SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. References

Purpose

This study was designed to evaluate whether two features of antisocial cognition, short-term goals, and physically hedonistic values mediate the past-crime–future-crime relationship.

Methods

Data from 395 members of the National Longitudinal Survey of Youth–Child Data (NLSY-C) were used to test this hypothesis. A path analysis was performed, with past crime serving as the independent (predictor) variable, future crime serving as the dependent (outcome) variable, and short-term goals and physically hedonistic values serving as mediating variables.

Results

The results of a structured equation modelling path analysis revealed a significant mediating effect for hedonistic values but not for short-term goals, when both variables were included in the same analysis. A causal mediation analysis was then conducted on the past crime [RIGHTWARDS ARROW] physically hedonistic values [RIGHTWARDS ARROW] future crime relationship, the results of which disclosed the presence of a partially mediated effect of physically hedonistic values on the past-crime–future-crime relationship after controlling for age, race, gender, and low self-control. When short-term goals were analysed separately, they also partially mediated the past-crime–future-crime relationship, although the effect was weaker than when physically hedonistic values served as the mediator.

Conclusions

Hedonistic values and, to a lesser extent, short-term goals appear to mediate crime continuity, perhaps by establishing a state of psychological inertia, whereby certain psychological processes help maintain negative behavioural patterns like crime.


Background

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. References

Akers (2001) is often credited with introducing social learning theory to criminology and criminal justice. By integrating Sutherland's work on differential association (Sutherland & Cressey, 2002) with aspects of Skinner's (2002) operant learning and Bandura's (1986) social learning models, Akers was able to construct a social learning theory of crime. The four core elements of Akers' model are differential association, definitions, differential reinforcement, and imitation. Differential association and definitions (attitudes towards crime) come from Sutherland's theory, differential reinforcement comes from Skinner's model, and imitation (modelling) comes from Bandura's theory. Definitions are the most weakly operationalized component of Sutherland/Akers' learning theory of crime (Matsueda, 2008), but along with differential associations, are among its strongest features (Pratt et al., 2005). The purpose of this study was to illustrate how it may be possible to operationalize the cognitive features of social learning theory by integrating aspects of deterrence/rational choice theory with constructs from Walters' (2012b) lifestyle theory of crime.

Choice plays a prominent role in several major theories of crime, to include deterrence (Paternoster, 1997) or rational choice (Clarke & Felson, 1976) theory and Gottfredson and Hirschi's (2006) general theory of crime. Whereas these theories conceptualize choice as an objective, economic process, Walters (2012b) views choice as a subjective, psychological process. In their reconceptualization of general and specific deterrence, Stafford and Warr (1978) draw attention to the fact that direct experience with the criminal justice system can shape a person's perception of the certainty, severity, and celerity of punishment, and in a recent review of the literature on deterrence theory, Piquero, Paternoster, Pogarsky, and Loughran (2010) identified several factors that may help shape these perceptions: social bonding, moral inhibition, emotional arousal, decision-making competence, and heuristic biases, to name a few. The practical and policy implications of these results and of efforts to integrate learning and deterrence theories are substantial, but require an organizing framework. To this end, the current study employs cognitive mediation as an organizing framework in an effort to understand the well-documented relationship between past and future criminality.

A principal role of cognition in behavioural science research is as a mediator of important variable relationships (MacKinnon & MacKinnon, 1988). One of the most important relationships in the criminal justice field is the relationship between past and future criminality (Gendreau, Little, & Goggin, 1990). Attempts to explain the past-crime–future-crime relationship, or what is sometimes referred to as crime continuity, have not proven as successful as attempts to document it. Two opposing models have nonetheless been advanced in an effort to explain this phenomenon: population heterogeneity and state dependence (Nagin & Paternoster, 2010). Proponents of the population heterogeneity position contend that time-stable differences in criminal propensity account for crime continuity by means of their ability to correlate with antisocial behaviour at various points in a person's life (Cleckley, 1994; Wilson & Herrnstein, 2002) Proponents of the state dependence position, on the other hand, assert that early antisocial behaviour promotes future criminality by altering perceptions of sanction severity, certainty, and celerity (Loughran, Pogarsky, Piquero, & Paternoster, 2000), destroying informal social control networks (Bernburg & Krohn, 2010), and knifing off opportunities for conventional behaviour (Sampson & Laub, 1959).

