Multinomial Logit vs. Multivariate regression

[Jesus Pulido]

A question on the choice of methodologies:

I want to obtain the determinants of adoption of a system of 3 practices that can be used individually or jointly by farmers (minimum-till, inorganic fertilizer and improved seed). Subjects can use any possible combination from these practices (8 different possibilities).

My first option is using the Multinomial-logit model. This I have seen allows me to use combinations as dependent variables but has the strong assumption of "independence of irrelevant alternatives", which might not be convenient when describing farmers' behaviour.

A second option is using multivariate regression. I've seen several applications of multivariate regression, but they always seem to regress practices on the x's individually, and not in the form of combinations of practices.

Any advice on how to capture combinations of practices as dependent variables would be much appreciated,

Jesus P.

[Joao Santos Silva]

Dear Jesus,

If you are modeling the choice between a set of alternatives you need a discrete choice model and I cannot see how a multivariate regression would help you. It is true that the MNL is very restrictive but it may be a decent starting point. Also, there are less restrictive discrete choice models that you can try.

Best wishes,

Joao

[Jesus Pulido]

Dear Joao,

Thank you for pointing me towards discrete choice models. Still I don't quite understand why choice modelling would be my option here, as it seems to deviate from my study strategy.

For this study, I will apply a large number of surveys to farmers. I want to regress y (the choice of combination of practices - I have 8 different choices) on the x's (their information on economic, demographic and social variables). I am trying to find the best method to do this.

I don't quite understand how a choice modelling strategy, where people state their preferences for/or value different available options, would fit in here. Isn't an econometrics approach more reasonable?

Cheers and thanks,

Jesus

[Richard Williams]

I also am not sure how you would use OLS regression. Perhaps the rologit or clogit commands would meet your needs.

[Joao Santos Silva]

Dear Jesus,

Using discrete choice models such as the multinomial logit is an econometric approach. So, I am not sure if I understand your concerns.

As I understand it, farmers have 8 choices and you want to model their decision; this is a typical example where discrete choice models would be used.

Best wishes,

Joao

[Joseph Coveney]

This is outside of my experience, but I can kind of see the OP's concern with multinomial logistic regression, because the eight choices are essentially combinations of only three underlying choices (not mutually exclusive). The OP didn't mention any specific research question that the survey is intended to address, but, for example, can multinomial logistic regression model fitted to the eight categories reveal, say, the preference for intensive tillage among farmers of a particular age category, or an association of minimum tillage and organic fertilizer usage in general? (Maybe a suggestion of what kind of mlogit postestimation command to use to assess a given question under circumstances where the set of choices is a combination of fewer underlying nonexclusive choices would help.) Would something like a gsem model of the three underlying choices be worth considering, or can it be dismissed out of hand for some reason?

[Jesus Pulido]

Dear Richard, Joao and Joseph,

Thanks for your elaborate replies. And thanks Joseph for your guess on what my concern is with this research. My research question focuses on the fact that despite these practices have been fostered by NGO's in Mexico for about 20 years (and despite obvious benefits they bring in terms of increased production/reduced costs), farmers have adopted them in different combinations, not always the 3 together as recommended. There's the obvious question of finding what combination is achieving utility maximization (i.e. profit / farmer's goals / farmer's perception of well-being) but that is not the topic of my research. I am focused on the determinants and patterns of adoption of each of the 8 combinations of practices.

I want to find what factors influence the adoption of the full package of practices. Is it the age category? Is it the ownership of the land? Is that the farmer's acquaintances? The frequency/quality of technical support? etc. There are several papers on this topic but they have never focused on the system of practices I talk about here...

Please allow me to dig a bit more into the methodologies and commands you kindly suggested here. Choice modelling and the gsem. And obviously rethink about the mlogit which seems to be the preferred choice in the literature (i.e. paper by Wu and Babcock, 1998). I'll keep you posted here, Cheers.

Jesus