This analysis would follow the design shown in Table 1—that is, a comparison of changes in turnover in selected groups of states pre- and post- HIPAA with the change in the control states in group D. Each monthly CPS provides data to measure job change in the past week. This is a rare event, and so very large sample sizes are needed to detect a difference over time. Consequently, this proposed analysis would not be useful in detecting small effects of HIPAA on turnover. However, by pooling data for each monthly CPS in a year to obtain both the pre- and post HIPAA measures, we estimate that the CPS does provide moderate power to detect differences within the range suggested by the literature. Some estimates of the effect of insurance on turnover have placed the estimate at a 25 percent increase. The sample obtained by pooling all CPS surveys in a year prior to and post-HIPAA would have about 65 percent power to detect an effect of this magnitude for state groups B or C (that is, a 25 percent increase in turnover in state group B or C relative to a constant rate in group D).
Ideally one would look at turnover among those in insured jobs. Unfortunately, the CPS does not collect information on insurance every month. Therefore, the evaluation would have to focus on turnover among all employees. Consequently, a 25 percent increase in turnover is probably much larger than one might reasonably expect. As we have noted elsewhere, one can improve power to detect smaller effects by pooling observations over multiple years in the CPS. Using two years in the pre- and post period, one could detect a 20 percent increase in turnover (relative to the control states) with about 65 percent power.