Questions about the asclogit command in Stata.

[Yan Song]

I read that there is a bug in stata 14's asclogit command. The link to the thread discussion this problem is here:https://www.statalist.org/forums/for...14-vs-stata-13. I am wondering if it's been fixed yet?

In addition to the above question, I have a few more questions.

I am running a model of student's matching to colleges. I am running a simplified model in stata 13 and 14. My model is the following:

asclogit group_chosen,case(v1) alternatives(alt_id) casevars(provid_* wave_*)

group_chosen: numeric variable indicating which school is chosen. 9 groups in total.

provid_* wave_* are dummies indicating a student's home province and year taking college entrance test

In stata 13, I got the message "collinear variables, convergence may not be achieved".
In stata 14, I got no warning message. But the log likelihood suggests nonconcavity. and it doesn't converge.

I am confused and could not find what the problem seems to be.

[William Lisowski]

The problem with asclogit apparently was fixed in the 19 May 2016 update to Stata 14.1. You should be sure your copy of Stata is updated to at least this level; use the about command to confirm your version.

Your problem with convergence may stem from using dummy variables rather than Stata's factor variable notation, see help factor variables for details. Factor variables give the commands using them additional information about the structure of your data: if for a given observation the entrance test was take in 2010, then for that observation it was not taken in 2009, 2011, etc. The algorithms used by the command might take advantage of this additional information.

[Yan Song]

The problem with asclogit apparently was fixed in the 19 May 2016 update to Stata 14.1. You should be sure your copy of Stata is updated to at least this level; use the about command to confirm your version.

Your problem with convergence may stem from using dummy variables rather than Stata's factor variable notation, see help factor variables for details. Factor variables give the commands using them additional information about the structure of your data: if for a given observation the entrance test was take in 2010, then for that observation it was not taken in 2009, 2011, etc. The algorithms used by the command might take advantage of this additional information.

Thanks. It's a very helpful answer.