Testing Theory and the Analysis of Time Series Data

Although the relationship between unemployment and crime has been a longstanding interest in criminology, there is little agreement about appropriate models for analyzing this relationship. David Greenberg's (this issue) discussion highlights two issues that raise questions about recent research on the unemployment–crime relationship. First, he extends the work of Hale and Sabbagh (1991) and argues that cointegration methods should be used instead of first-differenced regression models to analyze unemployment and crime time series data. Second, he argues that previous attempts to test his strain theory linking unemployment to the age distribution of crime rely on flawed hypotheses, inappropriate data, and faulty measurement. In this paper, I address both of Greenberg's claims. I begin with a discussion of the relative utility of cointegration analysis and of first-differenced regression models for the analysis of the unemployment–crime relationship, focusing on the link between theory and statistical model. I then discuss the possibility of ever testing and falsifying Greenberg's strain theory.     

LONG‐TERM CRIME DESISTANCE AND RECIDIVISM PATTERNS—EVIDENCE FROM THE ESSEX COUNTY CONVICTED FELON STUDY*

Two conflicting definitions of desistance exist in the criminology literature. The first definition is instantaneous desistance in which an offender simply chooses to end a criminal career instantaneously moving to a zero rate of offending ( Blumstein et al., 1986 ). The second definition views desistance as a process by which the offending rate declines steadily over time to zero or to a point close to zero ( Bushway et al., 2001 ; Laub and Sampson, 2001 ; Leblanc and Loeber, 1998 ). In this article, we capitalize on the underlying assumptions of several parametric survival distributions to gain a better understanding of which of these models best describes actual patterns of desistance. All models are examined using 18 years of follow‐up data on a cohort of felony convicts in Essex County, NJ. Our analysis leads us to three conclusions. First, some people have already desisted at the beginning of the follow‐up period, which is consistent with the notion of “instantaneous desistance.” Second, a three‐parameter model that allows for a turning point in the risk of recidivism followed by a long period of decline fits the data best. This conclusion suggests that for those offenders active at the start of the study period, the risk of recidivism is declining over time. However, we also find that a simpler two‐group model fits the data almost as well and gains superiority in the later years of follow‐up. This last point is particularly relevant as it suggests that the observed gradual decline in the hazard over time is a result of a compositional effect rather than of a pattern of individually declining hazards.     

Predicting recidivism in a communitarian society: China

Research on the prediction of recidivism has largely been an enterprise of Western criminology. Therefore, the identification and selection of predictors has tended to follow the individualistic traditions of the West. Important advances in models and methods have not been extended to non-Western societies such as China. This article explores the implications of communitarian features of Chinese urban communities for prediction of recidivism. The article applies the perspective of social capital to the specification of predictors. Available community social-capital measures are included in the prediction model to capture the effects of communitarian cultural features. The results indicate that social capital variables generally have significant effects.     

Survival models of recidivism among juvenile delinquents

Among juveniles, the probability of recidivism has a curvilinear relationship to age. Rates of reoffending do not simply increase or decrease with age, but rather, they increase as a function of age up to a certain point of peak activity and decrease with increasing age thereafter. Because of this, the forms of recidivism functions between cohorts of widely varying ages will differ considerably. This renders inapplicable one model for the analysis of recidivism rates which assumes proportional hazards (Cox proportional hazards regression models). Appropriate models will posit curved nonmonotonic hazard functions. This paper compares fits of the exponential, Weibull, and log-normal models to recidivism data collected on samples of chronic juvenile offenders and finds generally closer estimation of the recidivism functions using the latter model. Implications for the development of models of recidivism are discussed.     

