Skip to main content
Log in

Desistance or Displacement? The Changing Patterns of Offending from Adolescence to Young Adulthood

Journal of Quantitative Criminology Aims and scope Submit manuscript

Abstract

Research has devoted substantial attention to patterns of offending during the transition to early adulthood. While changes in offending rates are extensively researched, considerably less attention is devoted to shifts in the type of offending displayed during the transition to adulthood. Changes in the type of offending behavior suggest a pattern of “displacement” or shifts between various types of crime, rather than desistance from deviant behavior. In this paper, I integrate methods previously developed in stratification research and use longitudinal data from the National Survey of Youth that span the transition to adulthood to examine the extent to which desistance and displacement of deviant behavior are defining attributes of offending during the transition to early adulthood. The findings indicate that while desistance is clearly present, altering patterns of offending, or within-person displacement, rather than termination of illicit activity is most evident in the data.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1

Similar content being viewed by others

Notes

  1. To avoid redundancy, within person displacement is simply referred to as displacement for the remainder of the paper. I draw attention to the within-person, or individual nature of the behavior stressed in this paper as opposed to the developed research tradition on spatial displacement of crime, or the shifting of crime from one community location to another (Bursik and Grasmick 1993; Griffiths and Chavez 2004; Wilson 1987).

  2. I draw particular attention to the slow process of Stanley moving away from violence, as he went back to “Jack-Rolling” after prison, and often used violence to settle social disputes, such as work problems (Shaw 1966, p. 180). Stanley’s slow and uneven movement away from violence is suggestive of displacement as well as desistance.

  3. While an explicit discussion appears later in the paper, the empirical methods used in this analysis are also based on counts. However, the methods used in this paper analyze the association patterns as given by the cell frequencies in a cross classification of a variety of different offenses as well as the cross classification of latent offending classes.

  4. I conducted multiple tests of robustness to examine age variation in the results. The results reported here appear robust under multiple specifications. See Appendix 2 for a more complete discussion.

  5. Lauritsen concludes the NYS may suffer from testing effects (1998, p. 150). However, the main analysis in that paper is based on ordinal responses and further sub-analysis is based on frequencies and does not extend to the measures used in this analysis. However, in light of Lauritsen’s critique, and other work that advocates dichotomous measures (Piliavin et al. 1986, p. 105), prevalence indicators rather than rate measures are used in this analysis.

  6. The final number of categories in a latent variable is determined by specifying multiple models, each with a different number of latent categories, and then assessing the model fit of each specification. See Appendix 2 for a detailed discussion of model selection.

  7. Model fit is assessed through multiple indictors, including a BIC statistic, an index of dissimilarity, and chi-square tests (McCutcheon 1987). Model robustness is assessed in multiple ways. I used a different national sample of youthful offending, the National Longitudinal Survey of Youth (NLSY), I conducted sub-analysis within the National Youth Survey based on the age of respondents, and I made minor modifications to the offending measures used for analysis. In all cases, the results were substantively similar, both in terms of the number of latent classes and the characteristics of the offending classes. See Appendix 2 for a more detailed discussion.

  8. I conducted a formal test for an abstainer group, and there is not model support for this specification. This is consistent with other research using similar methods (see for instance, Laub et al. 1998; D’Unger et al. 1998). One implication of most latent class analysis is that individuals who commit very few crimes and individuals who commit no crime can be grouped in the same class. I thank a reviewer for pointing to this substantive implication of this aspect of the latent class methodology.

  9. Statistical significance is tested by fixing parameters and examining model fit statistics relative to an unrestricted model. In a series of model estimations, I set the vandalism, theft, and violence parameters in the predatory class to be equal to the estimated parameters in the normative class, and the results suggest a significantly poorer model fit. This indicates that the difference between the prevalence of these crimes in the adolescent predatory class and the adolescent normative class is statistically significant. This procedure was repeated multiple times to test for difference in behavior, for instance, whether the drug use class had elevated rates of substance use relative to the predatory group. Model fit is assessed through a df chi-square comparison of the restricted and unrestricted model. See McCuthchen (1987) and Clogg (1977) for more information.