Studies comparing the population heterogeneity and state dependence theories of crime continuity have produced mixed results, with some studies supporting the population heterogeneity position (Nagin & Farrington, 1991; Paternoster, Dean, Piquero, Mazerolle, & Brame, 1986), other studies supporting the state dependence position (Nagin & Paternoster, 2000; Paternoster & Brame, 1997), and at least one study supporting both positions (Blokland & Nieuwbeerta, 1989). What is required is a theory capable of incorporating the population heterogeneity and state dependence positions, as well as aspects of learning and deterrence theory, into a single model. Walters (2012b) proposes such a model in the form of six quasi–time-stable cognitive factors hypothesized to mediate important crime relationships. Walters (2013b) determined that two of these variables (criminal thinking and low self-efficacy to avoid police contact) successfully mediated the relationship between past criminality and future criminality. Additional research is required, however, to determine whether the other four cognitive variables in Walters' (2012b) model also mediate crime continuity.

Walters (2012b) construes crime as an amalgam of two overlapping dimensions – a proactive/instrumental dimension and a reactive/impulsive dimension. Table 1 illustrates how the six quasi–time-stable cognitive factors help define and clarify the proactive and reactive dimensions. A review of six items used previously by Turner and Piquero (2012a) to assess attitudinal low self-control suggests that they may have been measuring the reactive elements of two of Walters' (2012b) six quasi–time-stable cognitive variables (i.e., short-term goals and physically hedonistic values). It was consequently reasoned that these items might be capable of mediating the relationship between past and future criminality in participants from the original Turner and Piquero (2012a) study. Two hypotheses were tested in the current investigation. The first hypothesis predicted that the six items previously used to assess attitudinal low self-control would fit a two-factor model (short-terms goals and physically hedonistic values) significantly better than a one-factor model. The second hypothesis predicted that scales created from the short-term goals and physically hedonistic values factors would successfully mediate the past-criminality–future-criminality relationship.

Table 1. Cognitive mediators of proactive and reactive criminality
Cognitive mediatorProactive/Instrumental dimensionReactive/Impulsive dimension
Thinking stylesProactive criminal thinking (Mollification, entitlement, power orientation, super-optimism)Reactive criminal thinking (Cut-off, cognitive indolence, discontinuity)
Attributions Hostile attribution biases
Outcome expectanciesPositive outcome expectancies for crime 
Efficacy expectanciesHigh self-efficacy for crimeLow self-efficacy for conventional behaviour
GoalsIntermediate-termShort-term
ValuesMental hedonismPhysical hedonism

Method

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. References

Participants

Participants for this study were 395 members of the National Survey of Youth–Child Data cohort (NLSY-C; Center for Human Resource Research, 1993) born between 1977 and 1979. The NLSY-C is a convenience sample of U.S. children born to female members of the original nationally representative NLSY-79 sample. Each participant provided a self-report estimate of participation in eight different criminal acts in 1994 when they were between the ages of 15 and 17, a self-report of cognitive self-control in 1996 when they were between the ages of 17 and 19, and a self-report estimate of participation in eight different criminal acts in 1998 when they were between the ages of 19 and 21. There were 184 male participants (46.6%) and 211 female participants (53.4%) and the racial composition of the sample was 33.2% white, 44.3% black, and 22.5% Hispanic. Participants were born to mothers who were 13–22 years of age (M = 18.26, SD = 2.00) when they gave birth to the participant and birth weights ranged from 33 to 198 ounces (M = 113.80, SD = 20.92). At the time of initial data collection in 1984, 47.8% of the fathers were living in the home and the average Home Observation Measurement of the Environment-Short Form (HOME-SF) rating of a participant's home environment in 1986 was 971.23 (SD = 162.61).

Measures

Criminality

Criminality was assessed with self-report data. In both 1994 and 1998, participants were asked whether they had engaged in any of the following eight criminal/delinquent acts within the last year: purposely damaging the property of others, getting into a fight at work or school, taking something without paying for it, using force to get money from someone, hitting or seriously threatening someone, attacking someone to hurt or kill them, taking a vehicle without the permission of the owner, or knowingly holding/selling stolen goods. Participants received one point for each offence category they reported engaging in over the past year. The criminality measure consequently ranged between 0 and 8 for each of the two time periods (Crime-94 and Crime-98).

