01-18-2012 10:25 AM
i'm using proc logistic to model the probability of closure(0/1) in dependence of severel categorial variables on insurance data.
Due to the effect coding of the categorical variable sas doesn't compute estimates for the reference levels.
Even when using param=glm or the logit glm proc genmod there is still one estimate set to zero.
I know from software tools used in an insurance company that the logit glm there computes factors for all the levels of the categorical variables.
Is there any possibility to get paramter estimates for all categories of the categorical variable?
Is there any possibility to get estimates for all the cells build by the combinations of the categorical variables in SAS ?
If i include main effects only, i easily can compute the estimate of the reference level by using the design matrices.
But when including interactions of categorical variables more estimates are missing due to two reference levels.
I haven't figured out how to compute those missing levels or how to interpret the parameter estimates of interaction terms when using effect coding.
Thanks for your help.