Change in the Perceived Certainty of Punishment as an Inhibitor of Post-Juvenile Offending in Serious Delinquents: Deterrence at the Adult Transition

Abstract

The purpose of this study was to determine whether growth in perceived certainty of punishment explained the right leg of the age–crime curve. Using longitudinal data from the Pathways to Desistance study (N = 1,354), it was determined that offense variety achieved its steepest decline between the ages of 17 and 18, and offense frequency displayed its steepest decline between the ages of 19 and 20. Further analysis revealed that perceived certainty of punishment predicted the variety and frequency of offending from the periods of steepest decline but not during periods of offense growth or less optimal decline. These results provide preliminary support for the presence of sensitive periods in emerging adulthood whereby increased perceived certainty may inhibit future offending.

Keywords

perceived certainty of punishment deterrence theory serious delinquents sensitive periods

Rising rapidly during early to mid-adolescence, peaking in late adolescence, and then falling off gradually from here, the relationship between chronological age and crime is said to be invariant across a range of variables (Hirschi & Gottfredson, 1983). This is a characterization that is familiar to most criminologists. The problem is that there is enough deviation from the pattern based on data source, type of crime, nationality, and historical context to question the conclusion that the age–crime relationship is invariant across these variables (Ulmer & Steffensmeier, 2014). Still, it would be fair to say that most individuals stop or significantly reduce their involvement in crime at some point in late adolescence or early adulthood. Even many serious delinquents do not go on to become adult criminals (Laub & Sampson, 2001; Piquero, 2007). The current study was conducted in an effort to make sense of the age–crime curve and the large portion of individuals who drop out of crime in late adolescence or early adulthood, a period commonly referred to as emerging adulthood (Arnett, 2014). One possible mechanism supporting the deceleration of crime during the early or emerging adult years is maturation of the decision-making process (Steinberg & Cauffman, 1996). The current study sought to test this supposition using principles from deterrence theory.

According to deterrence theory, punishment that is swift, certain, and severe has the best chance of deterring future criminal conduct. Whereas the swiftness or celerity of punishment is difficult to incorporate into policy because of due process concerns, certainty and severity have been studied extensively by criminal justice experts, scholars, and policy makers. The results of these studies have fairly convincingly demonstrated that the certainty of punishment is a more effective deterrent to crime than the severity of punishment (Fader, 2016; Freiburger, Romain, Randol, & Marcum, 2017; Nagin, 2013). This appears to be particularly true of risk-affinity individuals, who, unlike risk-aversion individuals, are unresponsive to the severity of punishment but seem reasonably responsive to the certainty of punishment (Schulz, 2014). Given that most offenders can be classified as risk-affinity (Kuin, Masthoff, Kramer, & Scherder, 2015), it is assumed that as a group, offenders will be substantially more responsive to punishment certainty than to punishment severity. Because increased offending has been found to suppress the perceived risk of apprehension (Pogarsky, Kim, & Paternoster, 2005), there is a need to consider whether increased perceived risk of apprehension is capable of suppressing future offending.

Using data from the same delinquent sample as the current investigation (i.e., Pathways to Desistance: Mulvey, 2012), one study addressed the effect of punishment severity on rearrest and self-reported offending (SRO) variety and a second study examined the effect of perceived certainty on a dichotomous measure of SRO. In the first of these studies, Loughran et al. (2009) determined that the severity of punishment, operationalized as confinement versus probation, had no effect on future offending and that longer placements had no greater impact than shorter placements. This was followed by a second study in which Loughran, Pogarsky, Piquero, and Paternoster (2012) examined the effect of risk-certainty on a dichotomous measure of SRO (present vs. absent). Results from this study revealed the presence of a nonlinear relationship between perceived risk and offending. Increased perceived risk was associated with equivalent reductions in offending for youth in the midrange of the risk continuum (30%-90%), whereas it was associated with minimal reductions in offending for youth in the lower range (<30%), and accelerated offending for youth in the upper range (>90%). The Loughran et al. (2012) study demonstrated that the risk-offending relationship varies according to level of risk-certainty; the current study, by comparison, sought to determine whether the risk-offending relationship varied as a function of participant age in an effort to explain the age–crime relationship and late adolescent desistance from crime.

The concept of sensitive periods has been advanced by developmental psychologists in an attempt to explain the fact that infants and children have the greatest likelihood of achieving certain developmental milestones (e.g., attachment, language, self-control) during specific age periods (Bornstein, 1989). A sensitive period for attachment, for instance, has been proposed for human infants between the ages of 6 and 24 months (Bowlby, 1989). It is during these sensitive periods that the child becomes maximally sensitive to specific environmental stimuli, and while this sensitivity is partially a consequence of biological and neuropsychological preparedness, it is also shaped by social-environmental and cultural factors (Troller-Renfree & Fox, 2017). Sensitive periods may not be restricted to childhood. We know, for instance, that the prefrontal cortex is not fully functional until early adulthood and that development in this part of the brain in conjunction with certain social-environmental experiences is vital to the realization of a person’s ability to anticipate the likely consequences of his or her actions and calculate the odds of apprehension (Steinberg, 2008). This suggests that there may be a sensitive period in emerging adulthood wherein the perceived certainty of apprehension and punishment exerts its greatest influence on behavior and is most apt to encourage desistance from crime.