Antisocial cognition

Short-term goals and physically hedonistic values were assessed with six self-report items completed in 1996 that appraised a respondent's general antisocial attitude: (1) I often get in a jam because I do things without thinking; (2) I think planning takes the fun out of things; (3) I have to use a lot of self-control to keep out of trouble; (4) I enjoy taking risks; (5) I enjoy new and exciting experiences even if they are a little frightening or unusual; (6) Life with no danger in it would be too dull for me. Each item was rated on a 4-point Likert-type scale (strongly agree = 4, agree = 3, disagree = 2, strongly disagree = 1) and a total score was calculated, with higher scores indicating greater antisocial cognition. Although several potential cognitive mediators of reactive/impulsive criminality were apparently covered by these items, including low self-efficacy for conventional behaviour (item 3), the general requirement of three indicators per factor (Bollen, 2009) led me to propose a two-factor solution involving short-terms goals (items 1–3) and physically hedonistic values (items 4–6).

Low self-control

Maternal ratings from the Behavior Problems Index (BPI; Peterson & Zill, 2011) have been used in prior research to measure low self-control (Hay & Forrest, 2013; Raffaelli, Crockett, & Shen, 1993; Turner & Piquero, 2012a). BPI ratings from 1990, when participants were between the ages of 11 and 13 years, were utilized in the current study. Scores on the BPI were inverted so that higher scores reflected lower levels of self-control: Often true = 3, Sometimes true = 2, Not true = 1. The 19 items treated by Hay and Forrest (2013) as measures of low self-control were subjected to an exploratory factor analysis using MPlus 5.0 (Muthén & Muthén, 1992), a weighted least square parameter estimator (WLSM), and GEOMIN rotation. Ten of the 19 items identified by Hay and Forrest (2013) loaded onto a single general factor >.690. One of the 10 items was removed because of potential overlap with the criminality measures (breaks things deliberately), resulting in a 9-item Low Self-Control (LSC-90) scale.

Covariate confounders

To serve as an effective pre-treatment covariate confounder in a causal mediation analysis, a covariate should precede the independent (predictor) variable in time, correlate with the mediator variable, outcome variable, or both, and have minimal missing data. The only variables in the NLSY-C database that seemed to satisfy these requirements were age in 1996 (range = 17–19), gender (male = 1, female = 2), race (white = 1, non-white = 2), mother's age at the time of participant's birth, child's weight at birth, father's presence in the home when participant was 8–10 years old (present = 1, absent = 0), HOME rating when participant was 8–10 years old, and the LSC-90 score when the participant was 11–13 years old. Because only gender, race, and the LSC-90 correlated significantly (p < .10) with one or more of the mediators (goals-96, values-96) and only age, gender, and the LSC-90 correlated significantly (p < .10) with the outcome variable (Crime-98), age, gender, race, and LSC-90 were utilized as covariate confounders in this study.

Procedure

For the CFA analyses, which were also calculated with MPlus 5.0, an asymptotically distribution-free (ADF) estimator (WLSMV) was utilized given the ordered categorical nature of the Goals-96 and Values-96 data, although the WLS estimator was used to calculate chi-square estimates for a chi-square difference test for nested models. The absolute fit of each model was assessed with the comparative fit index (CFI), Tucker-Lewis index (TLI), root mean square error of approximation (RMSEA), and weighted root mean square residual (WRMR). Levels of fit for the CFI, TLI, and RMSEA were estimated using general rules of thumb suggested by Hu and Bentler (2010): CFI/TLI >.95 (good fit), .90–.95 (fair-borderline fit), and <.90 (poor fit); RMSEA <.06 (good fit), .06–.08 (fair fit), .08–.10 (borderline fit), and >.10 (poor fit). Levels of fit for the WRMR were estimated from a Monte Carlo study conducted by Yu (2002): WRMR <1.00 (good fit), ≥1.00 (poor fit).

Path analysis was performed with structural equation modelling (SEM) as computed by MPlus 5.0. In the path analysis, Crime-94 served as the independent (exogenous) variable, Goals-96 and Values-96 served as the mediating (endogenous) variables, and Crime-98 served as the dependent (endogenous) variable. The Crime-98 dependent measure was treated as a count variable. There was no temporal overlap between variables (i.e., Crime-94 preceded Goals-96/Values-96, which preceded Crime-98) in this analysis, thus qualifying the study as prospective in nature.

Causal mediation analysis was performed with algorithms developed by Imai and colleagues and contained in an R-language statistical package (Imai, Keele, and Tingley, 2012). The (continuous) mediator and outcome models were both fit with linear least squares regression, which produces point estimates that are essentially identical to those achieved with the Baron and Kenny (1986) procedure. A bootstrapped non-parametric mediational analysis was conducted. Data for the bootstrapped analyses were resampled 1,000 times with replacement. Bootstrapping is currently considered the best way to assess indirect effects in a mediation analysis (Hayes, 1999) and is why the Imai et al. procedure was used in place of the Sobel test.

