INTERFACING INDICATORS AND CONSTRUCTS IN CRIMINOLOGICAL RESEARCH: A NOTE ON THE COMPARABILITY OF SELF-REPORT VIOLENCE DATA FOR RACE AND SEX GROUPS

This paper examines issues in using self-reports of involvement in violent acts to measure a latent construct (propensity toward violence). Three issues are addressed. First, the meaning of latent variables and the factor-analytic approach to modeling latent constructs are discussed. Second, the comparability of self-report violence measures across race and sex groups is assessed. Third, the consequences are discussed of estimating factor analytic models on pooled samples of data where the measurement component of the model does not apply equally for different subsets of the sample.     

THE DYNAMICS OF DELINQUENT PEERS AND DELINQUENT BEHAVIOR*

This analysis examines the dynamic reciprocal relationship between delinquent peer associations and delinquent behavior. It tests the hypothesis, derived from learning and interactional theories, that delinquent peers and delinquent behavior are reciprocally related—delinquent peer associations foster future delinquency, and delinquency increases the likelihood of associating with delinquent peers. It also tests the competing hypothesis, derived from control theories, that delinquent peers do not cause delinquency, but instead, the relationship is (1) spurious due to individual criminal propensity, (2) a result of the effects of delinquent behavior on future associations with delinquent peers, or (3) an artifact of problems of measuring delinquent peers. To test these propositions, we use data from the National Youth Survey and estimate a cross-lagged panel model that corrects for measurement error in indicators of delinquent peers and delinquent behavior. The model species a covariance structure model for ordinal measures. Parameters are estimated by (1) estimating a threshold model relating ordinal measures to continuous latent variables; (2) estimating a matrix of polychoric correlations relating observed variables, and (3) using an asymptotic distribution-free estimator to estimate structural parameters. The results suggest that delinquent peer associations and delinquent behavior are reciprocally related, but the effect of delinquency on peer associations is larger than that of peer associations on delinquency.     

SELF‐CONTROL THROUGH EMERGING ADULTHOOD: INSTABILITY,MULTIDIMENSIONALITY, AND CRIMINOLOGICAL SIGNIFICANCE

This study assesses self‐control theory's stability postulate. We advance research on self‐control stability in three ways. First, we extend the study of stability beyond high school, estimating GBTMs of self‐control from ages 10 to 25. Second, drawing on advances in developmental psychology and social neuroscience, especially the dual systems model of risk taking, we investigate whether two distinct personality traits—impulsivity and sensation seeking—often conflated in measures of self‐control, exhibit divergent developmental patterns. Finding that they do, we estimate multitrajectory models to identify latent classes of co‐occurring developmental patterns. We supplement GBTM stability analyses with hierarchical linear models and reliable variance estimates. Lastly, using fixed effects models, we explore whether the observed within‐individual changes are associated with changes in crime net of overall age trends. These ideas are tested using five waves of data from the Family and Community Health Study. Results suggest that self‐control is unstable, that distinct patterns of development exist for impulsivity and sensation seeking, and that these changes are uniquely consequential for crime. We conclude by comparing our findings with extant research and discussing the implications for self‐control theory.     

Growth curves of deviant behavior in early adolescence: A multilevel analysis

Multilevel growth curve models provide a means of analyzing individual differences in the growth of deviance, allow a number of theories to be integrated in a single model, and can help to unify research on deviant/delinquent/criminal careers at different stages of the life cycle. Building on the distinction between population heterogeneity and state dependence as alternative explanations of persistent individual differences in deviance (Heckman, 1981; Nagin and Paternoster, 1991), we show that models with two levels can be used to represent and analyze a variety of criminological theories. The first level (level 1) uses repeated measurements on individuals to estimate individual-level growth curves. The second level treats the level 1 growth curve parameters (e.g., slope, intercept) as outcome variables and uses time-invariant factors to explain variation in these parameters across individuals. We illustrate this approach by estimating a model of growth in deviance drawn from Gottfredson and Hirschi's deviant propensity theory. An innovative feature is the assumption that adolescents' expected growth curves of deviance follow a classical Pearl-Verhulst logistic growth model (Pearl, 1930). The results suggest that five risk factors—parental psychiatric problems, lack of parental support, living arrangements with zero or one parent in residence, low family income, and male gender—have strongly positive effects on deviant propensity. For example, adolescents with no supportive parents, and no other risk factors, have expected asymptotic levels of deviance (peak levels attained at about age 18) that are about twice as high as those of adolescents with no risk factors. Yet more than two-thirds of the individual-level variability in growth curves is unexplained by the five risk factors. This unobserved heterogeneity would remain hidden in analyses using conventional structural equations models and the same explanatory variables.     