The principal objective of this study was to determine whether changes in the perceived certainty of punishment predicted a decrease in offending behavior. Because research indicates that offense variety and offense frequency produce different results, particularly at moderate levels of offending (Monahan & Piquero, 2007), separate analyses were performed on these two outcomes. Several control variables were included in the analyses: basic demographic variables (age, sex, and race), impulsivity, based on research showing that impulsivity may hold sway over the risk–offense relationship (Schulz, 2014), and three scales (Cost of Crime, Social Rewards of Crime, Personal Rewards of Crime) that are measured with the same inventory as the perceived certainty scale (i.e., the Social and Personal Rewards and Costs of Offending Inventory [SPRCOI]: Nagin & Paternoster, 1994). Although the Pathway study is composed of 11 waves of data, less than half the participants were in the study before age 16 and only a little more than half were still in the study by age 23. This limited the age range of the analyses. Fortunately, the periods of greatest offense deceleration (largest drop in offending from one age to the next) occurred in the middle of this range (17-20 years). It was therefore hypothesized that a change in perceived certainty would predict a change in offending for the period of greatest offense deceleration but not for an alternate period.

Method

Participants

The sample for this study consisted of all 1,354 members (1,170 males, 184 females) of the Pathways to Desistance study (Mulvey, 2012). Each member of the Pathways study was originally contacted about participating in the project shortly after being adjudicated delinquent or convicted of a felony in Maricopa County (Phoenix), Arizona or Philadelphia, Pennsylvania. A baseline interview was held sometime between November 2000 and January 2003 and follow-up interviews were conducted every 6 months for the next 3 years and then every 12 months for the last 4 years of the project. Data collection concluded in March 2010. The ethnic/racial composition of the sample was 20.2% White, 41.4% Black, 33.5% Hispanic, and 4.8% Other, and the average age at the time of the baseline interview was 16.04 years (SD = 1.14, range = 14-19).

Measures

Independent variable

Perceived certainty of punishment served as the independent variable in this study. This variable was assessed with the Certainty of Punishment (Self) scale from the SPRCOI (Fagan & Piquero, 2007; Nagin & Paternoster, 1994). Respondents used an 11-point rating scale (0 = no chance of being caught, 10 = absolute certainty of being caught) to estimate their perceived likelihood of being caught and punished if they were to engage in seven different criminal or antisocial acts (i.e., fighting, robbery with a gun, stabbing someone, breaking into a store or home, stealing clothes from a store, vandalism, and auto theft). The mean rating across the seven offenses was the score used to assess perceived certainty of punishment in this study. The internal consistency of the Certainty of Punishment scale in the current study was good to excellent (α = 0.85-0.92).

Dependent variable

The dependent variable for this study was general offending behavior as measured by the SRO scale (Huizinga, Esbensen, & Weihar, 1991). There were 22 categories of crime listed on the version of the SRO used in this study: damaged property, set fire, broke in to steal, shoplifted, bought/received/sold stolen property, used check/credit card illegally, stole car or motorcycle, sold marijuana, sold other drugs, carjacked, drove drunk or high, been paid by someone for sex, forced someone to have sex, killed someone, shot someone, shot at someone, took by force with a weapon, took by force without a weapon, beat up someone with serious injury, in a fight, beat someone as part of a gang, and carried a gun. The dependent variable was assessed as a variety score (proportion of crime categories engaged in during the recall period) and as a frequency count (total number of offenses the participant reported engaging in over the course of a year). The 2-year test–retest reliability for the total offending variety and frequency scores were .29 and .20, respectively.

Control variables

There were six control variables included in the current investigation. Two of these variables were demographic measures: sex (1 = male, 2 = female) and race (1 = White, 2 = non-White). The other four control variables were cognitive impulsivity, the cost of crime, the social rewards of crime, and the personal rewards of crime, all assessed at age 17 (offense variety) or 19 (offense frequency). The final three nondemographic control variables (cost of crime, social rewards of crime, and personal rewards of crime), like the perceived certainty measure, were assessed using various item combinations and scales from the SPRCOI.

Cognitive impulsivity was assessed with the eight-item Impulse Control (IC) scale of the Weinberger Adjustment Inventory (WAI; Weinberger & Schwartz, 1990). Each WAI item is rated by the respondent on a 5-point scale (1 = false, 5 = true), with higher scores reflecting poorer IC. Item content on the WAI-IC scale indicates lack of restraint over thoughts and actions (e.g., “I say the first thing that comes into my mind without thinking enough about it”; “I become ‘wild and crazy’ and do things other people might not like”; “I’m the kind of person who will try anything once, even if it’s not that safe”). The WAI-IC demonstrated adequate internal consistency in the Pathways study (α = 0.76-0.81).

The cost of punishment was assessed by having respondents indicate the material costs (“buying things”) and freedom costs (“choosing what to wear”) they endured as a result of criminal and juvenile justice supervision and/or incarceration. The 18 items on this scale were scored either yes or no and the yes responses were then summed. Scores on this measure could theoretically range from 0 to 18 and the internal consistency of the scale in the current sample of participants was excellent (Cronbach’s α = 0.96-0.99).

The social rewards of crime were also assessed using items from the SPRCOI. This consisted of three scales (stealing, fighting, and robbery) in which the social benefits of crime were assessed (e.g., “If I take things, other people my age will respect me more”) on a 4-point Likert-type scale (1 = strongly disagree, 2 = slightly disagree, 3 = slightly agree, 4 = strongly agree). The average score per item (five items per scale) was calculated and the average across the three scales was then used as a control measure in this study. These scales also achieved good internal consistency in the Pathways study (α = 0.80-0.90).