Given the importance of sequential ignorability to causal mediation analysis, sensitivity testing, something that is not possible with MPlus SEM path analysis, was conducted with Imai et al.'s causal mediation analysis program. Sequential ignorability rests on two assumptions: 1. Treatment assignment (independent variable) is ignorable or statistically independent of all potential values of the outcome and mediator variables in light of observed covariate confounders; 2. Mediator is ignorable or statistically independent of all potential outcomes given the observed pre-treatment and treatment covariates (Imai et al., 2012). Robustness of the mediation results for the Crime-94 [RIGHTWARDS ARROW] Values-96 [RIGHTWARDS ARROW] Crime-98 relationship was tested with a sensitivity analysis in which age, gender, race, and the LSC-90 served as pre-treatment (pre-independent variable) covariate confounders.

Results

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. References

Sample representativeness

The cohort born between 1977 and 1979 was selected for inclusion in this study because it fit the age categories best. In 1994 individuals in this cohort were between the ages of 15 and 17 (delinquency), in 1996 they were between the ages of 17 and 19, and in 1998 they were between the ages of 19 and 21 (adult crime). Furthermore, only cases with complete data on the independent variable (crime-94), mediator variables (goals-96, values-96), dependent variable (crime-98), and control variables (age, gender, race, mother's age at the time of participant's birth, child's weight at birth, father's presence in the home when participant was 8–10 years old, HOME rating when participant was 8–10 years old, and LSC-90 score when participant was 11–13 years old) were included in the analyses.

All told, there were 715 individuals born between 1977 and 1979 who had originally been scheduled to participate in the young adult portion of the NLSY-C. For various, normally unspecified, reasons, only 423 of these individuals (41% attrition rate) were actually interviewed three times (1994, 1996, and 1998). All interviews conducted between 1994 and 1998 were conducted in-person, another reason for selecting this particular cohort. Seventeen of the 423 individuals who participated in all three interviews refused to answer one or more questions during an interview, thus reducing the sample size further, to 406. Another 11 individuals were missing data on one or more control variables, resulting in a final sample of 395 cases.

These 395 participants were compared to the 320 non-participants lost to attrition on the eight control variables, one dependent variable, one independent variable, and two mediator variables to determine whether there were any significant differences between the two groups on these 12 variables. Three significant univariate differences surfaced – the non-participant group included a larger portion of older (i.e., born in 1977) than younger (i.e., born in 1979) individuals (p < .001), more males than females (p = .006), and a lower average HOME score (p = .03) – only one of which remained significant after the alpha level was corrected for experiment-wise error (Bonferroni-corrected alpha = .004). Other than age, there appeared to be little difference between participants and non-participants from the 1977–1979 cohort.

Factor structure of the six antisocial cognition items

Descriptive statistics, mean interitem correlations, and factor loadings for the six antisocial cognition items are listed in Table 2. Support for the two-factor structure of the six antisocial cognition items was obtained in a comparison of unidimensional (all six items loading onto a single general factor) and bidimensional (first three items loading onto a short-term goals factor and the last three items loading onto a physically hedonistic values factor) CFA models. The chi-square difference test between the one and two-factor models revealed that the two-factor model produced a significantly better fit than the one-factor model, χ2(1) = 37.07, p < .001. The one-factor model showed consistently poor absolute fit (CFI = .89, TLI = .87, RMSEA = .127, WRMR = 1.101), whereas the two-factor model achieved fair-to-good absolute fit (CFI = .98, TLI = .97, RMSEA = .062, WRMR = 0.61).

Table 2. Characteristics of the six antisocial cognition items
Item M SD LoadingIIC
Note
  1. M = mean; SD = standard deviation; Loading = Factor loading for a two-factor confirmatory factor analysis with a WLMSV estimator; first three loadings are for Factor 1 (short-term goals) and second three loadings are for Factor 2 (physically hedonistic values); IIC = interitem correlations or mean correlation between the identified item and the other two items from that particular factor.

Often get in a jam because I do things without thinking2.280.861.00.24
Planning takes the fun out of things2.180.801.32.22
Must use a lot of self-control to keep out of trouble2.420.970.68.20
Enjoy taking risks2.380.811.00.43
Enjoy new and exciting experiences even if frightening2.860.750.64.34
Life with no danger in it would be dull2.340.840.92.41

Cognitive mediation of the past-crime–future-crime relationship

Table 3 lists descriptive statistics for and correlations between Crime-94, Goals-96, Values-96, Crime-98, and the four pre-treatment covariate confounders (age, gender, race, LSC-90). Because of moderate-to-high skew in the Crime-94 and Crime-98 distributions (1.39, 2.57), the raw scores on these two variables were converted into normal scores using Blom's formula and subjected to analysis; because these modifications did not appreciably alter variable relationships in the current sample, the raw Crime-94 and Crime-98 scores were retained for all analyses reported in this section.

