Grant Miller, Diana Pinto and Marcos Vera-Hernandez present the results of an evaluation of publicly-financed health insurance for the poor in Colombia. The study is excellent and carefully conducted. However, the lack of statistical power that characterizes the study, particularly with respect to health outcomes, renders the study largely uninformative with respect to arguably the most important question of whether the 100% to 200% increase in the use of services that occurred was associated with better health. Standard errors in Table 4 for the four health outcomes are unable to detect reliably very large effect sizes. For example, the only statistically significant effect among the four health variables pertains to days lost to illness, and the standard error of the estimate is equal to the mean. So it is only because the effect size was over 200% of the mean that the estimate was statistically significant.

We very much appreciate these comments. We certainly agree that low power is a limitation of our study (which we raise on page 76). We also note that low power means that the relationships we are able to detect are likely to be true ones (and others which are statistically insignificant but suggested by our point estimates may be true relationships too). There are other nuances in interpretation as well. For example, given that our estimates identify treatment effects for the subset of compliers in the sample, full sample means may not be the only appropriate benchmark for assessing implied effect sizes. Comparisons with standard deviations as well as means are also reasonable. These are important issues for readers to bear in mind, and we are grateful to Bob for raising them.

- The Authors