A unified model of settlement and trial expenditures: The Priest–Klein model extended

Settlement and trial expenditures are crucially interrelated. The literature on settlement, however, takes no account of models of trial. In this paper, we develop a unified model of settlement and trial expenditures. We do this by discarding the usual assumption of settlement models that trial costs are constant across cases. Instead, we follow the literature on trial by permitting trial costs to vary with the legal merit of the plaintiff's case. Our approach can be used to extend standard models of settlement such as the well-known Priest–Klein model as well as models based on asymmetric information. As a demonstration, we extend the Priest–Klein model and generally overturn that model's canonical results. In particular, we show that even in a fully symmetric model, predicted win rates at trial can deviate substantially from 50 percent. Furthermore, win rates will vary in response to legal reforms that shift the decision standard.     

Financing Enterprises in the Czech Republic: Debt and Firm-specific Variables

This paper investigates the effects of firm specific variables on indebtedness, and thus financing decisions in the Czech Republic during the first years of transition. By estimating a Structural Equation Model with latent variables, the paper analyses the determinants of credit demand and supply in order to understand to what an extent they testify the transformation of both enterprises financing strategies and credit allocation policies and thus the emergence of a new financial discipline.     

THE FAMILIAL TRANSMISSION OF CRIMINAL CONVICTIONS*

This article examines the familial transmission of criminal convictions in families in the Cambridge Study in Delinquent Development. Our main analyses focus on the 344 families in the Cambridge Study with two or more children. Criminal convictions were highly familial because convictions in a parent increased the risk of convictions in a child. Correlations between siblings were stronger in same-sex siblings (.45 to .50) than in opposite-sex ones (.27). Sibling correlations did not vary by birth order. Convictions of mothers and father correlated .55. Parent-child correlations were about the same as within-generation correlations between siblings. LISREL models were used to assess whether the effect of parental convictions on child convictions was direct or mediated through the quality of the family environment (i.e., supervision. child rearing, and family size). The best fitting LISREL models suggested a direct effect of parental convictions on child convictions, without any mediation by family environment. These data on fill biological siblings, however, did not permit separate estimation of family environmental versus genetic effects. One environmental effect appeared, however—a socialization effect among siblings; in families with three sons, there appeared to be mutual influence of one sibling on another. Also, regression models based on the boys suggested that family environmental variables did add to parental criminality.     

BRIDGING CRIMINAL CAREERS,THEORY, AND POLICY THROUGH LATENT VARIABLE MODELS OF INDIVIDUAL OFFENDING*

Our goal is to build bridges between theoretical criminology, the study of criminal careers, and policy-relevant research. Insights from the criminal career and propensity positions lead us to seek (1) a comprehensive means of incorporating theoretical variables in research on criminal careers, (2) statistical models that yield meaningful projections relevant to public policy issues, and (3) methods for comparing findings for different measures of offending. We present a conceptual framework accomplishing this by applying the general linear model to the study of crime and criminal careers. This framework differentiates the elements of (1) a curvilinear function linking the scale of the linear model and the scale of the measure of offending, (2) a probabilistic relationship between a latent tendency to offend and the measure of offending, (3) a probability distribution of individual differences on the latent dimension, and (4) relationships among repeated observations for the same individual. We describe numerous versions of the general linear model that do not require special statistical expertise and are appropriate for the full range of measures of offending. We conclude by addressing strategies for comparing results across measures.     