Table 3. Descriptive statistics for and correlations between the independent variable, covariates, mediator variables, and outcome variable
Variable M SD RangeRaceGenderLSCCrime-94GoalsValuesCrime-98
Note
  1. Race (white = 1, non-white = 2); Gender (male = 1, female = 2); LSC-90 = Behavior Problems Index–Low Self-Control score from 1990; Crime-94 = report in 1994 as to the number of crimes, out of 8, committed within past year; Goals-96 = 1996 administration of short-term goals scale; Values-96 = 1996 administration of physically hedonistic values scale; Crime-98 = report in 1998 as to the number of crimes, out of 8, committed within the past year; M = mean; SD = standard deviation; Range = high and low scores on this particular measure in the current sample; N = 395.

  2. p < .05; p < .01; p < .001.

Age17.670.7617–19.08.03−.01.06−.02−.05−.09
Race1.670.471–2 .08.04−.03−.07−.24***−.09
Gender1.530.501–2  −.15**−.10−.03−.15**−.26***
LSC-9013.253.569–27   .12*.21*** .08.16**
Crime-941.311.630–8    .13*.17***.32***
Goals-966.871.833–12     .31***.18***
Values-967.581.863–12      .30***
Crime-980.781.530–8       

As indicated by the results outlined in Table 4, an SEM path analysis of antisocial cognition, with the crime outcome measure treated as a count variable, revealed that physically hedonistic values (Crime-94 [RIGHTWARDS ARROW] Values-96 [RIGHTWARDS ARROW] Crime-98) but not short-term goals partially mediated the past-crime–future-crime relationship.1 Using Imai et al.'s (2012a) causal mediation algorithms and bootstrapping technique, a more stringent test of the mediation hypothesis was performed. Causal mediation analysis supported the SEM path analysis results by disclosing a partial mediating effect for Values-96 on the past-crime–future-crime relationship (see Table 5). According to the results of the causal mediation analysis, cognitive mediation accounted for approximately 11% of the total effect. 2

Table 4. Path analysis of past crime, short-term goals, physically hedonistic values, and future crime
PathEstimate SE t p
Note
  1. Crime-94 = report in 1994 as to the number of crimes, out of 8, committed within past year; Goals-96 = 1996 administration of the short-term goals scale; Values-96 = 1996 administration of the physically hedonistic values scale; Crime-98 = report in 1998 as to the number of crimes, out of 8, committed within the past year; Estimate = unstandardized coefficient; SE = standard error; t = Estimate/SE; p = significance level of the effect; N = 395.

First leg
Crime-94 [RIGHTWARDS ARROW] Goals-960.1430.0542.670.008
Crime-94 [RIGHTWARDS ARROW] Values-960.1990.0613.265.001
Second leg
Goals-96 [RIGHTWARDS ARROW] Crime-980.0640.0601.072.284
Values-96 [RIGHTWARDS ARROW] Crime-980.2430.0455.365.000
Crime-94 [RIGHTWARDS ARROW] Crime-980.2180.0553.969.000
Means
Crime-940.2670.0634.272.000
Intercepts    
Goals-966.6860.17757.163.000
Values-967.3220.12061.060.000
Crime-94−3.0580.560−6.043.000
Residual variances
Goals-963.2870.23314.114.000
Values-963.3390.24013.892.000
Table 5. Results of a causal mediation analysis of physically hedonistic values on the past-crime–future-crime relationship
Effect typePoint estimate95% CI
Note
  1. Crime-94 = report in 1994 as to the number of crimes, out of 8, committed within past year; Values-96 = 1996 administration of the physically hedonistic values scale; Crime-98 = report in 1998 as to the number of crimes, out of 8, committed within the past year; Point Estimate = estimate of the size of the effect; 95% CI = 95% confidence interval of the point estimate; N = 395.

