R2 and adjusted R2 values in FMM estimated submodels STATA15.

[Jerzy Moczko]

Finite Mixture Models procedure in STATA 15 does not report in results R2 and adjusted R2 values in estimated submodels. Where or how can I find them in the simpliest way? ?

[Sean Fiedler]

From my limited understanding ...

There is no R2. The two regressions you see are not fit based on the usual least squares principle. No residuals are being minimized, no variance of y is being partitioned. If you run the regression with those variables you won't get those estimates, you of course won't even get two sets of estimates. These two models are determined jointly, they come from an iterative process that maximizes a certain likelihood function or EM algorithm. Just like any other regression that comes from maximizing likelihood, there is no R2. This is not particular to the new command in Stata 15, the user written command fmm also does not report an R2. No regression fit by maximum likelihood has an R2 value.

I would investigate other goodness of fit measures.

[Jerzy Moczko]

Thank you for your answer Sean. I found similar explanation on STATALIST FAQ (http://www.stata.com/support/faqs/statistics/r-squared/). What is your opinion on following procedure. Using fmm we can make predictions of the posterior probability of class membership (documentation fmm intro page 7). Using these probabilities we separate original data into two subsets and in each of them we perform regress command. In that way we obtain both R and R2. Is it correct approach?