Mixlogitwtp model not converging

[Paul Arnem]

Dear Statalisters,

I am currently analyzing stated choice data from an individualized choice experiment using the user-written commands "mixlogit" and "mixlogitwtp". The data looks like this:

* Example generated by -dataex-. For more info, type help dataex
clear
input int(id caseid) byte(Task alt) float(hochpreis_1 hochpreis_2 niedrigpreis_1 niedrigpreis_2) byte preiszonen int(kosten_angezeigt einsparung_angezeigt zusatzkosten_angezeigt) float(choice kosten_neg)
3 1 1 1 0 0 0 1 3 83 20  6 0 -83
3 1 1 2 0 1 1 0 3 76 18  6 0 -76
3 1 1 3 1 0 0 0 9 55 20 11 1 -55
3 2 2 1 0 1 0 1 9 83 14 23 0 -83
3 2 2 2 1 0 0 0 3 69 20  6 1 -69
3 2 2 3 1 0 0 1 3 55 20 17 0 -55
end

The model in preference space runs fine and yields the following results:

mixlogit choice kosten_angezeigt , rand(hochpreis_1 hochpreis_2 niedrigpreis_1 niedrigpreis_2 preiszonen einsparung_angezeigt zusatzkosten_angezeigt) group(caseid) id(id) nrep(1000) robust

Iteration 0:  Log likelihood = -6137.1659  (not concave)
Iteration 1:  Log likelihood = -6035.0273  
Iteration 2:  Log likelihood = -6015.9366  
Iteration 3:  Log likelihood = -6015.7484  
Iteration 4:  Log likelihood = -6015.7483  

Mixed logit model                                       Number of obs = 18,108
                                                        Wald chi2(8)  = 247.86
Log likelihood = -6015.7483                             Prob > chi2   = 0.0000

----------------------------------------------------------------------------------------
                       |               Robust
                choice | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-----------------------+----------------------------------------------------------------
Mean                   |
      kosten_angezeigt |  -.0576894   .0039003   -14.79   0.000    -.0653339   -.0500449
           hochpreis_1 |   .0441881   .0618708     0.71   0.475    -.0770764    .1654527
           hochpreis_2 |   .0022187   .0612087     0.04   0.971    -.1177482    .1221856
        niedrigpreis_1 |  -.0416261   .0484117    -0.86   0.390    -.1365112     .053259
        niedrigpreis_2 |  -.0079596    .051074    -0.16   0.876    -.1080628    .0921435
            preiszonen |  -.0388723    .008607    -4.52   0.000    -.0557417    -.022003
  einsparung_angezeigt |   -.013317   .0079175    -1.68   0.093    -.0288349     .002201
zusatzkosten_angezeigt |  -.0473624   .0047258   -10.02   0.000    -.0566249   -.0380999
-----------------------+----------------------------------------------------------------
SD                     |
           hochpreis_1 |   .8914174   .0783303    11.38   0.000     .7378929    1.044942
           hochpreis_2 |   .7835104   .0913115     8.58   0.000     .6045431    .9624777
        niedrigpreis_1 |   .3474652   .1461157     2.38   0.017     .0610837    .6338467
        niedrigpreis_2 |    .679681   .0840821     8.08   0.000     .5148831    .8444788
            preiszonen |   .1067137   .0167954     6.35   0.000     .0737953    .1396321
  einsparung_angezeigt |   .0703171   .0149687     4.70   0.000      .040979    .0996552
zusatzkosten_angezeigt |    .063285   .0069402     9.12   0.000     .0496825    .0768875
----------------------------------------------------------------------------------------
The sign of the estimated standard deviations is irrelevant: interpret them as
being positive

Sadly, this simple WTP space model does not converge:

mixlogitwtp choice , price(kosten_neg) rand(hochpreis_1 hochpreis_2 niedrigpreis_1 niedrigpreis_2 preiszonen einsparung_angezeigt zusatzkosten_angezeigt) group(caseid) id(id) nrep(1000) vce(robust)

We tried the following things each on their own and in combinations (mostly taken from this Statalist post: https://www.statalist.org/forums/for...vergence-issue), but nothing seems to work:

• Lower/higher nrep() values
• Used the burn() option
• Tried random starting values
• Used the wtp values obtained from the preference space model as starting values
• Used all the available options for technique()
• Used the "difficult" option
• Used "xlogit" in Python and "logitr" in R, both did not converge either using various options

I am looking forward to any tips and comments regarding this non-convergence issue.
Thank you!

[Xuejuan Zh]

Hi Paul, did you finally solve the problem? I also have the same. I wonder if it's a problem in the halton draws