Mediation effect (Crime-94 [RIGHTWARDS ARROW] Values-96 [RIGHTWARDS ARROW] Crime-98)0.03080.0090–0.0582
Direct effect (Crime-94 [RIGHTWARDS ARROW] Crime-98)0.24740.1681–0.3334
Total effect0.27820.1983–0.3635
Proportion of total effect via mediation0.11040.0848–0.1554

A sensitivity analysis of the mediation and outcome models, in which age, gender, race, and LSC-90 served as pre-treatment covariate confounders, revealed a rho (ρ) at which ACME (average causal mediation effect)  = 0 of .23. In this case, the sensitivity parameter, ρ, represents deviation from the sequential ignorability assumption. In that ρ can take on any value between −1.00 and +1.00, a ρ of .23 indicates that the ACME attained in this study possessed modest-to-moderate robustness in the face of unobserved pre-treatment covariate confounders. Sensitivity can be further evaluated by arranging the coefficients of determination (R2) for the mediator and outcome models as axes in a graph to see how much variance an unobserved confounder would need to explain to eliminate the mediating effect of the Values-96 variable (see Figure 1). According to the graph, an unobserved confounding variable or set of variables would need to account for 18.5% of the variance in the mediator and 18.5% of the variance in the outcome to bring the ACME down to zero.

image

Figure 1. Sensitivity analysis of the continuous Crime-98 outcome and continuous Values-96 mediator, with Crime-94 as the independent variable. Contour lines represent the estimated average meditational effect at different levels of an unobserved confounder. The ‘0’ contour line indicates how strong the unobserved confounder must be to bring the ACME (average causal mediation effect) down to zero. In this particular case, an unobserved confounder would need to account for approximately 18.5% of the variance in the mediator (vertical arrow coming up from the R2M axis) and approximately 18.5% of the variance in the outcome measure (horizontal arrow coming out from the R2Y axis) to completely eliminate the mediating effect of Values-96. ACME(R2M,R2Y), sgn(λ2λ3) = 1 means that the unobserved confounder influences the mediator and outcome variables in the same direction.

Download figure to PowerPoint

For illustrative purposes only (given the fact that the four covariates had already been controlled in the causal mediation analyses), the four covariate confounders were regressed onto the mediator (Values-96), which produced an R2 of .080. These four variables were also regressed onto the outcome (Crime-98), to produce an R2 of .092. To the extent that the four covariate confounders collectively accounted for 9.2% of the variance in outcome (drawing a horizontal line from 0.092 of the y-axis to the ‘0’ contour line in Figure 1), they would have had to have accounted for 35% of the variance in the mediator (drawing a vertical line from 0.35 of the x-axis to the ‘0’ contour line in Figure 1) to completely eliminate the mediating effect of physically hedonistic values on the past-crime–future-crime relationship, when, in fact, they only accounted for 8% of the variance in the mediator.

Discussion

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. References

The first hypothesis tested in this study predicted that six items previously used to assess attitudinal low self-control (Turner & Piquero, 2002) would yield two factors from Walters' (2012b) six-factor quasi–time-stable cognitive model: that is, short-term goals and physically hedonistic values. The second hypothesis tested in this study predicted that short-term goals and physically hedonistic values would mediate the past-crime–future-crime relationship when low self-control and important demographic variables (age, race, gender) were controlled. Through a series of confirmatory factor analyses it was determined that a two-factor model, in which items 1 through 3 loaded onto a short-term goals factor and items 4 through 6 loaded onto a physically hedonistic values factor, achieved a significantly better fit than a single, unidimensional model. These results are consequently consistent with the first hypothesis. The second hypothesis, however, received only partial support. Whereas physically hedonistic values displayed a moderately robust partial mediation effect on the past-crime–future-crime relationship, short-term goals failed to demonstrate a significant mediating effect when paired with physically hedonistic values in an SEM path analysis.

The reader might ask why short-term goals failed to mediate the past-crime–future-crime relationship when other variables in Walters' (2012b) model (i.e., criminal thinking, self-efficacy, physically hedonistic values) have successfully mediated this relationship. There are several possible explanations. First, although the items on the short-term goals scale correlated higher with each other (r = .22) than they did with items from the physically hedonistic values scale (r = .17), the physically hedonistic values items displayed a mean interitem correlation that was nearly twice that of the short-term goals items (r = .40). Hence, the physically hedonistic values scale may have done a better job of measuring the physically hedonistic values construct than the short-term goals scale did of measuring the short-term goals construct. Second, the short-term goals variable was found to partially mediate the past-crime–future-crime relationship when considered independent of the physically hedonistic values variable (see Footnote 2), but weakened to the point of non-significance when paired with the stronger physically hedonistic values variable, with which it correlated .31 (see Table 2). Additional or different items may accordingly be required to more effectively measure the short-term goals construct.

