Abstract
It was considered that the method of sampling is an important factor in the statistical evaluation of results obtained from observer studies in receiver operating characteristic(ROC)analyses. However, few papers have discussed the effect of selection. This study evaluated the effects of the method of selection on ROC analysis by using the results of past ROC experiments. Furthermore, a new approach to collecting case samples that can reduce the number of samples without decreasing statistical reliability was presented. Our study employed a combination of random sampling and stratified sampling. The statistical reliability of our method was evaluated by testing statistically significant differences between several combinations of sampling methods and populations of case samples.
