12 Adjustment by regression
Stratification controls confounding by comparing like with like, but it runs out of data. With five confounders of three levels each there are \(243\) strata, most of them nearly empty, and continuous confounders such as age must be cut into bands that leave residual confounding within them. Regression models solve both problems by assuming a form for how risk depends on the confounders, which lets information be shared across strata. The two models of this section, logistic regression for risks and odds and Poisson regression for rates, are the model-based counterparts of the Mantel–Haenszel odds ratio and rate ratio.
12.1 Logistic regression
12.2 Logistic regression for case-control data
12.3 Conditional logistic regression for matched data
12.4 Poisson regression for rates
12.5 Using models well
12.6 Problems
12.2 Logistic regression for case-control data
12.3 Conditional logistic regression for matched data
12.4 Poisson regression for rates
12.5 Using models well
12.6 Problems
Questions on this section
Stuck on something here? Ask below and it stays attached to this topic.