Xtprobit and weighted option, is it possible ?

[Jean Dewals]

Dear collegues,


I'm currently working on consumption survey data that concerns consumers WTP for labelled seafood. In this survey, we have a sample with an over-representation of young consumers.

However, for a purpose of a scientific article, I would like to correct my sample (to be the closest as possible to the real French population's age pyramid.)

1. To correct my sample, I'd like to weight my individuals according to their age. I thus identified the pweight option in stata.
2. Nevertheless, I would also like to use an xtprobit model to study the influence of different explanatory variables on my consumers' WTP and include individual heterogeneity ...

Finally, regarding these two conditions, I tried to use the following command.

 xtprobit ench cost catB catC catD genre deplittoral univ tradi sante enfant revenumid revenusup prix_crit_dom GMS VD Poiss PCE obj subj  ancrage ench1 [pweight=weight]

Unfortunately, I get the following error message :

pweight not allowed in random-effects case

So I'd like to know if there's a way to overcome this problem? Through the use of another model, or through a special command.



I remain available for any further information/precision !

Best regards,

Jean-François

[Maxence Morlet]

Cher Jean-François,

I can't answer your question directly, however you may want to reflect on three things (waiting for better advice):

- Is random effects rather than fixed effects what you really want? I understand that the incidental parameters problem can be a problem in panel data, so I would need to know more about your dataset before venturing too far on this topic. Why not try a linear probability model (with fixed effects) as well?

- xtprobit uses a joint likelihood function, if my memory serves, which is only consistent under no serial correlation.

- Finally, probability weighting only serves external, and not internal validity. An amazing paper discussing this issue is Solon, Haider and Wooldridge (2015).