Binge Drinking Among College Athletes and Non-Athletes

Abstract

Concerns about incidence, forms, and consequences of alcohol use among college students lack examinations of the lifestyles and predictors of college student athletes. This article, using a sample of student-athletes and non-athletes from four Southern universities, identifies the lifestyle predictors for each population, identifying patterns and sets of predictors of binge drinking behavior. Findings indicate that for both samples, binge drinking behavior is explained by sex, drinking in bars, number of male friends who drink, and cigarette smoking. Student-athletes' binge drinking is explained further by residing on campus. Non-athlete binge drinking is related to race and amount of study time per week. Implications for these findings are also discussed.

Notes

¹Note: the percentages do not add to 100 percent for any of the demographics due to missing data.

∗α ≤ .15 (variable carried forward).

χ² = 29.289 (α ≤ .00).

n = 98.

df = 4.

∗p < .05.

χ² = 121.755 (α ≤ .00).

n = 453.

df = 6.

∗p < .05.

²Factor analysis is another frequently used method for variable reduction. We do not use it here because factor analysis is only appropriate for ratio or interval variables, and our data is primarily nominal and ordinal. Further, factor analysis is not an efficient method of data reduction when the categories of the variables are not evenly distributed (SPSS, Inc. 1997); this is the case with our data.

³The equation, χ¹ − χ²; df1 − df2 can be used to evaluate which of two models using the same sample is the better one (SAS 1990; Lottes, Adler, and DeMaris 1996).

⁴Several statistics are presented in the regression tables to follow. The chi-square statistic provides an indication of the overall fit of the data to the model. A significant chi-square indicates that the variables, as a group, contribute significantly to the dependent variable. In addition, the tables report the logistic coefficients, and their standard errors (S.E.). The logistic coefficient (b) can be interpreted as the change in log odds for a one-unit change in the predictor. We also report Exp (b), which is an odds ratio. Also, the z-score measuring any significant difference in means for the groups analyzed is reported. This measure indicates which variables are significantly different across their values for the independent variables for both groups examined. Finally, each variable's tolerance is reported. This is a statistic that tests for multicollinearity among the independent variables in a model. Tolerances of more than .600 indicate no serious problem of collinearity. Variables with tolerances of less than .2 is cause for concern, and variables with tolerances of less than .100 almost certainly indicates a serious collinearity problem (Menard 1995). Variables are regarded as significant if α ≤ .05 using a two-tailed test.

⁵Paternoster et al. (1998) presented the z-test for difference as the proper test to examine the difference between regression coefficients that use maximum-likelihood estimation. The formula used to calculate the z-test is as follows: z = b1 − b2/square root of se1 + square root of se2. In this model, b1 is the logistic coefficient from the first measure, and the b2 is the logistic coefficient from the second measure. The square root of the seb1 is the standard error for the first slope coefficient and the square root of seb2 is the standard error for the second logistic coefficient.

⁶In Table 2, there are 6 variables with unstandardized regression coefficients of greater than 2. These variables, however, did not have tolerances that might indicate any multicollinearity. As such, we proceeded with the analysis. In the final model, the unstandardized coefficients are acceptable. Interestingly, one variable, membership in a greek social organization, has an extremely high unstandardized coefficient and an even higher standard error of the estimate. This indicates that there is great imprecision with this variable for the athlete sample. This might be expected for this particular variable and that particular sample, because the number of athletes who are also members of greek social organizations is quite small (n = 6). In any case, this variable was not a significant predictor of binge drinking among athletes or non-athletes so the variable was not in the final analysis.