  10. While the defining feature of this group is the predatory nature of their criminal involvement, the likelihood of involvement in violence, theft, and vandalism is much lower than the predatory group in adolescence, an issue I address in more detail below.

  11. Testing for differences over time is methodologically identical to the process outlined in footnote 7. Substantively the difference involves restricting, for instance, adolescent violent behavior in any given latent class to be identical to adult violent behavior in a given class, and then assessing model fit.

  12. In a later work, Bachman et al. 2002 conclude the use of all substances eventually declines by the time young adults reach their late 20s and early 30s (2002, p. 204).

  13. The transition table is a product of assigning individuals to latent classes in adolescence and young adulthood. This is generally done in one of two ways. One method, modal probability, assigns individuals to latent groups which they have the highest probability of being involved (for instance, Laub et al. 1998). The method used in this paper takes into account the uncertainty of latent class membership. The sample is assigned a latent class membership based on the exact probability of membership in each latent class. This is a two-stage process; first using the CDAS program (Eliason 1997), I calculated the exact probabilities of all 1,383 individual cases in the sample falling in each of the four latent classes. Then using random count techniques in SAS, each case is randomly assigned to one of the four latent classes based on the exact probabilities of falling in a latent class. Taken in sum, this reduces the uncertainty associated with latent class membership (see discussions by Laub et al. 1998; Roeder et al. 1999).

  14. When using age adjusted measures there is some conceptual overlap between stability and displacement of behavior. For example some individuals who were stealing from school in youth are stealing from work in adulthood, which may suggest some displacement of behavior. However, this pattern of behavior is most consistent with notions heterotypic continuity, thus it is treated as stability of behavior.

  15. In this case, the ordering of the latent variable is not explicit. Clearly, the normative class is the least serious, and pervasive class is the most serious. However, whether the drug class or the violent class is more serious, and the scaling or the distance between latent classes is unclear.

  16. I test for zero, one, two and three dimensional solution, and the data support a one dimensional solution, which is presented in Fig. 1.

  17. The dimensional axis is bound between −1 and 1. Fig. 1 presents the adjusted or standardized scores.

  18. The formula for the pairwise test of significance is: \(z=(\hat{\nu}a-\hat{\nu}b)/\sqrt{V(\hat{\nu}_a)+V(\hat{\nu}_b)- 2\hbox{cov}(\hat{\nu}_a\hat{\nu}_b)}\) where \(\hat{\nu}_a\) and \(\hat{\nu}_b\) are the estimated RC association model scores for latent classes a and b respectively, \(V(\hat{\nu}_a)\) and \(V(\hat{\nu}_b)\) give the estimated variances of \(\hat{\nu}_a\) and \(\hat{\nu}_b\), and \(\hbox{cov}(\hat{\nu}_a\hat{\nu}_b)\) gives the covariance. The null hypothesis is that the true distance between the two latent classes is zero. In the present analysis a significant pairwise association indicates that moving from one latent class to another represents a significant change in behavior.

References

  • Bachman JG, O’Malley PM, Schulenberg JE, Johnston LD, Bryant AL, Merline AC (2002) The decline of substance use in young adulthood: changes in social activities, roles, and beliefs. Lawrence Erlbaum Associates, Mahwah, NJ

    Google Scholar 

  • Bachman JG, Wadsworth KN, O’Malley PM, Johnston LD, Schulenberg JE (1997) Smoking, drinking, and drug use in young adulthood: the impacts of new freedoms and new responsibilities. Lawrence Erlbaum Associates, Mahwah, NJ

    Google Scholar 

  • Becker MP (1990) Algorithm AS 253: maximum likelihood estimation of the RC(M) association model. Appl Stat 39:152–167