The sensitivity test results for the Crime-94 [RIGHTWARDS ARROW] Values-96 [RIGHTWARDS ARROW] Crime-98 mediation analysis revealed that early low self-control assessed behaviourally after the age when low self-control is believed to stabilize (i.e., 8–10 years: Gottfredson & Hirschi, 2006) was incapable of erasing the mediating effect of physically hedonistic values on the past-crime–future-crime relationship. When viewed in light of previous research (Walters, 2013b), there would appear to be a growing body of support for the argument that cognitive variables partially mediate the strong and consistent relationship between past and future criminality. The old adage that the best predictor of future behaviour is past behaviour may be true but is not, in and of itself, an explanation. Cognitive mediation provides an explanation in the form of psychological inertia (the tendency for cognitive variables like criminal thinking and values to be self-perpetuating: Walters, 2012b). Early criminal involvement shapes a person's thinking, which, in turn, influences a person's propensity to engage in future criminality. The model proposed by Walters (2012b), in which six quasi–time-stable-mediating factors give rise to crime continuity through psychological inertia is not incompatible with the existence of additional mediators of the past-crime–future-crime relationship, to include official and unofficial labelling (Bernburg & Krohn, 2010) and cumulative disadvantage (Sampson & Laub, 1959), both of which require direct contact with the criminal justice system. In fact, the interaction of direct (labelling and cumulative disadvantage) and indirect (psychological inertia) learning factors (Stafford & Warr, 1978) could be the driving force behind crime continuity.

In the current study a behavioural measure of low self-control administered at age 11–13 failed to account for the mediating effect of a cognitive measure of physically hedonistic values completed between the ages of 17 and 19. What is the significance of this finding for population heterogeneity explanations of crime continuity and Gottfredson and Hirschi's (2006) general theory of crime? Population heterogeneity may still play a role in crime continuity because the six cognitive mediators are shaped to some extent by genetic and early environmental factors. Twin studies, for instance, reveal that genetic factors account for a significant portion of the variance in people's outcome expectancies for alcohol (Slutske et al., 2005), and developmental studies denote that strong child–mother attachment at age 36 months is associated with fewer hostile attributions when the child is in first grade (McElwain, Booth-LaForce, Lansford, Wu, & Dyer, 1998). Research also indicates that these cognitive variables undergo significant developmental change over time (Bandura, 1986; Weiner, 1985). The same could be said of Gottfredson and Hirschi's (2006) general theory of crime. Some individuals have a stronger predisposition to low self-control than others, but the degree of stability in low self-control proposed by Gottfredson and Hirschi (2006) has not always been confirmed empirically (Burt, Simons, & Simons, 2006; Hay & Forrest, 2013; Turner & Piquero, 2012a). Prior research, in which self-report measures of low self-control were administered to juveniles and young adults, may have confounded the predisposition to low self-control and certain cognitive factors that shape and mediate low self-control over the life course.

Deterrence theory's view of choice as an objective, economic appraisal of the certainty, severity, and celerity of punishment has its origins in the early writings of Becarria and Bentham. Although deterrence theory was not directly tested in this study, a cognitive model composed of subjective, psychological factors with clear implications for decision making (i.e., goals and values) was found to partially mediate the relationship between past criminality and future criminality. This would seem to suggest that the decision making that leads to crime may be at least as psychological as it is economic. A practical implication of the current findings, then, is that through problem-solving training and other forms of cognitive intervention it may be possible to improve the decision-making skills of offenders and in the process make their choices more rational and objective and less impulsive and reactive. Moral values must also be assimilated into the decision-making process, however, because emotionless proactive decisions can be just as destructive as reactive impulsive ones. Therefore, in addition to conducting a proper economic analysis of one's options it is also important that values, goals, and expectancies be incorporated into one's decisions.

A policy implication of the current results is that we must be mindful of the subjective aspects of decision making when imposing sanctions on those who violate society's laws. Prison, for instance, is not generally viewed to be particularly aversive by many high-rate and criminally sophisticated offenders. Many such offenders, in fact, view prison as a means of cultivating new criminal contacts and enhancing their credentials as bona fide law breakers (Muntingh, 2008). Given the apparent criminogenic effect of prison (Bales & Piquero, 2012) it may be wise to avoid prison except in cases where the individual presents a physical threat to the community or where the committed offence demands the most severe non-capital sanction available to society, diverting a majority of offenders, particularly those with no history of violence, to community-based programs. Because rates of offending in the current sample were relatively low, it is important that the current results be replicated in a more criminally involved sample.