Recidivism among drug offenders: A survival analysis of the effects of offender characteristics,type of offense,and two types of intervention

The determinants of recidivism are increasingly becoming the focus of public concern. This study explores the relative effect of type of intervention, offender characteristics, and type of incident offense on time to a petition to revoke probation and time to a probation revocation. Our analysis of intervention effects includes both parametric and nonparametric estimation procedures. Estimating five distributional forms of survival and a proportional hazard model for each measure of recidivism, the analysis indicates no difference in the effect of a program of drug monitoring and treatment, compared to drug monitoring only, for either of the two measures of recidivism. In addition, findings indicate that younger offenders and African American offenders have a shorter time to a petition to revoke probation. We also found a reduced time to failure for a probation revocation for African American offenders and offenders with a prior arrest record. Our findings offer empirical support for a reconsideration of the type of intervention effective in deterring offenders while on probation.     

A Time Series Modeling Approach in Risk Appraisal of Violent and Sexual Recidivism

For over half a century, various clinical and actuarial methods have been employed to assess the likelihood of violent recidivism. Yet there is a need for new methods that can improve the accuracy of recidivism predictions. This study proposes a new time series modeling approach that generates high levels of predictive accuracy over short and long periods of time. The proposed approach outperformed two widely used actuarial instruments (i.e., the Violence Risk Appraisal Guide and the Sex Offender Risk Appraisal Guide). Furthermore, analysis of temporal risk variations based on specific time series models can add valuable information into risk assessment and management of violent offenders.     

The parole performance of offenders released from shock incarceration (boot camp prisons): A survival time analysis

The parole performance of offenders who were released after successfully completing a shock incarceration program was examined and compared to the performance of offenders who were serving time on probation or parole after a period of incarceration. Separate survival analyses were performed for recidivism as measured by (1) arrests and (2) failures (jailed, absconded, or revoked). Prior incarceration, age, age at first arrest, and risk assessment score were related to recidivism but type of sentence was not. Intensity of supervision was significantly related to recidivism but this relationship was eliminated when risk level was controlled. There was no evidence that shock incarceration reduces recidivism. Future research should focus on methods of reducing failures during community supervision for these young, nonviolent offenders within the framework of either a shock incarceration program or some other sentence.An earlier version of this paper was presented at the American Probation and Parole Association 14th Annual Training Institute, Milwaukee, Wisconsin, August 1989.     

Theory, Operationalization, Identification, and the Interpretation of Different Differences in Time Series Models

The principals in this exchange, David Cantor, David Greenberg, and Kenneth Land, have each contributed outstanding methodological and substantive scholarship to criminology and to the social sciences more generally. In exchanges such as this we are engaged in a collective process of refining and advancing our knowledge in the areas of both criminology and statistical methods. The issues raised by Greenberg provide an opportunity to reflect on time series modeling and to revisit some issues concerning operationalization and model specification. I do not expect any of us to agree fully on these issues, but we may be able to shed some light on these issues, and collectively this may help the larger community of criminologists and social scientists.     

What methods are most frequently used in research in criminology and criminal justice?

Articles published in seven leading criminology and criminal justice journals were coded with regard to the research methods used, focusing on the general research designs, data-gathering methods, and statistical analysis techniques employed. The results indicated that survey research was by far the dominant mode of acquiring criminological information, that cross-sectional nonexperimental designs still predominated, and that multivariate statistical methods were the norm. The findings could aid criminology and criminal justice faculty in devising graduate methods curricula that reflected the state-of-the-art as currently practiced by criminological researchers.     

Social ecology, individual risk, and recidivism: A multilevel examination of main and moderating influences

Purpose

Despite a longstanding tradition in criminology to consider the impact of neighborhood context on crime-related outcomes, criminologists have largely ignored the influence of social ecology on recidivism until recently. The purpose of the present study was to examine the main and moderating influences of social ecology on recidivism.

Materials and Methods

The present study used hierarchical nonlinear modeling to estimate the effects of concentrated disadvantage, immigrant concentration, and residential stability on recidivism for a sample of offenders released from custody/supervision in 2006 and nested within Iowa counties. We controlled for individual-level risk for recidivism using the Level of Service Inventory-Revised (LSI-R), a validated risk assessment instrument. We also examined whether the relationship between LSI-R score and recidivism varied across counties, and if so, whether this variation can be explained by social structural characteristics.

Results

Results indicate that residential stability was the only contextual variable significantly related to recidivism. The relationship between individual-level risk and recidivism did not vary across contexts.