    Article  Google Scholar 

  • Bohrnstedt G, Knoke D (1994) Statistics for social data analysis, 3rd edn. E. Peacock Publishers, Itasca, IL

    Google Scholar 

  • Brame R, Paternoster R (2003) Missing data problems in criminological research: two case studies. J Quant Criminol 19:55–78

    Article  Google Scholar 

  • Brame R, Piquero A (2003) Selective attrition and the age-crime relationship. J Quant Criminol 19:107–128

    Article  Google Scholar 

  • Brame R, Bushway SD, Paternoster R (2003) Examining the prevalence of criminal desistance. Criminology 41:423–448

    Article  Google Scholar 

  • Bursik R Jr, Gramick HG (1993) Neighborhoods and crime: the dimensions of effective community control. Lexington Books, California

    Google Scholar 

  • Bushway SD, Thornberry TP, Krohn MD (2003) Desistance as a developmental process: a comparison of static and dynamic approaches. J Quant Criminol 19:129–153

    Article  Google Scholar 

  • Bushway S, Piquero A, Broidy L, Cauffman E, Mazerolle P (2001) An empirical framework for studying desistance as a process. Criminology 39:491–516

    Article  Google Scholar 

  • Caspi A (1998) Personality development across the life course. In: Eisenberg N (eds) Handbook of child psychology, vol. 3: social, emotional, and personality development. Wiley, New York

    Google Scholar 

  • Clogg CC (1977) Unrestricted and restricted maximum likelihood latent structure analysis: a manual for users. Working paper no. 1977–09. Pennsylvania State University, Population Issues Research Center, University Park, PA

  • Clogg CC (1995) Latent class models. In: Arminger G, Clogg C, Sobel ME (eds) Handbook of statistical modeling for the social and behavioral sciences. Plenum Press, New York

    Google Scholar 

  • Clogg CC, Eliason SR, Wahl RJ (1990) Labor-market experiences and labor-force outcomes. Am J Sociol 95:1536–1576

    Article  Google Scholar 

  • Clogg CC, Eliason SR, Leicht K (2002) Analyzing the labor force: concepts, measures, and trends. Plenum Press, New York

    Google Scholar 

  • Clogg CC, Shihadeh ES (1994) Statistical models for ordinal variables. Sage Publication, Beverly Hills

    Google Scholar 

  • Cohen J (1986). Research on criminal careers: individual frequency rates and offense seriousness. In: Blumstein A, Cohen J, Roth J, Visher C (eds) Criminal careers and “career criminals.” Report to the National Academy of Sciences Panel on Research on Criminal Careers. National Academy Press, Washington, DC

    Google Scholar 

  • D’Unger AV, Land KC, McCall PL, Nagin DS (1998) How many latent classes of delinquent/criminal careers? Results from mixed Poisson regression Analyses. Am J Sociol 103:1593–1630

    Article  Google Scholar 

  • Eliason SR (1993) Maximum likelihood estimation: logic and practice. Sage Publication, Beverly Hills

    Google Scholar 

  • Eliason SR (1995). Modeling manifest and latent dimensions of association in two-way cross-classifications. Sociol Methods Res 24:30–67

    Article  Google Scholar 

  • Eliason SR (1997) The categorical data analysis system user’s manual. Version 4.0

  • Elliott D, Huizinga D, Ageton S (1985) Explaining delinquency and drug use. Sage Publications, Beverly Hills

    Google Scholar 

  • Elliott D, Huizinga D, Menard S (1989) Multiple problem youth: delinquency, substance abuse, and mental health problems. Springer-Verlag, New York

    Google Scholar 

  • Farrington D, Hawkins D (1991). Predicting participation, early onset, and later persistence in officially recorded delinquency. Crim Behav Ment Health 1:1–33

    Google Scholar 

  • Felson RB, Haynie DL (2002) Pubertal development, social factors, and delinquency among adolescent boys. Criminology 40:967–988

