Religious Background and Gambling Among Young Adults in the United States

We also hypothesized that persons would be more likely to gamble when living in areas with high concentrations of Catholics, who occasionally sponsor low stakes games of chance (Diaz 2000). We found no evidence of this in our ancillary analysis and thus do not pursue this line of argument here.

We also conducted exploratory analyses with each of the three types of gambling as a separate dependent variable. The results were substantively similar for our religion predictor variables across the models. Religious predictors were slightly more effective at explaining lottery gambling then casino or other types of gambling.

In exploratory analyses we also examined money lost with an ordered logistic regression (predicting the full range of the variable) and found that there was little significant variation in money loss above $500, at least as predicted by our religion measures.

In exploratory analyses we also combined the five problem gambling measures described above into an index measure of problem gambling (alpha = .79; tetrachoric = .89). This index was similar to measures of problem gambling used in other studies (Hoffmann 2000). We experimented with a number of modeling techniques, including count-based (poisson, negative binomial) and ordered logistic regressions. In each case the results were heavily driven by the variable for money lost. Ultimately, we felt the index measure obscured important variation among the problem gambling measures.

We removed the black Protestant category from RELTRAD to prevent potential collinearity issues with the race controls and because we think that preserving race as an analytical category is more important to this analysis than any advantages which would be gained by using the black Protestant category.

Unfortunately, the items we use in our thrill-seeking index are not included in Wave I of Add Health; they only appear in Wave III. Because we use this measure to tap risk-propensity as an underlying personality characteristic we believe our use of this Wave III variable is reasonable. Wave III also includes a shorter list of items to measure risk-propensity but this set of items was asked only of a small subset of Wave III participants.

The percentage of conservative Protestants in county only predicts a small amount of the (considerable) county level variation in the dependent variable (results not shown; available upon request). This large amount of variation is likely due to differential availability of gambling from one county to another. Unfortunately, Add Health provides only very crude measures of region (South, West, Northeast, Midwest) so we were unable to incorporate regional controls, even at the state level. In ancillary analyses using the available measures of region as level 2 predictors, we found no significant effect of region. We also experimented with county-level measures of socio-economic status and these too proved poor predictors. We expect that much of the county-level variation could be explained with measures tapping the availability of legal gambling in the area (such as the proximity to casino measure used by Hoffmann (2000) or a measure of proximity to a state with a lottery). Unfortunately, such measures were not available to us in Add Health.