime-related characteristics. Sampson and Laub (1993) found that local life circumstances (e.g., job stability and attachment to spouse) were inhibitors of arrest frequency during both the 17 to 25 and 25 to 32 age periods for the Glueck men.’ Warr (1998) used data on respondents aged 15 to 24 from the National Youth Survey to explore how changes in marriage and delinquent peers related to desistance from crime. His results suggested that the transition to marriage was followed by a dramatic decline in time spent with friends as well as exposure to delinquent peers, and that these factors largely explained the association between marital status and delinquent behavior (p. 183).

Two additional studies are particularly important in discussing the

1. Although in some models, the local life circumstance effects varied across crime type with job stability predicting both property and violent crime and marital attachment predicting property crime only (Sampson and Laub, 1993:174-175, 278).

142 PIQUERO ET AL.

impact of changing local life circumstances on crime; however, their results are not limited to the emerging adulthood period. Horney et al. (1995) used life history calendars to analyze month-to-month variations in offending and life circumstances for a period of 25 to 36 months. Employ- ing hierarchical linear models to examine within-individual changes, they found that meaningful short-term changes in crime were strongly related to variation in several local life circumstances. In estimating separate models for property and assaultive offending, they found different effects among the life circumstance variables. For example, for property crime, they found that heavy drinking, illegal drug use, and employment were positively related to further crime, and although being married was nega- tively associated with property crime, the coefficient was not statistically significant. In predicting assault, they found that although illegal drug use increased assault, marriage served as an inhibitory factor. Also, separate analyses for drug use continued to uncover different effects among the various local life circumstances.

Laub et al. (1998) used a semiparametric mixed Poisson model to examine how investment in marriage related to desistance with a sample of 500 white men from Boston, followed from childhood to age 32. Unlike the approach used by Horney et al., Laub et al. employed a method that uncovers distinctive trajectories of criminal offending. Three important findings arise from their effort. First, they found that desistance from crime was related to the development of quality marital bonds, that the influence was gradual and cumulative over time, and that the deterrent effect was independent of persistent individual differences. Second, the marriage effect varied across the trajectory groups such that “good marriages” exhibited a deterrent effect on future crime for some trajectories but not others. Third, when they examined violent and property crime separately, they found that a good marriage had a significant lagged pre- ventive effect (two and three years out for violent and property crime, respectively).

LIMITATIONS OF PRIOR RESEARCH AND CURRENT FOCUS

Although the aforementioned studies are important for further understanding continuity and change in offending, several limitations should be noted. First, although the Horney et al. study controlled for street time, the Laub et al. study did not. Thus, it is unknown if the marriage effect detected in the latter study remains after controlling for street time, and if such controls influence the marriage effect differently across distinct offender trajectories. Second, both the Horney et al. and Laub et al. studies estimated separate models for distinct offense types. This approach


treats criminal outcomes as independent and does not recognize the shared but imperfect covariance across distinct types of offenses. Third, although employing studies of relatively serious offenders, both studies provide little information on the factors related to persistence/desistance in crime after release from correctional institutions among a serious group of offenders, a point that Laub and Sampson (2001) suggest is worthy of future research.

We build on prior research in a number of important ways. First, we use prospective data on male releasees from California Youth Authority (CYA) institutions to study the relationship between local life circumstances and involvement in criminal activity for a seven-year post-parole period. The use of a serious offending population is important for several reasons. First, the characteristics that distinguish persistence/desistance within groups of high-risk offenders are generally unknown, largely because of the lack of long-term studies involving individuals in the criminal justice system (Laub and Sampson, 2001:1, 65). Second, amidst evidence that many criminals are likely to return to correctional facilities (Beck and Shipley, 1989), knowledge of the correlates of persistence/desistance would aid in the development of effective prevention and treatment programs and highlight whether such programs necessitate an individual- or offender-specific approach (Tremblay and Craig, 1995). Third, the application of a trajectory analysis to a serious offending population is consistent with the recognition that there is heterogeneity within the offending population (Cohen and Vila, 1996). Fourth, because serious offenders are likely to be absent from general population studies (Cernkovich et al., 1985), offending populations are useful for studying the kinds of offenders that are explicated earlier in the article. Thus, isolating distinct, relatively homogenous criminal trajectories among serious offenders can help us better understand the processes related to persistence/ desistance from various types of crime that may be unique to distinct offending trajectories.

Second, the statistical model we employ is based on the semiparametric model developed by Nagin and Land (1993) to study trajectories of criminal activity over the life course. This statistical model departs from the growth curve and hierarchical analyses primarily in its treatment of individual heterogeneity. For example, hierarchical modeling captures individual variation in developmental trajectories via a random coefficients modeling strategy, whereas latent growth curve modeling relies on covariance structure methods. These two approaches model variation in the parameters of developmental trajectories using continuous multivariate density functions. The models used in the present study employ a multinomial modeling strategy that approximates the heterogeneity in offending trajectories with a finite number of distinctive groups that vary not only in

144 PIQUERO ET AL.

terms of the level of offending, but also in the rate of offending over time. This approach captures the cumulative impact of change and the time path by which change is achieved.

This model has a number of useful features that extend prior efforts. First, it does not require that we build the mixture from any specific probability distribution. In other words, we are free to choose any probability distribution that makes sense for our specific problem. This is not the case with hierarchical and covariance structure modeling. Second, because each individual is observed at multiple time periods, it is unlikely that arrest counts at different time periods are independent of each other. Thus, we allow for this within-subject dependence. Third, in light of research showing that the incidence of criminal activity within individuals changes over the life course, it is reasonable to expect that very different parameters govern the growth of offending in different subpopulations. Thus, the models used assume that these parameters are drawn from a multinomial distribution whose shape is estimated from the data. In this sense, they can be viewed as semiparametric as opposed to fully parametric.

Third, in the analyses that follow, we not only describe the relation between life circumstances and violent and nonviolent crime separately, but we also examine the joint covariation of these behaviors over time as they relate to life circumstances. The primary advantage of this model extension is that it allows us to estimate a statistical model whose parameters govern the joint longitudinal distribution of violent and nonviolent crime (see Brame et al., 2001). This is important because prior efforts in this area have estimated outcome models separately by crime type, thereby ignoring the likely shared covariation across crime types. This is unfortunate because the cost of not allowing for the potential covariation seems to be higher than is the inverse. To the best of our knowledge, this estimation procedure cannot be accomplished with hierarchical or covariance structure methods.

Finally, our data take into consideration exposure time, or the amount of time individuals are incapacitated such that they are able to engage in crime while on the street, a feature that was unavailable in the Laub et al. (1998) study. The relevance of this issue was recently demonstrated by Piquero et al. (2001). Using a population of serious offenders, they found that controlling for exposure time reveals different conclusions about the number of offenders who are classified as persisters/desisters. For example, without controls for exposure time, 92% of their sample incurred salient declines in offending throughout the late 20s and early 30s; however, with controls for exposure time, 72% of the sample exhibited this decline, whereas the remainder of the sample remained active in crime.


The analyses that follow attempt to integrate and overcome the limitations of prior research in studying the effect of local life circumstances on criminal activity among a sample of serious offenders. Additionally, the analysis will allow for a thorough examination of whether changes in local life circumstances affect changes in criminal activity during emerging adulthood while controlling for criminal propensity. Because group differences in time-varying causal factors and offending can only appear longitu- dinally, analyzing group differences on such factors requires explicit comparison of covariates and offending over time (McDermott and Nagin, 2001:285). To accomplish this task, data measuring the timing and sequence of changes in local life circumstances and criminal activity are needed.