The personal rewards of crime scale consisted of seven items (e.g., “How much ‘thrill’ or ‘rush’ is it to break into a store or home?”) rated on an 11-point scale (0 = no fun or kick at all, 10 = a great deal of fun or kick) designed to assess the personal benefits of crime. The average score across seven different potential rewards served as an indicator of personal rewards of crime in the current study. This scale displayed good to excellent internal consistency in the Pathways sample (α = 0.88-0.91).

Analytic Plan

Rather than analyzing the Pathways data in waves, data were organized into age groups and then analyzed. Control and precursor measures were assessed at age 17 in the offense variety analysis and at age 19 in the offense frequency analysis. The independent variable was assessed at age 18 in the offense variety analysis (Certainty-18) and at age 20 in the offense frequency analysis (Certainty-20). The dependent variable was assessed at age 19 in the offense variety analysis (Variety-19) and at age 21 in the offense frequency analysis (Frequency-21). The offense variety scores were assessed in a multiple regression analysis using a maximum likelihood (ML) estimator, whereas offense frequency scores were assessed in a negative binomial regression analysis using a maximum likelihood with robust errors (MLR) estimator. Because the length of the recall period varied between participants, time at risk (in months) for the recall period in which the dependent variable was measured (i.e., age 19 or 21) was included in each regression equation as a control measure. All analyses were performed with Mplus 5.2 (Muthén & Muthén, 1998-2007).

Precursors to both the independent and dependent variables were measured at age 17 or 19 as a means of establishing the causal direction of the results. This was done to rule out the alternate hypothesis that changes in offending was responsible for changes in certainty beliefs rather than changes in certainty beliefs being responsible for changes in offending. It is important for the reader to understand that with the inclusion of precursor measures it was change in the independent and dependent variables that was being assessed. Conditioning on a precursor to an outcome, however, can create endogenous selection bias or collider effects (Elwert & Winship, 2014). Additional analyses were accordingly performed in which the precursor to the outcome (Variety-17 or Frequency-19) was removed from the regression equation to determine whether this led to a reduction in the independent variable effect size.

The rationale for using different age groups for the offense variety and offense frequency analyses was a graph showing different age–crime curves for offense variety and frequency in the Pathways sample. As indicated by the two age–crime curves depicted in Figure 1, offense variety displayed the greatest decline in slope between the ages of 17 and 18 and offense frequency dropped the most between ages 19 and 20. Consequently, the variety analyses were conducted at ages 17, 18, and 19 and the frequency analyses were conducted at ages 19, 20, and 21. An alternate set of analyses were performed to determine whether results changed when the risk–variety relationship was evaluated after the optimal drop in variety of offending (i.e., ages 19, 20, and 21) and the risk–frequency relationship was evaluated prior to the optimal drop in frequency of offending (i.e., ages 17, 18, and 19).

Figure 1.

Average offense variety and offense frequency scores for Pathways to Desistance participants assessed annually between the ages of 16 and 23 years.

Missing Data

Nearly two thirds of participants (62.9%) in this study had complete data on all 11 variables for the multiple regression analysis and a little more than one third of participants (34.0%) had complete data on all 11 variables for the negative binomial analysis. Of the remaining participants in the multiple regression analysis, 18.3% were missing data on one variable, 7.9% were missing data on two to four variables, 9.0% were missing data on six or seven variables, and 2.3% were missing data on eight or nine variables. Of the remaining participants in the negative binomial analysis, 43.0% were missing data on one variable, 13.9% were missing data on two to four variables, 4.2% were missing data on six or seven variables, and 4.9% were missing data on eight or nine variables. The variables with more than 10% missing data were Impulsivity-17 (10.8%), Costs-17 (26.4%), Costs-19 (55.6%), Social Rewards of Crime-17 (10.8%), Personal Rewards of Crime-17 (10.8%), Certainty-17 (11.8%), Certainty-19 (11.2%), Certainty-20 (15.0%), Variety-17 (10.7%), and Frequency-21 (15.6%).

Missing data for both analyses were handled with full information maximum likelihood (FIML), a procedure that estimates model parameters and standard errors from nonmissing data and then applies them to the entire database. This has been found to produce significantly less biased estimates than simple imputation or listwise deletion (Allison, 2012; Peyre, Leplége, & Coste, 2011). FIML rests on two assumptions: the untestable missing at random (MAR) assumption and the multivariate normality assumption. Although MAR cannot be tested because the required data are, by definition, missing, there was no reason to believe that any of the data were missing not at random (MNAR). Multivariate normality was tested by comparing standard errors from analyses calculated using an ML estimator versus standard errors conducted using an MLR estimator. The results of this analysis produced moderate support for the multivariate normality assumption (namely, a mean difference between the two sets of standard errors of 11.0% and a range of 0.0%-37.0%).

Results

Offense Variety

Descriptive statistics and correlations for the 11 variables included in the offense variety analyses can be found in Table 1. Nearly two thirds of the zero-order correlations in this intercorrelational matrix were significant using a Bonferroni-corrected alpha level, including the correlation between Certainty-18 and Variety-19 (r = −.18, p < .001). Collinearity diagnostics failed to show evidence of multicollinearity: Tolerance =.692-.982, variance inflation factor (VIF) = 1.018-1.444.

Table 1.

Descriptive Statistics and Correlations for the 11 Variables Included in the Offense Variety Analyses.