    Article  Google Scholar 

  • Fergusson DM, Horwood LJ, Nagin DS (2000) Offending trajectories in a New Zealand birth cohort. Criminology 38:525–551

    Article  Google Scholar 

  • Francis B, Soothill K, Fligelstone R (2004) Identifying patterns and pathways of offending behaviour: a new approach to typologies of crime. Eur J Criminol 1:48–87

    Article  Google Scholar 

  • Glueck S, Glueck E (1940) Juvenile delinquents grown up. Oxford University Press, New York

    Google Scholar 

  • Glueck S, Glueck E (1968) Delinquents and nondelinquents in perspective. Harvard University Press, Cambridge, MA

    Google Scholar 

  • Goodman LA (1987) New methods for analyzing the intrinsic character of qualitative variables using cross-classified data. Am J Sociol 93:529–583

    Article  Google Scholar 

  • Gottfredson M, Hirschi T (1990) A general theory of crime. Stanford University Press, Stanford, CA

    Google Scholar 

  • Griffiths E, Chavez JM (2004) Communities, street guns and homicide trajectories in Chicago, 1980–1995: merging methods for examining homicide trends across space and time. Criminology 42:941–978

    Article  Google Scholar 

  • Heimer K, Matsueda RL (1994) Role taking, role commitment, and delinquency: a theory of differential social control. Am Sociol Rev 59:365–390

    Article  Google Scholar 

  • Hindelang M, Hirschi T, Weis JG (1981) Measuring delinquency. Sage Publication, Beverly Hills

    Google Scholar 

  • Hirschi T, Gottfredson M (1983) Age and the explanation of crime. Am J Sociol 89:552–584

    Article  Google Scholar 

  • Jang SJ, Johnson BR (2001) Neighborhood disorder, religiosity, and adolescent use of illicit drugs: a test of multilevel hypotheses. Criminology 39:109–144

    Article  Google Scholar 

  • Laub JH, Nagan D, Sampson RJ (1998) Trajectories of change in criminal offending: good marriages and the desistance process. Am Sociol Rev 63:225–239

    Article  Google Scholar 

  • Laub JH, Sampson RJ (2003) Shared beginnings, divergent lives: delinquent boys to age 70. Harvard University Press, Cambridge MA

    Google Scholar 

  • Lauristen J (1998) The age-crime debate: assessing the limits of longitudinal self-report data. Soc Forces 77:127–155

    Article  Google Scholar 

  • Loeber R, Le Blanc M (1990) Toward a developmental criminology. In: Tonry M, Morris N (eds) Crime and justice. University of Chicago Press, Chicago

    Google Scholar 

  • Loeber R, Stouthamer-Loeber M, Van Kammen W, Farrington D (1991) Initiation, escalation, and desistance in juvenile offending and their correlates. J Crim Law Criminol 82:36–82

    Article  Google Scholar 

  • Maruna S (2000) Making good: how ex-convicts reform and rebuild their lives. American Psychological Association, Washington, DC

    Google Scholar 

  • Matsueda R (1992) Reflected appraisals, parental labeling, and delinquency: specifying a symbolic interactionist theory. Am J Sociol 97:1577–1611

    Article  Google Scholar 

  • McCutcheon A (1987) Latent class analysis. Sage Publications, New York

    Google Scholar 

  • McDermott S, Nagin DS (2001) Same or different? Comparing offender groups and covariates over time. Sociol Methods Res 29:282–319

    Article  Google Scholar 

  • Moffitt T (1993) Life-course-persistent and adolescence limited antisocial behavior: a developmental taxonomy. Psychol Rev 100:674–701.

