Crime propensity and lifestyle risk: The interplay of personal morality and self-control ability in determining the significance of criminogenic exposure

Abstract

According to Situational Action Theory, acts of crime are the result of a propensity—exposure interaction, with greater effects of criminogenic exposure for people of higher crime propensity—the latter determined by personal morality and the ability to exercise self-control. The present study delves into this interplay using individual-level survey data from adolescents in Slovenia. It goes beyond extant research by explicitly testing the three-way interaction of personal morality, self-control ability and a criminogenic lifestyle. Employing a respondent's lifestyle risk as proxy measure of the extent of exposure to crime-conducive settings, supportive results are obtained. An individual's lifestyle risk fosters criminal activity particularly among those with higher crime propensity. There is also evidence of a complex three-way interaction according to which self-control ability protects against detrimental lifestyle risk effects mainly among adolescents exhibiting weak morality. This suggests that personal morality and the capacity for self-control jointly and interdependently condition the significance of criminogenic exposure.

Keywords

Lifestyle risk morality propensity–exposure interaction self-control situational action theory

Research interest

In their most recent monograph ‘Character, Circumstances, and Criminal Careers,’ Wikström et al. (2024) emphasize the significance of the person–environment interplay for the explanation of unlawful behavior. At the heart of their Situational Action Theory (SAT) is a propensity–exposure interaction suggesting that ‘human action is an outcome of the interaction between personal characteristics and environmental features’ (Wikström et al., 2024: 38). The theory's PEA-hypothesis (Wikström and Treiber, 2016) posits that propensity and exposure jointly shape action. Acts of crime are a function of who is where. People do what they do because of who they are (their character or action-propensities) and where they are (the settings or microenvironments they participate in).¹ Criminal behavior results from the interaction of a person's crime propensity and his or her criminogenic exposure. However, depending on the level of analysis, interaction can mean different things—namely convergence or moderation (Hardie, 2020). At the situational level, criminal activity is assumed to be most likely when crime-prone individuals encounter crime-conducive surroundings. At the person level, a moderation relationship is expected according to which the impact of one influencing factor depends on the level of the other determining factor, with greater setting effects in the company of high crime propensity and propensity mattering more in criminogenic settings (Wikström et al., 2012).

Drawing on survey data collected from adolescents in Slovenia, this study seeks to test the presence of propensity–exposure interactions. Similar to previous works (Kokkalera et al., 2020, 2023; Kokoravec Povh et al., 2024; Svensson and Pauwels, 2010; Wikström and Butterworth, 2011; Wikström and Svensson, 2008), it utilizes a lifestyle risk measure to determine the extent of involvement in criminogenic settings. However, while these studies relied on linear regression models to examine the postulated interactive effects of ‘character and circumstances’ on skewed measures of crime frequency or variety, the present inquiry conducts the crucial interaction analyses in a more appropriate negative binomial framework. It also goes beyond previous works by addressing the detailed interplay of lifestyle risk, self-control ability and personal morality. The question of whether an individual's capacity for self-control or a person's morality or both characteristics jointly determine the significance of a crime-facilitating way of life has been neglected yet. Therefore, we will investigate whether the strength of the association between lifestyle risk and self-reported offending depends on the level of criminal propensity and how the components of an individual's propensity for crime interact in shaping the size of the lifestyle risk effect. The present study represents the first test of the three-way interaction between personal morality, self-control ability and a lifestyle-induced exposure risk. It also contributes to the cross-cultural testing of SAT by applying the theory to a Southern European context.

Literature review

SAT is a ‘complex integrative mechanism-based action theory’ (Pauwels et al., 2018: 33) that combines person-oriented and environment-oriented reflections. It aims at explaining all violations of moral rules of conduct, including breaches of the law (Wikström and Treiber, 2016). For this purpose, the theory submits a very detailed description of how personal characteristics and outer circumstances interact in the process of crime causation. The centerpiece of the theory is a person–environment or propensity–exposure interaction (Wikström et al., 2012). The probability that a criminal act occurs is expected to peak when an individual with a strong disposition for crime takes part in a crime-facilitating setting. Thereby, criminogenic exposure is hypothesized to exert a greater effect on behavior when the individual's propensity to offend is higher. This proposition has already received nonnegligible empirical support (Hirtenlehner and Treiber, 2017; Kabiri et al., 2020, 2023; Kennedy, 2024; Kokkalera et al., 2020, 2023; Kokoravec Povh et al., 2024; Pauwels et al., 2018; Wikström et al., 2010, 2012, 2018).

According to SAT, an individual's crime propensity—his or her ‘tendency to see and choose crime as an action alternative’ (Wikström et al., 2024: 63)—is shaped by personal morality and the ability to exercise self-control. Personal morality describes an individual's conception of what is right or wrong (acceptable or unacceptable) to do in response to a given motivation in a particular setting. It involves both moral beliefs and associated moral emotions (particularly feelings of shame and guilt). Law-relevant personal morality denotes the extent to which a person's morals correspond to the rules of conduct stated in penal law (i.e., the individual's level of legal norm acceptance). Self-control ability on the other hand captures an individual's capacity to withstand situational temptations and provocations to act against internalized moral rules. The theory's notion of self-control focuses on the moral management of the inducements to offend that are present in a setting. Thereby, SAT distinguishes between trait self-control (an enduring capability) and state self-control (a situational activity) (Wikström et al., 2012). It is assumed that a well-developed ability to exercise self-control enables individuals to resist the lures of crime-conducive surroundings.

The criminogeneity of a setting—its ‘tendency to lead people to view and choose crime as an action alternative’ (Kokoravec Povh et al., 2024: 2)—depends on the moral norms that apply to it (moral context) and the degree to which these norms are enforced (deterrence) (Wikström, 2010). Which moral norms prevail in a setting, is contingent, among other things, on the legal order of the country concerned and the type of people present in this setting (Wikström et al., 2024). Regarding the latter, it must be noted that the moral values and behavioral expectations shared by the participants of a setting form an important constituent of its moral context. The deterrent quality of a setting hinges on the setting's capacity to enforce its moral norms by creating fear of sanctions. It is assumed that individuals refrain from breaking the law when discernible environmental cues (e.g., police presence, security cameras or watchful citizens) instill fear of punishment (Wikström et al., 2012).

