Confidence Limits to the Risk Ratio

For the problem of obtaining confidence limits to the risk ratio, or the ratio of two binomial probabilities, Katz et al. (1978, Biometrics 34, 469-474) proposed a method based on the logarithm of the observed ratio and using Fieller's result. We propose a method based on a power of the observed ratio, the power being chosen to minimize the skewness of the pivotal random variable. The method is simple to use and more stable than that of Katz et al. A continuity correction is suggested if a conservative interval is desired.