Considering hierarchical models for research on inmate behavior: Predicting misconduct with multilevel data

Penologists recognize that both inmate- and prison-level characteristics are relevant to an understanding of individual inmates' behaviors; yet extant studies have focused only on unilevel models with either individual- or aggregate-level predictors and outcomes. To explore the potential of multilevel modeling for related research, we examine empirical relationships predicting the likelihood of inmate misconduct with individual-level (inmate) variables and aggregate levels of prison population crowding. The framework for the model borrows from both individual- and aggregate-level theories of informal social control. We examine three secondary data sets, using information common to each set. We compare results from hierarchical logistic models with those from stepwise pooled logistic regression models to see whether results differ significantly by method of estimation. The pooled models reveal inconsistency in the significance of inmate predictors (social demographics and criminal histories) across the three samples, and non-significant relationships involving prison crowding and an interaction between crowding and an inmate's age for all samples. By contrast, the hierarchical models reveal much more consistency in prediction (or a lack thereof) at either level across all three models, as well as significant aggregate-level main and interaction effects. The theoretical and methodological implications of these results are discussed.     

Optimal antitrust enforcement: information cost and deterrent effect

This study analyzes the optimal antitrust enforcement rule and, in doing so, presents a model that illuminates two important issues. First, it compares the per se legal and illegal judicial standards to the rule of reason judicial standard in terms of information costs and general social welfare. Second, it seeks to derive the optimal judicial standard that minimizes the problems of under- and over-deterrence. These two issues are closely related because the benefit of additional information can only be measured by its deterrent effects. In this respect, this work synthesizes economic models from decision theory and the public enforcement of law. Lastly, in addition to discussing the optimal information level, we derive the optimal permissiveness of the judicial standard, the optimal burden of proof, and the optimal punishment level. We also analyze how these policy variables are interrelated.     

Comparing NOMINATE and IDEAL: Points of Difference and Monte Carlo Tests

Empirical models of spatial voting allow us to infer legislators' locations in an abstract policy or ideological space using their roll‐call votes. Over the past 25 years, these models have provided new insights about the U.S. Congress, and legislative behavior more generally. There are now a number of alternative models, estimators, and software packages that researchers can use to recover latent issue or ideological spaces from voting data. These different tools usually produce substantively similar estimates, but important differences also arise. We investigated the sources of observed differences between two leading methods, NOMINATE and IDEAL. Using data from the 1994 to 1997 Supreme Court and the 109th Senate, we determined that while some observed differences in the estimates produced by each model stem from fundamental differences in the models' underlying behavioral assumptions, others arise from arbitrary differences in implementation. Our Monte Carlo experiments revealed that neither model has a clear advantage over the other in the recovery of legislator locations or roll‐call midpoints in either large or small legislatures.     

Modeling the Deviant Y in Criminology: An Examination of the Assumptions of Censored Normal Regression and Potential Alternatives

Many dependent variables of criminological interest have censored distributions. Investigations that use such variables increasingly have turned to the Tobit model, a censored regression technique that is specified based on a latent dependent variable. When used under suitable circumstances, this model provides appropriate estimates. This paper discusses key assumptions of the Tobit model. It then highlights the risk of violating these assumptions and reviews alternative flexible parametric and semiparametric modeling techniques, currently used sparingly in criminology, which researchers may find helpful when assumptions regarding the error terms are untenable. By using an empirical example focused on sentencing outcomes and comparing estimates across analytic methods, this study illustrates the potential utility of simultaneously estimating the Tobit model along with some alternatives.

Analyzing block randomized studies: the example of the Jersey City drug market analysis experiment

Objectives

While block randomized designs have become more common in place-based policing studies, there has been relatively little discussion of the assumptions employed and their implications for statistical analysis. Our paper seeks to illustrate these assumptions, and controversy regarding statistical approaches, in the context of one of the first block randomized studies in criminal justice—the Jersey City Drug Market Analysis Project (DMAP).