SAT conceptualizes behavior as a function of perception and choice (Wikström, 2006, 2010). When an individual with a certain propensity for crime encounters a setting with a certain level of criminogeneity, motivation (in the form of temptation or provocation) evolves. This motivation initiates a sequential perception–choice process that directly governs action. Perception (guided by personal and contextual morality) determines which action alternatives are taken into consideration as a means of responding to a given motivation. The application of a person's morals to the moral context of the current setting yields a moral filter that regulates whether the individual sees crime as an acceptable action alternative. Choice (sometimes influenced by controls) determines which of the considered alternatives is actually selected for execution. Under conflicting rule guidance, when—due to a weakened moral filter—a deliberating actor perceives both criminal and noncriminal response options, self-control or deterrence may affect behavioral decisions. Controls are only relevant when an individual ponders over various action alternatives, of which at least one violates the law. A criminal response option will be selected if it is viewed as the best morally acceptable alternative in the present situation (Wikström et al., 2024).

As mentioned above, various studies provide evidence of the hypothesized propensity-exposure interaction (Pauwels et al., 2018). Some researchers employ space-time budget data to capture the extent of involvement in crime-facilitating settings at a situational level (Beier, 2018; Hardie, 2019; Kennedy, 2024; Wikström et al., 2010, 2012, 2018). Their works show that the likelihood of crime events culminates when crime-prone individuals are faced with crime-conducive settings. Other scholars merge people's peer delinquency and deterrence perceptions to obtain a person-level measure of criminogenic exposure (Hirtenlehner and Treiber, 2017; Kabiri et al., 2020, 2023). Still others draw on measures of an individual's lifestyle risk to quantify the degree of his or her criminogenic exposure (Kokkalera et al., 2020, 2023; Kokoravec Povh et al., 2024; Svensson and Pauwels, 2010; Wikström and Butterworth, 2011; Wikström and Svensson, 2008). These operationalizations rest on the assumption that individuals with a ‘dangerous’ way of life more frequently come across crime-conducive microenvironments. People who often socialize with delinquent others, spend much time at ‘risky’ places (e.g., bars, parks, the city center) or with unstructured peer-oriented leisure activities and frequently abuse alcohol or drugs are more likely to find themselves faced with criminogenic outer circumstances. In support of this presumption, the corresponding investigations consistently find crime-promoting effects of a person's lifestyle risk that increase with the individual's propensity to offend. Last but not least, several scenario studies also back the significance of the interplay of people and settings (Haar and Wikström, 2010; Pauwels, 2018a, 2018b; Wikström et al., 2012). These works present situational-level evidence of greater exposure effects for individuals characterized by a higher propensity for crime.

Aside from specific tests of SAT's PEA-hypothesis, numerous inquiries indicate interindividual heterogeneity in the susceptibility to environmental influences. Setting features appear to have different effects on different people (Treiber, 2017). Among other things, the influence of exposure to criminal opportunities (Grasmick et al., 1993; Hay and Forrest, 2008), association with delinquent peers (Mears et al., 1998; Wright et al., 2001), neighborhood characteristics (Meier et al., 2008; Zimmerman et al., 2015) and deterrence variables (Hirtenlehner, 2020; Walters, 2020) on offending has been shown to depend on measures of criminal propensity.

In this context, the interworking of the key components of an individual's propensity for crime must be addressed. Several studies have examined the interplay of self-control ability and personal morality. The majority observed that the size of the self-control effect is contingent on the actor's law-relevant morality, with the ability to exercise self-control being more predictive of offending among individuals characterized by weak morality (Craig 2019; Hirtenlehner and Kunz, 2016; Ivert et al., 2018; Kroneberg and Schulz, 2018; Svensson et al., 2010; Wikström and Svensson, 2010). This interaction pattern can be derived from the outlined functioning of the moral filter. Weak law-relevant morals render the actor's moral filter permeable for crime, resulting in his or her contemplation of criminal action alternatives. Only when the individual perceives crime as a real option to satisfy a given motivation does self-control come into play. ‘For individuals with strong morality, the ability to exercise self-control is not important, … because they do not see crime as an action alternative, and, hence, there is no need to exercise self-control’ (Wikström and Svensson, 2010: 405). In accordance with this reasoning, a few scenario studies capturing directly whether crime is perceived as alternative in a hypothetical situation demonstrated that the respondent's capacity for self-control matters solely for those who actually contemplate crime (Brauer and Tittle, 2017; Kroneberg and Nägel, 2024).

In a seminal study, Hirtenlehner et al. (2022) investigated the three-way interaction of perceived peer delinquency, trait self-control, and personal morality. Inspired by SAT, they assumed that self-control ability mitigates the impact of exposure to delinquent peers chiefly among people who, because of weak morality, are prone to see crime as a selectable option. The obtained results support this theorizing. Based on survey data from adolescents in Austria, they found that an individual's capacity for self-control conditions the criminogenic peer effect mainly among those with lower levels of law-relevant morality. Among youths of strong morality, the ability to exert self-control hardly affects the magnitude of the delinquent peer effect.²

The impact dynamic outlined above may be extended to the understanding of individual differences in the vulnerability to criminogenic environmental influences generally. At the point of action, personal morality and self-control ability may act as a buffer against detrimental implications of crime-conducive exposure, and this interdependently. Self-control has the potential to defuse criminogenic influences of the current microenvironment, but this counteracting process only comes into operation when a permeable moral filter enables the perception of crime as one among several acceptable alternatives (Hirtenlehner and Leitgoeb, 2021). The latter is more likely to occur among individuals who have poorly internalized the rules of law. Such an interworking implies that setting effects peak for people uniting weak personal morality with low self-control ability.

Hypotheses

The deliberations developed above lead to a catalog of four hypotheses that will guide the ensuing analyses:

Hypothesis 1: The criminogenic lifestyle risk effect is greater for individuals with higher crime propensity.

Hypothesis 2: The criminogenic lifestyle risk effect is greater for individuals with lower self-control ability.

Hypothesis 3: The criminogenic lifestyle risk effect is greater for individuals with weaker personal morality.

Hypothesis 4: Self-control ability decreases the criminogenic lifestyle risk effect especially among individuals possessing weaker personal morality.

Methods

Data

The present research utilizes data gathered in the course of the Slovenian Study of Parental Monitoring and Adolescent Delinquency in spring 2011 in Ljubljana. Ljubljana is the centrally located capital of Slovenia. It has approximately 300,000 residents. Sampling followed a three-stage selection process. At Stage 1, nine out of 32 high schools in Ljubljana were incorporated in the inquiry. To ensure an accurate portrayal, a purposive sampling strategy oriented at type and size of school was implemented. At Stage 2, a random selection of classes of the 10th grade (second-year high school) was conducted: three classes in large schools and two classes in small schools were drawn by chance, which yielded a total sample of 19 classes. At Stage 3, all students attending the chosen school classes were included in the study. Altogether, 409 adolescents, mostly aged 16, completed a questionnaire. Fifty-one percent of the respondents are females, 49% are males.

