Displaying variance-covariance matrix with SAS GAMPL

[pm68]

Hi,

I am using PROC GAMPL to perform GAM analyses, in which I am mixing spline() and param() effects. I am looking for a way to get the covariance matrix of the parameter estimates for additional calculations, but I could not find how to do that. For GLM analyses performed with PROC GENMOD, the covariance matrix may be retrieved by requesting the covB table with the ods output statement. Is there not an equivalent way to proceed with PROC GAMPL?

Thanks in advance for your help

[StatDave]

Instead of using GAMPL, consider using a modeling procedure that offers the EFFECT statement (GLIMMIX, LOGISTIC, ORTHOREG, and others) which can add spline effects to your model. You could then use the COVB (or other option depending on the procedure) to display and/or save the parameter covariance matrix. Or use the STORE statement in the procedure (if available) to save the model and then use PROC PLM to test additional hypotheses.

[pm68]

Hi StatDave,

Thanks for your quick answer and for drawing my attention on these
procedures; I did not realize these could add spline effects.
Unfortunately, I am exploring models with Tweedie distributions (I
should probably have stated this earlier), and I do not think any of
these procedures can accomodate this?

[StatDave · Accepted solution]

As noted in the list of Frequently Asked-for Statistics (FASTats) in the Important Links section of the Statistical Procedures Community page, the Tweedie distribution is available in GENMOD and HPGENSELECT, but these do not offer the EFFECT statement (the Tweedie is also available in some procedures in SAS Viya). One way around this is to use a procedure that supports the EFFECT statement to build and save the design matrix columns needed for your model and then use them as input in GENMOD or HPGENSELECT to fit your Tweedie model. PROC LOGISTIC can do that by using the OUTDESIGN= and OUTDESIGNONLY options (see this note). For example, these statements create a data set, OD, that contains the design matrix containing columns for the spline effect on the variable BLAST. It then uses those columns (SB: is shorthand for all variables starting with SB) in the model. 

proc logistic data=remiss outdesign=od outdesignonly;
effect sb=spline(blast/naturalcubic);
model infil = smear sb ;
run;
proc genmod data=od;
model infil = smear sb: / dist=tweedie;
run;

[pm68]

Thanks very much StatDave, it seems to work!