Reconsidering the odds ratio as a measure of 2×2 association in a population

Authors

  • Helena Chmura Kraemer

    Corresponding author
    1. Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, U.S.A.
    • Department of Psychiatry and Behavioral Sciences MC 5717, 401 Quarry Road, Stanford University, Stanford, CA 94305, U.S.A.
    Search for more papers by this author

Abstract

The odds ratio (OR) is probably the most widely used measure of 2×2 association in epidemiology, but it often produces results that are puzzling or misleading. Receiver operating characteristic (ROC) methods are used to take a fresh look at the OR and show where and why such puzzling results arise. When researchers choose to report a summary measure of association, the OR is one of many measures of association that might be considered, not one that should be considered the ‘gold standard’ of 2×2 measures of association. In a randomized clinical trial with binary outcome for success, either the success or failure rates in treatment and control groups might be reported separately or the number needed to treat to achieve one extra success, to emphasize the cost of unnecessary treatment needed to achieve a success. In studies assessing reliability or heritability, we recommend the intraclass kappa. In studies in which one binary variable is assessed against a binary criterion, we recommend the weighted kappa. Copyright © 2004 John Wiley & Sons, Ltd.

Ancillary