The process of creating mathematical models to assess the association(s) between exposures, outcomes, and confounders (See Aschengrau & Seage, pp. 299-300). The mathematical model employed depends on the data in your study and include: linear regression, logistic regression, Cox proportional hazard, and Poisson models. Typically, investigators evaluate potential confounders first by stratification. If, upon stratification, the crude estimate is changed by 10% or more, the variable is retained and controlled for using multivariate techniques.