This article is only available in the PDF format. Download the PDF to view the article, as well as its associated figures and tables.
Drs Fried and Appel state that requiring independent samples to compare performance characteristics unnecessarily doubles the sample size. We do state that independent samples are needed to retain mathematical correctness, because the 2 × 2 contingency configurations on which our tables are based are modeled on independent samples. In practice, however, using paired data in a 2×2 contingency table is equivalent to using independent data and equally efficient. One only sacrifices the opportunity for further statistical efficiency. Therefore, we maintain that the sample size requirements (SSRs) of 67 obtained in our exercise 1 need not be doubled.Their other concern is our failure to minimize SSRs through paired analysis where one considers only the discordant data pairs, ie, those results positive by method A and negative by method B or vice versa. If methods A and B are independent of one another, paired analysis neither increases statistical power
Arkin CF, Wachtel MS. How Many Patients Are Necessary to Assess Test Performance?-Reply. JAMA. 1990;264(16):2075. doi:10.1001/jama.1990.03450160042021