The survey was performed during class hours in paper-and-pencil form. Questionnaires were distributed to groups of five to eight students at a time. The participants were seated in classrooms in a manner that prevented them from chatting or copying the answers from each other. Trained interviewers led them through the questionnaire by introducing each section of the survey and then asking them to complete the section by themselves.

Measurement

Offending

Self-reported offending was measured in terms of a crime variety scale that depicts the number of different types of crime a respondent has committed in the year before the survey. We rely on such a measure because it has been shown that variety scales outperform frequency scales in several ways (Bendixen et al., 2003; Sweeten, 2012). Variety scores exhibit a higher reliability than frequency scores, are less skewed and show higher correlations with other measures of criminal conduct. Furthermore, they are not biased by a ‘guessing’ of the frequency of one's perpetration of a specific offense and less dominated by petty crimes. Here, information was gathered on eight types of crime (shoplifting, theft from a person, theft from a car, theft of a car, burglary, vandalism, robbery, hitting someone or beating somebody up). Twenty percent of the respondents report having committed at least one of these offenses in the past 12 months. The skewness parameter of the variety score amounts to 4.47. A Kolmogorov–Smirnov test discloses a significant (p = .000) deviation from the normality assumption.

Personal morality

To assess the students’ law-related moral beliefs, participants were asked to indicate how wrong they think it is for someone of their age to commit six specific acts of crime (paint graffiti on a house wall/smash a street light/hit another young person/use a weapon or force to get money or things from another young person/steal a CD from a shop/break into a building to steal something). The four available response options ranged from ‘very wrong’ to ‘not wrong at all.’ The individual answers were added to a composite score, which was coded so that high values denote weak moral beliefs (alpha = .80).

Self-control ability

The respondents’ capacity for self-control was measured by an abridged version of the attitudinal scale developed by Grasmick et al. (1993) and refined by Wikström et al. (2012). The employed items focus on the risk-taking, impulsivity and temper components of the construct. Here, the participants were asked to assess ten statements³ on a four-point Likert format ranging from ‘strongly agree’ to ‘strongly disagree.’ The individual answers were summed to a total score, which was coded in a way that high values indicate low self-control ability (alpha = .82).

Crime propensity

In accordance with SAT's contention that a person's morality and his or her ability to exercise self-control are the key elements of an individual's propensity for crime, Z-scores of the measures of both concepts were added up to create an index variable depicting the respondent's crime propensity.

Peer delinquency

Perceived peer delinquency has been established as one of the best predictors of criminal conduct (Hoeben et al., 2016). Students’ perceptions of peer delinquency were measured by asking ‘Does it often happen that some of your friends (steal things/destroy things that do not belong to them/beat up or get into fights with others)?’. The answer categories were ‘no, never,’ ‘yes, sometimes,’ ‘yes, often,’ and ‘yes, very often.’ Items were added to a sum score, with greater values denoting more perceived peer delinquency (alpha = .76).

Lifestyle risk

People with different lifestyles are unequally exposed to crime-conducive settings. Although the outer circumstances in which an individual operates change frequently across the day, a person's lifestyle risk describes his or her general tendency to enter highly criminogenic microenvironments. Guided by Wikström and Butterworth (2011: 181), we created a composite measure of lifestyle risk ‘based on …peer delinquency, time spent in high-risk public environments and … substance use’ to quantify an individual's level of criminogenic exposure. Our operationalization of peer delinquency has been described above. Inspired by the notion that unstructured leisure activities in the company of peers and under the influence of alcohol provide ample opportunities for crime, exposure to crime-facilitating public environments was captured by the question ‘How often do you and your friends spend time in the evenings in the city center?’ (four response options between ‘never’ and ‘six or seven days a week’).⁴ The extent of alcohol consumption was determined by the open question ‘How many times did you drink alcohol so that you felt drunk in the last year?’. To obtain an overall lifestyle risk measure, all three variables were median-dichotomized and then summed up.

Sex

In some models, sex (coded 0 for females and 1 for males) is included as control variable. The underlying rationale is that sex is associated with numerous other predictors of criminal activity, which is why it qualifies as a proxy measure for many unobserved third variables (Moffitt et al., 2001; Steffensmeier and Allan, 1996). However, we admit that this procedure is not perfectly in line with SAT which considers sex as attribute and not as cause in the proper sense (Hirtenlehner and Treiber, 2017).

Table 1 reports descriptive statistics and bivariate correlations for all concepts employed in the ensuing analyses.

Table 1.

Descriptive statistics and product–moment correlations.

	1	2	3	4	5
1. Crime variety	1
2. Lifestyle risk	.32***	1
3. Crime propensity	.34***	.33***	1
4. (Weak) personal morality	.30***	.33***	.76***	1
5. (Low) self-control ability	.22***	.17***	.76***	.15**	1
Mean/Std Dev	0.31/0.82	1.45/1.04	0.00/1.51	2.99/3.09	15.40/5.23
Range	0–8	0–3	−3.21–6.88	0–18	0–30

*** p ≤ .001, ** p ≤ .01, * p ≤ .05.

Analytic strategy

Testing interactive effects on skewed count variables represents a delicate endeavor (Hardie, 2020). Crime variety or frequency measures usually violate crucial assumptions of ordinary least squares (OLS) regression (i.e., normality of the error terms and homoscedasticity). This leads to biased inference statistics and the risk of detecting spurious interaction effects when utilizing linear regression for this kind of data. Thereby, the spurious interaction may reflect the unique distribution of the response variable, in particular floor or ceiling effects (Osgood et al., 2002).

Negative binomial models (Hilbe, 2011) are better suited to analyze the determining factors of adolescents’ crime variety. However, methodological works have recently revealed that the convenient practice of interpreting product terms added to the model equations—which functions well in the context of OLS regression—cannot be applied in a nonlinear framework (McCabe et al., 2022; Mize, 2019). The product term is not instructive here, neither for the presence nor the direction of a potential interaction.

