Confidence intervals that match Fisher's exact or Blaker's exact tests

• DOI: 10.1093/biostatistics/kxp050
• Publication Date: 2009-11-30T20:48:59Z
• Authors: Michael P. Fay
• Keywords: Exact statistics; Credible interval; Coverage probability; Robust confidence intervals; Tolerance interval; Confidence region; Confidence distribution

When analyzing a 2 x table, the two-sided Fisher's exact test and usual confidence interval (CI) for odds ratio may give conflicting inferences; example, rejects but associated CI contains an of 1. The problem is that inversion if either one-sided tests at half nominal significance level. Further, set not be interval, so following Blaker (2000, Confidence curves improved intervals discrete distributions. Canadian Journal Statistics 28, 783-798), we define "matching" as smallest set. We explore these versions well suggested by (2000) provide R package exact2x2 which automatically assigns appropriate matching to each 3 tests.