[Limdep Nlogit List] How to apply sampling weights in choice model?

[Mikołaj Czajkowski]

Dear Richard,

I do not know how are weights implemented in NLOGIT exactly, but the 
scaling variable w is probably scaled for mean = 1.
If I understand your question correctly - note that w*log(p) is not 
log(p*w) but rather log(p^w).

Best regards,
Mik



On 2018-01-08 17:31, Richard Turner wrote:
> Hello,
>
> I am trying to apply sampling weights in a choice model. I have a variable
> that is the inverse probability that certain survey respondents were
> sampled from the population.
>
> To apply this type of weight in to a choice model in NLOGIT, do I simply
> use the command: ;wts= <name of inverse probability weight> ?
>
> If so, I read in the LIMDEP/NLOGIT manual that in maximum likelihood
> estimation, the terms in the log likelihood and its derivatives, and not
> the data themselves, are multiplied by the weighting variable. I.e. Sum_i
> w_1log(f_i). Is this the way NLOGIT applies sampling weights?
>
> Finally, why would it be incorrect to multiply the predicted probabilities
> of each choice set by the inverse probability weight and then rescale the
> probabilities in each choice set to sum to 1?
>
> Regards,
>
> Richard
> _______________________________________________
> Limdep site list
> Limdep at mailman.sydney.edu.au
> http://limdep.itls.usyd.edu.au
>