Controlling for multiple testing in an investigation of the association between occupation and mortality from diabetes.
Li-J; Feng-HA; Robinson-CF; Walker-JT
Joint Statistical Meetings, July 30 - August 4, 2011, Miami Beach, Florida. Alexandria, VA: American Statistical Association, 2011 Jul; :302169
Multiple testing problems arise frequently in statistical data analysis. Choice of the appropriate procedure to correct for these problems is important in order to avoid erroneous inferences. The National Occupational Mortality Surveillance System (NOMS) contains death certificate data from up to 28 states for the years 1984-1998 with coded information about usual occupation. In this study, we used data from the NOMS system to examine the association between occupation, sex, and death from diabetes. The problem of multiple testing is relevant, because occupation was coded in detailed categories and inferences were made simultaneously for the set of categories. In the present work, we illustrated the use of the false discovery rate (FDR) approach to the multiplicity problem with two examples. We addressed the differences in interpretation between familywise error rate (FWE) and FDR, and illustrated use of FDR control in the case of both independent tests and tests under dependency. We concluded that the FDR approach is appropriate for data analysis with a large number of hypothesis tests to maintain a balance between type I and type II errors.
Statistical-analysis; Statistical-quality-control; Demographic-characteristics; Surveillance-programs;
Author Keywords: multiple testing; familywise error rate; false discovery rate
Joint Statistical Meetings, July 30 - August 4, 2011, Miami Beach, Florida