All,
I've estimated a conditional logit model to explain schooling choice decisions by high school pupils. I'm now running simulations to predict what other option pupils choose if their chosen alternative would no longer be available. For example, I eliminate all university campuses from the choice set and calculate the predicted choice probabilities for the remaining options (which are the vocational alternatives and the outside option).
I use predictnl to obtain the choice probabilities and their 95% confidence intervals:
predictnl ACphat=predict(p) if ACAD==0, se(phat_se) ci(ci1 ci2) iterate(10000)
Two questions:
1) Is this, i.e. the use of the "if ACAD==0"-clause, the right way to implement a simulation where all academic schooling alternatives (marked by the dummy ACAD==1) are abolished? By doing this, the choice probabilities redistribute over the remaining alternatives so the choice probabilities sum to 1 for every individual. This seems fine, I'm just wondering if anyone has used other techniques.
2) I use predictnl instead of just predict because I want the confidence intervals (CI) for the predicted probabilities. Is the obvious way to do it? In prinicpal, the delta method should be ok, but it is quite slow if the estimated model has many parameters, so any alternatives? In relation to this: is it possible in some way to supply predictnl with an analytic expression for the derivative of the choice probability to avoid the numerical derivative calculation (very slow)?
Thanks,
Stijn Kelchtermans
Catholic University Leuven
Belgium
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