Due to the logarithmic link function, negative binomial models are multiplicative by nature, which implies that they can contain model immanent and product term interaction. The two interaction components mutually affect each other and add up to the total interaction effect. This fact creates the risk of producing methodological artifacts when focusing solely on the sign and significance of the slope parameter of the multiplicative term to establish the pattern of interaction inherent in the data. Therefore, in a negative binomial framework the product term coefficient may be a misleading estimate of the interplay of two predictor variables (Berry et al., 2010; Bowen, 2012).

To address the overall interaction, scholars can compare conditional marginal effects obtained from negative binomial regression models (Williams, 2012). A marginal effect relates a continuous independent variable to the predicted change of a dependent variable, given specific values of other explanatory factors (Mize, 2019). In the case of a negative binomial regression, it expresses how the expected value of the count response changes with a one-unit increase in the predictor variable, with other independent variables held constant (Hilbe, 2011). Differences in the conditional marginal effects of a focal variable (here: lifestyle risk) at representative values of a moderator variable (e.g., crime propensity) indicate the presence of interaction. The calculated conditional marginal effects can be tested for equality using the Z-test proposed by Paternoster et al. (1998).⁵ According to Hardie (2020: 62), this procedure ‘seems most fruitful for the further study of … interaction within a nonlinear framework.’

In order to assess the existence of a three-way interaction between personal morality, self-control ability and lifestyle risk, we relied on a difference-in-differences approach (Hirtenlehner et al., 2022). Separate negative binomial models were fitted for subsamples of individuals with weaker and stronger morality. To this end, the sample was split into two subgroups at the median value of the personal morality measure. For both groups, we estimated negative binomial models with lifestyle risk, self-control ability and their product as predictors. In line with the difference-in-differences strategy, we determined the lifestyle risk effect differentials arising from the respondents’ level of self-control for each subsample individually, and subsequently compared these differentials across groups. A greater difference in the (self-control-related) conditional lifestyle risk effects in the ‘weaker morality’ group can be interpreted as evidence of the hypothesized three-way interaction.

All regression models were fit using Stata 15. Predictor variables were z-standardized before computing the corresponding interaction terms (Aiken and West, 1991). Due to the hierarchical organization of the employed data (students nested in school classes), the model estimations were based on cluster-robust standard errors. Because of the directional nature of the underlying interaction hypotheses, we rely on one-tailed inference statistical testing.

Results

SAT suggests that ‘crime tends to occur when crime-prone individuals face temptations or provocations in criminogenic settings’ (Kokoravec Povh et al., 2024: 2). Thus, our analysis departs from a look at the mean crime variety scores for different combinations of criminal propensity and lifestyle risk (Table 2). For this purpose, both explanatory variables were dichotomized at the median.⁶ It is immediately apparent that criminal activity peaks when a higher crime propensity meets a greater lifestyle risk. The least crime involvement occurs when a lower crime propensity joins a smaller lifestyle risk. The other groups lie between these poles. According to an analysis of variance, the offending scores differ significantly across the four propensity-exposure combinations (η = .31; p = .000). These observations are compatible with the PEA-hypothesis.

Table 2.

Mean crime variety at different combinations of crime propensity and lifestyle risk.

Lifestyle risk	Crime propensity	Mean crime variety score
Lower	Lower	0.05
Lower	Higher	0.25
Higher	Lower	0.32
Higher	Higher	0.68

The interplay of lifestyle risk with crime propensity

In a next step, we examine whether the strength of the association between lifestyle risk and criminal activity varies with the level of an individual's propensity for crime. Figure 1 gives a first impression of the investigated moderation relationship, again with the median-dichotomized propensity and exposure variables. The vertical axis denotes the mean crime variety score. From the interaction diagram, it can be seen that an individual's lifestyle risk shapes his or her self-reported offending to a greater extent among those of higher crime propensity. Among those of lower crime propensity, lifestyle risk makes less difference regarding the scope of criminal activity. The relationship pattern inherent in the raw data certainly accords with the hypothesized person–environment interaction.

Figure 1.

Interaction diagram.

To determine the statistical significance of the observed interplay, we draw on comparisons of conditional marginal effects derived from a negative binomial regression analysis. Altogether, we estimated a whole series of negative binomial models. Table 3 reports the results.

Table 3.

Negative binomial regression analyses.

	Model 1			Model 2			Model 3
	B	SE	P	B	SE	p	B	SE	p
Lifestyle risk	0.775	0.173	.000	0.766	0.169	.000	0.835	0.171	.000
Crime propensity	0.447	0.125	.000	0.414	0.124	.000	0.629	0.141	.000
Sex				0.395	0.247	.110
Lifestyle*propensity							−0.197	0.078	.006^#
Overall model	n = 409; R²_pseudo = .137; p = .000			n = 409; R²_pseudo = .142; p = .000			n = 409; R²_pseudo = .142; p = .000

B: regression coefficient; SE: cluster-robust standard error; p: error probability; n: number of observations.

The coefficient must be considered insignificant because its sign is at odds with the theoretical expectations.

Model 1 includes lifestyle risk and crime propensity as independent predictors—both variables in z-standardized form. Its findings reveal that both concepts are significantly and positively associated with self-reported offending. However, lifestyle risk (IRR = 2.17) contributes slightly more to the explanation of criminal activity than crime propensity (IRR = 1.56).⁷ These results remain stable when sex is added to the equation (Model 2).

Finally, Model 3 serves as basis for the examination of the hypothesized propensity–exposure interaction. Here, the crime variety score is regressed on lifestyle risk, crime propensity and their product. Conditional marginal lifestyle risk effects at representative crime propensity values were calculated from the findings of this analysis. The specific lifestyle risk effects were determined at three levels of criminal propensity: its mean, one standard deviation below the mean and one standard deviation above the mean. To assess whether these conditional marginal effects differ significantly from each other, we draw on pairwise Z-tests (Paternoster et al. 1998). Table 4 presents the results.

Table 4.

Conditional effects of lifestyle risk at representative values of crime propensity (marginal effects from negative binomial models).

Conditional lifestyle risk effects at	ME (p)
Lower crime propensity (µ − 1 SD)	0.163 (.001)
Medium crime propensity (µ)	0.210 (.000)
Higher crime propensity (µ + 1 SD)	0.264 (.000)

Equality tests	Z (p)
Lower/higher	2.05 (.020)
Lower/medium	2.69 (.004)
Medium/higher	1.68 (.046)

ME: marginal effect; Z: Z-value; p: error probability.

