```Date: Sat, 21 Jun 2008 15:05:55 -0400 Reply-To: Peter Flom Sender: "SAS(r) Discussion" From: Peter Flom Subject: Re: NLMIXED - Class variable - p-value Comments: To: "Ingrid K. Friberg" Content-Type: text/plain; charset=UTF-8 "Ingrid K. Friberg" wrote >Hello, > >I am getting ready to do a survival analysis in NLMIXED and have worked >with my statistician to figure out how to code it. Now I have a new, >simple problem and don't want to bother him. I know that NLMIXED cannot >use variables with three levels in it. Therefore I have recoded this with >2 dummy variables. How do I calculate a p-value for that which takes into >account both of the dummies? My table below gives me the p-value of beta1 >as .33 and of beta2 as .20. I would like to know the importance of adding >both at the same time!!! For some of my variables (not this one) one part >may be significant and the other not... > >Parameter Estimate Error DF t Value Pr > |t| Alpha >Lower Upper Gradient > > beta0 4.4968 0.3082 191 14.59 <.0001 0.05 >3.8889 5.1047 0.000962 > beta1 -0.2841 0.3259 191 -0.87 0.3845 0.05 - >0.9269 0.3588 0.00016 > beta2 -0.3254 0.3021 191 -1.08 0.2827 0.05 - >0.9212 0.2704 0.000763 > sigma0 0.8779 0.06126 191 14.33 <.0001 0.05 >0.7571 0.9987 -0.00066 > lambda0 1.0108 0.2705 191 3.74 0.0002 0.05 >0.4772 1.5443 -0.00043 > > >Thanks in advance. I really appreciate always getting such helpful >thoughts and comments. > First of all, p-values don't measure importance --- Importance is hard to define, but, whatever it is, it isn't measured by p-values. Second, I'd be very surprised if there were a way to combine these p-values Third, to your question --- One way to look at 'importance' is to fit the model with and without those two variables, then plot the predicted survival times for the two models against each other. You could even, if you liked, square and sum these differences, but I am not sure how that sum would be distributed. Or, you could compute the differences and plot those, or take means and quantiles and see if the differences were 'large' --- for some definition of large. HTH Peter Peter L. Flom, PhD Statistical Consultant www DOT peterflom DOT com ```

Back to: Top of message | Previous page | Main SAS-L page