Methods

Using DMAP data, we show that there are multiple approaches that can be used in analyzing block randomized designs, and that those approaches will yield differing estimates of statistical significance. We develop outcomes using both models with and without interaction, and utilizing both Type I and Type III sums-of-squares approaches. We also examine the impacts of using randomization inference, an approach for estimating p values not based on approximations using normal distribution theory, to adjust for possible small N biases in estimating standard errors.

Results

The assumptions used for identifying the analytic approach produce a comparatively wide range of p values for the main DMAP program impacts on hot spots. Nonetheless, the overall conclusions drawn from our re-analysis remain consistent with the original analyses, albeit with more caution. Results were similar to the original analyses under different specifications supporting the identification of diffusion of benefits effects to nearby areas.

Conclusions

The major contribution of our article is to clarify statistical modeling in unbalanced block randomized studies. The introduction of blocking adds complexity to the models that are estimated, and care must be taken when including interaction effects in models, whether they are ANOVA models or regression models. Researchers need to recognize this complexity and provide transparent and alternative estimates of study outcomes.

     

LEARNING TO BE BAD: ADVERSE SOCIAL CONDITIONS,SOCIAL SCHEMAS,AND CRIME*

In this article, we develop and test a new approach to explain the link between social factors and individual offending. We argue that seemingly disparate family, peer, and community conditions lead to crime because the lessons communicated by these events are similar and promote social schemas involving a hostile view of people and relationships, a preference for immediate rewards, and a cynical view of conventional norms. Furthermore, we posit that these three schemas are interconnected and combine to form a criminogenic knowledge structure that results in situational interpretations legitimating criminal behavior. Structural equation modeling with a sample of roughly 700 African American teens provided strong support for the model. The findings indicated that persistent exposure to adverse conditions such as community crime, discrimination, harsh parenting, deviant peers, and low neighborhood collective efficacy increased commitment to the three social schemas. The three schemas were highly intercorrelated and combined to form a latent construct that strongly predicted increases in crime. Furthermore, in large measure, the effect of the various adverse conditions on increases in crime was indirect through their impact on this latent construct. We discuss the extent to which the social‐schematic model presented in this article might be used to integrate concepts and findings from several major theories of criminal behavior.     

A Latent Variable Analysis of General Strain Theory

Strain theory has returned to the forefront of criminological theory and research, due primarily to the general strain model developed by Robert Agnew. Agnew posits that a broad range of negative social relations comprises strain and that these straining mechanisms lead to delinquent behavior and other maladaptive functioning. Moreover, strain has its strongest effect on delinquency when certain coping strategies are attenuated or when delinquent peers reinforce perceptions of strain. Although several studies have now shown the utility of general strain theory as an explanation of delinquency, they have relied mainly on cross-sectional effects or two-wave panel designs using methods that fail to consider measurement error or autocorrelated errors. In this study we extend these analyses by estimating a latent variable structural equation model that examines the effects of strain—operationalized as negative life events—on conventional attachment and delinquency over a 3-year period. Furthermore, we directly assess Agnew's coping strategies hypotheses by stratifying the models by self-efficacy, self-esteem, and peer delinquency. The results indicate that significant longitudinal effects of strain on delinquency emerge during year 3 but that these effects are not conditioned by self-efficacy or self-esteem. Changes in strain also affect changes in delinquency, but only among those who report no delinquent peers. We do find, however, that over the initial 2 years strain has a negative effect on delinquency among those high in self-efficacy, self-esteem, or delinquent peers. The findings are discussed in terms of Agnew's theory.     