HYPOTHESES

Two hypotheses related to the relationship between changes in local life circumstances and changes in criminal activity are examined. The first hypothesis compares the two general theories and concerns the extent to which local life circumstances significantly predict crime. For Gottfredson and Hirschi (1990:154-168), life events occurring after childhood/early adolescence are expected to have little (if any) impact on future patterns of crime. On the other hand, Sampson and Laub would argue that once controls are introduced for individual differences in criminal propensity, local life circumstances will still exert a causal influence on crime.

The second hypothesis concerns the extent to which patterns of persistence/desistance vary by crime type or offender characteristics. The theories put forth by Gottfredson and Hirschi (1990) and Laub and Sampson (2001:15, 63) are consistent with the view that similar factors predict the covariation of violent and nonviolent crime over the life course and that this expectation will not vary within and across offender trajectories. On the other hand, developmental theories proposed by Moffitt (1993) and Loeber and Stouthamer-Loeber (1998) argue that different factors predict persistence/desistance in violent and nonviolent crime uniquely, and that this pattern of relationships varies within and across distinct offending trajectories. If developmental theories are correct, different factors should predict different crime types within and across different groups of offenders.

In sum, we examine whether changes in local life circumstances lead to changes in criminal activity. To the extent that it occurs for all offenders, support for Sampson and Laub would be observed; to the extent that it does not occur after controlling for persistent individual differences, support for Gottfredson and Hirschi would be observed; and to the extent that it occurs for certain offenders and not for others as well as for certain

146 PIQUERO ET AL.

crime types but not others, support for developmental models such as Moffitt’s and Loeber and Stouthamer-Loeber’s would be observed.

DATA

The data used in the present effort were part of a larger study of the criminal careers of serious offenders in CYA institutions collected between 1965 and 1984 (Haapanen, 1990). Herein, we analyze the effect of local life circumstances on the joint distribution of violent and nonviolent crime for 524 male parolees. These individuals were released from the CYA at various ages around the late teens and early 20s, but were followed for a seven-year post-parole period. As such, any observed declines in criminal activity are not a function of sample attrition or exposure time. To illustrate, consider two separate males, one released from the CYA at age 17 and the other released at age 20. The first individual is followed for seven consecutive years post-CYA release until age 24 (beginning with age 18), whereas the second individual is followed for a different seven consecutive years post-CYA release until age 27 (beginning with age

In California, once a ward is committed to the CYA, an arrest history is initiated. Any adult arrest(s) or subsequent incarceration(s) is reported by law enforcement to the California Department of Justice. For the small percentage of individuals who were not of adult age at the time of their release, subsequent arrests are reported to the California Department of Justice by the CYA while the ward is on parole.

For each individual, we obtained information on counts of criminal arrests as well as information on exposure time. Arrest information was obtained from California Criminal Identification and Investigation (CII) rap sheets. In this paper, we focus on the joint distribution of violent and nonviolent arrests. Violent arrests included murder, rape, aggravated assault, robbery, and other person offenses, such as extortion and kidnap- ping. Nonviolent arrests included burglary, receiving stolen property,

21).*

2. One of our anonymous reviewers correctly noted that our data are based on information that is more than 20 years old. This raises the possibility that our results are generated by period-specific influences that are no longer operational. Although this is a plausible-and a reasonable-hypothesis, it is also difficult to test. In response, we would note that our findings can generally be reconciled with results from the extant literature, also using “aged” data (see Laub et al., 1998). Moreover, analysis of similar data sets in the future can shed light on the extent to which the results we have observed are purely historical in nature. Finally, given that collection of the kind of data we employ is time-consuming and costly, we believe that our analytic approach and subsequent results can help chart the course for future data collection and research efforts.


grand theft, forgery, and grand theft auto. As with any measure of offending, the use of official data is subject to potential biases affecting the arrest decision. Although we do not have self-report measures, these too have been the subject of concern (Lauritsen, 1998; Piquero et al., 2002). Still, scholars have shown that official and self-report records produce “compa- rable and complementary results on such important topics as prevalence, continuity, versatility, and specialization in different types of offenses” (Farrington, 1989:418).

Data on exposure times were also obtained from the CII information. Within each year time period, individuals were coded free for the number of months that they were not serving time in jail, prison, or in CYA deten- tion; otherwise, they were coded as being under some form of correctional supervision. So, an individual who was in prison for eight months during a particular year would be coded as having exposure time equal to four months.

Data on life circumstances were collected from CYA case files. Specific information was collected on (1) alcohol dependence, (2) heroin dependence, (3) full-time employment, and (4) marriage. Two variables associated with nonrecidivism are steady (full-time) employment and marriage (Blumstein et al., 1986:206; Sampson and Laub, 1993). Two variables that have been associated with recidivism are heroin and alcohol dependence (Chaiken and Chaiken, 1990). Although the effects of marriage and employment are anticipated to be general in the sense that they should inhibit both violent and nonviolent crime, the effects of alcohol/drug abuse have shown otherwise. A body of research has shown that except during withdrawal periods, heroin users in need of funds tend to avoid violent crimes if nonviolent (albeit criminal) alternatives are available (Reiss and Roth, 1993:191). On the other hand, alcohol, generally, and long-term heavy alcohol use, in particular, have been shown to be predisposing factors for violence (Reiss and Roth, 1993:13). In fact, heavy/problem drinkers are significantly more likely to accumulate arrests for violent crimes than are nonheavy/nonproblem drinkers (Collins, 1986:106-111). Infor- mation for marriage and employment came from CYA case files, whereas heroin and alcohol dependence measures came from prison and probation records, including self-report and collateral information. Because only substance dependence was of interest, minor alcohol and heroin use was not coded. During the course of each of the seven years of post-parole observation, individuals were given a score of 1 if they were involved in each of the four respective local life circumstances; if they were not involved in a particular local life circumstance, they received a score of 0 on that particular indicator. Thus, the local life circumstances were coded in terms of change in status. Offenders were assumed to maintain the

148 PIQUERO ET AL.

same status unless the change was noted in the California Department of Corrections files.

Following previous research, we developed a stakes in conformity index (DeJong, 1997; Pate and Hamilton, 1992; Paternoster, Brame et al., 1997; Sherman and Smith, 1992). This index combines the life circumstances of marriage and full-time employment. Individuals possessing neither of these circumstances were coded 0, individuals possessing one or the other were coded 1, and individuals possessing both were coded 2. This particular index of informal social control is consistent with Sampson and Laub’s (1993:21) position that “social ties to jobs and family. . . are the key inhibitors to adult crime and deviance.”3 Such an index allows for a more direct examination of the impact of cumulative amounts of informal social control on criminal activity, and research has shown that the independent effects of informal social controls are not as important as their cumulative effect (Sherman and Smith, 1992).

RESULTS DESCRIPTIVE ANALYSIS

Tables 1A and 1B present a frequency distribution of the number of violent and nonviolent arrests at each age, respectively. Two features of this table need to be highlighted. First, the number of individuals observed at each age varies. There are two reasons for this: (1) Each individual was followed for a maximum of seven years, and the ages at the first year of the study ranged from 16 to 22; and (2) individuals were not necessarily free in the community to commit crimes during the entire follow-up. Second, for both violent and nonviolent arrests, the analysis reveals that arrests seem to rise to a peak in the early 20s and decline thereafter.

