ABSTRACT
We can now return to the examples of Chapter 7 with renewed confidence. Let us take a fresh look at the association between coffee drinking and pancreatic cancer. Could this be explained by a confounding variable such as sex? Is some of the effect of behavior type on CHD due to confounding by body weight, say? In the previous chapter, we indicated that, having measured factors D, E , and a potential confounder, C , on individuals sampled through one of the design mechanisms, stratification of the data provides a strategy for the control of confounding by C . In this chapter, we provide a full description of how to extend the ideas of Chapters 6 and 7 to accommodate stratified data. In particular, we wish to both assess the association of D and E , and quantify the extent of the relationship through a measure of association, in both cases having removed or controlled for the confounding effects of one or more extraneous variables. We first turn to testing the association.