    Article  Google Scholar 

  • Moffitt TE, Krueger RF, Caspi A, Wright R (2000) Partner abuse and general crime: how are they the same? How are they different? Criminology 38:199–232

    Article  Google Scholar 

  • Nagin D, Farrington D, Moffitt T (1995) Life-course trajectories of different types of offenders. Criminology 33:111–140

    Article  Google Scholar 

  • Nagin D, Tremblay RE (2001) Analyzing developmental trajectories of distinct but related behaviors: a group based method. Psychol Methods 6:18–34

    Article  Google Scholar 

  • Paternoster R, Brame R (1998) The structural similarity of processes generating criminal and analogous behaviors. Criminology 36:633–670

    Article  Google Scholar 

  • Paternoster R, Brame R (2000) On the association between self-control, crime, and analogous behaviors. Criminology 38:971–982

    Article  Google Scholar 

  • Piliavin I, Gartner R, Thornton C, Matsueda R (1986) Crime, deterrence, and rational choice. Am Soc Rev 51:101–119

    Article  Google Scholar 

  • Piquero AR, Brame R, Mazerolle P, Haapanen R (2002a) Crime in emerging adulthood. Criminology 40:137–170

    Article  Google Scholar 

  • Piquero A, MacIntosh R, Hickman M (2002b) The validity of a self-reported delinquency scale: comparisons across gender, age, race, and place of residence. Sociol Methods Res 30:492–529

    Article  Google Scholar 

  • Roeder K, Lynch K, Nagin D (1999) Modeling uncertainty in latent class membership: a case study in criminology. J Am Stat Assoc 94:766–776

    Article  Google Scholar 

  • Sampson RJ, Laub JH (1990) Crime and deviance over the life course: the salience of adult social bonds. Am Sociol Rev 55:609–627

    Article  Google Scholar 

  • Sampson R, Laub J (1993) Crime in the making: pathways and turning points through life. Harvard University Press, Cambridge MA

    Google Scholar 

  • Shaw C (1966) The Jack-Roller, 5th edn. University of Chicago Press, Chicago

    Google Scholar 

  • Shover N (1993) Great pretenders: pursuits and careers of persistent thieves. Westview Press, Boulder

    Google Scholar 

  • Silver E (2000) Extending social disorganization theory: a multilevel approach to the study of violence among persons with mental illnesses. Criminology 38:1043–1074

    Article  Google Scholar 

  • Steffensmeier D, Allan E, Harer M, Streifel D (1989) Age and the distribution of crime. Am J Sociol 94: 803–831

    Article  Google Scholar 

  • Uggen C (2000) Work as a turning point in the life course of criminals: a duration model of age, employment, and recidivism. Am Sociol Rev 65:529–549

    Article  Google Scholar 

  • Warr M (1998) Life-course transitions and desistance from crime. Criminology 36:183–215

    Article  Google Scholar 

  • Wilson W (1987) The truly disadvantaged. The inner city, the underclass and public policy. University of Chicago Press, Chicago

    Google Scholar 

  • Wright BR, Caspi A, Moffitt TE, Silva PA (1999) Low self-control, social bonds, and crime: social causation, social selection, or both? Criminology 37:479–514

    Article  Google Scholar 

  • Wright BR, Caspi A, Moffitt TE, Silva PA (2001) The effects of social ties on crime vary by criminal propensity: a life course model of Interdependence. Criminology 39:321–351

    Article  Google Scholar 

  • Yamaguchi K (1987) Event-History analysis: its contributions to modeling and causal inference. Sociol Theor Method 2:61–82

    Google Scholar 

Download references

Acknowledgments

This research is supported by the National Institute of Mental Health [#MH19893]. I am especially indebted to Christopher Uggen, Scott Eliason, and Ryan King for their constructive advice. I also thank Shawn Bushway, Valerie Clark, Jennifer Lee, Ross Macmillan, Jeylan Mortimer, Raymond Paternoster, Alex Piquero, Eric Silver, Jeremy Staff, Sara Wakefield, David McDowall and three anonymous reviewers for their comments on an earlier version of this paper.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Michael Massoglia.

Additional information

An earlier version of this paper was presented at the annual meetings of the American Sociological Association, San Francisco, August 2004.