The interaction analysis points toward a greater relevance of exposure for individuals characterized by a stronger disposition for crime. Although they are without exception predictive of offending, the conditional lifestyle risk effects increase in size with the level of the person's crime propensity. The lifestyle risk effect in the ‘higher propensity’ condition is 1.6 times as large as in the ‘lower propensity’ condition; the parameter estimate at medium crime propensity lies in between. According to the conducted Z-tests, all pairwise effect differences are statistically significant. These observations clearly indicate that the individual's lifestyle matters more for people with a higher propensity to offend.

As a sensitivity analysis (and in order to tie up with previous studies), Model 3 was reestimated as OLS regression (likewise with cluster-robust standard errors). Its results provide a significant interaction term whose sign suggests that lifestyle risk becomes more influential when crime propensity rises (B = 0.15; p = .020). The introduction of the product term adds 3.2% points explained variance.

The interplay of lifestyle risk with personal morality and self-control ability

Hitherto, we have shown that crime propensity determines the significance of an individual's lifestyle risk. However, it is also interesting to explore which components of the person's propensity to offend affect the size of the lifestyle risk effect and how morality and self-control ability collaborate in this regard.

To approach this issue, we first investigate whether the magnitude of the lifestyle risk effect varies with the level of self-control ability, personal morality, or both. Here we again draw on comparisons of conditional marginal effects obtained from negative binomial regression analyses. Two models were estimated: one with self-control ability, lifestyle risk and their product as predictors, and another one with personal morality, lifestyle risk and the corresponding multiplicative term as regressors. The detailed results of the model estimations can be found in Appendix 1. In the current case, the conditional marginal effects were computed at three representative values of self-control or morality: one standard deviation below the mean, the arithmetic mean and one standard deviation above the mean. Table 5 informs about the effect differentials.

Table 5.

Conditional effects of lifestyle risk at representative values of personal morality or self-control ability (marginal effects from negative binomial models).

Conditional lifestyle risk effects at	ME (p)	Conditional lifestyle risk effects at	ME (p)
Higher self-control (µ − 1 SD)	0.197 (.001)	Stronger morality (µ − 1 SD)	0.156 (.000)
Medium self-control (µ)	0.253 (.000)	Medium morality (µ)	0.213 (.000)
Lower self-control (µ + 1 SD)	0.323 (.000)	Weaker morality (µ + 1 SD)	0.289 (.000)

Equality tests	Z (p)	Equality tests	Z (p)
Lower/higher	1.81 (.020)	Weaker/stronger	2.19 (.014)
Lower/medium	1.59 (.056)	Weaker/medium	1.96 (.025)
Medium/higher	2.19 (.015)	Medium/stronger	2.60 (.005)

ME: marginal effect; Z: Z-value; p: error probability.

From the first two columns of Table 5, it is apparent that the lifestyle risk effect increases as the individual's capacity for self-control decreases. Criminogenic exposure is most influential among adolescents of lower self-control ability and least consequential among youths of higher self-control ability. Two out of three examined pairwise effect differences turn out to be significant. The third one very narrowly misses the critical significance threshold (p < .06).

From the last two columns of Table 5, it is discernible that the lifestyle risk effect rises when personal morals become more conducive to crime. Criminogenic exposure is most influential among adolescents of weaker morality and least consequential for youths of stronger morality. All pairwise effect differences prove to be significant.

So far, we have presented evidence that the magnitude of the lifestyle risk effect varies as a function of both self-control ability and personal morality. Still open is the question of the concrete interplay of these individual characteristics in shaping the significance of criminogenic exposure. To examine the complex interworking of all three concepts, we divided the sample at the median value of the morality measure into two subgroups. Subsequently, negative binomial regression models including self-control ability, lifestyle risk and their product as predictors were fitted separately for the ‘weaker morality’ and the ‘stronger morality’ subsample (see Models 3a and 3b in Appendix 1). Self-control-specific marginal lifestyle risk effects obtained from these models form the basis of a difference-in-differences analysis. First, we study the differences between the conditional exposure effects at representative values of the self-control trait (one standard deviation below the mean and one standard deviation above the mean) within groups characterized by a given level of morality. Then we compare the observed effect differentials across the two morality groups. Table 6 displays the results.

Table 6.

Conditional effects of lifestyle risk at selected combinations of self-control ability and personal morality (marginal effects from negative binomial models).

Self-control	Morality	ME (p)	Δ \| Z (p)	ΔΔ (p)
Lower	Weaker	+0.447 (.001)
Higher	Weaker	+0.243 (.017)	0.204 \| 1.69 (.045)
Lower	Stronger	+0.101 (.008)
Higher	Stronger	+0.158 (.050)	−0.057 \| 0.60 (.274)	0.261 \| 1.71 (.044)

ME: marginal effect; Δ: effect difference; ΔΔ: difference of differences; Z: Z-value; p: p-value.

The main finding of this analysis is a systematic three-way interaction according to which well-developed self-control ability attenuates the detrimental effects of lifestyle risk particularly among individuals who have poorly internalized the legal rules. In the ‘weaker morality group,’ there is significant variation in the conditional lifestyle risk effects, with criminogenic exposure being more predictive of offending for people who lack self-control ability. In the ‘stronger morality group,’ the capacity for self-control does not modify the predictive power of the individual's lifestyle risk. Comparing the observed effect differentials across the investigated morality groups reveals significantly larger effect differences among adolescents of weaker personal morality. This suggests that the protective impact of an individual's self-control ability unfolds predominantly when poor personal morals enable the perception of crime as a viable action alternative.

By far the largest environmental influence emerges for youths who combine low self-control ability with weak personal morality. Here, for each additional standard deviation of lifestyle risk the predicted crime variety score increases by nearly 0.5 offenses.

Conclusions

Based on survey data from adolescents in Slovenia, the present study examines whether an individual's predisposition for crime conditions the relevance of his or her lifestyle risk for the extent of offending. The results harmonize with the research-guiding hypotheses. Personal disposition moderates the significance of the microenvironment. Criminogenic exposure matters particularly for people characterized by an elevated propensity for crime. The effect of lifestyle risk on self-reported offending increases with the level of criminal propensity. This interaction relationship holds for both personal morality and self-control ability. The latter implies that both components of the individual's propensity for crime contribute to the explanation of the size of the lifestyle risk effect. Personal morality and self-control ability affect the role an adolescent's lifestyle-induced exposure risk plays in crime causation.