In order to better understand the time trend in both violent and nonviolent arrests, we estimated a Poisson regression model that specifies the

3. Sampson and Laub have argued that it may not necessarily be participation in social bonds that are salient for inhibiting criminal activity; rather, the attachment or level of involvement may be more indicative of social capital. Unfortunately, our data do not contain such measures. Nevertheless, being married, being employed, and so on should imply some level of social bonding; after all, one would probably not be married if one did not have affective ties to one’s spouse (Nielson, 1999). As Hindelang (1973) argued, there is likely to be some overlap between these proxies and the theoretical constructs underlying their examination (e.g., persons who are employed are more likely to be committed to their job simply because they have jobs to be committed to). In a similar fashion, Horney et al. (1995:657-658; see also Nagin and Paternoster, 1994) note that upon entry into such social institutions, one’s “social investment in these institutions accumulates from that point on.” In addition, researchers using “participation” measures have found effects consistent with the social bond predictions emanating from Sampson and Laub’s theory (DeJong, 1997).


Table 1A. Violent Arrest Frequency Distributions (N = 524)

Number of Average Arrest Number of Arrests & 0 1 2 3 4 5+ Individuals Frequency 16 1 0 0 0 0 0 1 .oo 17 36 2 0 0 0 0 38 .05 18 157 34 7 5 2 0 205 .35 19 267 56 20 13 0 0 356 .38 20 313 73 23 7 1 1 418 .36 21 332 69 18 11 5 1 436 .38 22 342 58 17 8 5 3 433 .36 23 300 63 27 6 0 4 400 .40 24 312 37 23 8 1 1 382 .30 25 185 22 5 3 2 1 218 .25 26 7 0 1 0 3 3 0 1 87 .36 27 23 2 1 0 0 0 26 .15 28 3 2 0 0 0 0 5 .40

Table 1B. Nonviolent Arrest Frequency Distributions (N = 524)

& 16 17 18 19 20 21 22 23 24 25 26 27 28

Number of Arrests

0 1 2 3 4 5 + 0 1 0 0 0 0

1 8 1 4 4 0 2 0 61 59 27 17 15 26 87 69 69 39 23 69

115 85 60 44 39 75 133 80 78 42 30 73 131 95 55 49 32 71 132 93 50 39 30 56 132 81 65 28 28 48 89 43 29 23 9 25 34 19 16 9 4 5 1 0 8 4 1 1 2 2 2 0 1 0 0

- - - - - - Number of Individuals

1 38

205 356 418 436 433 400 382 218 87 26 5

Average Arrest Frequency

1 .oo .79

1.94 2.51 2.53 2.31 2.36 2.18 1.99 1.71 1.51 1.73 1 .oo

average number of arrests for an individual at a particular age as a log- quadratic function of age and the number of months an individual is “on

150 PIQUERO ET AL.

the street” at that age (up to 12 months in the year) (see Table 2). We write the expected number of violent arrests at age t as:

1 t t 2 E ( V , ) = a, = exp + a”, - + log, (s,) 7

100

where t ranges from 16 to 28 and s, denotes the number of months an individual is not incarcerated at year t. In similar fashion, we write the expected number of nonviolent arrests as:

1 t t 2 100 ~ ( n , ) = A,,, = exp + a,,,- +@,(st ) .

For both of these equations, the vector a comprises maximum likelihood estimates of the time trend parameters in the population from which our sample is drawn. These estimates are obtained by maximizing the likelihood function, assuming a simple Poisson probability mass function, assuming independent time periods both within and across individuals:

Table 2. Street Time Distribution (In Months) By Age

Number of Average Number of Individuals Months on Street

16 1 7.00 17 38 8.71 18 205 8.40 19 356 9.03 20 418 8.81 21 436 8.88 22 433 8.88 23 400 9.08 24 382 9.17 25 218 9.21 26 87 9.14 27 26 8.96 28 5 10.60

Table 3 presents the results of this analysis with standard errors adjusted for overdispersion in the violent and nonviolent arrest distributions (see McCullagh and Nelder, 1989:124-128,174-175). Figure 1 presents a graph


of the violent and nonviolent arrest trends based on the parameter estimates presented in Table 3. As can be seen, both violent and nonviolent arrests rise to a peak during the late teens and early 20s and they fall from that point on.

Figure 1 Comparison of Actual and Expected Arrest Rates for Violent and Nonviolent Offenses

Nonviolent Arrests - Log-Quadrafic Poisson Model Assuming 12 Months Street Time

Nonviolent Arrests -Actual

Nonviolent Arrests - Log-Quadrafic Poisson Model wth Vanation in Street Time

Arrest Role

Violent Arrests - Log-Quadratic Poisson Model Assummg 12 Months Street Tune

f Violent Arrests - Actual

---- Vanation in Street Time Violent Arrests ~ Log-Quadrahc Poisson Model with

16W 18M 2 o W 22W 24W 28W 2800

17W 1800 21W 23W 2 5 W 27W Age (In Years)

HETEROGENEITY IN ARREST TRENDS

Although the analysis presented in the previous section is helpful, it has

Table 3. Parameter Estimates For Log-Quadratic Poisson Trend Model

Parameter Estimate - S.E. I z I -ratio Violent Arrests

a0 -11.042 5.001 2.21 a1 7.506 4.634 1.62 a2 -1.786 1.066 1.68

Scale Factor 1.787 Nonviolent Arrests

a0 -9.953 2.506 3.97 a1 8.354 2.326 3.59 a 2 -2.015 0.536 3.76

Scale Factor 2.223 Note: Standard errors and z-ratios are adjusted for overdispersion using methods described in McCullagh and Nelder (1989).

152 PIQUERO ET AL.

some important limitations. First, the overall trends in violent and nonviolent arrests are summaries of what might be a more complex pattern of arrest activity (Nagin, 1999). A model that takes the possible heterogeneity of trends in arrest activity into account would provide a more complete and accurate description. Second, the descriptive model assumes that the violent and nonviolent arrest trends are independent of each other. In light of research showing that offenders tend not to specialize, however, this assumption seems unrealistic (Brame et al., 2001; Nagin and Tremblay, 1999). Third, research on longitudinal patterns of offending suggests that individuals exhibit stable differences in their proclivity to offend (Nagin and Land, 1993).

A useful method for addressing all of these issues involves the use of a more complicated version of the Poisson model that allows for a finite mixture of Poisson processes along the lines discussed by Nagin and Land (1993). The likelihood function for this mixture model is given by:

[ [ 28 ,exP(-~~,l),:]~[exP(-a~,,,)~~,,,])

L = f i r=l Z n , j = l n t=16 V l t n I t !

where the parameters now depend on the support of the mixing distribution. The mixing distribution is multinomial and can have any shape. A key issue in estimating such models is determining the optimal number of components in the mixing distribution. The most widely used method involves evaluation of the Bayesian Information Criterion (BIC) (D’Unger et al., 1998; Nagin, 1999). The BIC measures the posterior probability that a model specification is correct when there are equal a priori probabilities associated with that specification and an alternative specification. As such, the BIC provides researchers with a means to assess the most probable model from a set of candidate specifications. For a particular model, the BIC is given by:

where K is the number of components in the mixing distribution, N is the sample size, and log(L) is the natural logarithm of the likelihood function. We follow the standard approach of choosing the model that maximizes BIC.

Candidate specifications from one to five components were considered. We were unable to achieve convergence with a five-component model. This typically means that there is not enough heterogeneity in the observed data to support a more complicated model. Out of the other models considered, a four-component model maximized BIC, and we,


therefore, focused our interpretation on that specification. Table 4 presents the parameter estimates and standard errors associated with the four component model.

