Double-Hurdle corner solution model

[Jonas Ngouhouo]

Good evening

I am running a Double-Hurdle corner solution model in other to characterise the preferences of local and indigenous households for elephant conservation.
After wrinting the command below, the maximization procedure failed to converge to a solution. Indeed, I have the following result.

Hope some one should help me.

Regards

Command

*TWO-PART USING CRAGGIT
global yhurdle1 discrete
global xhurdle1 lb sex age hsize educlvl expensemth autochbaka smalfarm tgoldmin hunt_gath fmu_foad othadmin hum_el ldarea dist_narea densit
global yhurdle2 maxwtp
global xhurdle2 sex age hsize educlvl expensemth autochbaka smalfarm tgoldmin hunt_gath fmu_foad othadmin hum_el ldarea dist_narea densit

craggit $yhurdle2 $xhurdle2, sec($yhurdle1 $xhurdle1)



Estimating Cragg's tobit alternative
Assumes conditional independence

initial: log likelihood = -<inf> (could not be evaluated)
feasible: log likelihood = -953.88316
rescale: log likelihood = -953.88316
rescale eq: log likelihood = 213988.65
numerical derivatives are approximate
flat or discontinuous region encountered
could not calculate numerical derivatives
discontinuous region with missing values encountered
r(430);

end of do-file

r(430);

[Friedrich Huebler]

I am only writing to note that craggit is a user-written command described in the Stata Journal, vol. 9, no. 4. Please see the Statalist FAQ, section 12.

[Jeff Wooldridge]

Jonas: It is easy to estimate Craigg's model "by hand." First, you estimate a probit model for the y = 0 versus y > 0 decision. Next, use a truncated normal regression model for y > 0 (with truncation at zero). That's what craggit should be doing, but you might check using the probit and trunreg commands. JW

[Jonas Ngouhouo]

Hi Jeff Wooldridge and Friedrich Huebler.
Thank you for your reply.
I finally succeeded to run it using "craggit" and to calculate the various values of interest (the partial effects of the indepvar on the probability of participation, the partial effects of the indepvar on the conditional and unconditional expected preferences).

[Jonas Ngouhouo]

.

[Enrique Pinzon (StataCorp)]

Dear Jonas,

You could also use the churdle command in Stata

[Jonas Ngouhouo]

Thank you Enrique .

[Sir Wentemi]

Hi Jonas Ngouhouo,

I am curious to know how you solved the problem with the craggit function, I am facing the same problem. Thank you in advance.

[Alfonso Sánchez-Peñalver]

Hi, churdle is only available if you have Stata 14 or later. If you have a previous version, and you don't want to do the estimation in the two steps described by Jeff, you can also consider the twopm package available on SSC. It is a very flexible command in that it allows many different specifications for two-part (hurdle) models.

[Enrique Pinzon (StataCorp)]

Hello Jonas and Alfonso,

Another alternative is to use gsem to estimate the hurdle model. Charles Lindsey and I have just written a post on the Stata Blog about the topic.

http://blog.stata.com/2016/06/07/mul...ts-using-gsem/

[Jonas Ngouhouo]

Dear Sir Wentemi,

Hope you will find this paper helpfull to succeed your double hurdel estimations.

The author is : William J. Burke
The tittle of its paper is : Fitting and interpreting Cragg’s tobit alternative using Stata
And the link to the paper is : http://ageconsearch.umn.edu/bitstream/143014/2/sjart_st0179.pdf

Best Regards

[Erik Thibaut]

There are are a number of (double) hurdle approaches that can be used:

• Craggit (type: "findit craggit", install user-written command and use the helpfile) --> very straightforward
• The Burke paper (as indicated by mister Jonas Ngouhouo)
• If these do not fit (which is sometimes the case, don't ask me why :-) ), you can also use a log normal hurdle approach, as explained by prof. Wooldridge in his "Introductory Econometrics: A Modern Approach by Wooldridge" --> Section 17.2 .

[esther Kemigisha]

Dear Statalist users, I am facing the same problem as Jonas Ngouhouo,


I am using double hurdle model using the craggit command, in consultation with the article: Burke, W. J. (2009). Fitting and interpreting Cragg's tobit alternative using Stata. Stata Journal,
9 (4): 584-592.

From this article I am able to get the average partial effects APEs for The probabilities regarding whether y is positive (Tier 1) and The expected value of y, conditional on y > 0 (Tier 2a) and Finally, the “unconditional” expected value of y (Tier 2b). I am getting these APEs from the three commands bellow and later using summarize command to get the mean APEs.

generate dpw1_dagehea=[Tier1]_b[agehead ]*normalden(x1g)
generate dEyyx2_dagehead=[Tier2]_b[agehead]*(1-IMR*(x2b/sigma+IMR))
generate dEy_dagehead= [Tier1]_b[agehead]*normalden(x1g)*(x2b+sigma*IMR)+[Tier2]_b[agehead]*normal(x1g)*(1-IMR*(x2b/sigma+IMR))

To get the standard deviations for each variable, I am using the nlcom command for Tier 1 and Tier 2a .

How ever I have two questions
1. How can I obtain p-values that corresponds ton each APEs for Tier 1, Tier 2a and Tier 2a .
1. How can I obtain standard errors for the Tier 2b (the “unconditional” expected value of y) since the nlcom does not specify whether it is for the conditional (Tier 2a) or unconditional (Tier 2b) expectation .

I look forward to your kind response

Esther

[FernandoRios]

Hi Esther
Not sure what version of Stata you are using, however, I think Stata introduced an official version of the craggit model (type help churdle) that does what you are trying to accomplish.
HTH
Fernando

[esther Kemigisha]

Hi Fernando
Thank you very much!
This is very helpful and I have found the answers to my questions.