PARTIAL PROPORTIONAL ODDS MODELS AND THE LOGIST PROCEDURE

Thirteenth Annual SAS Users Group International Conference

Bercedis Peterson ; Duke University Medical Center ; Durham ; NC

PID  :  86431

来源: CEUR

【 摘 要 】

The LOGIST procedure fits the multiplelogistic regression model of Walker and Duncan(1967) to ordinal response variables that take onk+l possible values. In this model, logits ofcumulative probabilities are expressed as linearfunctions of the predictor variables. Theresulting regression coefficients are identical foreach of the k cumulative probabilities, except forthe intercept terms, CXJI j=l,,k, which areordered to reflect the order of the cumulativeprobabilities. The assumption of identicalregression coefficients, referred to by McCullagh(1980) as proportional odds, implies that the logodds ratio for the relationship between adichotomized response variable (Y ~j vs. Y -->