Discrete choice labour supply: lslogit

[Olivier Delaruelle]

I use the user-written Stata package lslogit (available at https://github.com/mloeffler/lslogit) to estimate a discrete choice model of labour supply. My data is in long version. The dataset consists of about 760 individuals having 5 choice alternatives, and therefore the dataset entails 5*760=3600 observations.
When I run the following line:

lslogit choice, group(idcode) quadratic consumption(ydisp) leisure(leis)

Stata output gives LR chi2(0)=0.00 and Prob>chi2=.

When trying to compare the predictive accuracy (predictions via lslogit predict <60% of surveytakers' choices correctly) I also want to use the boxcox or the translog utility function, as is allowed for in the lslogit commando. However, when I use boxcox or translog instead of quadratic utility function, Stata issues an error

<istmt>:  3200  conformability error

What could be causing this?

Secondly, is the fact that my LR chi2(0) equals zero a problem? Thirdly, is anyone aware of a way to include extra dependent variables like, the number of children of a surveytaker, in the commando without having to add it to an interaction effect/taste shifter? I would like to simply see the effect of a variable - e.g. number of children - on its own and not combined with leisure or consumption.

Any help is greatly appreciated,
Olivier

[Agathe Simon]

Dear Olivier,
Unfortunately I am not able to help you but I have the same problem. I have this error even with the quadratic utility function (as well as with the other forms, i.e. boxcox and translog). I was wondering if you had since found a solution to your problem?
I think there must be a problem with my data. (I managed to run the model with a simulated database only).

Thank you very much!

[Laman Orujova]

Hi everyone,

I have also experienced a problem with this command. Although in my case, it says that "1,744 (group size) take 538 (# positives) combinations results in numeric overflow; computations cannot proceed" (I have 3441 observations and four choice points". Were you able to solve the issue?

[Agathe Simon]

Hi Laman,

I do not know about this type of error but did you check that there is only 1 positive outcome by households across the 4 alternatives?
I am not sure that I can help you but maybe could you paste some of your code here?

In my case, I deleted all observations for which the disposable income is below 0 rather than delete the entire household with the 4 alternatives, then I had some hh with only 3 alternatives left, which obviously can not work.

Best,