Variable	n	M	SD	Range	2	3	4	5	6	7	8	9	10	11
1. Sex	1,354	1.14	0.34	1-2	−.06	−.06	−.16*	−.16*	−.11*	.19*	.19*	−.12*	−.12*	.01
2. Race	1,354	1.80	0.40	1-2		−.13*	.15*	.08	−.10	−.13*	−.13*	.00	−.08	.03
3. Impulsivity-17	1,208	2.90	0.95	1-5			.10	.24*	.39	−.14*	−.15*	.34*	.20*	.01
4. Costs-17	996	9.19	6.99	0-18				.19*	.06	−.19*	−.15*	.21*	.08	−.06
5. Social Rew-17	1,208	1.92	0.51	1-3.33					.34*	−.22*	−.21*	.22*	.15*	−.04
6. Personal Rew-17	1,208	2.12	2.39	0-10						−.16*	−.18*	.33*	.22*	−.03
7. Certainty-17	1,194	5.38	2.92	0-10							.49*	−.23*	−.15*	−.01
8. Certainty-18	1,246	5.42	3.06	0-10								−.20*	−.18*	.01
9. Variety-17	1,209	0.09	0.13	.00-.82									.29*	−.05
10. Variety-19	1,230	0.06	0.10	.00-.86										.06
11. Risk-19	1,232	11.89	1.04	6-16

Note. Variable = study variables; n = number of participants with nonmissing data; M = mean, SD = standard deviation; range = range of scores in the current sample; sex = 1 (male) or 2 (female); race = 1 (White) or 2 (non-White); Impulsivity-17 = cognitive impulsivity measured when participant was 17 years of age; Costs-17 = count of the perceived material and freedom costs of punishment measured when participant was 17 years of age; Social Rew-17 = social rewards of crime measured when participant was 17 years of age; Personal Rew-17 = personal rewards of crime measured when participant was 17 years of age; Certainty-17 = perceived certainty of punishment measured when participant was 17 years of age; Certainty-18 = perceived certainty of punishment measured when participant was 18 years of age; Variety-17 = offense variety when participant was 17 years of age; Variety-19 = offense variety when participant was 19 years of age; Risk-19 = number of months covered by age 19 outcome measure.

p < .00091 (Bonferroni-corrected alpha level; .05/55 correlations).

The results of a multiple regression analysis with an ML estimator revealed that Certainty-18 predicted offense variety at age 19 net the effects of sex, race, Impulsivity-17, Costs-17, Social Rewards-17, Personal Rewards-17, Certainty-17, offense variety at age 17, and time at risk (β = −.11). Controlling for Certainty-17 and Variety-17 means that a change in certainty of punishment between ages 17 and 18 successfully predicted a change in offense variety between ages 17 and 19. These results are summarized in Table 2.

Table 2.

Maximum Likelihood Multiple Regression Analysis of Offense Variety at Age 19.

Predictor	b (95% CI)	β	t	p
Offense Variety-19 (Outcome)
Sex	−0.017 [−0.033, −0.001]	−.058	−2.12	.034
Race	−0.026 [−0.039, −0.012]	−.099	−3.62	<.001
Impulsivity-17	0.006 [−0.001, 0.013]	−.053	1.68	.092
Costs-17	0.000 [−0.001, 0.001]	.017	0.54	.588
Social Rewards-17	0.006 [−0.006, 0.018]	.030	0.96	.339
Personal Rewards-17	0.004 [0.001, 0.007]	.091	2.82	.005
Certainty-17	0.000 [−0.003 0.002]	−.011	−0.33	.741
Certainty-18	−0.004 [−0.006, −0.002]	−.111	−3.36	<.001
Offense Variety-17	0.169 [0.120, 0.219]	.210	6.66	<.001
Time at Risk-19	0.008 [0.002, 0.013]	.076	2.82	.005

Note. Offense Variety-19 (Outcome) = offense variety score measured when participant was 19 years of age; sex = 1 (male) or 2 (female); race = 1 (White) or 2 (non-White); Impulsivity-17 = cognitive impulsivity measured when participant was 17 years of age; Costs-17 = count of the perceived material and freedom costs of punishment measured when participant was 17 years of age; Social Rew-17 = social rewards of crime measured when participant was 17 years of age; Personal Rew-17 = personal rewards of crime measured when participant was 17 years of age; Certainty-17 = perceived certainty of punishment measured when participant was 17 years of age; Certainty-18 = perceived certainty of punishment measured when participant was 18 years of age; Offense Variety-17 = offense variety score when participant was 17 years of age; Time at Risk-19 = number of months covered by age 19 offense variety score; b (95% CI) = unstandardized coefficient and the lower and upper limits of the 95% confidence interval for the unstandardized coefficient (in parentheses); β = standardized coefficient; t = asymptotic t test; p = significance level of the asymptotic t test; N = 1,354.

To evaluate for the possibility of a collider effect, this multiple regression analysis was recomputed without Variety-17. Results indicated that the effect of Certainty-18 on Variety-19 improved slightly with the removal of Variety-17 from the regression equation (β = −.12, asymptotic t test = −3.53, p < .001). This suggests that conditioning on the precursor to offense variety did not create a noticeable collider effect.

To determine whether perceived certainty of punishment continued to predict offense variety scores several years after the steepest decline in variety offending, the certainty–offense variety relationship was reevaluated at ages 19, 20, and 21. The results revealed that there was no effect when perceived certainty of punishment was assessed at ages 19 and 20, and offense variety was measured at age 21 (β = −.04, asymptotic t = −1.25, p = .212).

