Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "William Verheul" <W.Verheul@nivel.nl> |
To | <statalist@hsphsun2.harvard.edu> |
Subject | st: discrete choice model with case-specific variables using xtlogit |
Date | Tue, 10 Aug 2010 09:56:48 +0200 |
Dear listers, I am using xtlogit to estimate the effect of 4 dichotomous attributes for choosing treatment A or treatment B. 200 patients choose 8 times between A or B with varying combinations of attribute values. B is always the exact opposite of A: where attributes are 1 for A, they are 0 for B, etc. I have 200 subjects *8 choices= 1600 observations. I also want to assess if those attribute effects vary with age and gender. The data is as follows: Id=patient id, Alt=id of combination of alternatives Sex=gender, age= age, cho=1 for choice A, 0 for choice B At* are the attributes id alt sex age cho at1 at2 at3 at4 1 1 m 28 1 1 0 0 0 1 2 m 28 0 1 1 0 0 1 3 m 28 1 0 0 0 1 1 4 m 28 1 0 1 0 1 1 5 m 28 0 1 0 1 1 1 6 m 28 0 1 1 1 1 1 7 m 28 1 0 0 1 0 1 8 m 28 0 0 1 1 0 2 1 f 41 1 1 0 0 0 2 2 f 41 1 1 1 0 0 2 3 f 41 1 0 0 0 1 2 4 f 41 0 0 1 0 1 2 5 f 41 1 1 0 1 1 2 6 f 41 1 1 1 1 1 2 7 f 41 0 0 0 1 0 2 8 f 41 0 0 1 1 0 For the basic model I use: xtlogit cho at1 at2 a3 at4, i(nr1) re noskip Now I want to include gender and age (in two separate models). I tried using asclogit, which has an easy way to include such variables, but asclogit doesn't take into account that subjects made multiple choices. If I use xtlogit in the same way as clogit http://www.ats.ucla.edu/stat/stata/seminars/stata10/choice_models.htm I think I should only include the interaction term in the model, so for gender: xtlogit cho at1 at2 a3 at4 at1*sex at2*sex a3*sex at4*sex, i(nr1) re noskip Is this correct? Or should I also include sex as a main effect (possible because it is a random effects model)? I understand that this main effect is not of interest (it would say only if men or women are more likely to choose option A or B regardless of the attributes), but I wonder whether this is necessary for obtaining correct estimates for the interactions? My second questions is concerning age, a continuous variable: do I get correct estimates by running: xtlogit cho at1 at2 a3 at4 at1*age at2*age a3*age at4*age, i(nr1) re noskip or does it work differently for continuous variables? Your advice is very much appreciated! William This message contains information that may be privileged or confidential and is the property of NIVEL. It is intended exclusively for the person to whom it is addressed. If you are not the intended recipient you are not authorized to use this message or any part thereof. If you receive this message in error please notify the sender immediately and delete all copies of this message. NIVEL rules out any and every liability resulting from any electronic transmission. No rights may be derived from the contents of this e-mail message. * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/