Images of a Loving God and Sense of Meaning in Life

Recent analyses indicate that the precision of parameter estimates using sample data are minimally related to response rates (American Association for Public Opinion Research 2008; Singer 2006). Nonetheless, the BRS response rate is within the normal parameters for recent RDD samples.

In order to more precisely locate respondents within religious traditions, the affiliation of the respondent’s congregation was also taken into account (see Dougherty et al. 2007).

SAS computes standardized coefficients for logistic regression using the logit distribution’s standard deviation (π²/3). Although interpreting coefficients for standardized dummy variables is difficult, dummy variable standardized coefficients are useful in comparing their effects to those of continuous predictors (see Pampel 2000, pp. 72–73).

To calculate the percent change in odds for those measures with negative odds ratios, due to their being bounded between 0 and 1, we divide 1 by each ratio. Therefore, Mainline Protestants/Catholics = 1/0.67 = 1.49; unaffiliated = 1/0.46 = 2.17.

The r-square in model 3 is on par with other studies in the images of God research literature (see e.g., Driskell et al. 2008; Froese and Bader 2008; Mencken et al. 2009; Unnever et al. 2010).

As an indication of model 3’s performance, we provide the hit ratio ((correct_event + correct_non-event)/N × 100 = 70.9%).

The association between loving God and sense of purpose is also robust to the inclusion of other potentially relevant factors. Lim and Putnam (2010) find subjective religiosity predictive of life satisfaction and Froese and Bader (2007) find authoritative God images (α = 0.85) and engaged God images (α = 0.88) predictive of a variety of outcomes. We control for these measures in ancillary models. We also control for an index of volunteering (α = 0.65), index of church activities (α = 0.70), index of community friendships (α = 0.54), and friendships at work. None of these predictors reach statistical significance in the final model, nor do they meaningfully change the findings presented here.

Predicted probabilities were calculated by setting age, prayer, and congregational friends to their means and then solving for the loving image of God composite measure.

We are grateful to a reviewer for pointing out that with a relatively large number of categorical parameters and some of the religion variables having moderate correlations with the loving God index, it is possible that a problematic influence of multicollinearity may be present. While the bivariate correlations between variables do not exceed the classic cut-off point of 0.70, another way of assessing multicollinearity in binary logistic models is to inspect variance inflation factor scores from an ordinary least squares model (Allison (2001). Using Allison's strategy, we find that the largest VIF score in the model is 2.32646 for the church attendance variable. All remaining VIF scores are less than 1.88667. When attendance is removed from the model, the highest VIF becomes 1.78683, and the effect for loving images of God retains the largest standardized beta. Since such VIF scores are generally considered non-problematic, we summarize that no strong evidence for problems associated with multicollinearity was found. Although several of our religion control variables are correlated, we elect to retain them in our models for the sake of continuity with prior research on the effects of images of God (Bader and Froese 2005; Bader et al. 2010; Draper 2008; Froese and Bader 2007, 2008; Froese et al. 2008; Mencken et al. 2009; Unnever et al. 2010; Whitehead 2012, 2010). In each of these studies, measures of religious affiliation, behavior, and belief are included as controls to ensure that the association between the dependent variable and the image of God is robust.