RELIGIOSITY, PEERS, AND DELINQUENCY: DOES RELIGIOSITY REDUCE THE EFFECT OF PEERS ON DELINQUENCY?

Abstract

Do peer influences have the same effect on religious and nonreligious youth, or does religiosity reduce the effect of peers on delinquency? Using data from the National Youth Survey, we examined the interaction of religiosity and peer influences on marijuana use, alcohol use, hitting, and property offenses. The results suggest that, for marijuana use and alcohol use, three measures of peer influence—peer attitudes, behaviors, and pressure—have weaker effects on religious adolescents. Thus, even when religious youth are exposed to peers who encourage substance use, religiosity may serve as a protective factor that reduces the effect of peers. In contrast, religiosity does not seem to condition the effect of peers on hitting and property offenses.

Acknowledgments

This research uses data from Waves III and IV of the National Youth Survey (NYS). The NYS data were made available, in part, by the Inter-university Consortium for Political and Social Research (ICPSR). The data were originally collected by Delbert Elliott. Neither the collector of the original data nor the consortium bear any responsibility for the analyses or interpretations presented here.

Notes

¹We recognize a sample of BYU college students is hardly representative. We cite this study simply to illustrate the point, documented by additional studies (Burkett 1993; Burkett and Warren 1987; Simons et al. 2004), that religiosity influences adolescent peer networks.

²The difference between mediators and moderators is discussed by Baron and Kenny (1986) and Wu and Zumbo (2008). In short, mediation, or an indirect effect, occurs when a variable (the mediator or intervening variable) accounts for the relationship between an independent variable and a dependent variable (x influences y through z). With moderation, also called an interaction or conditional relationship, a variable (the moderator) influences the strength of the relationship between an independent variable and a dependent variable (the strength of the relationship between x and y depends on the level of z). Since previous research has already established that delinquent peers mediate the effect of religiosity on delinquency (Burkett 1993; Burkett and Warren 1987; Simons et al. 2004), we examine how religiosity might moderate the effect of peers on delinquency.

³In addition to social control, research also suggests that, compared to non-religious youth, religious adolescents have higher levels of self-control (Welch et al. 2006). When exposed to delinquent peer influences, the greater self-control of religious adolescents may help them to avoid the temptation to commit delinquent acts.

⁴For this project, we considered using more recent data sets. Although some of the other data sets we examined contained good measures of religiosity, the alternatives often did not include items that captured multiple dimensions of peer influence (i.e., behavior, attitudes, and pressure) and/or important control variables, in particular moral beliefs. Since peer influences are the focus of our paper, we elected to use the NYS, even though more recent data sets are available.

We do not think the age of the data influences the results. Research suggests religious beliefs and behaviors in the United States have remained relatively stable since the late 1970s. For example, Gallup polls indicate the percentage of people who claim they have attended church in the past seven days (Saad 2009) and that religion is “very important” (Newport 2006) to them has varied by only 5 percent since 1980. Furthermore, we have no reason to believe religious groups have softened their stance against adolescent drug use and delinquency. Therefore, we would expect the same results if our study was replicated using more recent data (see footnote 14).

⁵Although the NYS includes numerous items that can be used to measure peer influences, the data contains very few measures of religiosity. Unfortunately, the NYS does not include a measure of religious affiliation/denomination or parents' religiosity.

⁶According to Hirschi (1969), attachment refers to “sensitivity to the opinions of others” (i.e., do you care what other people think about you?). Attachment is often measured using items that assess caring, trust, communication, intimacy, and openness. Therefore, unlike the other measures of peer influence we use for the analysis (peer attitudes towards delinquency, association with delinquent peers, and peer pressure to use marijuana and alcohol), peer attachment provides no indication of the delinquency of peers.

⁷Given the longitudinal design of the NYS, we could also control for the effect of previous delinquency (i.e., a lagged dependent variable model). Controlling for the effect of previous delinquency, however, would change the interpretation of the results. If previous delinquency was included in the models, a significant effect would represent the influence of an independent variable (or an interaction term) on change in delinquency from wave three to wave four. In contrast, we would argue that adolescents' current religiosity should reduce the effect of their current friends on their immediate behavior, rather than how much their behavior changes over the next year. That is, the protective effects of religiosity are mostly contemporaneous. We would not expect static measures of religiosity, peer influence, and their interaction, measured at a single point in time, to be significantly related to a measure of change in delinquency over time. Therefore, since we think controlling for previous delinquency would be an overspecification, we did not control for previous delinquency in our models.

⁸Our measure of religiosity ranges from 0 to 8 with a mean of 4.45. For these comparisons, adolescents with a score of four or less on the religiosity index were classified as below average, while those with a score of five or higher were classified as above average.

⁹The correlations between religiosity and peer attitudes, peer associations, and peer pressure for marijuana use were .353, −.340, and −.345, respectively. Although all of the correlations are significant, they are modest in size. Again, this suggests the relationship between religiosity and peer influences is less than perfect. Religious youth are less likely to be exposed to peers who promote delinquency, but peer influences are not completely absent from their lives.

*p < .05. **p < .01.

