Proc Genmod Gamma with Log When to Scale

[cxc810]

Hello Community,

When using a GLM gamma model with log link, is there a rule of thumb for when to specific the scale as the default, Pearson or deviance?   

I've been reading up online, but they're always answering the question for a different distribution of GLM, and I don't think any of the answers i have found are relevant for gamma models.

Moreover, in college, I did all my statistical analysis in python, and all the packages we use had Pearson as the default scale, so that has me wondering if I've been doing it wrong the whole time!

In case it helps, here is the model setup.

ods graphics on;

proc genmod data=data;

model y = x

      /dist=gamma link=log  ;

run;

ods graphics off;

Thanks in advance! 

[StatDave]

Scaling with the SCALE= option is only needed as a way to attempt to deal with overdispersion as discussed in this note. 

[Ksharp]

Gamma distribution is for Y>0 ,which means dependent variable Y must be positive .