Table 4. Parameter Estimates For Log-Quadratic Poisson Mixture Model ( N = 524)

Violent Arrests Nonviolent Arrests Parameter Estimate S.E. I z I -ratio Estimate S.E. I z 1 -ratio

Group #1 a, a1

Group #2 a0 a1

a2

a2 Group #3 a0

a2

Group #4 a0 al

a2

a1

IT, = p(Group #1) = .145 n2 = p(Group #2) = .511 n3 = p(Group #3) = .124

7.540

2.798

9.054 -2.247 -8.044

5.331 -1.098

-30.630

-1 1.704

-12.457

24.885 -5.622

12.198 .62 11.146 1.05 2.530 1.11 5.811 2.14 5.492 1.65 1.291 1.74 6.247 1.29 5.833 .91 1.355 .81 9.243 3.31 8.525 2.92 1.956 2.87

-3.220 .445

-.144 -5.134

3.515 -.865

-10.556 9.618

-2.406 -18.354

16.709 -3.892

7.370 6.740 1.53 1 2.31 1 2.175 SO5

4.943 4.731 1.124 2.357 2.189

SO7

.44

.07

.09 2.22 1.62 1.71 2.14 2.03 2.14 7.79 7.63 7.67

n4 = p(Group #4) = .220

Because of the number of nonlinear terms in this model, it is somewhat difficult to interpret the numerical values of the parameter estimates. Consequently, Figure 2 presents a graph of the violent and nonviolent arrest trajectories for each component of the mixture under the assumption of 12 months of street time at each age. There are two interesting features of this analysis. First, it reveals substantial heterogeneity in the long-term outcomes of this sample. Therefore, it would not be realistic to simply assume that all of these parolees are at high risk for future problems or that they all have similar outcomes. Instead, it is apparent that some of these parolees go on to essentially desist from further crime, whereas others exhibit more persistent tendencies to commit crime. Sec- ond, the analysis reveals a positive but imperfect association between variation in violent and nonviolent arrests. In general, those who rank low on violent arrests also tend to rank low on nonviolent arrests. This cross- behavior stability notwithstanding, there is also a group of individuals exhibiting a moderate ranking on violent arrests but a relatively high ranking on nonviolent arrests, so the analysis helps to illustrate how trends in one behavior can be used to help predict trends in another behavior. We

154 PIQUERO ET AL.

nevertheless must keep in mind that such predictions will not be perfect."

Figure 2 Summary of Violent and Nonviolent Arrest Trajectories Under Assumption of 12 Months Street Time Each Year

Expected Violent Arrest Rate

Expected Nonviolent Arrest Rate

Age (In Years)

To better describe the within-group, between-crime associations, Figure 3 presents the predicted and actual joint trajectories of violent and nonviolent arrests (adjusted for exposure time). As this graph shows, there appears to be a mixture of specialization and generality in arrests across and within trajectories. For example, although all four groups evidence more nonviolent than violent arrests, the disparity in the mean number of arrests varies greatly. For TI and TZ, their arrest patterns appear to be characterized by low nonviolent and violent arrests; however, the mean number of arrests for T3 and T4 is large and varies by crime type. For example, for T3, their level of both nonviolent and violent arrests is fairly high, and this is especially the case for the mean number of violent arrests incurred by individuals in this trajectory. Although the other three trajectories incur less than .5 violent arrests per year, individuals in T3 are well over one violent arrest throughout much of the observation period. Indi- viduals in T4 exhibit a peak of five nonviolent arrests per year in the early 20s. What is even more interesting is the number of violent arrests for T4; unlike their high level of nonviolent arrests, they incur very few violent arrests such that these offenders tend to be more specialized in the nonviolence domain. In sum, Figure 3 suggests that although individuals ranking low on one crime dimension are typically ranking low on another

4. The basic theme of the graphs for trajectories TI and T2 is one of relatively little change. Moreover, the up-tick in offending for TI should not be overanalyzed because there are a small number of offenders at ages 27/28 (see Table 2).


3 6 - 3 3

3 2 1 - 2 4 2 7 1 8 - 1 5 1 2 ~ o s - 0 6 03-

0 -

crime dimension, this association is far from perfect because the rankings do not appear to be the same for all four trajectories. Two trajectories, in particular, T3 and T4, exhibit different rankings on violent and nonviolent arrests, thereby suggesting that there is an imperfect covariation between violent and nonviolent arrests that needs to be taken into consideration when studying the longitudinal sequence of criminal careers.

Figure 3 Predicted and Actual Joint Trajectories of Violent and Nonviolent Arrests (Predictions Adjusted for Exposure Time)

Mean # of hats

Trajectory System # I

Nonviolent 0.7

0 6 a m - 0 5 1 ' , b

0 4

17 18 19 20 21 22 23 24 25 26 27 Age (in years)

Mean # of Arrcsts

Trajectory System Y3

*/'Y\t Nonviolent / 2, a

I

17 18 19 20 21 22 23 24 25 26 27 28

~ g a (m ysars)

2

1 8

1 6

1 4

7 2 1

0.8

0.6

0 4

0.2

0 Lzzz2zd 16 17 18 19 20 21 22 23 24 25 26 27 28

~ g e (in yeam)

Mean # 01.4rrests Trajector) System lt4

3 5

2 5

1 5

0 5

7 7 18 19 2 0 21 22 23 24 25 26 27 28 Age (in years)

NOTE: Smooth lines represent model predictions. Lines with symbols represent actual data points.

EFFECTS OF COVARIATES ON POST-PAROLE ARRESTS AFTER ADJUSTING FOR TREND HETEROGENEITY

An important focus of our analysis involves an assessment of how variation in several covariates is associated with arrests over the course of the follow-up period. Specifically, we investigated the association between arrests and the following covariates: (1) race, coded 1 = white (48.5%), 0 = nonwhite (African Americans = 33%, Hispanics = 16.6%, and Other = 1.9%); (2) stake in conformity; (3) heroin dependence; and (4) alcohol dependence. Table 5 provides summary statistics for these covariates.

156 PIQUERO ET AL.

Because it is possible that individual time-stable characteristics may simul- taneously influence variation in these covariates (with the exception of race) and arrests, it is important to try to adjust for the influence of these stable individual differences. Our objective is to be able to assess whether and to what extent these covariates are associated with violent and nonviolent arrests after conditioning on stable individual differences.

Table 5. Distributions of Covariates At Each Age

Number of & Observations 16 1 17 38 18 205 19 356 20 418 21 436 22 433 23 400 24 382 25 218 26 87 27 26 28 5

Stake in Heroin Alcohol Race = White Conformity Dep. Dep.

1 .ooo .ooo 1 .ooo .ooo SO0 .lo5 .263 .132 .468 .268 .239 .161 .447 .284 .292 .191 .483 .349 .337 .220 .486 .440 .388 .239 .480 .460 -413 .266 SO5 .523 .418 .275 .516 .558 .393 .288 .550 .601 .413 .307 .609 .644 .414 .356 .731 SO0 SO0 .231 .600 .400 .200 .400

To accomplish this task, we adopt the methods described by Laub et al. (1998). Their basic approach is to use information from a trajectory model like the one estimated in the previous section to actually sort individuals into one of the four trajectory groups based on their observed arrest history. This classification scheme is based on calculating the posterior probability of trajectory group membership for each individual in the sample and for each trajectory group. For each group, the calculation is given by:

Pr(Individua1 i is in trajectory group j I vi, ni) = (Lilj x ni)Ej(Lilj x n,), where Lilj is the likelihood function for the trajectory model in the previous section for individual i, assuming that the individual actually is a member of trajectory group j ; nj is the estimated unconditional probability that individual i is a member of trajectory group j , and the outcome of the calculation is the (posterior) conditional probability that individual i is a member of trajectory group j , given the available data, vi and ni. For pur- poses of our analysis, each individual will have four of these posterior


probabilities, one for each trajectory group. We then assign each individual to the trajectory group to which he has the highest estimated posterior probability of belonging. Table 6 presents the frequency distribution of this new “trajectory group” variable along with the average posterior probability of each individual’s being assigned to the group that he has the highest probability of belonging. This analysis suggests that most individuals in our analysis have a very high probability of being assigned to the group that maximizes this posterior probability.

