a broad class of statistical procedures that allow variables to be related in a prediction or
regression analysis by taking into account the variance of each. The generalized linear model extends ordinary
least squares regression to include other procedures—such as
logistic regression and
Poisson regression—that allow for the use of categorical dependent variables having very lopsided distributions, in which most people endorse a particular category (e.g., no heroin use) or fail to endorse a particular category (e.g., heroin use).