A more nuanced analysis of the interplay of the investigated concepts provides evidence of a three-way interaction: self-control ability attenuates the impact of a criminogenic way of life chiefly among individuals of weaker morality. The protective self-control effect is greatest among adolescents with few moral objections against crime, (presumably) because they seriously consider crime for action. Among youths of strong morality, the capacity for self-control does not shape the magnitude of the lifestyle risk effect, (probably) because they do not view crime as an acceptable, personally selectable option. Taken together, these findings suggest that personal morality constitutes a first and self-control ability a second line of defense against criminogenic environmental influences (Hirtenlehner et al., 2022). Self-control becomes relevant solely under conditions of a crime-permeable moral filter (Hirtenlehner and Leitgoeb, 2021). Such an interplay is consistent with SAT's reflections on the moral filtering of action alternatives (Wikström et al., 2012).

More broadly, these results corroborate that the significance of crime-facilitating environmental conditions is contingent on personal characteristics (Wikström et al., 2024). Individuals who face the same setting differ in their likelihood to respond with criminal behavior, depending on somewhat stable intraindividual properties (e.g., personal morality and self-control ability). Admittedly, here this conclusion must be drawn with caution: in deviation from a truly situational analysis, our study uses the individual-level variable ‘lifestyle risk’ as proxy measure of (the extent of) criminogenic exposure. Such a procedure rests on the assumption that people with a specific lifestyle more frequently encounter crime-conducive settings. Examining situational processes with data collected at the individual level also relies on the auxiliary assumption that a person's self-reported offenses were committed when he or she was exposed to criminogenic microenvironments. Analyses of space–time budget data lend credence to this supposition (Kennedy, 2024; Wikström et al., 2010, 2012, 2018). In line with these presumptions, we found the lifestyle risk measure to be predictive of criminal conduct and observed the highest level of criminal activity for respondents who combine a heightened crime propensity with an elevated lifestyle risk. The latter is exactly what we would expect to see, were the pertinent situational processes operating as conjectured. Such a convergence is obviously compatible with the PEA hypothesis. Nonetheless, to exclude ecological fallacies, more situational-level research on the spatiotemporal configuration of individuals, settings and actions is definitely needed (Hardie, 2020; Wikström and Kroneberg, 2022).

Aside from the individual-level nature of the employed data, the cross-sectional format of the study might be seen critically, because it implies problems with the temporal ordering of the constructs. The response variable (criminal activity in the last 12 months) dates before some of the pivotal predictor variables (personal morality and self-control ability), which were assessed at the time of the survey. This fact challenges causal interpretations of the findings. However, in accordance with Wikström et al. (2024: 157) we assume that ‘[c]rime involvement in the previous year reflects current crime involvement’ and that personal morality and self-control ability are somewhat stable characteristics that change only slowly. On the other hand, the temporal ordering of our measures of self-reported offending and lifestyle risk is less clear-cut. All the indicators used to determine the respondent's lifestyle risk also refer to past activities. Here, the ‘chicken-or-the-egg’ question remains unsolved.

Our parsimonious use of control variables can be warranted in theoretical and methodological terms. From the perspective of SAT, propensity and exposure as well as their interaction are the only genuine causes of criminal conduct (Wikström et al., 2012). Recent methodological research recommends including only theoretically justified predictor variables in regression models because controlling for the ‘wrong’ third variables can result in effect estimates that are more biased than uncontrolled ones (Wysocki et al., 2022). Thereby, the amount of bias depends partly on the true causal structure between the involved variables. In addition, McClelland and Judd (1993) have shown that nonexperimental survey research often suffers from difficulties in detecting interaction effects. These problems largely arise from measurement errors and scant variance in the variables involved. They are enhanced by the fact that the theoretical expectations usually only refer to ordinal interactions. The low power to establish interactions in observational studies suggests refraining from including many control variables into regression equations aimed at testing moderation relationships.

In this context, a short remark on statistical power issues seems appropriate. A study's power—the probability of rejecting a null hypothesis that is false—generally depends on the sample size, the accepted Type I error and the magnitude of the true effect in the population (Cohen, 1988). Here, low power comes down to a limited ability to establish the significance of the tested moderation relationships. In order to achieve a high probability of detecting interaction effects of the (low) strength typically observed in the social sciences, large samples are necessary (Aiken and West, 1991). Wikström and Kroneberg (2022: 186) argue that ‘one needs at least four times the sample size to estimate an interaction effect with the same accuracy as a main effect.’ In our case, 400 respondents sufficed to substantiate the hypothesized moderation relationships. However, a sample size like this equals the minimum value to obtain a statistical power of .80 (at α = .05) in the absence of any measurement error in the independent variables (Cohen, 1988). When reliability drops, the power goes down. Given imperfect reliability of the predictors, the demands on sample size increase dramatically (Aiken and West, 1991). Therefore, future studies seeking to test the interaction effects proposed by SAT will be well-advised to either draw on much larger samples or at least rely on an oversampling of extreme cases—that is, people with very low and very high levels of crime propensity and criminogenic exposure (McClelland and Judd, 1993).

Summating personal morality and self-control ability to a composite measure of criminal propensity—although only correlated with .15—needs justification. This procedure follows a common practice (Gerstner and Oberwittler, 2018; Hirtenlehner and Treiber, 2017; Kokkalera et al., 2023; Kokoravec Povh et al., 2024; Walters, 2020; Wikström et al., 2012, 2018) that can be warranted in theoretical terms: morality and self-control ability constitute the key elements of an individual's propensity to offend (Wikström, 2010). However, in light of SAT's conceptualization of self-control ability as an individual's capacity to withstand external inducements to breach internalized rules of conduct, totaling it with personal morality to an aggregate measure of crime propensity is not unproblematic (Pauwels et al., 2018). When a person possesses a crime-encouraging morality, a well-developed capability to exert self-control could increase his or her inclination to offend (Kroneberg and Schulz, 2018). Wikström et al.'s (2024: 60) counter this argument by pointing out that even under conditions of an antisocial personal morality, a strong self-control ability promotes an elevated deterrence sensitivity, which would then reduce the individual's tendency to offend. Here, the additive merger of the two concepts works because the employed measurement of self-control ability closely resembles Grasmick et al.'s (1993) attitudinal self-control scale designed to capture the construct in the sense of Gottfredson and Hirschi (1990). The present operationalization of the capacity for self-control focuses (inversely) on the individual's readiness for temperamental-impulsive risk-taking, a trait that doubtlessly fosters criminal action. This type of self-control proved to be most effective in neutralizing criminogenic setting influences among respondents of weak morality—a pattern of results allowing the construction of a summative score.

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.

ORCID iDs

Helmut Hirtenlehner

Gorazd Mesko

Notes

Appendix 1: Negative binomial models with product terms.

