You could just report the marginal effects for both -probit- and -truncreg-.
The idea behind a hurdle model is that there are basically 2 decisions. For the 2nd decision (the continuous portion) to come into play, you have to have a yes value on the first decision.
The Katchova and Miranda paper (that I referenced in the earlier post) notes that some variables may have, for instance, a positive effect on the binary choice to use a contract (yes/no) and a negative effect on the decision of how much to contract. The authors never tried to add or multiply the effects from both models together to get an aggregate effect, because it is two decisions.
My understanding is context specific (specifically the contracting decisions mentioned above), but I believe that the contribution of the hurdle model is that it allows us to differentiate between the influences of a variable on the binary and continuous decisions. If, for instance, you applied OLS to a dataset of including both farmers with contracts and farmers without contracts, you would confound these effects. Sometimes, researchers examine just the second question (the continuous portion of the hurdle model) using a truncated regression if that is the research question of true interst. However, if they do not also perform a probit or logit they get no information regarding the binary choice.
Hope this helps.
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Edoardo Di Porto
Sent: Thursday, July 02, 2009 1:36 PM
To: [email protected]
Subject: Re: st: RE: How to deal with Double hurdles post estimation...
And then I have to multiply the mfx...?...or just I have to read it
indipendently?
by the way many thanks
Ed
2009/7/2 Franken, Jason R. <[email protected]>:
> Ed,
>
> My understanding (from the publications I've seen applying a double hurdle model) is that you report the results for both probit and truncreg (see Katchova and Miranda cited below as an example). If you want marginal effects computed at the means, they can be obtained by following -probit- or -truncreg- commands with the -mfx- command.
>
> Best,
>
> Jason
>
> Katchova, A.L., M.J. Miranda. "TWO-STEP ECONOMETRIC ESTIMATION OF FARM CHARACTERISTICS AFFECTING MARKETING CONTRACT DECISIONS." American Journal of Agricultural Economics, 86,1(February 2004):88-102.
>
>
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of Edoardo Di Porto
> Sent: Thursday, July 02, 2009 4:31 AM
> To: [email protected]
> Subject: st: How to deal with Double hurdles post estimation...
>
> I've estimated a double hurdle model (Cragg,1971) following the
> procedure explained in Mc Dowell et al.
> So with a combination of command...
> I've estimated a firts stage with "probit" and a second with a "truncreg"
> I've the same covariates in both the stage. The dependent variables
> are: a dummy in the first stage (of course)
> while in the second I've a real positive, Let's call it "Y".
> I get my estimates for betas in both the stages...but now I need to compute:
> a) the betas for the double hurdle model
> b) predict Y for the double hurdle model
> c) compute the elasticities Y/Xi for the double hurdle model.
>
> I guess that: a) it should be the sum of the betas in the first and
> second stage ("B1 from probit+B1 from truncreg"...is it true?)
> b) could be obtained predicting the probabilities at the first stage
> (predict xb) and multiplying those for the predicted values of Y.?..
> No idea for c)!...someone knows anything about it.
>
> Ed
>
> *
> * 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/
>
> *
> * 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/
>
*
* 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/
*
* 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/
