No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Choice models -cmxtmixlogit- command for including random parameter for one of the alternatives (the base/status quo alternative)

    I am having trouble with the new stata 16 cmxtmixlogit command.

    I am trying to estimate a choice model for repeated-choice 'unlabeled' discrete choice data, where one of my alternatives is essentially a status quo alternative and I would like to include an alternative specific constant (for one out of my three alternatives) as a normally distributed random parameter. I have other random parameters in the model, including correlations and constraints.

    It seems that cm- commands are more applicable for 'labeled' choice data but I cannot do my specification using mixlogit.

  • #2
    I haven't used -cmxtmixlogit- but based on my reading of the help menu for the command, I would like to suggest the following. You can tell -cmxtmixlogit- that you don't have any alternative specific constant by using the -noconstant- option. Then, you can include your alternative-specific constant for the status quo option by the back door, by listing it under -rand(.)-.


    • #3
      Thank you Hong. That strategy worked for me but I had to restart STATA for it to work... not sure why.

      I have a new issue when using starting values with cmxtmixlogit. Once I create a matrix of starting values from a cmxtmixlogit (that includes correlated random variables)
      {matrix x=e(b)}
      I am having trouble saving the matrix using
      {matsave x, saving}
      it gives me an error {invalid syntax r(198)}. I think there is an issue with the svmat code when saving the matrix? I can replicate the problem using STATA example data.


      • #4
        I haven't used -matsave- before so I can't say much. If you can elaborate on what you are trying to do, I may have something useful to say about it. I don't quite get why setting starting values requires a command that I have not used because I have been setting starting values in Stata for several years without knowing the command!