Empirical verification of confidence intervals computed from dental data

Sampling experiments were conducted on five distributions of dental epidemiological data which varied in degree of positive skew. The tendency of nominal confidence intervals to be underestimated and asymmetric in the two tails increased with skew and decreased with increasing sample size. The findings conform to theoretical expectations and demonstrate the importance of the normality assumption regarding sampling distributions. A rule is presented which identifies a sample size so that the 95% confidence probability statement would be incorrect no more than 6% of the time.