st: Sample bias in ASCLOGIT?

[Simon Falck <simon.falck@abe.kth.se>]

Dear Statalist, 

I am running a firm location model on all regions within a country, using an ASCLOGIT regression model since my economic model contains both case-specific and alternative-specific variables that seek to explain firms´ investment behavior in choosing between 72 regions, i.e. investment locations.

Over the period of study, only 56 out of 72 regions have received investments. Consequently, in the organization of the data used in the ASCLOGIT regression model, does this imply that I, apart from the case-specific variables, include alternative-specific regressors exclusively from regions that have been chosen at least once? To my understanding, then the results will suffer from a sample bias since the regression only considers 56 out of 72  possible investment locations existing in the country.

Ideal is to include all 72 regions as possible choices since then firm have considered all possible location choices. However, it seems that the ASCLOGIT regression model cannot not handle alternatives which have not been chosen at least once, since the model do not converge if all 72 regions are included in the regression.

Any suggestions on what I may do wrong? Whats the strategy here?

Thanks in advance!
/Simon




 





*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/