	Model 1 (full sample)			Model 2 (full sample)			Model 3a (“weaker morality” group)			Model 3b ('stronger morality” group)
	B	S.E.	p	B	S.E.	p	B	S.E.	p	B	S.E.	p
Lifestyle risk	+0.92	0.16	.000	+0.81	0.17	.000	+0.74	0.19	.000	+0.96	0.24	.000
Crime propensity
Weak personal morality				+0.41	0.13	.001	—	—	—	—	—	—
Low self-control ability	+0.49	0.12	.000				+0.41	0.14	.002	+0.47	0.20	.008
Lifestyle risk*crime propensity
Lifestyle risk*morality				−0.05	0.09	.274^#
Lifestyle risk*self-control	−0.13	0.08	.058^#				−0.05	0.11	.325^#	−0.41	0.17	.008^#
Model fit	n = 409; x² = 61.44; d.f. = 3; p = .000			n = 409; x² = 78.46; d.f. = 3; p = .000			n = 181; x² = 56.91; d.f. = 3; p = .000			n = 228; x² = 19.37; d.f. = 3; p = .000

B: unstandardized regression coefficient; S.E.: standard error; p: p-value.

The coefficient must be considered insignificant because its sign is at odds with the theoretical expectations.

References

Aiken

West

(1991) Multiple Regression: Testing and Interpreting Interactions. London: Sage.

Barton-Crosby

Hirtenlehner

(2021) The role of morality and self-control in conditioning the criminogenic effect of provocation. A partial test of situational action theory. Deviant Behavior 42(10): 1273–1294.

Beier

(2018) Situational peer effects on adolescents’ alcohol consumption: The moderating role of supervision, activity structure, and personal moral rules. Deviant Behavior 39(3): 363–379.

Bendixen

Endresen

Olweus

(2003) Variety and frequency scales of antisocial involvement: Which one is better? Legal and Criminological Psychology 8(2): 135–150.

Berry

DeMeritt

Esarey

(2010) Testing for interaction in binary logit and probit models: Is a product term essential? American Journal of Political Science 54(1): 248–266.

Bowen

(2012) Testing moderating hypotheses in limited dependent variable and other nonlinear models: Secondary versus total interactions. Journal of Management 38(3): 860–889.

Brauer

Tittle

(2017) When crime is not an option: Inspecting the moral filtering of criminal action alternatives. Justice Quarterly 34(5): 818–846.

Cohen

(1988) Statistical Power Analysis for the Behavioral Sciences. Hillsdale: LEA.

Craig

(2019) Extending situational action theory to white-collar crime. Deviant Behavior 40(2): 171–186.

10.

Gerstner

Oberwittler

(2018) Who’s hanging out and what’s happening? A look at the interplay between unstructured socializing, crime propensity and delinquent peers using social network data. European Journal of Criminology 15(1): 111–129.

11.

Gottfredson

Hirschi

(1990) A General Theory of Crime. Stanford: Stanford University Press.

12.

Grasmick

Tittle

Bursik

, et al. (1993) Testing the core empirical implications of Gottfredson and Hirschi’s general theory of crime. Journal of Research in Crime and Delinquency 30(1): 5–29.

13.

Haar

Wikström

P-O

(2010) Crime propensity, criminogenic exposure and violent scenario responses: Testing situational action theory in regression and Rasch models. European Journal of Applied Mathematics 21(4): 307–323.

14.

Hardie

(2019) Why monitoring doesn’t always matter: The interaction of personal propensity with physical and psychological parental presence in a situational explanation of adolescent offending. Deviant Behavior 42(3): 329–352.

15.

Hardie

(2020) Studying Situational Interaction. Cham: Springer.

16.

Hay

Forrest

(2008) Self-control theory and the concept of opportunity: The case for a more systematic union. Criminology; An Interdisciplinary Journal 46(4): 1039–1072.

17.

Hilbe

(2011) Negative Binomial Regression. Cambridge: Cambridge University Press.

18.

Hirtenlehner

(2020) The interaction between self-control and perceived sanction risk. An analysis from the viewpoint of different theories”. Criminal Justice Review 45(1): 104–128.

19.

Hirtenlehner

Kunz

(2016) The interaction between self-control and morality in crime causation among older adults. European Journal of Criminology 13(3): 393–409.

20.

Hirtenlehner

Leitgoeb

(2021) Differential self-control effects: Moral filtering and the subsidiary relevance of self-control. International Criminology 1(2): 91–106.

21.

Hirtenlehner

Treiber

(2017) Can situational action theory explain the gender gap in adolescent shoplifting? Results from Austria. International Criminal Justice Review 27(3): 165–187.

22.

Hirtenlehner

Bacher

Leitgoeb

, et al. (2022) Do morality and self-control protect from criminogenic peer influence? Testing multidimensional person-environment interactions. Justice Quarterly 39(1): 78–112.

23.

Hoeben

Meldrum

Walker

, et al. (2016) The role of peer delinquency and unstructured socializing in explaining delinquency and drug use: A state-of-the-art review. Journal of Criminal Justice 47(1): 108–122.

24.

Ivert

Andersson

Svensson

, et al. (2018) An examination of the interaction between morality and self-control in offending: A study of differences between girls and boys. Criminal Behaviour and Mental Health 28(3): 282–294.

25.

Kabiri

Shadmanfaat

Perkins

, et al. (2023) Aggressive spectators in sporting milieus: A test of situational action theory. Global Crime 24(3): 195–214.

26.

Kabiri

Shadmanfaat

Samuels-Wortley

, et al. (2020) Does moral identity matter in situational action theory? Some evidence of Iranian fans’ cyberbullying perpetration. International Criminal Justice Review 30(4): 406–420.

27.

Kennedy

(2024) Prioritise propensity: A multi-method analysis of peer influence and school-based aggression. Deviant Behavior 45(2): 139–168.

28.

Kokkalera

Marshall

(2020) How exceptional is India? A test of situational action theory. Asian Journal of Criminology 15(3): 195–218.

29.

Kokkalera

Marshall

(2023) Does national moral context make a difference? A comparative test of situational action theory. International Journal of Comparative and Applied Criminal Justice 47(3): 235–253.

30.

Kokoravec Povh

Marshal

Mesko

(2024) Comparing rural and urban juvenile delinquency in Slovenia: Testing for situational action theory. Victims & Offenders (online first). https://doi.org/10.1080/15564886.2024.2387003.

31.

