Stratified analysis of means

Speaker:  David Clayton, Institute of Public Health

Mantel–Haenszel procedures for combining stratum-specific estimates of effects are widely used for rate ratios and odds ratios, but hardly at all for differences between means. This is curious, because it is the ability to see the stratum-specific estimates directly that makes the procedures useful, and this is just as important for a metric response as for any other.

The command mhmean is introduced as a metric response equivalent to mhodds for odds ratios and stmh for rate ratios. Stratum specific differences between means are displayed, and then combined using a weighted average. The within group variance is estimated and confidence limits can be based on either the normal or t distributions.

This command is a useful way of exploring data on a metric response at a more intuitive level than fitting regression models. An example of its use will be given.