Table 6. Trajectory Group Classification Distribution and Posterior Group Assignment Probabilities

Trajectory Number of Percent Mean Posterior Group Individuals of Total Probability

TI 73 13.9 .905 T2 277 52.9 .899 7-3 64 12.2 .852 7-4 110 21.0 .924

Total 524 100.0

The next step of our analysis is to estimate a Poisson regression model for each group in which the dependent variables are the number of violent and nonviolent arrests, respectively. The independent variables in this analysis are the covariates described above. Following Laub et al. (1998), the strength of this analysis is based on the fact that we condition on group membership before estimating the effects of the covariates. This feature of the analysis provides a strong control for persistent individual differences in arrest activity that could bias parameter estimates of the effects of these covariates.

Table 7 presents the results of this analysis. Scale factors are also pro- vided because these are used to adjust the standard errors of the parameter estimates for overdispersion (McCullagh and Nelder, 1989:124-128, 174-175). For violence, it appears that whites have a significantly lower risk of arrest for T2 and T4 but not for the other groups. Race appears to have no effect at all for any of the trajectory groups for nonviolent arrests. The sign of the stake in conformity effect is negative in most of the models (the models for T3 are the exception) presented in Table 7 but is only statistically significant (two-tailed p < .05 level) for nonviolent arrests in T2. Heroin dependence appears to increase the risk of nonviolent arrests for all four groups but is only statistically significant at the two-tailed p < .05 level in T2 and T4. Finally, alcohol dependence is positively associated

PIQUERO ET AL.

with violent arrests in T4, but its effect is not statistically significant at the two-tailed p < .05 level in the other analyses.

Table 7. Estimated Effects of Covariates on Arrest Activity After Conditioning on Group Membership

Violent Arrest Activity

Group I (N = 73) Group 2 (N = 277) Group 3 (N = 64) Group 4 (N = 110) Parameter Estimate I z I -ratio Estimate I z I -ratio Estimate I z I -ratio Estimate I z 1 -ratio - - - - -~ -~

Intercept 13.773 1.00 -13.056 2.03 -5.930 0.75 -32.918 3.08 AgellO -17.170 1.36 9.746 1.63 3.541 0.49 26.989 2.75 Age’l100 3.971 1.39 -2.369 1.71 -A97 0.42 4.096 2.72 Race = White -.389 1.05 - m 4.19 -.210 0.85 -.w 2.15 Stake in Conformity -.069 0.25 -.032 0.29 ,060 0.31 -.331 1.69 Heroin Dep. ,509 1.08 -.148 1.07 -.080 0.38 ,267 1.29 Alcohol Dep. ,797 1.81 -.ox 0.24 -.IU 0.55 ,478 2.33

Scale Factor 1.277 1.398 1.885 1.29 1 Nonviolent Arrest Activity

Intercept -7.455 0.94 -4.640 1.67 -7.636 1.39 -18.036 4.51 Age/lO 4.003 0.55 3.028 1.17 6.843 1.33 16.383 4.42 Age’l100 -382 0.54 -.758 1.28 -1.772 1.47 -3.816 4.48 Race = White ,011 0.06 ,022 0.36 -.ON 0.13 -.ox 0.30 Stake in Conformity -.214 1.53 -.lo9 2.05 ,075 0.61 -.lo4 1.35

,396 1.58 ,140 2.19 ,212 1.62 .196 2.27 Heroin Dep. Alcohol Dep. -.029 0.10 ,132 1.89 -.OlO 0.08 -.004 0.04

Scale Factor 1.401 1.688 1.600 2.203

The major theme of the results presented in Table 7 is that most of the parameter estimates for the covariates being studied are not statistically significant at conventional levels. Although, in general, the signs for the stake in conformity variable are negative and the signs for heroin and alcohol dependence are positive, the sampling distributions for most of these effects include zero. Thus, it is difficult to infer a great deal about the signs of these effects based on this evidence. It might be argued that reduced statistical power is responsible for the wide sampling distributions on these parameter estimates; however, it is important to keep in mind that each of these groups of individuals is observed over multiple time periods. Therefore, the likelihood function in each analysis is evaluated from a minimum of 276 times for T3 to a maximum of 1,624 times for T2. All of the analyses, therefore, meet the large sample requirements for obtaining desirable properties from the maximum likelihood estimates.

We then conducted an exploratory analysis in which we allowed for interaction terms between the trajectory group variable and the intercept, age, and age-squared variables while imposing the constraint that the effects for race, stakes in conformity, and heroin and alcohol dependence


are the same across groups. The results of this exploratory analysis are presented in Table 8. Because of the increased power of this analysis, the effect of race is statistically significant at the two-tailed p < .05 significance level for violent arrests, although none of the other variables are. In addition, the effects of stake in conformity and heroin dependence are statistically significant at the two-tailed p < .05 significance level for nonviolent arrests. In light of these results, an important question is “how large are these effects?” To answer this question, we exponentiated the parameter estimates associated with each of the covariates. This is a useful calculation because it tells us the factor by which the mean of the dependent variable is expected to change for a unit change in the independent variable. As Agresti (1996:81) notes, “The mean of Y at x+l equals the mean of Y at x multiplied by [the exponentiated parameter estimate].” Viewed in this light, the largest effect by far in this table is the effect of race on violent arrests. The effects of stake in conformity and heroin dependence on nonviolent arrests appear to be relatively small even though they are statistically significant.

DISCUSSION

The life-course criminology literature reveals several contradictory predictions regarding the influence of local life circumstances on criminal activity. Herein, we sought to provide further evidence on this relationship in a way that helped improve on five gaps in the research literature: First, little was known about what specific experiences and life events are important in altering upward or downward trajectories of criminal activity (Nagin and Paternoster, 2000); second, little was known about how such experiences distinguished persistence/desistance within a group of high- risk offenders (Laub and Sampson, 2001). Third, little was known about how these experiences varied during emerging adulthood. Fourth, prior research has not examined these issues within the context of controlling for exposure time, and fifth, prior research has not examined the influence of local life circumstances on the joint covariation between violent and nonviolent crime.

Four key findings emerge from our analysis. First, like other studies (D’Unger et al., 1998; Nagin and Land, 1993), our application of the semiparametric model yielded four criminal trajectories. Key differences in the shape, rate, and type of arrest activity across all four trajectories were observed. Further, we found between-group differences in both arrest trajectories and covariate effects such that the trajectories appeared to differ in how covariates predicted the degree of increase/decrease in arrests (McDermott and Nagin, 2001). Second, our results suggest that some of the releasees go on to desist from-while others persist in-crime.

160 PIQUERO ET AL.

Table 8. Estimated Effects of Covariates on Arrest Activity After Conditioning on Group Membership and Imposing Equality Constraints On Covariates Across Groups

Violent Arrests Nonviolent Arrests Estimate I z I -ratio exp(Estimate) Estimate I z I -ratio exp(Estimate) -___ Parameter

Intercept Group 1 Group 2 Group 3 Group 4 Age/lO*Group 1 Age/lO*Group 2 Age/lO*Group 3 Age/lO*Group 4 AgeZ/lOO*Group 1 Age2/100*Group 2 Age2/100*Group 3 Age2/100*Group 4 Race = White Stake in Conformity Heroin Dep. Alcohol Dep.