Kroneberg

Nägel

(2024) When violence is not an option: Perceived choice sets and differential deterrability among adolescents in Germany. Crime & Delinquency (online first). https://doi.org/10.1177/00111287241258730.

32.

Kroneberg

Schulz

(2018) Revisiting the role of self-control in situational action theory. European Journal of Criminology 15(1): 56–76.

33.

McCabe

Halvorson

King

, et al. (2022) Interpreting interaction effects in generalized linear models of nonlinear probabilities and counts. Multivariate Behavioral Research 57(2): 243–263.

34.

McClelland

Judd

(1993) Statistical difficulties of detecting interactions and moderator effects. Psychological Bulletin 114(2): 376–390.

35.

Mears

Ploeger

Warr

(1998) Explaining the gender gap in delinquency: Peer group influence and moral evaluations of behavior. Journal of Research in Crime and Delinquency 35(3): 251–266.

36.

Meier

Slutske

Arndt

, et al. (2008) Impulsive and callous traits are more strongly associated with delinquent behavior in high risk neighborhoods among boys and girls. Journal of Abnormal Psychology 117(2): 377–385.

37.

Mize

(2019) Best practices for estimating, interpreting, and presenting nonlinear interaction effects. Sociological Science 6(1): 81–117.

38.

Moffitt

Caspi

Rutter

, et al. (2001) Sex Differences in Antisocial Behaviour. Cambridge: Cambridge University Press.

39.

Mood

(2010) Logistic regression: Why we cannot do what we think we can do, and what we can do about it. European Sociological Review 26(1): 67–82.

40.

Osgood

Finken

McMorris

(2002) Analyzing multiple-item measures of crime and deviance II: Tobit regression analysis of transformed scores. Journal of Quantitative Criminology 18(4): 319–347.

41.

Paternoster

Brame

Mazerolle

, et al. (1998) Using the correct statistical test for the equality of regression coefficients. Criminology; An Interdisciplinary Journal 36(4): 859–866.

42.

Pauwels

(2018a) Analysing the perception-choice process in situational action theory. A randomized scenario study. European Journal of Criminology 15(1): 130–147.

43.

Pauwels

(2018b) The conditional effects of self-control in situational action theory. A preliminary test in a randomized scenario study. Deviant Behavior 39(11): 1450–1466.

44.

Pauwels

Svensson

Hirtenlehner

(2018) Testing situational action theory: A narrative review of studies published between 2006 and 2015. European Journal of Criminology 15(1): 32–55.

45.

Steffensmeier

Allan

(1996) Gender and crime: Toward a gendered theory of female offending. Annual Review of Sociology 22(1): 459–487.

46.

Svensson

Pauwels

(2010) Is a risky lifestyle always “risky”? The interaction between individual propensity and lifestyle risk in adolescent offending. Crime & Delinquency 56(4): 608–626.

47.

Svensson

Pauwels

Weerman

(2010) Does the effect of self-control on adolescent offending vary by level of morality? A test in three countries. Criminal Justice and Behavior 37(6): 732–743.

48.

Sweeten

(2012) Scaling criminal offending. Journal of Quantitative Criminology 28(3): 533–557.

49.

Treiber

(2017) Biosocial criminology and models of criminal decision making. In: Bernasco

van Gelder

J-L

Elffers

(eds) The Oxford Handbook of Offender Decision Making. Oxford: Oxford University Press, pp. 87–120.

50.

Walters

(2020) Criminal thinking as a moderator of the perceived certainty – offending relationship. Psychology, Crime & Law 26(3): 267–286.

51.

Wikström

P-O

(2006) Individuals, settings and acts of crime. Situational mechanisms and the explanation of crime. In: Wikström

P-O

Sampson

(eds) The Explanation of Crime: Context, Mechanisms and Development. Cambridge: Cambridge University Press, pp. 61–107.

52.

Wikström

P-O

(2010) Explaining crime as moral action. In: Hitlin

Vaisey

(eds) Handbook of the Sociology of Morality. New York: Springer, pp. 211–240.

53.

Wikström

P-O

Butterworth

(2011) Adolescent Crime. New York: Routledge.

54.

Wikström

P-O

Kroneberg

(2022) Analytic criminology: Mechanisms and methods in the explanation of crime and its causes. Annual Review of Criminology 5(1): 179–203.

55.

Wikström

P-O

Svensson

(2008) Why are English youths more violent than Swedish youths? A comparative study of the role of crime propensity, lifestyles and their interactions. European Journal of Criminology 5: 309–330.

56.

Wikström

P-O

Svensson

(2010) When does self-control matter? The interaction between morality and self-control in crime causation. European Journal of Criminology 7(5): 395–410.

57.

Wikström

P-O

Treiber

(2016) Situational theory: The importance of interactions and action mechanisms in the explanation of crime. In: Piquero

(ed) The Handbook of Criminological Theory. Chichester: Wiley, pp. 415–444.

58.

Wikström

P-O

Ceccato

Hardie

, et al. (2010) Activity fields and the dynamics of crime. Advancing knowledge about the role of the environment in crime causation. Journal of Quantitative Criminology 26(1): 55–87.

59.

Wikström

P-O

Mann

Hardie

(2018) Young people’s differential vulnerability to criminogenic exposure: Bridging the gap between people- and place-oriented approaches in the study of crime causation. European Journal of Criminology 15(1): 10–31.

60.

Wikström

P-O

Oberwittler

Treiber

, et al. (2012) Breaking Rules. The Social and Situational Dynamics of Young People’s Urban Crime. Oxford: Oxford University Press.

61.

Wikström

P-O

Treiber

Roman

(2024) Character, Circumstances, and Criminal Careers. Towards a Dynamic Developmental and Life-Course Criminology. Oxford: Oxford University Press.

62.

Williams

(2012) Using the margins command to estimate and interpret adjusted predictions and marginal effects. Stata Journal 12(2): 308–331.

63.

Wright

Caspi

Moffitt

, et al. (2001) The effects of social ties on crime vary by criminal propensity: A life-course model of interdependence. Criminology; An Interdisciplinary Journal 39(2): 321–351.

64.

Wysocki

Lawson

Rhemtulla

(2022) Statistical control requires causal justification. Advances in Methods and Practices in Psychological Science 5(2): 1–19.

65.

Zimmerman

Botchkovar

Antonaccio

, et al. (2015) Low self-control in ‘bad’ neighborhoods: Assessing the role of context on the relationship between self-control and crime. Justice Quarterly 32(1): 56–84.