Comparison of coefficients obtained in mixed logit regressions from two subsamples

[Aurelie Bonein]

Dear all stata users,

I’m wondering if it is possible to compare the estimates obtained in mixed logit models, estimated from 2 different subsample.
I will try to explain in more details
I have data from a stated choice experiment with 162 individuals who answered from 5 sets of situations each.
Each set of situation requires from individuals to choose between a contract A, a contract B or the reference contract.
In each contract, there are 3 attributes : price, duration and loss. The level of attributes vary across contract.
I would like to know if choices made by individuals who experience strong risk aversion differ from choices made by individuals who experience small risk aversion.
I have conducted mixlogit regressions, one for each subsample (strong risk averse subjects and small risk averse subjects). To do so, I have used the User-written command mixlogit to fit mixed logit models (Hole, 2007). I have used ssc install mixlogit- to install -mixlogit
I get estimates for each subsample and I would like to know if it is possible to compare the estimates I have obtained in the two subsamples. For instance the estimate for the price variable obtained in the regression conducted for high risk averse subject and the one obtained for the pool of subjects who have a small risk aversion; the same for the coefficient for the length of the contract ….
I hope someone could help me,
Thanks
Aurélie

[ben earnhart]

I might be wrong here, but given that it's arguably two independent samples, any reason you can't use a T-test for two independent samples?

[Aurelie Bonein]

Yes, I try to do it but I don't know how.
I get the estimates for the two independent equation but I don't know how to store the results and next to use the T-test (if the t-test can be applied with mixlogit)
Thanks
Aurélie

[Joseph Coveney]

Start Stata, and then type in the command line "search chow" (without the quotation marks). Then take a look at the pages that are hyperlinked in the window that pops up.

Focus at first on those that are labeled "FAQ", especially the first two of them.

The general method is illustrated there with linear models, but as the first FAQ states, "the technique is really more general than that".

[Stephen Jenkins]

Out of curiosity: to what extent is the "Chow" test of coefficient differences in the context of a mixed logit model dependent on assumptions about the [lack of] differences in the error variances? For the plain binary logit model (i.e. not the mixed logit), it is "well known" that tests about differences in coefficients across groups for which the model is estimated rely on assumptions about the error variances. A nice paper by our very own Rich Williams covers the literature about that well: Williams, R. (2009). ‘Using heterogeneous choice models to compare logit and probit coefficients across groups’, Sociological Methods & Research , 37 (4), 531-559.

[Joseph Coveney]

I'm just guessing here, obviously, but I'd wager that it's strongly dependent upon the assumption, just as for the plain logit model. I agree about Richard's paper as an excellent review of the practical options in interpretation.

[Aurelie Bonein]

Thank you for the reference. I will look at this paper
Aurelie