Stratified Analysis

Stratification is used both to evaluate and control for confounding and requires separating your sample into subgroups, or strata, according to the confounder of interest (e.g., by age, gender, race/ethnicity, etc.). Because each stratum is homogeneous with regard to the confounder of interest, one can then evaluate the association between exposure and disease within each stratum (e.g., the odds ratio for women only and the odds ratio for men only). Once you have conducted stratified analyses you have the option to combine your data to derive a summary, or pooled estimate. One of the most common techniques for pooling data is the Mantel-Haenszel procedure. This statistic combines information across partial tables and enables you to calculate one common odds ratio, as opposed to many for each strata (See Aschengrau & Seage, pp. 297-300).