Offense Frequency

Descriptive statistics and correlations for the 11 variables included in the offense frequency analyses are listed in Table 3. Over half the zero-order correlations in this table were statistically significant using a Bonferroni-corrected alpha level. Similar to the offense variety results, the zero-order correlation between Certainty-20 and Frequency-21 was statistically significant (r = −.10, p < .001). Collinearity analysis revealed no evidence of multicollinearity: Tolerance = .692-.982, VIF = 1.016-1.473.

Table 3.

Descriptive Statistics and Correlations for the 11 Variables Included in the Offense Frequency Analyses.

Variable	n	M	SD	Range	2	3	4	5	6	7	8	9	10	11
1. Sex	1,354	1.14	0.34	1-2	−.06	−.05	−.16*	−.14*	−.15*	.19*	.22*	−.08	−.08	.06
2. Race	1,354	1.80	0.40	1-2		−.16*	.08	.09	−.07	−.13*	−.14*	−.01	.01	−.07
3. Impulsivity-19	1,230	2.76	0.98	1-5			.05	.27*	.34*	−.13*	−.10*	.10*	.07	.02
4. Costs-19	601	10.96	6.98	0-18				.09	.03	−.13*	−.05	.02	−.06	.07
5. Social Rew-19	1,230	1.84	0.53	1-4					.30*	−.26*	−.21*	.12*	.14*	−.01
6. Personal Rew-19	1,230	1.62	2.23	0-10						−.15*	−.15*	.06	.11*	−.00
7. Certainty-19	1,203	5.56	2.95	0-10							.55*	−.12*	−.11*	.03
8. Certainty-20	1,151	5.74	2.97	0-10								−.12*	−.10*	.03
9. Frequency-19	1,230	95.09	350.28	0-4,263									.20*	.00
10. Frequency-21	1,143	71.29	281.34	0-5,985										−.05
11. Risk-21	1,150	11.46	1.88	4-14

Note. Variable = study variables; n = number of participants with nonmissing data; M = mean, SD = standard deviation; range = range of scores in the current sample; sex = 1 (male) or 2 (female); race = 1 (White) or 2 (non-White); Impulsivity-19 = cognitive impulsivity measured when participant was 19 years of age; Costs-19 = count of the perceived material and freedom costs of punishment measured when participant was 19 years of age; Social Rew-19 = social rewards of crime measured when participant was 19 years of age; Personal Rew-19 = personal rewards of crime measured when participant was 19 years of age; Certainty-19 = perceived certainty of punishment measured when participant was 19 years of age; Certainty-20 = perceived certainty of punishment measured when participant was 20 years of age; Frequency-19 = offense frequency when participant was 19 years of age; Frequency-21 = offense frequency when participant was 21 years of age; Risk-21 = number of months covered by age 21 outcome measure.

p < .00091 (Bonferroni-corrected alpha level; .05/55 correlations).

Because the offense frequency measure was a count variable, a Poisson class regression was performed. Preliminary analyses revealed that the distribution of frequency scores was overdispersed (α = 10.71, p < .001) but not zero-inflated. A non-zero-inflated negative binomial regression analysis was consequently conducted. As indicated by the results found in Table 4, Certainty-20 predicted offense frequency at age 21 after controlling for sex, race, Impulsivity-19, Costs-19, Social Rewards-19, Personal Rewards-19, Certainty-19, offense frequency at age 19, and time at risk (standardized coefficient = −.09). Again, controlling for Certainty-19 and Frequency-19 indicates that a change in certainty of punishment from age 19 to age 20 successfully predicted a change in offense frequency from age 19 to 21.

Table 4.

Negative Binomial Regression Analysis of Offense Frequency at Age 21.

Predictor	Estimate (95% CI)	t	p
Offense Frequency-21 (Outcome)
Sex	−1.425 [−2.089, −0.761]	−4.21	<.001
Race	−0.237 [−0.723, 0.249]	−0.96	.338
Impulsivity-19	−0.015 [−0.250, 0.220]	−0.12	.900
Costs-19	−0.032 [−0.071, 0.008]	−1.56	.118
Social Rewards-19	0.282 [−0.096, 0.660]	1.46	.143
Personal Rewards-19	0.112 [0.021, 0.202]	2.42	.016
Certainty-19	−0.019 [−0.100 0.062]	−0.46	.646
Certainty-20	−0.087 [−0.168, −0.005]	−2.08	.037
Offense Frequency-19	0.001 [0.001, 0.002]	4.03	<.001
Time at Risk-21	0.035 [−0.077, 0.147]	0.61	.540

Note. Offense Frequency-21 (Outcome) = offense frequency score measured when participant was 21 years of age; sex = 1 (male) or 2 (female); race = 1 (White) or 2 (non-White); Impulsivity-19 = cognitive impulsivity measured when participant was 19 years of age; Costs-19 = count of the perceived material and freedom costs of punishment measured when participant was 19 years of age; Social Rew-19 = social rewards of crime measured when participant was 19 years of age; Personal Rew-19 = personal rewards of crime measured when participant was 19 years of age; Certainty-19 = perceived certainty of punishment measured when participant was 19 years of age; Certainty-20 = perceived certainty of punishment measured when participant was 20 years of age; Offense Frequency-19 = offense frequency score when participant was 19 years of age; Time at Risk-21 = number of months covered by age 21 offense frequency score; Estimate (95% CI) = point estimation and the lower and upper limits of the 95% confidence interval for the point estimation (in parentheses); t = asymptotic t test; p = significance level of the asymptotic t test; N = 1,354.