¹⁰The items we use to measure delinquency are not normally distributed. Since measures of crime and delinquency are often skewed, alternatives to OLS regression are often used by criminologists. Unfortunately, testing interaction terms with a nonlinear model (e.g., logistic regression, ordered logistic regression, negative binomial, logit and probit models) creates statistical problems that can lead to faulty conclusions (Ai and Norton 2003; Allison 1999; Norton et al. 2004). In particular, with nonlinear models interaction terms will often “reverse” direction. In fact, when we replicated our analysis using negative binomial regression, compared to the OLS results, the interaction terms were completely opposite. After closely inspecting the data (i.e., a series of crosstabs), we determined the OLS results are correct. That is, the direction of the OLS interaction terms fit the pattern that is found in the data, whereas the results for the negative binomial do not. Therefore, although some might argue an alternative method should be used for the analysis, since we are testing interaction terms, we think OLS is a more trustworthy method.

¹¹When the peer variables are not included in the model, religiosity has a significant direct effect on marijuana use (coefficient = −.106, p = .000). The effect for religiosity on marijuana use is almost significant when moral beliefs are not included in the model (coefficient = −.059, p = .073). Therefore, since peer influences and moral beliefs mediate much of the effect of religiosity on marijuana use and other forms of delinquency, religiosity does not have a significant direct effect on marijuana use when peer influences and moral beliefs are included in the model.

¹²For the analysis we used unstandardized interactions. As expected, regression diagnostics indicated that multicollinearity was a problem. The variance inflation factors (VIFs) for models containing an unstandardized interaction ranged from 3.63 to 37.73. The VIFs for all models with an interaction term except one were above the threshold of 4.0 which is often used as an indicator of serious multicollinearity. Since using unstandardized interactions increases multicollinearity, thereby inflating the standard errors of the coefficients, it is harder to find significant interaction effects. In order to check our results, we replicated the analysis using standardized interactions. Standardizing the interactions eliminated any problems with multicollinearity. The pattern of results for the standardized interactions was the same as the results using the unstandardized interactions (reported in the text).

¹³As can be seen in the figure, for nonreligious youth who associate with friends who strongly approve of marijuana use the predicted value for marijuana use is between 2.5 and 3.0. For marijuana use, a score of 3 is equivalent to “once a month.” Rather than focusing on the predicted values, when discussing the figures we refer to the response format for the dependent variable in order to make the results more interpretable and meaningful.

*p < .05. **p < .01.

*p < .05. **p < .01.

*p < .05. **p < .01.

¹⁴We considered using more recent data sets for this project, but decided to use the NYS because it contains items that can be used to measure multiple dimensions of peer influence and better control variables (see footnote 4). While working on a separate project, we partially replicated the analysis using the first two waves of the National Longitudinal Study of Adolescent Health (Add Health), which were collected between 1994 and 1996. Consistent with our results, we found that adolescent religiosity, measured as a combination of church attendance, importance of religion, and frequency of prayer, significantly reduced the effect of associating with delinquent peers on both alcohol use and marijuana use. Unfortunately, we could not attempt to replicate the results for additional dimensions of peer influence (peer attitudes and peer pressure) or delinquent behaviors (hitting and property offenses) because the necessary items were not included in the Add Health data. Nonetheless, our partial replication using a more recent data set supports our argument that adolescent religiosity significantly reduces the effect of delinquent peers on delinquency and our results are not influenced by the age of the NYS data.

¹⁵In addition to marijuana use, alcohol use, hitting, and property offenses, we conducted a supplemental analysis using two additional items: cheating on exams and selling hard drugs. We did not make these items a central part of the analysis because cheating on exams is not a form of delinquency (it's not a legal issue) and very few adolescents report selling hard drugs. Nonetheless, the interaction for religiosity and peer attitudes was significant for cheating, but the interaction for religiosity and peer associations was not. The pattern was opposite for selling hard drugs. The interaction for religiosity and peer associations was significant, but the interaction between religiosity and peer attitudes was not. Although the results for cheating and selling hard drugs are not as consistent as the results for marijuana use and alcohol use, the supplemental results for cheating and selling hard drugs offer further support for the antiascetic hypothesis. That is, the protective effects of religiosity seem to be greater for so called “victimless” crimes.

¹⁶We are not able to measure the religiosity of friends directly, since questions about friends' religiosity are not included in the data. In making this argument, we are assuming that delinquent friends are less religious than non-delinquent friends. Given the extensive literature that shows religious youth are less likely to be delinquent, we think this is a safe assumption.

¹⁷We conducted a supplemental analysis for subgroups based on gender, race, and age. The results were not what we expected. For gender, five of the six interactions for marijuana use and alcohol use were significant for boys (all but religiosity and peer associations on alcohol use), whereas none of the six interactions were significant for girls. Five of the six interactions for marijuana use and alcohol use were significant for white youth (all but religiosity and peer associations on alcohol use), but only two were significant for nonwhite youth (religiosity by peer attitudes and peer pressure for alcohol use). None of the interactions for marijuana use and alcohol use were significant for adolescents younger than 16 years of age, while four of the six interactions were significant for adolescents 16 years of age or older (all but religiosity by peer associations for both marijuana use and alcohol use). Therefore, religiosity seems to protect boys, white youth, and older adolescents from the influence of peers more than girls, nonwhite youth, and younger adolescents.