ABSTRACT
At the beginning of Chapter 12, we noted that matching represents a design alternative to regression analysis that mitigates the loss of precision due to adjustment for several potential confounding variables. In this chapter, we describe matched designs and statistical techniques to analyze data arising from matched studies. We consider both frequency matching, where there are only a few distinct levels of the confounding variables in question, and pair matching, where there are very large numbers of possible confounding variable patterns. These designs represent extreme examples of the approach, and it should be recognized that there are intermediate variations.
