Means, standard deviations and correlations among the variables are displayed in . In order to ensure consistency between the bivariate and multivariate analyses, the statistics reported in this table are based on the list-wise deletion of observations with missing data on any of the control variables. T test analyses comparing the drinking behavior and absence of those dropped from the analyses (n=49) and those remaining (n=421) indicate no significant differences between the two groups (p>0.10). The bivariate results indicate a positive relationship between absenteeism and both drinking measures i.e., log of modal consumption (r=.13, p<0.01) and frequency of episodic heavy drinking (r=.24, p<0.01), with both drinking measures being, not surprisingly correlated (r=.47). It is also interesting to note the significant inverse correlation between supervisory support and days absent (r=−.11, p<.05).
The results of our multivariate analyses testing Hypotheses 1a, 1b, and 1c (which specified that the frequency of episodic heavy drinking: (a) is positively associated with the number of absence days, (b) is a more robust predictor of absence than modal consumption, and (c) explains a greater share of the variance in absence than modal alcohol consumption even when taking into account the possible curvilinear relationship of the latter with absence) are presented in Models 2-5 of . Although neither of the two modal consumption parameter estimates specified in Model 2 were statistically significant, it is notable that this model still explained 2 percent more variance than the control model and that the contrast analysis indicated that this difference is statistically significant (χ2df=2 = 9.39, p<.01). The lack of significant parameter estimates for the two modal consumption variables likely stems from the high correlation between the log of modal consumption and its squared term (r=0.92). Consequently, we ran an additional analysis specifying a linear effect only. In this model (Model 3), the log of modal alcohol consumption had a significant, positive effect on absenteeism (B=0.07, p<.01). Moreover, a contrast analysis indicates that the addition of a curvilinear effect for modal consumption fails to significantly add to the predictive utility of the model (χ2df=1 = 1.26, n.s.).
Negative Binomial Analyses Testing the Influence of Modal Alcohol Consumption and the Frequency of Episodic Heavy Drinking on Absenteeism, and the Moderating Effects of Peer & Supervisor Support (N=421)
We nevertheless tested Hypotheses 1a-c on the basis of the more conservative specification including a curvilinear effect for modal consumption. The results of these tests (displayed in Model 4) indicate that, as proposed in Hypothesis 1a, the frequency of episodic heavy drinking is positively associated with the number of absence days (B=.11, p<.01). Additionally, as proposed in Hypothesis 1b, even when contrasted against a strictly linear effect of modal consumption (Model 5), the effect of heavy drinking is of a larger magnitude (B=.12, p<.01) than that of the log of modal consumption (B=.02, n.s.). Finally, consistent with Hypothesis 1c, the model including the frequency of heavy drinking (Model 4) explains a significantly greater share of the variance in absence days (R2= .10) than that explained by modal consumption alone (Model 2 -- R2= .07) (ΔR2=.03, χ2df=1 = 10.21, p<0.01). Consequently, Hypotheses 1a - 1c were fully supported.
In order to test Hypotheses 2a (positing that the positive association between episodic heavy drinking and the number of days absent would be attenuated as a function of peer support) and 2b (positing that the positive association between episodic heavy drinking and the number of days absent would be amplified as a function of supervisor support), we first centered the two interaction terms, namely the three alcohol measures and two support measures (Aiken & West, 1991
). These interactions terms were then incorporated into the full model. As shown in Model 6 of , the generally positive association between episodic heavy drinking and the number of days absent was found to be attenuated as a function of peer support (B for the interaction= −.08, p<0.01) and amplified as a function of supervisor support (B for the interaction= .22, p<0.01). The inclusion of the interaction terms resulted in a further increase in the total effect size relative to Model 4 (ΔR2
= .04, χ2df=4
= 18.72, p<0.01). An expansion of this same model to include the interaction of peer and supervisor support with the centered log of modal alcohol consumption failed to explain a significantly greater degree of variance in absenteeism, and neither of the modal consumption interaction terms was significant.
To further examine the effect of peer and supervisor support on the link between the frequency of heavy episodic drinking and absenteeism, we graphically illustrated the interaction utilizing a procedure similar to the one recommended by Stone and Hollenbeck (1989)
. Specifically, we plotted three slopes of each of the two moderating variables (i.e., peer support, supervisory support): one at one standard deviation below the mean, one at the mean, and one at one standard deviation above the mean. The slopes presented in these graphs are not necessarily linear in that the regression model upon which they are based assume the log of the expected value of absenteeism.
As illustrates, while under conditions of low and average levels of peer support there is the expected positive association between heavy drinking and the number of days absent, under conditions of high peer support, the link between heavy drinking and absenteeism is largely invariant. In addition, as illustrates, while the highest level of absenteeism was obtained under conditions of a high level of heavy drinking and a high level of supervisor support, the generally positive association of heavy drinking and the number of days absent is attenuated and, indeed, reversed as a function of low levels of supervisor support.
The association between Heavy Drinking and Absenteeism as a Function of Peer Support: Curves for 3 Different Levels of the Moderator (−1 STD, mean, and +1 STD of peer support).
The association between Heavy Drinking and Absenteeism as a Function of Supervisor Support: Curves for 3 Different Levels of the Moderator (−1 STD, mean, and +1 STD of supervisor support).
Simple slopes analyses were conducted on the heavy drinking-absence relationship under 9 different conditions determined by the combination of varying levels of peer support and supervisory support (each at −1 SD below the mean, mean, and +1 SD above the mean). Consistent with our hypothesis, the effect of heavy drinking on absence is positive and significant under conditions of high (+1SD) supervisory support, regardless of the level of peer support (estimates of 0.23, 0.32 and 0.40, respectively for high, mean and low levels of peer support, all at a significance level of p<0.01). The effect of heavy drinking on absence is also significant at mean levels of supervisory support when the level of peer support is at and below mean levels (estimates of 0.13 and 0.22 respectively for mean and low levels of peer support, both at a significance level of p<0.01). However, consistent with our hypotheses, assuming a mean level of supervisory support for those perceiving a high level of peer support (+1SD), the effect of heavy drinking on absence (estimate = 0.03) is not statistically significant. Moreover, assuming a low level of supervisory support (− 1 S.D.), for those perceiving a high level of peer support (+1SD), the effect of heavy drinking on absence is negative (estimate = −.14), although not statistically significant (p=.08). Consequently, Hypotheses 2a and 2b were also supported.