-34.624 47.300 21.725 28.582

.Ooo -16.146

9.563 3.635

28.821 3.780

-2.343 -.720

4.533 -.459 -.045 -.006

.063

2.96 2.50 1.63 2.18

1.18 1.58 0.66 2.68 1.22 1.67 0.58 2.66 4.69 0.58 0.07 0.65

-18.023 11.293 13.565 10.317

,000 3.324 2.856 6.881

16.350 -.731 -.716

-1.766 -3.809

0.632 .001 0.956 -.094 0.994 ,177 1.065 .055

5.65 1.08 3.16 1.51

.36 1.06 1.21 5.54 .36

1.16 1.34 5.61 .02 1.001

2.47 ,910 3.93 1.194 1.12 1.057

Scale Factor 1.423 1.765

This result appears to call into question the policy of locking up offenders for significant periods of time while forcing certain theories to perhaps revisit the claim that there are “life-course-persistent’’ offenders. As our results imply, many of the CYA offenders appear to be on a trajectory toward desistance, at least as measured by arrest records. Third, using a model that was specifically developed to analyze the joint distribution of violent and nonviolent arrests, our analyses revealed a positive but imperfect association between the two outcomes. On the one hand, our results suggest that, for the most part, those individuals (in TI and T2) who rank low on violent arrests are the same individuals who rank low on nonviolent arrests. On the other hand, there was also a group of individuals (in T3 and T4) who exhibited divergent rankings on violent and nonviolent arrests. Although the glass appears to be more full than empty with regard to the generality (rather than specialization) hypothesis, there still remains a group of individuals whose arrests are concentrated in the nonviolence domain (e.g., Wolfgang et al., 1972).

Fourth, the present effort was the first to study the effect of both stable


and time-varying covariates on the joint distribution of violent and nonviolent arrests. By estimating the effects of covariates on arrests after conditioning on trajectory group membership, this allowed us to control for differential propensities to offend, which may have confounded the relationships. Our findings showed that although race failed to exert an effect for any of the trajectory groups for nonviolent arrests, it exerted a significant effect on violent arrests. This result is consistent with studies report- ing higher violent arrest rates for nonwhites (Blumstein et al., 1986). With respect to stakes in conformity, the results were mixed. Although stakes exhibited its anticipated negative effect, it was only statistically significant for two trajectory groups, T4 for violent arrests and T2 for nonviolent arrests. Heroin dependence exhibited a significant and positive effect for T2 and T4 for nonviolent arrests. In an exploratory analysis, we estimated a model in which we allowed for interaction terms between the trajectory group variables and the intercept, age, and age-squared variables while imposing the constraint that the effects for race, stakes in conformity, heroin dependence, and alcohol dependence were the same across groups. These results indicated that race was significantly related to violent arrests, whereas heroin dependence and stakes in conformity were significantly related to nonviolent arrests.

The finding that stakes in conformity predicted nonviolent but not violent offending is unique. One potential reason for this difference is the idea that these types of informal social controls can inhibit instrumental (nonviolent) offenses that entail some (however minimal) degree of plan- ning and aforethought but cannot inhibit expressive offenses that may escalate spontaneously from interpersonal conflicts. Another potential but related reason for this finding concerns the nature of the offenses exhibited and the types of persons engaging in such offenses. For example, Moffitt’s developmental taxonomy proposes that unlike their adolescence-limited counterparts, life-course-persistent offenders are failed socialization products: as a result, such offenders are likely to not possess the characteristics (i.e., cognitive and verbal abilities) that make them eli- gible for effective socialization and social control efforts. Such individuals, then, fail to be influenced by social control agents because they have not been socialized themselves. As a function of the forces of cumulative continuity, failed social skills, and few (if any) academic achievements, the life-course trajectory for life-course-persistent offenders is bleak and their bonding levels and antisocial behavior become even more resistant to change (Cernkovich and Giordano, 2001). Thus, it may be the case that both weak bonding and involvement in violent crime are consequences and indices of some sort of underlying trait, and it is likely the case that among this small, select group of offenders, the formation or sustenance of

PIQUERO ET AL.

strong social attachment to conventional others and institutions is compro- mised (see also Cernkovich and Giordano, 2001:398).

In sum, the results of our study indicate that local life circumstances exert different effects across distinct offender trajectories and offense types. This pattern of findings seems to support theoretical statements that are somewhat more complicated than the generalhtatic approach advocated by Gottfredson and Hirschi or the general/dynamic approach put forth by Sampson and Laub. Although these findings do not necessarily confirm the existence of distinct offender groups, they do offer some support to the notion that different local life circumstances operate differently across offenders and crime types during the emerging adulthood period.

We should identify several limitations to the current effort. First, our data come from a sample of CYA parolees; thus, although we believe that the pattern of results obtained herein would be similar to those observed for other jurisdictions, this remains an empirical question. Second, our data contained information solely for males. Future efforts should attempt to collect similar data for females to determine how local life circumstances influence their patterns of criminal activity. Third, our outcome measures relied on official arrest records. Although scholars continue to debate the merits of official and self-report data, it is likely the case that a more complete study would include data from both sources.

Several projects lay on the horizon for those interested in further study of the relationship between local life circumstances and criminal activity. The first concerns the collection of more and different types of local life circumstances. Previous empirical tests, as well as our own, have not examined the full range of specific experiences and life events to determine how they influence the joint distribution of violent and nonviolent arrests. Second, given the interesting substance abuse effects uncovered herein, it would be useful to examine the extent to which persistence/desistance trajectories vary across offending groups stratified by substance abuse histories. Third, because we assessed the effects of local life circumstances concurrently, future studies may wish to examine the lagged effect of local life circumstances on crime. In addition, future research should consider the possibility that an arrest experience reduces offenders’ access to full-time employment and, as a result, lowers their marriagibility pre- mium (Cohen, 1999). Thus, nonrecursive modeling of local life circumstances and arrests may uncover more interesting findings. Fourth, although our results detected patterns of criminal activity (in rate, shape, and type) that were somewhat similar to those found in other studies, it would be useful for researchers to apply our modeling protocol to other data sets to determine the usefulness of the joint covariation model, especially with controls for exposure time. Fifth, because our analyses


weighted violent and nonviolent arrests equally, future studies may wish to incorporate a weighting scheme in which violent arrests are weighted more than are nonviolent arrests. Sixth, future studies should strive to collect similar types of data for offenders sentenced to different correctional experiences. For example, researchers could collect life circumstances data for offenders assigned to incarceration, probation, and intensive probation supervision. This would allow researchers to compare how local life circumstances vary across all three experiences in predicting the joint distribution of violent and nonviolent arrests. Seventh, because little is known about how the relationship between local life circumstances and criminal activity varies across race and sex, especially in light of evidence to suggest that such groups differentially experience and interpret life events (Broidy and Agnew, 1997; Nielson, 1999), future efforts should strive to conduct subgroup analyses. Finally, although our presentation focused on a cumulative stakes in conformity measure, there are reasons to believe that employment and marriage operate independently for offenders (e.g., the need to be employed prior to marriage). In an exploratory fashion, we disaggregated the stakes measure and found that full- time employment, as opposed to marriage, was inhibiting criminal activity in the sample. Thus, it may be that marriage and full-time employment affect different aspects of offenders’ lives, and these differences may be worth pursuing, especially as they relate to desistance.

On the whole, our results indicate that the criminal trajectory of most parolees is decreasing as they approach their late 20.5, thereby challenging proponents of “Three Strikes” policies. Although the desistance curve was much sharper and dramatic for declining rates of nonviolent arrests, the trend was in a similar direction among three of the four groups for violence. In fact, we observed that more than 87% of the CYA releasees experienced a violent arrest rate of less than .60 by their late 20s. The 65 (12.4%) offenders who continued to accumulate violent arrests, around 1.5 per year by their late 20s, therefore, although comprising a small number of individuals, highlight the importance of decomposing aggregate age- crime curves into distinct offending trajectories and studying in a more in- depth fashion the determinants of violent arrests among this small select group.

