ABSTRACT
This study aims to assess the longitudinal patterns of multifaceted religious profiles and their relationships with illegal substance abuse among young people transitioning from late adolescence to early adulthood. A novel longitudinal approach integrating the cutting-edge unsupervised and supervised learning techniques is proposed to analyze the data from the National Longitudinal Survey of Youth 1997. The results show that emerging adults who are highly religious in either subjective (e.g., religious beliefs) or objective (e.g., religious attendance) domain are much less likely to abuse illegal substances than their religiously disengaged peers. Religiosity, regardless of subjective or objective, tends to be protective, but its effect is most prominent among young people most profoundly devoted to both religious beliefs and behaviors. Nevertheless, possessing strong commitment to religious beliefs without accompanying frequent religious behaviors may put emerging adults at greater risk for illicit substance abuse, compared to those who hold high level of religious beliefs but do not engage in corresponding religious behaviors frequently.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 To account for missing data, we also conducted mean and multiple imputation to re-estimate the models. Overall, the results are substantively similar to those using listwise deletion presented in the study. Full results are available upon request.
2 In the unchanged group, we observed that 2,299 adults remained unmarried, while 215 adults remained married. We conducted supplemental analyses by separating the unchanged group into two subgroups: stay married and stay unmarried and including them in the model. The results are substantially consistent with those presented in the study, which suggest that the stability of marital status may not significantly impact the risk of substance abuse.
3 The reported findings lack adjustments for weighting, clustering, and missing data, but we acknowledge the presence of these issues within the dataset. To address these concerns, we reanalyzed the models by incorporating customized longitudinal weights provided by the NLSY, which account for selection probability, nonresponse, and oversampling. Additionally, we considered the clustering effect by household, as the data includes all eligible youth within each household. Furthermore, we conducted additional analyses using only the cross-sectional sample (excluding the oversample of Blacks and Hispanics) while still accounting for household clustering. As discussed above, we re-estimated the models by employing mean and multiple imputations to address the missing values. All the findings are consistent with the results presented in the study.
Additional information
Funding
Notes on contributors
Siying Guo
Siying Guo is an Assistant Professor in the Department of Criminology and Criminal Justice at Wayne State University. Her research focuses on cybercrime, juvenile delinquency and justice, religion and crime, and program evaluation.
Jianxuan Liu
Jianxuan Liu is an Assistant Professor of Statistics and a Research Affiliate at the Center for Policy Research at Syracuse University. Her research focuses on causal inference, complex high-dimensional data, measurement errors, and semiparametric modeling.
Chen Meng
Chen Meng is an Assistant Professor in Economics at Kean University and Research Affiliate at Boston University and UC Berkeley. Her research focuses on Applied Microeconomics with causal inference methods, with specialization in Labor, Education, Law and Economics, and Health Economics.
Hyejoon Park
Hyejoon Park is an Associate professor at the School of Social Work at Western Michigan University. Her primary research areas include maternal/women health, child health,community/public health, and evaluating program services (e.g., mentoring, childcare, and afterschool programs). The secondary areas are coping mechanisms and racial and health disparities.