https://orcid.org/0000-0003-2228-158X
![Sample our Law journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5942/TOC-Subject-Sample-2021-Law.jpg"})
ABSTRACT
Although prior research documents a positive relationship between religious involvement and emotional well-being among incarcerated individuals, the salutary effect of religion on mental health remains in need of scientific scrutiny. To examine this understudied issue, we hypothesized that prisoner religiosity is positively associated with virtues, which are in turn inversely related to negative emotions. To test this hypothesis, we applied structural equation modeling to analyze data from a survey with a convenience sample of 139 males housed at a Colombian prison. As hypothesized, we found that both public (religious service attendance) and private religious behaviors (praying and reading a sacred book) were positively associated with the virtues of forgiveness, self-control, and gratitude. Additionally, we found that forgiveness and self-control were inversely related to state anger, depression, and anxiety. Finally, self-control was central in explaining the relationship between religious service attendance and state anger. Implications and limitations of the present findings are discussed.
Acknowledgements
The authors are grateful to: Prison Fellowship International and its administration and staff (including Mr. David Van Patten and Ms. Rae Wood); Prison Fellowship Colombia (Confraternidad Carcelaria de Colombia) and its administration and staff (including Mr. Lácides Hernandez and Mr. Joshua Ramirez); the administration and staff at the Montería Prison (Carcel Las Mercedes) in Colombia; all of the volunteers who made this study possible (including Dr. Michael Joseph); and all participants in this study.
Disclosure statement
No potential conflict of interest was reported by the authors.
Supplementary material
Supplemental data for this article can be accessed online at https://doi.org/10.1080/0735648X.2023.2249439
Notes
1. One possible reason is a selection effect given that public religiosity may reflect, to some extent, extrinsic religiosity (i.e., participation in religious activities that serves other ultimate ends than religious conviction per se, such as access to local volunteers and snacks, exemption from daily chores, or protection from a potential danger from other inmates) compared to private religiosity, which is more likely to indicate intrinsic religiosity. As a result, we may find more empirical support for hypothesized relationships involving praying and reading a sacred book than service attendance.
2. Because our endogenous variables were ordinal variables, it was necessary to apply an estimation method specifically designed for categorical variables (Distefano & Finney, Citation2013). While a robust weighted least squares (WLS) – implemented as weighted least squares mean and variance adjusted (WLSMV) in Mplus – was an option, we chose the estimator of maximum likelihood (ML) with robust standard errors (called MLR in Mplus) because this robust ML was found to be better than the WLSMV method in terms of error-free convergence rates and estimating variance parameters and their standard errors (Bandalos, Citation2014; Newsom & Smith, Citation2020). Furthermore, MLR allows missing data with categorical variables to be better handled with FIML compared to the WLSMV method that uses a pairwise deletion (Muthén et al., Citation1997; Asparouhov & Muthén, Citation2010).
3. We examined correlations among variables to see whether there was any indication of collinearity problem. We found no such indication with the largest correlation being .575 between state depression and anxiety. The measures of religiosity were not correlated higher than .500: specifically, service attendance was correlated with praying and reading a sacred book at .405 and .311, respectively, whereas correlation between the measures of private religiosity was .495 (complete results are available upon request).
4. A supplemental analysis showed that praying privately was not significantly related to any of the virtues even when it was included without religious service attendance and reading a sacred book in the model.
5. While the virtues were expected to be positively associated with one another, we found that only forgiveness and gratitude were correlated in the expected direction (β = .236), and self-control was not related to either of them (β = ‒.038 and .120, p > .05, respectively).
6. When gratitude was included with no other virtues in the model, it was found to be inversely related only to state anger (b = ‒.140, S.E. = .080, p < .05, one-tailed test; not shown in the table). This may indicate the gratitude-anger relationship is indirect via either forgiveness or self-control, which was significantly related to state anger. For example, the more grateful, the more forgiving or self-controlled, and the less likely to feel angry. In addition, after virtues were added to the baseline model, the negative emotional states remained significant in their relationships, though their size somewhat reduced (β = .289, .197, and .589 compared to β = .337, .296, and .611).