Our results also show that some of the change in nonviolent arrests is a function of informal social bonds. This highlights the importance of strengthening offenders’ ties to informal social control agents, and suggests that investment in social bonds appears to provide some sort of “looking-to-the-future’’ view. This is also consistent with the view that as individuals transition out of emerging adulthood in their late 20s, instabil- ity ceases and more enduring choices in love and work are made (Arnett,

164 PIQUERO ET AL.

2000). The above also provides indirect support for Maruna’s (2001) contention that offenders undergo some sort of identity change in their move- ment away from crime and, thus, begin to take an active role in shaping their developmental pathways (e.g., Brandstadter, 1989). As offenders move through and beyond emerging adulthood, they likely begin to take stock of their past and future lives such that local life circumstances exert a much more different meaning in their 20s than they did in their teens. For example, at age 18, the value of stakes in conformity was .268, but at age 26, it was .644, an increase of 58%. This suggests that even among some serious offenders, investment in social bonds is possible, and that such an investment provides an inhibitory effect on nonviolent arrests, independent of the persistent individual differences captured by our trajectory method. Thus, serious offenders can, in the parlance of Moffitt et al. (1996), “recover” from their criminal pathways and move toward more prosocial outcomes as they enter adulthood. Identification of these factors remains a priority for researchers and policymakers.

REFERENCES

Agresti, Alan

Arnett, Jeffrey Jason

1996 An Introduction to Categorical Data Analysis. New York: Wiley.

Emerging adulthood: A theory of development from the late teens through the twenties. American Psychologist 55:469-480.

Recidivism of Prisoners Released in 1983. Washington, D.C.: Bureau of Jus- tice Statistics.

2000

Beck, Allen and Bernard Shipley 1989

Blumstein, Alfred, Jacqueline Cohen, Jeffrey A. Roth, and Christy A. Visher 1986 Criminal Careers and “Career Criminals.” Washington, D.C.: National

Academy Press.

On the development of different kinds of criminal activity. Sociological Methods and Research 29:319-341.

Brame, Robert, Edward Mulvey, and Alex Piquero 2001

Brandstadter, Jochen 1989 Personal self-regulation of development: Cross-sequential analyses of devel-

opment-related control beliefs and emotions. Developmental Psychology 25196-108.

Broidy, Lisa and Robert Agnew 1997 Gender and crime: A general strain theory perspective. Journal of Research

in Crime and Delinquency 34:275-306.

Stability and change in antisocial behavior: The transition from adolescence to early adulthood. Criminology 39:371-410.

Cernkovich, Stephen A. and Peggy C. Giordano 2001


Cernkovich, Stephen A., Peggy C. Giordano, and Meredith D. Pugh 1985 Chronic offenders: The missing cases in self-report delinquency research.

Journal of Criminal Law and Criminology 76:705-732.

Drugs and predatory crime. In Michael Tonry and James Q. Wilson (eds.), Drugs and Crime. Chicago, Ill.: University of Chicago Press.

Self-control and social control: An exposition of the Gottfredson-Hirschil Sampson-Laub debate. Studies on Crime and Crime Prevention 5:125-150.

Racial-ethnic and gender differences in returns to cohabitation and marriage: Evidence from the Current Population Survey. U.S. Census Bureau. Population Division Working Paper No. 35.

The relationship of problem drinking to individual offending sequences. In Alfred Blumstein, Jacqueline Cohen, Jeffrey A. Roth, and Christy A. Visher (eds.), Criminal Careers and “Career Criminals.” Vol. 2. Washington, D.C.: National Academy Press.

Survival analysis and specific deterrence: Integrating theoretical and empirical models of recidivism. Criminology 35561-575.

D’Unger, Amy, Kenneth C. Land, Patricia McCall, and Daniel S. Nagin How many latent classes of delinquentkriminal careers? Results from mixed Poisson regression analyses of the London, Philadelphia, and Racine cohort studies. American Journal of Sociology 103:1593-1630.

Self-reported and official offending from adolescence to adulthood. In Mal- colm Klein (ed.), Cross-National Research in Self-Reported Crime and Delinquency. Dordrecht, the Netherlands: Kluwer.

Farrington, David P., Bernard Gallagher, Lynda Morley, Raymond J. St. Ledger, and

Unemployment, school leaving, and crime. British Journal of Criminology

Chaiken, Jan M. and Marcia R. Chaiken 1990

Cohen, Lawrence E. and Bryan J. Vila 1996

Cohen, Philip N. 1999

Collins, James J. 1986

DeJong, Christina 1997

1998

Farrington, David P. 1989

Donald J. West 1986

26~335-356. Gottfredson, Michael and Travis Hirschi

Graham, John and Benjamin Bowling

Haapanen, Rudy

1990

1995

1990

A General Theory of Crime. Stanford, Calif.: Stanford University Press.

Young People and Crime. London: Home Office Research Study 145.

Selective Incapacitation and the Serious Offender: A Longitudinal Study of Criminal Career Patterns. New York: Springer-Verlag.

Causes of delinquency: A partial replication and extension. Social Problems 20471-487.

Control theory and the life-course perspective. Studies on Crime and Crime Prevention 4131-142.

Hindelang, Michael 1973

Hirschi, Travis and Michael Gottfredson 1995

PIQUERO ET AL.

Homey, Julie D., D. Wayne Osgood, and Ineke Marshall 1995 Criminal careers in the short-term: Intra-individual variability in crime and

its relation to local life circumstances. American Sociological Review 60:655-673.

Labouvie, Erich 1996 Maturing out of substance use: Selection and self-correction. Journal of

Drug Issues 26:457-476.

Understanding desistance from crime. In Michael Tonry (ed.), Crime and Justice: An Annual Review of Research. Chicago, Ill.: University of Chicago Press.

Good marriages and trajectories of change in criminal offending. American Sociological Review 63:225-238.

The age-crime debate: Assessing the limits of longitudinal self-report data. Social Forces 77:127-155.

Development of juvenile aggression and violence: Some common miscon- ceptions and controversies. American Psychologist 53:242-259.

Making Good: How Ex-Offenders Reform and Reclaim Their Lives. Wash- ington, D.C.: American Psychological Association Books.

Generalized Linear Models. 2d ed. London: Chapman and Hall.

Same or different? Comparing offender groups and covariates over time. Sociological Methods and Research 29:282-318.

Desistance from a delinquent way of life? In Elmar G. M. Weitekamp and Hans-Jurgen Kerner (eds.), Cross-National Longitudinal Research on Human Development and Criminal Behavior. Dordrecht, the Netherlands: Kluwer.

Adolescence-limited and life-course persistent antisocial behavior: A developmental taxonomy. Psychological Review 1 00:674-701.

Moffitt, Terrie E., Avshalom Caspi, Nigel Dickson, Phil A. Silva, and Warren Stanton Childhood-onset versus adolescent-onset antisocial conduct problems in males: Natural history from ages 3 to 18 years. Development and Psychopa- thology 8399424.

Laub, John H. and Robert J. Sampson 2001

Laub, John H., Daniel S. Nagin, and Robert J. Sampson 1998

Lauritsen, Janet 1998

Loeber, Rolf and Magda Stouthamer-Loeber 1998

Maruna, Shadd 2001

McCullagh, Peter and John A. Nelder

McDermott, Shaun and Daniel S. Nagin

1989

2001

Mischkowitz, Robert 1994

Moffitt, Terrie E. 1993

1996

Nagin, Daniel S. 1999 Analyzing developmental trajectories: A semiparametric, group-based

approach. Psychological Methods 4:139-157.

Age, criminal careers, and population heterogeneity: Specification and estimation of a nonparametric. mixed Poisson model. Criminology 31:327-362.