Endogenous selection bias was tested as a possible explanation for these results by recalculating the negative binomial regression analysis with Frequency-19 removed from the equation. The outcome of this analysis revealed that the overall effect of Certainty-20 on Frequency-21 was even stronger once Frequency-19 was removed from the regression equation (estimate = −0.11, asymptotic t test = −2.74, p = .006). This implies that conditioning on the precursor to offense frequently did not create a collider effect.

Moving the time frame back 2 years to before the optimal period for crime deceleration identified in Figure 1, it was noted that a change in perceived certainty of punishment from age 17 to age 18 failed to predict offense frequency at age 19 (estimate = −0.04, asymptotic t test = −0.94, p = .347).

Changes in Certainty Over Time

Figure 2 graphs the mean certainty scores achieved by participants in the Pathways study when they were between the ages of 16 and 23. Comparing these results with the two crime trajectories in Figure 1, we can see that perceived certainty of punishment began to rise around the same time offending began to fall and that certainty continued to rise up through age 23.

Figure 2.

Average perceived certainty of punishment scores for Pathways to Desistance participants assessed annually between the ages of 16 and 23 years.

Discussion

Perceived certainty of punishment or risk of apprehension was selected for analysis in this study because it is the aspect of deterrence theory that has received the greatest amount of support (Fader, 2016; Freiburger et al., 2017; Nagin, 2013). The results of the current investigation further indicate that perceived certainty of punishment is an effective deterrent to both offense variety and offense frequency, although at slightly different points in the adolescence-to-adulthood transition. An increase in perceived certainty between the ages of 17 and 18 predicted a reduction in offense variety between the ages of 17 and 19 and an increase in perceived certainty between the ages of 19 and 20 predicted a reduction in offense frequency between the ages 19 and 21. By contrast, a rise in perceived certainty between the ages of 19 and 20 failed to predict a reduction in offense variety between the ages of 19 and 21 and a rise in perceived certainty between the ages 17 and 18 failed to predict a reduction in offense frequency between the ages of 17 and 19. In both instances, a change in perceived risk predicted a decrease in offense variety or frequency at the point of maximum offense deceleration but not at an alternate point, even though in one of the negative instances crime dropped noticeably (offense variety, ages = 19-21). Congruent with predictions, a sensitive period seemed to exist in the transition between adolescence and adulthood. Thus, while perceptions of punishment certainty continue to grow over the course of emerging adulthood (see Figure 2), their impact on future offending may be concentrated in a relatively narrow time frame.

A major implication of these results is that sensitive periods for perceived certainty in emerging adulthood contribute to reduced levels of offending as part of the right leg of the age–crime curve. There were actually two different age–crime curves identified in the present study, one for offense variety and the other for offense frequency. Although the frequency curve peaked 2 years later than the variety curve, the two curves displayed the distinctive declining right leg traditionally observed in the age–crime curve (Hirschi & Gottfredson, 1983; Ulmer & Steffensmeier, 2014). As previously noted, sensitive periods are the consequence of both neurobiological and social-environmental factors. The neurobiological contributions to the putative sensitive period for perceived certainty encompass ongoing development of cortical structures, particularly those found in the prefrontal cortex (Steinberg, 2008). The social-environmental contributions to the putative sensitive period for perceived certainty have been discussed by Sampson and Laub (1993) and cluster around higher education, employment, marriage, and the military. These developing social bonds may increase perceptions of punishment certainty or highlight social capital the individual is risking by continuing to engage in criminal offending. The intersection of neurobiological and social-environmental factors in the developmental phase known as emerging adulthood may consequently provide suitable conditions for inhibition of offending through an increase in perceived certainty of punishment.

The results of this study have potentially important policy and practical implications. Policy is often the means by which deterrence theory is applied to the criminal justice system. As such, the idea of redirecting funds earmarked for prisons (sanction severity) to policing and the courts to increase apprehension and conviction rates (sanction certainty) is consistent with research showing that sanction certainty is superior to sanction severity in reducing offending behavior (Nagin, 2013). According to the present results, these certainty-promoting policies should be most effective when used with emerging adults. From a practical standpoint, improved decision-making skills acquired through problem solving and social skills training have been found to be effective in augmenting decision-making competence and reducing antisocial behavior in children and adolescents (Biggam & Power, 2002; Kazdin, Siegel, & Bass, 1992; Knight, Dansereau, Becan, Rowan, & Flynn, 2015). Providing participants in these programs with objective risk information could increase the perceived certainty of punishment given research showing that youth use objective risk information when constructing their subjective estimates of certainty (Scheider, 2001). The results of the present investigation indicate that information and programs will have the most impact on individuals making the transition from adolescence to adulthood, perhaps in part because they realize that adult sanctions will have a more lasting effect on their future lives than sanctions received in adolescence.

Missing data are a common problem in longitudinal research and can be considered a limitation of the current investigation. Approximately one third of participants were missing data on one or more variables from the multiple regression variety analysis and about two thirds were missing data on one or more variables from the negative binomial frequency analysis. Missing data were handled in this study with FIML, a well-respected procedure that rests on two principal assumptions: MAR and multivariate normality. The MAR assumption could not be evaluated because the data required to determine whether data are MAR are, by definition, missing. MAR would be violated, for instance, if high income respondents systematically failed to answer a question about their annual income. There was no evidence that such a situation existed in the current study, and even if it did, research denotes that FIML is robust to moderate violations of the MAR assumption (Collins, Schafer, & Kam, 2001; Young & Johnson, 2013). The second assumption, multivariate normality, was tested by comparing the standard errors obtained from analyses using an ML estimator against standard errors achieved using an MLR estimator. The comparison disclosed that the multivariate normality assumption was reasonably satisfied in this sample. It should also be noted that the only variable with more than 16% missing data was a control variable (i.e., cost of crime).