7. Although religious service attendance and reading a sacred book were positively associated with forgiveness (b = .065 and .045, respectively), which was in turn inversely related to state anger (b = ‒.495), the mediation of forgiveness was only marginally significant (b = ‒.032 and ‒.022, S.E. = .022 and .015, respectively, both p < .10, one-tailed test; not presented in the table).
8. The three types of forgiveness were positively associated with one another. That is, survey respondents who forgave themselves for what they had done wrong were more likely to ask for forgiveness from those whom they had hurt and to forgive those who hurt them (β = .445 and .456, respectively; see underlined coefficients in Supplemental Table 1). Forgiving others who hurt them and asking others to forgive them were also positively related (β = .534).
9. The χ2 statistic of baseline (186.685, d.f. = 143, p < .05) and full models (281.279, d.f. = 219, p < .05) were both significant despite the relatively small sample size.
10. In addition, our finding public religiosity to be significantly related to negative emotional states via self-control, unlike private religiosity, implies that the possibility of service attendance reflecting extrinsic religiosity (i.e., a selection bias) was minimal (see note 1).
11. Having said that, although the items of negative emotions were “previous measures” (i.e., emotional states during the last one or two weeks prior to the survey) and those of religiosity and virtues were “current measures” (i.e., religious behaviors and traits at the time of survey), estimated relationships among the three are practically concurrent given that the religiosity and virtue items were likely to measure behaviors and traits “during the last one or two weeks prior to the survey” as well as at the very moment of survey administration. To the extent that this reasoning is valid, the trivariate relationships could be causally interpreted based on theories and prior research. For example, interpreting the religiosity-virtue relationship as religiosity fostering virtue seems more plausible than virtue leading to religiosity given that virtuous people are not necessarily religious. Relationships between virtues and negative emotional states are also more likely to indicate virtuous traits reducing negative emotional states rather than vice versa, while the religiosity-negative emotional state relationship is likely to be reciprocal as it may reflect psychological distress weakening religious behaviors as well as religiosity reducing negative emotional states.
12. Results from one of the three Monte Carlo studies are summarized in Supplemental Table 3. The first column gives parameter estimates presented in the bottom panel of , which were used as population parameter, whereas the next three columns show the average of the parameter estimates, the standard deviation of the parameter estimates, and the average of the estimated standard errors across the replications of the Monte Carlo studies, respectively. The “M.S.E.” column presents the mean square error for each parameter, and the column labeled “95% Cover” gives the proportion of replications for which the 95% confidence interval contains the population parameter value. An estimate of observed power is reported in the “% Sig Coeff” column, which shows the proportion of replications for which the null hypothesis that a parameter is equal to zero is rejected for each parameter at the .05 level (two-tailed test). The next two columns show the percentage of parameter and standard error biases, whereas the last four columns report the mean and standard deviation of “95% Cover” and “% Sig Coeff” values from the simulation studies with small standard deviations indicating the stability of results across the simulation studies. We found that the parameter and, to a lesser extent, standard error biases tended to not exceed the maximum percentage used to determine sample size in power analysis for prospective studies (10% and 5%, respectively) and that all the coverage (95% Cover) remained between .91 and .98 (Muthén, L. K. & Muthén, Citation2002). While these conditions were satisfied, the post hoc power of significant coefficients varied from .427 to .998, whereas that of non-significant coefficients ranged from .064 to .426, showing a functional relationship between p-value and observed power (i.e., the observed significance level of a test determining the observed power). Given this one-to-one relationship between p-value and observed power, some scholars argue that post hoc power analysis does not provide any new information or is even “fundamental flawed” (Hoenig & Heisey, Citation2001:19; Quach et al., Citation2022). Regardless, while not all the significant coefficients were observed to have power greater than a commonly accepted value for sufficient power (.800), overall results confirmed a suspicion that the non-significance of at least some coefficients might have been in part due to our small sample, implying a need for replication of the present study using a larger sample.
13. Despite the lack of applicability beyond the present sample, our hypotheses generally received empirical support, consistent with previous studies using different samples.