Nagin, Daniel S. and Kenneth C. Land 1993


Nagin, Daniel S. and Raymond Paternoster 1994

2000

Personal capital and social control: The deterrence implications of a theory of individual differences in criminal offending. Criminology 325814506. Population heterogeneity and state dependence: State of the evidence and directions for future research. Journal of Quantitative Criminology 16117-144.

Nagin, Daniel S. and Richard Tremblay 1999 Trajectories of boys’ physical aggression, opposition, and hyperactivity on

the path to physically violent and nonviolent juvenile delinquency. Child Development 70:1181-1196.

Testing Sampson and Laub’s life course theory: Age, race/ethnicity, and drunkenness. Deviant Behavior 20129-151.

Nielson, Amy 1999

Pate, Anthony M. and Edwin E. Hamilton 1992 Formal and informal deterrents to domestic violence: The Dade County

Spouse Assault Experiment. American Sociological Review 57:691-697.

Do fair procedures matter? The effect of procedural justice on spouse assault. Law and Society Review 31:163-204.

Paternoster, Raymond, Charles Dean, Alex Piquero, Paul Mazerolle, and Robert

Continuity and change in offending careers. Journal of Quantitative Crimi- nology 13:231-266.

A model for early onset of delinquent behavior. In Sheilagh Hodgins (ed.), Crime and Mental Disorder. Newbury Park, Calif.: Sage.

The validity of a self-reported delinquency scale. Sociological Methods and Research. In press.

Piquero, Alex R., Alfred Blumstein, Robert Brame, Rudy Haapanen, Edward Mulvey,

Assessing the impact of exposure time and incapacitation on longitudinal trajectories of criminal offending. Journal of Adolescent Research 1654-74.

Transitional life events and desistance from delinquency and crime. In Mar- vin E. Wolfgang, Terence P. Thornberry, and Robert M. Figlio (eds.), From Boy to Man, From Delinquency to Crime. Chicago, Ill.: University of Chi- cago Press.

Understanding and Preventing Violence. Washington, D.C.: National Acad- emy Press.

Crime in the Making. Cambridge, Mass.: Harvard University Press. Understanding variability in lives through time: Contributions of life-course criminology. Studies on Crime and Crime Prevention 4:143-158.

Paternoster, Raymond, Robert Brame, Ronet Bachman, and Lawrence W. Sherman 1997

Brame 1997

Patterson, Gerald and Karen Yoerger 1993

Piquero, Alex R., Randall MacIntosh, and Matthew Hickman 2002

and Daniel S. Nagin 2001

Rand, Alicia 1987

Reiss, Albert J. and Jeffrey A. Roth 1993

Sampson, Robert J. and John H. Laub 1993 1995

PIQUERO ET AL.

Sherman, Lawrence W. and Douglas A. Smith 1992 Crime, punishment, and stakes in conformity: Legal and informal control of

domestic violence. American Sociological Review 57:680-690.

A test of latent trait versus life-course perspectives on the stability of adolescent antisocial behavior. Criminology 36:217-243.

Developmental crime prevention. In Michael Tonry and David P. Farring- ton (eds.), Building a Safer Society: Strategic Approaches to Crime Preven- tion. Chicago, Ill.: University of Chicago Press.

Work as a turning point in the life course of criminals: A duration model of age, employment, and recidivism. American Sociological Review

Simons, Ronald L., Christine Johnson, Rand D. Conger, and Glen Elder, Jr. 1998

Tremblay, Richard and Wendy M. Craig 1995

Uggen, Christopher 2000

65529-546.

Warr, Mark

Wolfgang, Marvin E., Robert M. Figlio, and Thorsten Sellin

Wright, Bradley R., Avshalom Caspi, Terrie E. Moffitt, and Phil A. Silva

1998 Life-course transitions and desistance from crime. Criminology 36183-216.

Delinquency in a Birth Cohort. Chicago, Ill.: University of Chicago Press.

Low self-control, social bonds, and crime: Social causation, social selection, or both? Criminology 37:479-514.

1972

1999

Alex R. Piquero is Associate Professor of Criminology and Sociology in the Center for Studies in Criminology and Law at the University of Florida, Member of the National Consortium on Violence Research, and Network Associate with the MacAr- thur Foundation’s Research Network on Adolescent Development and Juvenile Justice. His research interests include crime over the life course, criminological theory, and quantitative research methods. Please address all correspondence to: Alex R. Piquero, University of Florida, Center for Studies in Criminology and Law, P.O. Box 115950,201 Walker Hall, Gainesville, FL 32611-5950 (E-mail: [email protected]%edu).

Robert Brame is Assistant Professor in the College of Criminal Justice at the Univer- sity of South Carolina. H e is also a Fellow with the National Consortium on Violence Research and Network Associate with the MacArthur Foundation’s Research Network on Adolescent Development and Juvenile Justice. His research interests include juvenile delinquency and legal issues.

Paul Mazerolle is Senior Lecturer in the School of Social Science at the University of Queensland, Australia. H e is also a Research Associate at the Center for Youth at Risk at St. Thomas University in Canada. His research examines theoretical and empirical dimensions of crime and delinquency, including specific issues related to criminal careers and crime over the life course.

Rudy Haapanen is currently a research manager with the California Youth Authority (CYA), serving as Chief of the Ward Information and Parole Research Bureau. He has conducted a number of federally funded research projects in the areas of delinquency prevention, program evaluation, methodology development, prediction, classification


development, drug testing, mental health, and criminal careers. He is currently Princi- pal Investigator on a large-scale study of mental health status and institutionalized adjustment of 1,OOO wards in the CYA.

170 PIQUERO ET AL.

A COMPARISON OF CHANGES IN POLICE AND GENERAL HOMICIDES: 1930-1998*

ROBERT J. KAMINSKI National Institute of Justice

THOMAS B. MARVELL Justec Research, Inc.

This paper presents a new data series for homicides of law enforcement officers. Available for more than two centuries, it is much longer than series previously examined. Police killings had two extreme peaks, one in the 1920s and another in the 1970s. We use the post-l930part of the series in a time-series regression to explore structural conditions that affect police killings in the short term. Economic conditions, prison populations, and World War II have considerably larger impacts on police killings than on homicide generally. Police killings are less affected by demographic changes and by the crack epidemic.

The purpose of this paper is threefold. First, we present a data series for nationwide felonious killings of police that is new to criminology research. This series, which begins in 1794, was compiled by the National Law Enforcement Officers Memorial Fund (NLEOMF) (1998). Second, we explore factors that affect police murders with a time-series analysis for 1930 to 1998. Third, we compare their impact with factors affecting total homicide rates.

Table 1 lists findings from the several studies that use multivariate regression to analyze structural determinants of police murders. There is little similarity in variables entered or in results, and most of the regressors are not significant. Even the variables used most, racial composition and percentage below poverty, do not evidence a consistent pattern. The theory concerning what affects police homicides is almost always theory that is applicable to homicides in general (exceptions are Lott, 2000; Southwick, 1998). Thus, one may predict that the factors have similar relationships with police and total homicides. Only one study, however, compares impacts on police and total murder regressions: Peterson and Bailey (1988) conclude that results for the two murder types are only partially

~~ ~

*We thank Craig Floyd, Director of the National Law Enforcement Officers Memorial Fund, for providing us the data on officers killed in the line of duty, and Bernie Spence, Director of Research for the NLEOMF, for her assistance. We also thank the anonymous reviewers for their helpful comments. Points of view are those of the authors and do not necessarily represent the views of the U.S. Department of Justice or the National Institute of Justice.

CRIMINOLOGY VOLUME 40 NUMBER 1 2002 171

Date post:	19-Nov-2023
Category:	Documents
Upload:	independent
View:	1 times
Download:	0 times

CRIME IN EMERGING ADULTHOOD*

Documents