If the variable is not involved in interactions, and if it is either continuous or is listed in the CLASS statement with reference (also called indicator or dummy) coding for it by specifying the PARAM=GLM or PARAM=REF option, then the odds ratio estimate for that variable will be equal to its exponentiated parameter estimate.

However, if the variable in question is listed in the CLASS statement, and you do not select reference coding, then the default effects coding is used internally for the variable. Under effects coding, the odds ratio is not estimated by simply exponentiating a single parameter estimate. This is discussed in "Odds Ratio Estimation" in the Details section of the LOGISTIC procedure documentation. Under effects coding, the LOGISTIC and SURVEYLOGISTIC procedures determine the appropriate contrast of parameters that defines the difference in log odds between two levels of the variable. Exponentiating this contrast's estimated value yields an estimate of the odds ratio that compares the two levels of the variable. This estimate is displayed in the Odds Ratio Estimates table.

Also, if the variable is involved in interactions, then its odds ratio depends on the variable(s) that it interacts with and there are therefore multiple odds ratios. Odds ratio estimates for a variable involved in interactions can be obtained using the ODDSRATIO statement and is illustrated in this note. For discussion of how to compute odds ratios in a model that includes interaction under both indicator coding and effects coding, see Examples of Writing CONTRAST and ESTIMATE Statements. 

Operating System and Release Information