I plan to fit a GAM or GAMM. There is one categorical variable which I think is important for explaining Y (or Y*), but it is not in my dataset - it is measurable but has not been measured.

Can I use a mixture model (GAMM) to compensate for the omission of this variable? If I had this variable, I would have just used a GAM.

Can you recommend some software for me to use? Do you have examples? Is the gamm4 R package "the way to go"?

Other details: Y is count data, so I plan to use a Negative Binomial model or "family". The purpose of this taks is to investigate the relationships between the variables. Although the dataset, with N rows, fits into memory of my machine, a matrix with NxN rows will not fit into memory.

  • 1
    $\begingroup$ How would a mixture model deal with the problem illustrated by Simpson's paradox -- essentially a consequence of omitted variable bias? $\endgroup$ – Glen_b May 23 '14 at 4:13
  • 1
    $\begingroup$ @Glen_b I have not thought of a way how. I am starting to think that the short answer to this question is "no". $\endgroup$ – power Jun 2 '14 at 7:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.