by less than half the length of one confidence interval arm, then the p-value between the confidence intervals is at least below the level of significance (i.e., less than .05). Previous studies show this method to be sufficiently accurate when two conditions are met: 1. When sample sizes are at least 10, and 2. When the two intervals do not differ in width by more than a factor of 2 [30]. This applies to both dichotomous and continuous variables alike. All comparisons in this study met both assumptions. Further, our sample size is considerably large (in excess of 200,000). Thus, post-estimation F tests are more conservative than traditional group difference tests (e.g., t-tests and ANOVAs) and may protect against type 1 errors when analyze a larger sample [30]. It should be noted that this method has been used in other studies which utilized large samples [31, 32].
