imrmc: Size and Analyze Multi-reader Multi-case Reader Studies, 5.4 Hypothesis Testing and Confidence Intervals:
5.4 Hypothesis Testing and Confidence Intervals:
The null hypothesis for comparing two modalities is AUC_A - AUC_B = 0.00 (or AUC = 0.5 for signal modality).
The test statistic is the (AUC_A - AUC_B) / se(AUC_A - AUC_B) for all methods.
For the Large Sample Approx(Normal) results, we assume the test statistic comes from a normal distribution. This is referred to as a Wald or z-test.
For the T-test with df(BDG) results: we assume the test statistic comes from a t-distribution with df reported (df = degrees of freedom are presented in Obuchowski et al. 2012 [11]). This is a t-test.
For the T-test with df(Hillis) results: we assume the test statistic comes from a t-distribution with df reported (df = degrees of freedom are presented in Hillis et al. 2008 [10]). This is a t-test.
Each method results include: p-value of the null hypothesis and the confidence interval on the difference (or the confidence around the single modality AUC).
