Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: double hurdle model with Fractional logit second stage estimation

 From Jeph Herrin To statalist@hsphsun2.harvard.edu Subject Re: st: double hurdle model with Fractional logit second stage estimation Date Mon, 06 May 2013 13:30:08 -0400

```
```
Federico is correct, as long as your continous variable is 0 (and not missing) when the binary variable is 0. If it the continous variable is missing when the binary variable is 0, you can create a new depvar that has the correct properties:
```
gen newdepvar=cond(var1==0,0,var2)

and use this as the depvar.

hth,
Jeph

On 5/6/2013 10:40 AM, Federico Belotti wrote:
```
```Dear Beatrice,

if my understanding of your problem is correct, you can solve your issue by directly specifying as depvar your
continuous variable. -tpm- will automatically generate (for the first part regression) a dummy equal to 1 when the
continuous variable is greater than 0. My suggestion is to normalize your continuous variable in order to get a
depvar that is bounded between 0 and 1.

HTH, Federico

On May 6, 2013, at 4:08 PM, Beatrice Muriithi wrote:

```
```Thank you Jeph, I have installed tpm and tried to execute the model. Unfortunately tpm requires that the dependent
variables for both parts are the same (error:"The dependent variable must be the same in both equations" r(110)).
In my case, the dependent variable are different, the first stage has a discrete variable (0,1) and the second part
has continuous variable but include 0 and 100 (proportion). I will appreciate any further insights. Regards
Beatrice

On Mon, 06 May 2013 09:10:31 -0400 Jeph Herrin <stata@spandrel.net> wrote:
```
```Look at ssc describe tpm hth, Jeph On 5/6/2013 7:29 AM, Beatrice Muriithi wrote:
```
```Hey,

I have case of sample selection for which i would wish to use double hurdle model. The dependent variable for
the first stage is a discrete variable therefore i can use Probit model. The dependent variable for the second
stage is fractional response variable (contains 0-100  - percent). Is it ok if i use GLM (applied mainly for
fractional response variable problem) in the second stage? and if yes, is there a do program that can assist me
in this procedure to get the correct standard errors accounting for the first stage estimation?

Thank you Muriithi Universität Bonn

* *   For searches and help try: *   http://www.stata.com/help.cgi?search *
http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/ *   http://www.ats.ucla.edu/stat/stata/
```
```
Beatrice wambui Muriithi Junior Researcher, Centre for Development Research, ZEF; Universität Bonn
Walter-Flex-Street 3, 53113,Bonn Germany * *   For searches and help try: *   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/
```