3+ DUI categories were too small to permit analysis the categories were combined to form a 2+ DUI category. Interactions of gender with each variable were tested. When there was a trend for a gender interaction (p≤.10), results were reparameterized using an equivalent model (i.e., same number of parameters) and were reported separately by gender to enable direct comparison of risk ratios for males and females. Differences between the 1, 2, and 3+ DUI groups were determined using post-hoc Wald tests, the results of which are reported in tables. We used a p-value of .05 for all tests of significance except for interaction terms, as noted above. All regressions were corrected for familial clustering using the Huber–White robust variance estimator in the Stata statistical program (Stata Corporation, 2004).