A second limitation of this study is that it used perceived certainty rather than actual certainty as the independent variable. Research indicates that actual risk and perceived certainty are, at best, weakly related (Kleck & Barnes, 2008). Even in the Scheider (2001) study, where objective information moderately influenced the construction of subjective perceptions of risk, indices of actual and perceived certainty were far from perfectly correlated. Consequently, we should exercise extreme caution when applying findings from the current study to policy questions in criminal justice. This is because policy decisions, such as putting more police on the streets or increasing court efficiency, are based on objective estimates of certainty, whereas the current study was grounded exclusively in subjective perceptions of risk. When it comes to working with individual offenders, however, the distinction between objective and subjective risk is less important. In the event the individual perceives a high likelihood of apprehension and punishment if he or she pursues a particular criminal course of action, it is the subjective estimate of apprehension and punishment that will guide his or her decision, not the actual odds of getting caught.

Whereas the notion of a time-limited sensitive period where a change in perceived certainty is differentially associated with a reduction in future offending received preliminary support in this study, there are alternative explanations for these results. It could be argued, for instance, that because the period in which crime decelerated the most was selected for analysis, the methodology used to identify sensitive periods produced a confound. In that periods were selected based on a maximal decrease in the crime rate and crime and certainty are negatively correlated this may have created an ideal environment for a rise in perceptions of certainty. Less extreme changes in offending, regardless of direction, would have offered less variation in scores and therefore less opportunity for significant correlations. Hence, these results may have less to do with sensitive periods than with they do with greater variation in scores leading to higher correlations. To control for this possibility, future researchers might want to consider selecting the sensitive periods a priori rather than using the data to guide the selection process and extend the within-person analysis by measuring certainty and offending more than just twice. Of course, this means that more pilot work will need to be done to locate the sensitive periods across specific types of crime (person vs. property; felony vs. misdemeanor) and specific crime measures (e.g., self-report vs. official data; variety vs. frequency).

The age–crime curve is easier to describe than it is to explain. Descriptions may vary, but most, if not all, describe an asymmetrical bell-shaped curve that rises rapidly during adolescence, peaks in late adolescence/emerging adulthood, and then demonstrates a gradual downward slope in early to middle adulthood (Tremblay & Nagin, 2005). A full explanation of the age–crime curve, however, continues to elude scholars in the criminology field. It was the right leg or decelerating portion of the curve that was of principal interest in the current study. Why do so many emerging adults appear to drop out or significantly curtail their involvement in crime during late adolescence or early adulthood? The answer to this question could involve increased neuropsychological and emotional maturity, reduced status anxiety, enhanced opportunities for adult-level freedoms and responsibilities, alterations in the risk and severity of apprehension for adult crime, and a dawning realization that crime does not “pay” (Ulmer & Steffensmeier, 2014). Findings from the current investigation suggest that a combination of these factors—neuropsychological maturity and alterations in the risk of apprehension and severity of sanctions, in particular—may partially explain a portion of the desistance that gives rise to the right leg of the age–crime curve by means of an increase in the perceived certainty of punishment.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biography

Glenn D. Walters, Ph.D., is an associate professor in the Department of Criminal Justice at Kutztown University in Kutztown, Pennsylvania where he teaches classes in corrections, criminology, substance abuse and crime, and research methods. His current research interests include criminal decision-making, crime desistance, and the development of an overarching psychological theory of offending behavior.

References

Allison

P. D.

(2012). Handling missing data by maximum likelihood (SAS Global Forum, Paper 312-2012). Cary, NC: SAS Institute.

Arnett

J. J.

(2014). Emerging adulthood: The winding road from the late teens through the twenties (2nd ed.). New York, NY: Oxford University Press.

Biggam

F. H.

Power

K. G.

(2002). A controlled, problem-solving, group-based intervention with vulnerable incarcerated young offenders. International Journal of Offender Therapy and Comparative Criminology, 46, 678-698.

Bornstein

M. H.

(1989). Sensitive periods in development: Structural characteristics and causal interpretations. Psychological Bulletin, 105, 179-197.

Bowlby

(1989). Secure and insecure attachment. New York, NY: Basic Books.

Collins

L. M.

Schafer

J. L.

Kam

C. M.

(2001). A comparison of inclusive and restrictive strategies in modern missing data procedures. Psychological Methods, 6, 330-351.

Elwert

Winship

(2014). Endogenous selection bias: The problem of conditioning on a collider variable. Annual Review of Sociology, 40, 31-53.

Fader

J. L.

(2016). “Selling smarter, not harder”: Life course effects on drug sellers’ risk perceptions and management. International Journal of Drug Policy, 36, 120-129.

Fagan

Piquero

A. R.

(2007). Rational choice and developmental influences on recidivism among adolescent felony offenders. Journal of Empirical Legal Studies, 4, 715-748.

10.

Freiburger

T. L.

Romain

D. M.

Randol

B. M.

Marcum

C. D.

(2017). Cheating behaviors among undergraduate college students: Results from a factorial survey. Journal of Criminal Justice Education, 28, 222-247.