Appendices

Appendix 1: Measurement of Offending

Prior research on desistance uses dichotomous measures, particularly research in the behavioral desistance tradition (see, for instance, Farrington and Hawkins 1991; Loeber et al. 1991; Warr 1998). Aside from the general utility of dichotomous measures, the use of such indicators to test ideas of displacement or desistance warrants discussion. I independently model involvement in six distinct types of criminal behavior. In doing so, I capture a greater range of offending behavior than much prior work and leverage the unique information from multiple types of illicit and anti-social behavior, rather than masking variation in a single summed indicator. The strength of the method is the range of behaviors captured, while a possible weakness is the lost information on the frequency of certain behaviors.

However, there is some debate as to the overall utility of frequency measures. Some research questions the accuracy of asking the “number” of times an offense happens, arguing that indicators of whether an event “ever” happened are more reliable (Piliavin et al. 1986; Hindelang et al. 1981). In this study, the vandalism measure included damaging family property, school property, and other property. These three response categories were collapsed into one dichotomous variable representing overall involvement in vandalism. Thus, consistent with prior work (Fergusson et al. 2000) in the analysis, people who reported involvement in any of the three crimes are coded “1” for involvement in vandalism and “0’ for no involvement in vandalism.

There are also weaknesses inherent to using dichotomous indicators, for instance an individual who commit a crime multiple times in adolescence and only once in adulthood would be coded the same at both time points. Moreover, someone who commits multiple crimes is treated the same as someone who commits a single act, hypothetically making individuals who are heavily involved in crimes, for instance frequent cocaine use, look similar to those whose cocaine use is infrequent. However, the coding scheme used in the latent class analysis cross classifies six types of crime to create 64 different offending patterns. Under such a scheme, for frequent and isolated cocaine use to look identical, someone heavily involved in cocaine use would have to display no antisocial behavior in other areas of the life course, for example binge drinking or workplace problems, as individual assignment into latent classes is based on the co-occurrence of the range of behaviors measured. While any measurement decisions have potential drawbacks, the research design used allows for a precise examination of movement in out of various crimes that adds to the exiting body of research that has used summary frequency scales.

To capture theft, a dichotomous measure asks respondents if they have stolen money, stolen from family, or stolen things at school. A dichotomous measure of involvement in violence is created by collapsing respondents’ frequency of hitting parents, hitting students, and hitting others. Similarly, marijuana use is coded into a single dichotomous variable. Involvement in other drug use (hallucinogen, amphetamines, and cocaine) is collapsed into a dichotomous variable, called hard drug use. In adolescence, the general deviance indicator includes whether respondents have been drunk, publicly disorderly and/or skipped class. In adulthood the measure captured whether respondents had been drunk/bought alcohol for minors, been publicly disorderly, and whether they had been fired for cause, measured by asking if respondents had lost their job for violation of work rules such racial or sexual discrimination or drug use at work. The measures used for the general deviance indicator were driven by theory (Gottfedson and Hischi 1990) and past empirical work (Glueck and Glueck 1968). The descriptive statistics are reported for each of the 6 offending measures in Appendix 1, Table 5.

Table 5 Variables and descriptive statistics

Appendix 2. Model fit, Selection, and Robustness

The latent class analysis indicates that four distinct offending patterns best represent criminal behavior in youth. I test for model specifications using two, three, four, and five offending patterns. Multiple indicators are used to assess model fit. As shown in Appendix 2, Table 6, in all cases, the fit statistics best support a four class model. When assessing fit under different offending specifications, a low BIC statistic, low index of dissimilarity, and non-significant p-values generally indicate a proper model specification.

Table 6 Fit statistics

In adolescence, models with two or three offending patterns are not well supported by the data. That is, neither the likelihood ratio chi-square statistic, the index of dissimilarity, nor the BIC statistic suggests the two or three class models adequately represent the data. When examining the five pattern specification, again, there is no support over the specification with four offending patterns. In contrast, all indicators of fit suggest a model with four offending patterns.

