Hello,I use JMP and SAS daily. One of my current project is an analysis of Survival data. To train myself, I studied the example provided in the SAS manual:https://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_phreg_sect0... When I fit the same model with the exact data in JMP, I got different parameter estimates for the categorical variables (Cell, Prior, Therapy) even though I set the same reference category for these variables.
I wonder why JMP get different estimates for the categorical covariates?
PS. I fit the same model in R using the suvival package. SAS and R provide almost the similar estimates for the parameters.
SAS and JMP parameterize categorical variables differently, so the estimates are not the same. That is, you enter one term for a variable with three levels leads to different dummy variables in the regression and different estimates. See Help > Books > Reliability and Survival. Look at the chapter with the Statistical Details. The book Fitting Linear Models also explains the parameterization of the linear predictor that you are using.
I bet, though, that the SAS model and the JMP model yield the same predicted values.
You cannot change the model parameterization in JMP.
Thank you for your explanation. I already assumed that they may be using different parametrization for the categorical variables, so the parameter estimates would be different. I also agree with you that they should provide the same predicted values.
But in this example, the overall likelihood test (Joint test in SAS) are also different. They should be similar which ever parameterization (or dummy variables) you use. SAS says Therapy is a significant covariate, JMP says not.
Can you show the SAS and the JMP output for these effect tests?
The joint tests from SAS and JMP agree if you intend to use the Wald test.
The difference in parameter estimates is because of the difference in parametrization of the categorical variables in SAS and JMP.
SAS default is REF whereas JMP uses EFFECT ( in SAS - model statement, param option).
Here is the outputs from JMP and SAS (with param=EFFECT), they are same.
Thank you Mark :)
Each parameterization is internally consistent. That is why the effect tests are the same. The parameter estimates are different, of course.
Glad that we sorted it out!
There are no labels assigned to this post.