Conclusions

The findings suggest that the social structural context has limited influence on recidivism, while the LSI-R is a robust predictor of recidivism across contexts. We discuss the implications of our findings for theory, practice, and future research.     

The effect of time served on recidivism: An interdisciplinary theory

The research reported in this paper is concerned with the functional relation between the probability that an individual offender will recidivate and the time that the offender has served in prison as a consequence of the offense. Theory suggests that time served affects recidivism through both specific deterrent and social bonding effects. The former is hypothesized to produce an inverse relation between recidivism and time served; the latter, a direct relation. The paper reviews the literature concerning these two theoretical relations. The review provides the basis for the establishment of a multiequation model whose principal result is to suggest that, more than likely, the recidivism/time served relation is a U-shaped function. The empirical model used to evaluate the hypothesis that the function is U-shaped uses the occurrence of a new arrest as its dependent variable. The model is estimated using OLS and LOGIT procedures. The data used for estimation relate to 1425 prisoners released from the North Carolina prison system in 1980. Recidivistic outcomes are measured at the end of the first and second years subsequent to their release. The research is unique in that it represents the first effort to synthesize the economist's specific deterrence model with the sociologist's social bonding model. The theory provides an explanation for the failure of past empirical work to establish a relation between time served and recidivism. The research is also unique in its a priori assumption of a U-shaped behavioral relation and its attempt to measure that particular relation. The data support the hypothesis that time served affects postprison recidivism rates and that the direction of the effect varies by offense class.     

Religious programs and recidivism among former inmates in Prison Fellowship programs: A long-term follow-up study

In the mid-1990s, Prison Fellowship (PF), a nonprofit religious ministry to prisoners, commissioned a study to determine the relationship, if any, between religious programming and recidivism. Subsequent research found no difference between PF and non-PF inmates on measures of recidivism. Inmates most active in PF Bible studies, however, were significantly less likely to be arrested during a 1-year follow-up period. This study extends and improves on previous research by: (1) increasing the recidivism window from 1 to 8 years; (2) incorporating new approaches to measuring program participation; (3) including two measures of recidivism—rearrest and reincarceration; and (4) using survival analysis and proportional hazards modeling to present and analyze the data. Results from survival analyses indicate: (1) no difference in median time to rearrest or reincarceration between PF and non-PF groups throughout the 8-year study period; (2) participants with higher levels of participation in Bible studies were less likely to be rearrested at 2 and 3 years after release, though the effect diminishes over time; (3) statistical differences across groups only border significance at 2 and 3 years for reincarceration; and (4) proportional hazards modeling shows that high participation in Bible studies significantly reduces the hazard of rearrest at years 2 and 3.     

The importance of recidivism research in understanding criminal behavior

Robert Martinson believes that a new epoch in criminology research has arrived, “the essence of which is a change in the dependent variable from recidivism to crime rate (combined with cost).” However, no study or evaluation should be discounted simply because it measures recidivism as a parameter of criminality. If adhered to, the Martinson thesis could retard scientific inquiry and clog the normal processes of data generation and theory construction. The new epoch of criminology research will involve the construction and testing of higher-order theories derived from more basic facts and empirical generalizations.     

Competing Risks, Persistence, and Desistance in Analyzing Recidivism

A statistical procedure is developed to analyze recidivism in samples whichare subject to the presence of desisters and to multiple modes ofreconviction. This allows for a more accurate study of individuals'transition and hazard in the type and timing of offenses following aspecific type of conviction. The use of a nonparametric approach forinvestigating failure in the presence of other acting causes is shown;initial estimators of the probabilities of reconviction for different typesof offenses are obtained, and the method can be used both to display thedata and to choose an appropriate parametric family for the survivaltimes. An exponential mixture model for competing risks is presented insuch a way that it allows us to adjust for concomitant variables and toassess their effects on the probabilities both of reconviction forpredetermined types of offenses and desistance and of the hazards ofreconviction; a method for assessing calibration of predicted survivalprobabilities is suggested. A 21-year follow-up of persons convicted ofindecent assault on a female in 1973 illustrates the methods; we find ahigh probability of sexual reconviction for individuals with previoussexual convictions and evidence of diversity and a raised hazard ofreconviction for young chronic offenders.     

