I have a question regarding the use of linktest after glm.
I am trying to replicate the results from the linktest command, and while I
am (unsuccessfully) trying to achieve this, I am also wondering what
predict-command/option should be used to generate the _hat _hatsq for glm
with non-identity-link functions.
My understanding is that the linktest tests how well the predicted values
relate to the observed values and whether there is a fundamental problem
with the form of the relationship (which is why the squared term is used).
therefore, I thought the re-transformed predictions should be used (the
default setting of the predict command). Unfortunately, those results don't
match. But even when I use "_predict prediction, xb" to generate the
predicted values (and from reading the linktest command that should be the
command used), I can't get the same results that linktest is showing me.
Does anyone have experience with the linktest and can tell me what
predict-command to use after the regression to generated predicted values
to replicate that linktest command? Should it be re-transformed or linear
predictions?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