In early adulthood, the data again best support a specification with four offending patterns. As with adolescence, the models with two or three offending patterns do not adequately fit the data. The BIC statistic and index of dissimilarity are supportive of a specification with four offending patterns. The likelihood ratio chi-square statistic indicates some slight mis-fit in both the four and five class models. However, other indicators support a four class model. Both the BIC statistic and the index of dissimilarity best support four class model. Perhaps more importantly, examination of the Freeman–Tukey residuals indicates a four class model mis-fits the data in only one of a possible sixty-four cells. This cell contains only five individuals. Thus, a four-class model well represents over 99.5% of the sample.

Additionally I conduct multiple tests to examine the robustness of the findings. First, I analyze two sub-samples within the NYS to test for age variation in offending patterns. The first consist of individuals who were 11, 12, and 13-year-olds in adolescence and the second sub-sample is made up of 14, 15, 16, and 17-year-olds in adolescence. This cut point essentially breaks the sample into individuals who were in junior high school and respondents who were in high school during wave 1, and examines whether the results observed in the full sample are influenced by age variation within the sample. For each sub-sample, I then examine the latent structure of the offending patterns in both adolescence and adulthood for each sub-group. Again, while the behavioral probabilities are not identical to the full sample measures, the results do not change substantively for either group at either time point, with a four class model again providing the best fit to the data in all cases.

I then create a transition table of adolescent and adult offending for both sub-groups. These disaggregated results were compared to the full results presented in Table 3. During the transition to adulthood, there is slightly more desistance among the older respondents at wave 1, those 14–17, and slightly more escalation among the younger respondents at wave 1, the 11 to 13-year-olds. However, the patterns of displacement and stability were remarkably similar for both the full results and the disaggregated results. For instance, in the full sample, approximately 43.5% of respondents had identical latent class assignment at both points in time, whereas 43% of the younger respondents and 44% of the older responds had stable classification at both points in time. Thus the patterns of displacement and stability evident in the full sample appear to hold for more narrow age ranges as well.

Secondly, I randomly remove specific indicators from the analysis, and again assess the latent offending structure. This procedure is done to test the extent to which any single indicator is influencing the overall results of the model. This procedure is repeated multiple times, and the results were substantively consistent with the findings presented in this paper. Accordingly, the results appear robust across different samples as well as to age and offense variation within the sample.

Finally, in light of recent work that examines sample attrition in criminology, supplementary analyses were conducted to see how sample attrition might impact the results of the paper. Elliot et al. (1989) argue that offending is unrelated to host of social process and Brame and Piquero (2003) find offending does not significantly increase the likelihood of dropping out of the NYS. However, if certain assumptions are made, Brame and Paternoster (2003) conclude sample attrition is related to social process such as receiving public assistance. Given the equivocal evidence on sample attrition in the NYS, additional analysis were undertaken. Using the wave 1 full sample (n = 1,725) of the NYS, the latent structure of offending was estimated. Differences in the wave one structure of offending between the full sample and the sample used, with cases missing at wave seven omitted, speak to possible attrition bias. Again the behavioral structure is substantively similar, with normative, predatory, drug, and pervasive groups evident in the four class structure. One difference does emerge, in the analysis using the full sample, the percentage of the overall population that would fall into the predatory increased approximately 7%, from 27% to 34%. Such a finding suggests that some of the dramatic decline in the rates of predatory crimes may be a function of sample attrition. It is important to note that this is purely speculative, as that data do not exist to definitively answer this question. This finding does point to the need for research to consider the potential effects of sample attrition on our models and inferences. All tests of robustness are available from the author upon request.

Rights and permissions

Reprints and permissions

About this article

Cite this article

Massoglia, M. Desistance or Displacement? The Changing Patterns of Offending from Adolescence to Young Adulthood. J Quant Criminol 22, 215–239 (2006). https://doi.org/10.1007/s10940-006-9009-8

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10940-006-9009-8

Keywords

Navigation