The current study is not without limitations, one of which is the use of mediational analysis itself. Spencer, Zanna, and Fong (1993) contend that mediational analysis has been misused and oversold as the primary means of establishing causal relationships. They maintain that a well-planned series of experimental studies are superior to mediational analysis in documenting causal links when the psychological processes under investigation are easy to manipulate but difficult to measure. They nonetheless acknowledge that when the psychological processes under investigation are difficult to manipulate but easy to measure then mediational designs may be the preferred method for establishing causal relationships. The psychological processes central to crime continuity are difficult to manipulate, but many are relatively easy to measure. It would appear, then, that mediational procedures have a place in crime continuity research, with the understanding that causality cannot be unequivocally established in a single mediational study. This is because three conditions must be met before it can be concluded that a causal relationship exists: (1) correlation; (2) direction; and (3) absence of viable alternative explanations. Mediational analysis addresses the first two conditions, but the third condition is only partially addressed in a single study, principally through sensitivity testing.

Whereas all viable alternative explanations of the cognitive mediation effect observed in this and the previous Walters (2013b) investigation have yet to be ruled out, results from the current study cast serious doubt on early low self-control as one such viable alternative explanation. To the extent that the current study involved analysis of secondary data it was limited to variables from the original NLSY-C database. Additional viable alternative explanations that could not be tested because they were not part of the NLSY-C database include criminal associates, labelling effects, cumulative disadvantage, intelligence, and the effect of incarceration on values, goals, and the other four cognitive components of Walters' (2012b) model. In the current study goals and values measured at age 17–19 served as the mediating variable, but had roots extending back before delinquency was measured at age 15–17. Hence, it is possible that goals and values shaped delinquency rather than the other way around. Ruling out this particular alternative hypothesis requires childhood measures of goals and values that could be included as confounding covariates in the Imai et al. procedure. Further research is also required to determine whether the other two cognitive variables in Walters' (2012b) model, attributions and outcome expectancies, mediate important crime-relevant relationships, if the six quasi–time-stable cognitive variables interact with one another to mediate the past-crime–future-crime relationship and other important relationships in the criminal justice field, and if the 2-year time intervals that separated the independent and mediating variables and mediating and dependent variables in this study were optimal.

The extension of learning theory to crime was an important development back when Sutherland and Akers first introduced differential association theory and the social learning model of crime to the fields of criminology and criminal justice. The next step in the development of a comprehensive theory of crime is integrating learning theory with models that emphasize cognitive skills (decision making) and cognitive restructuring (criminal thinking/expectancy challenge). Criminological theory has been dominated by single variable models for too long (Ericson & Carriere, 1996); the time has come to integrate the many disparate ideas that have been developed over the years into more comprehensive models. Statistical mediation and moderation (Baron & Kenny, 1986; Walters, 2013a, 1985; Walters, 1985) offer researchers an opportunity to explore the manner in which variables currently used to explain crime interact with one another. This may not only help explain individual differences in the propensity to commit crime but could also shed light on possible avenues of intervention, prevention, and change.

  1. 1

    Removal of cases with incomplete data did not appreciably alter the results of these analyses. There were 132 cases of the original sample of 715 cohort members (birth year between 1977 and 1979) that could not be analysed even with Full Information Maximum Likelihood (FIML) estimation (75 cases had a missing dependent variable, which in the case of a count measure precludes analysis with FIML; 57 cases were missing the independent variable and both mediating variables). FIML estimation analysis of the remaining 583 cases revealed significant Crime-94 [RIGHTWARDS ARROW] Goals-96, Crime-94 [RIGHTWARDS ARROW] Values-96, Crime-94 [RIGHTWARDS ARROW] Crime-98, and Values-96 [RIGHTWARDS ARROW] Crime-98 paths, and a non-significant Goals-96 [RIGHTWARDS ARROW] Crime-98 path.

  2. 2

    When the Goals-96 variable was included as the sole mediator in the SEM path analysis, it achieved a significant leg one (Crime-94 [RIGHTWARDS ARROW] Goals-96) effect (z = 2.67, p < .01) and a significant leg two (Goals-96 [RIGHTWARDS ARROW] Crime-98) effect (z = 2.58, p < .05). In addition, a causal mediation analysis of Goals-96 as a mediator of the Crime-94 [RIGHTWARDS ARROW] Crime-98 relationship, controlling for age, race, gender, and low self-control proved significant: mediation effect = .0130 (95% CI = .0003–.0313), accounting for 4.7% of the variance in the total effect.

References

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. References