Hello,
I am trying to explain the choice of agricultural companies to site investment in 11 different destinations (provinces) of three countries, as a function of attributes of these destinations and of the companies themselves, with special attention to the role of environmental regulations (cf. the "pollution haven hypothesis" literature). I have been using the nested logit (nlogit) for this, with countries as nests and provinces as alternatives within nests, using the company-level variable ("pollution_level") as an interaction term with one of the alternative-specific variables ("environmental regulations stringency") for the lower-level model. So my model looks like this (removing some variables for simplicity):
nlogit occurrence distance yield stringency_PL || toc: , base(Argentina) || to: , case(idp_int)
with occurrence being the occurrence of an investment for that company and destination; stringency_PL = environmental_regulations_stringency*pollution_lev el; toc = countries; to = provinces.
Now I would like to see the effect of "pollution_level" separately from "stringency". But nlogit doesn't seem to accommodate variables that are not alternative-specific, and if I add "pollution_level" into the model it gives me the following message:
"variable pcforest is not alternative-specific: it has no within-case variability".
If I understand correctly, this is not a limitation of the nested logit per se, but rather a limitation of its implementation in STATA, is that right? I can of course specify "pollution_level" as a variable in nest (country) choice, but this specification has a different meaning.
So I would like to know:
- Is there a way to use individual-specific variables for nested logit models in STATA that I am not aware of?
- Supposing there isn't, is there another type of discrete choice model that 1) relaxes the IIA, 2) accommodates both individual-specific and alternative-specific variables? It sounds like mixed logit might be an option but having limited econometrics background I am not quite clear on how that would work out for this model.
I found an older post about this (http://www.stata.com/statalist/archi.../msg01144.html) that never seems to have been answered (I contacted the author, who confirmed this). Any help would be greatly appreciated.
Best,
Yann le Polain.
Note: I use STATA 14 for mac.
I am trying to explain the choice of agricultural companies to site investment in 11 different destinations (provinces) of three countries, as a function of attributes of these destinations and of the companies themselves, with special attention to the role of environmental regulations (cf. the "pollution haven hypothesis" literature). I have been using the nested logit (nlogit) for this, with countries as nests and provinces as alternatives within nests, using the company-level variable ("pollution_level") as an interaction term with one of the alternative-specific variables ("environmental regulations stringency") for the lower-level model. So my model looks like this (removing some variables for simplicity):
nlogit occurrence distance yield stringency_PL || toc: , base(Argentina) || to: , case(idp_int)
with occurrence being the occurrence of an investment for that company and destination; stringency_PL = environmental_regulations_stringency*pollution_lev el; toc = countries; to = provinces.
Now I would like to see the effect of "pollution_level" separately from "stringency". But nlogit doesn't seem to accommodate variables that are not alternative-specific, and if I add "pollution_level" into the model it gives me the following message:
"variable pcforest is not alternative-specific: it has no within-case variability".
If I understand correctly, this is not a limitation of the nested logit per se, but rather a limitation of its implementation in STATA, is that right? I can of course specify "pollution_level" as a variable in nest (country) choice, but this specification has a different meaning.
So I would like to know:
- Is there a way to use individual-specific variables for nested logit models in STATA that I am not aware of?
- Supposing there isn't, is there another type of discrete choice model that 1) relaxes the IIA, 2) accommodates both individual-specific and alternative-specific variables? It sounds like mixed logit might be an option but having limited econometrics background I am not quite clear on how that would work out for this model.
I found an older post about this (http://www.stata.com/statalist/archi.../msg01144.html) that never seems to have been answered (I contacted the author, who confirmed this). Any help would be greatly appreciated.
Best,
Yann le Polain.
Note: I use STATA 14 for mac.
Comment