11.

Hirschi

Gottfredson

(1983). Age and the explanation of crime. American Journal of Sociology, 89, 552-584.

12.

Huizinga

Esbensen

Weihar

(1991). Are there multiple paths to delinquency? Journal of Criminal Law and Criminology, 82, 83-118.

13.

Kazdin

A. E.

Siegel

T. C.

Bass

(1992). Cognitive problem-solving skills training and parent management training in the treatment of antisocial behavior in children. Journal of Consulting and Clinical Psychology, 60, 733-747.

14.

Kleck

Barnes

J. C.

(2008). Deterrence and macro-level perceptions of punishment risks: Is there a “collective wisdom”? Crime and Delinquency, 59, 1006-1035.

15.

Knight

D. K.

Dansereau

D. F.

Becan

J. E.

Rowan

G. A.

Flynn

P. M.

(2015). Effectiveness of a theoretically-based judgment and decision making intervention for adolescents. Journal of Youth and Adolescence, 44, 1024-1038.

16.

Kuin

Masthoff

Kramer

Scherder

(2015). The role of risky decision-making in aggression: A systematic review. Aggression and Violent Behavior, 25, 159-172.

17.

Laub

J. H.

Sampson

R. J.

(2001). Understanding desistance from crime. Crime and Justice: A Review of Research, 28, 1-69.

18.

Loughran

T. A.

Mulvey

E. P.

Schubert

C. A.

Fagan

Piquero

A. R.

Losoya

S. H.

(2009). Estimating a dose-response relationship between length of stay and future recidivism in serious juvenile offenders. Criminology, 47, 699-740.

19.

Loughran

T. A.

Pogarsky

Piquero

A. R.

Paternoster

(2012). Re-examining the functional form of the certainty effect in deterrence theory. Justice Quarterly, 29, 712-741.

20.

Monahan

K. C.

Piquero

A. R.

(2007). Investigating the longitudinal relation between offending frequency and offending variety. Criminal Justice and Behavior, 36, 654-674.

21.

Mulvey

E. P.

(2012, November). The Pathways to Desistance study: Design and methods. Paper presented at the American Society of Criminology Annual Meeting, Chicago, IL. Retrieved from http://citation.allacademic.com/meta/p574246_index.html

22.

Muthén

(1998-2007). Mplus user’s guide (5th ed.). Los Angeles, CA: Author.

23.

Nagin

D. S.

(2013). Deterrence in the twenty-first century. Crime and Justice, 42, 199-263.

24.

Nagin

D. S.

Paternoster

(1994). Personal capital and social control: The deterrence implications of individual differences in criminal offending. Criminology, 32, 581-606.

25.

Peyre

Leplége

Coste

(2011). Missing data methods for dealing with missing items in quality of life questionnaires: A comparison by simulation of personal mean score, full information maximum likelihood, multiple imputation, and hot deck techniques applied to the SF-36 in the French 2003 decennial health survey. Quality of Life Research, 20, 287-300.

26.

Piquero

A. R.

(2007). Taking stock of developmental trajectories of criminal activity over the life course. In Liberman

A. M.

(Ed.), The long view of crime: A synthesis of longitudinal research (pp. 23-78). New York, NY: Springer.

27.

Pogarsky

Kim

Paternoster

(2005). Perceptual change in the National Youth Survey: Lessons for deterrence theory and offender decision-making. Justice Quarterly, 22, 1-29.

28.

Sampson

R. J.

Laub

J. H.

(1993). Crime in the making. Cambridge, MA: Harvard University Press.

29.

Scheider

M. C.

(2001). Deterrence and the base rate fallacy: An examination of perceived certainty. Justice Quarterly, 18, 63-86.

30.

Schulz

(2014). Individual differences in the deterrence process: Which individuals learn (most) from their offending experiences? Journal of Quantitative Criminology, 30, 215-236.

31.

Steinberg

(2008). A social neuroscience perspective on adolescent risk-taking. Developmental Review, 28, 78-106.

32.

Steinberg

Cauffman

(1996). Maturity of judgment in adolescence: Psychosocial factors in adolescent decision making. Law and Human Behavior, 20, 249-272.

33.

Tremblay

R. E.

Nagin

D. S.

(2005). The developmental origins of physical aggression in humans. In Tremblay

R. E.

Hartup

W. H.

Archer

(Eds.), Developmental origins of aggression (pp. 83-106). New York, NY: Guilford.

34.

Troller-Renfree

Fox

N. A.

(2017). Sensitive periods of development: Implications for risk and resilience. In Luby

J. L.

(Ed.), Handbook of preschool mental health development, disorders, and treatment (2nd ed., pp. 3-26). New York, NY: Guilford.

35.

Ulmer

J. T.

Steffensmeier

(2014). The age and crime relationship: Social variation, social explanations. In Beaver

K. M.

Barnes

J. C.

Boutwell

B. B.

(Eds.), The nurture versus biosocial debate in criminology: On the origins of criminal behavior and criminality (pp. 377-396). Thousand Oaks, CA: Sage.

36.

Weinberger

D. A.

Schwartz

G. E.

(1990). Distress and restraint as superordinate dimensions of self-reported adjustment: A typological perspective. Journal of Personality, 58, 381-417.

37.

Young

Johnson

D. R.

(2013). Methods for handling missing secondary respondent data. Journal of Marriage and Family, 75, 221-234.