Modeling sequential choices

[Florian Seliger]

Dear Statalist,
I am wondering whether there is any command or implementable procedure to estimate a model (maybe related to Multinomial Logit or Nested Logit) where I can give the decision tree of alternatives a sequential interpretation.

Basically, I have the following data:

Firm i decides in t whether to invest in a, b or c.
Dependent on this choice it decides in t+1 again between a, b, c, etc. etc.
I consider 5 periods and have a set of explaining variables for t, t+1... t+5 (but I am thinking about only using variables from t to explain subsequent sequential decision making).

I have been using Multinomial Logit (mlogit) so far, but then I have to aggregate the information somehow because I cannot consider so many choices at the same time.

Therefore, I am looking for something more flexible that is explicitly suitable for sequential choices (Nested Logit is not).


Thanks for your consideration!

[Phil Bromiley]

If you searched for sequential logit on this listserve, you would have found: http://www.statalist.org/forums/foru.../2418-seqlogit

Apparently, there is a program for the kind of model. I do have not used the program so I cannot vouch for it.

[Florian Seliger]

Phil, of course I read the threads for seqlogit beforehand, but it does not seem the appropriate model.

The issue is that it models binary choices at each node and that the values cannot be reached through multiple paths.

My choices in the final period should be possible through all previous paths (however, I noticed that 5 periods give me a too complex tree in any case so I have to aggregate information somehow).


Any help is appreciated.