Instrumental variables methods in experimental criminological research: what, why and how

Quantitative criminology focuses on straightforward causal questions that are ideally addressed with randomized experiments. In practice, however, traditional randomized trials are difficult to implement in the untidy world of criminal justice. Even when randomized trials are implemented, not everyone is treated as intended and some control subjects may obtain experimental services. Treatments may also be more complicated than a simple yes/no coding can capture. This paper argues that the instrumental variables methods (IV) used by economists to solve omitted variables bias problems in observational studies also solve the major statistical problems that arise in imperfect criminological experiments. In general, IV methods estimate causal effects on subjects who comply with a randomly assigned treatment. The use of IV in criminology is illustrated through a re-analysis of the Minneapolis domestic violence experiment. The results point to substantial selection bias in estimates using treatment delivered as the causal variable, and IV estimation generates deterrent effects of arrest that are about one-third larger than the corresponding intention-to-treat effects.     

Profiles of Academic Activities and Careers: Does Gender Matter? An Analysis Based on French Life Scientist CVs

The aim of this paper is to analyze the factors that influence the length of time to promotion for male and female academics. Promotion is defined as elevation to a professorship. We examine the role of academic profiles, which are based not only on publications, but also include activities such as fund raising, consulting, teaching, and managerial appointments (dean of a department for instance). The paper examines the factors that speed up or slow down the progress of an academic career for males and females, respectively, to explore the “glass ceiling” effects. Survival and duration models are used to test whether the gender differential persists after controlling for observed and unobserved heterogeneity. The originality of this paper lies in the use of duration models to track sex differences in promotion criteria. It highlights that the different criteria of promotion for male and female academics: women have to demonstrate higher involvement in different networks in order to be promoted.     

统计学方法在微量物证数据处理中的应用

统计学方法在微量物证理化检验工作中受到越来越多的重视,采用有效的统计学方法对检验数据进行分析,不仅能够深度挖掘复杂数据背后信息,还能不断提高数据的应用价值和证据本身的证据价值。本文对目前微量物证领域常用统计学方法进行归纳总结,简要介绍了常用统计方法的应用,对各类方法优缺点进行说明。最后,对法庭科学工作者都比较感兴趣的贝叶斯统计与似然比方法在理化检验数据统计分析中的应用进行了介绍。     

Are Risky Youth Less Protectable as They Age? The Dynamics of Protection during Adolescence and Young Adulthood

Research on recidivism in criminal justice and desistance in criminology are not integrated. Yet, both fields are moving towards models that look at how positive elements in a person’s environment can impact a person’s behavior, conditional on different levels of risk. This study builds on this observation by applying interactional theory and the concept of Risk–Needs–Responsivity to theorize that both Needs and Responsivity will change over time in predictable ways. We then use a novel empirical approach with the Rochester Youth Development Study to show that even in late adolescence, individuals who are at risk for violence can be protected from future violence and risky behavior like gun carrying with positive events in their environment and personal life. In young adulthood, fewer people are still at risk for violence, and those who are at risk are harder to protect from future violence and gun carrying.     

Predicting pretrial misconduct with drug tests of arrestees: Evidence from eight settings

Identifying defendants at high risk of pretrial misconduct is a major problem for the judiciary. Currently, some have argued that testing arrestees for recent drug use is one way to distinguish between those who will and those who will not commit pretrial misconduct. The research reported here questions whether the incremental predictive power resulting from drug testing always improves predictions of pretrial misconduct. Using survival analysis to study time until rearrest and a probit model to analyze the occurrence of a failure to appear, we show that urine test results have no consistent power to predict pretrial misconduct after accounting for defendant's criminal records, community ties, and other factors commonly known by the court. These results are based on our analysis of eight data sets from different locales, time periods, and age groups.