A New Approach of Juvenile Age Estimation using Measurements of the Ilium and Multivariate Adaptive Regression Splines (MARS) Models for Better Age Prediction

Juvenile age estimation methods used in forensic anthropology generally lack methodological consistency and/or statistical validity. Considering this, a standard approach using nonparametric Multivariate Adaptive Regression Splines (MARS) models were tested to predict age from iliac biometric variables of male and female juveniles from Marseilles, France, aged 0–12 years. Models using unidimensional (length and width) and bidimensional iliac data (module and surface) were constructed on a training sample of 176 individuals and validated on an independent test sample of 68 individuals. Results show that MARS prediction models using iliac width, module and area give overall better and statistically valid age estimates. These models integrate punctual nonlinearities of the relationship between age and osteometric variables. By constructing valid prediction intervals whose size increases with age, MARS models take into account the normal increase of individual variability. MARS models can qualify as a practical and standardized approach for juvenile age estimation.     

Developmental explanations of delinquent behavior among siblings: Common factor vs. transmission mechanisms

We compare the statistical fit of two developmental explanations of delinquent behavior using longitudinal sibling data. The transmission effects model relates future delinquency to prior delinquency, delayed sibling effects, and unique environment. The common-factor effects model adds to these influences a latent variable representing persistent causes shared by siblings. These two models were fit to longitudinal data on 470 sibling pairs interviewed on three occasions. The common-factor effects model fit the data more closely than the transmission model. Nonmutually exclusive interpretations of the common effects model include (a) personality dispositions and (b) unchanging aspects of the social environment such as concurrent sibling effects and siblings' common friends.     

A Comparison of Methods for Analyzing Criminological Panel Data

We demonstrate that fixed- and random-effects models for pooled cross-sectional and time series data, and latent growth curve models for panel data are special cases of a more general model. We compare the estimates obtained from each type of model for a data set consisting of homicide rates and a vector of explanatory variables for 400 US counties over a 15-year period. Most, but not all, estimates are similar in the two models. We identify circumstances under which one approach may be advantageous to the other.

GENETIC AND ENVIRONMENTAL COMPONENTS OF ANTISOCIAL BEHAVIOR: A STUDY OF 265 TWIN PAIRS*

This study investigates the common-family environmental (CE), within-family environmental (WE), and hereditary (H) components of antisocial behavior and its correlates using a twin study design. The subjects are 265 adolescent twin pairs who reported in a mail survey on their antisocial behavior, deceitfulness, parental rejection (as perceived), anger, impulsivity, and value placed on school achievement. These six variables are intercorrelated in two ways: between-families (twin pairs' sums) and within-families (twin pairs' differences). The former covariance structure captures the twins' resemblances: the latter, the twins' differences in behavior. LISREL is used to model the observed relationships using structural equations containing CE, WE, and H factors. The best-fitting model requires only H and WE factors to explain the variables' relationships. Within this population, delinquent behavior is unaffected by CE influences such as social class, child rearing styles, parental attitudes, parental religion, and other factors equally affecting the twins. The principal genetic correlates of delinquency appear to be deceitfullness and temperamental traits.     

CONTRIBUTIONS OF FAMILIES AND PEERS TO DELINQUENCY*

A model is presented that explains the contribution of parents and peers to adolescent delinquent behavior. It is hypothesized that during adolescence a failure in Parent Monitoring and deficits in Social Skills increase the likelihood that a youngster associates with Deviant Peers. Poor Parent Monitoring Deviant Peers, and low levels of Academic Skills are hypothesized to contribute directly to an adolescent's engagement in delinquent behavior, The present model was tested on a sample of 136 seventh and tenth grade male adolescents by using the structural modeling approach in the LISREL IV analysis program (Jöreskog & Sörbom, 1978). The major revision of the model was that the correlation between Academic Skills and Parent Monitoring was set to zero. A chi square goodness-of-fit test for the revised model showed adequate agreement between the hypothesized model and the observed covariance structure of the data. It was suggested that a number of requirements be completed before accepting the above model: (1) replication of this model on a new set of data, (2) longitudinal analyses showing the hypothesized relations through time, and (3) experimental testing by manipulation of one or more independent variables, as is possible in clinical intervention studies.