JMP’s GLM, JMP Pro’s GenReg and SAS Proc GLMSome of you have asked me about JMP’s GLM, JMP Pro’s GenReg and SAS Proc GLM. To help more SEs better assess your customer’s needs I am sharing my replies.
JMP’s GLM has been around for many years. Before GenReg was added, GLM was the only modeling personality that supports fitting Poisson-distribution based count data models (e.g., # of defects). Also, if you want to fit a Probit model as opposed to a Logistic regression model GLM is the place to go. Other important options that are still only available in GLM are the Firth bias-adjusted estimates and over-dispersion tests. For more on the Firth bias-adjusted estimates in modeling rare events please check out this blog post https://statisticalhorizons.com/logistic-regression-for-rare-events. The author, Prof. Paul Allison, has written several widely popular SAS books including Logistic Regression using SAS.
Today, you can fit count data models as well as many other non-normal distribution based models using GenReg. GenReg also provides many extremely useful variable selection methods. However, if someone wants Firth-bias adjustments or Probit models GLM is the place to go.
SAS PROC GLM is very different from JMP’s GLM or JMP Pro GenReg. It only fits continuous response variables. And It doesn’t provide variable selection options like LASSO, etc. A relatively new SAS procedure, called Proc GLMSELECT, performs variable selections.The screenshot below lists key capabilities in Proc GLM:
