Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.

# Re: st: Modeling proportion with endogenous treatment indicator

 From Guy Grossman To statalist@hsphsun2.harvard.edu Subject Re: st: Modeling proportion with endogenous treatment indicator Date Sat, 11 Sep 2010 09:24:03 -0400

```Thanks - this is helpful!

May I follow up with another question - how will the response change
when Yij is binary? treatreg and iverg2 assume continuous dependent
variable and ivprobit assumes continuous endogenous regressor - so
none would be appropriate.

By contrast ssm (wrapper for gllamm which is a user-written program)
can fit an IV model with endogenous binary treatment and binary
response. however it does not seem to allow clustering SEs at the
group level.

regards,
Guy

On Fri, Sep 10, 2010 at 5:33 PM, Austin Nichols <austinnichols@gmail.com> wrote:
> Guy Grossman <guygrossman1@gmail.com>:
> I would start with -ivreg2- (on SSC) and use robust SEs (or
> cluster-robust); the
> coefficients will certainly be easy to interpret as dp/dX.  But see
> http://www.stata.com/meeting/snasug08/abstracts.html#wooldridge
> for panel models with fractional outcomes and instruments.
>
> On Fri, Sep 10, 2010 at 5:25 PM, Guy Grossman <guygrossman1@gmail.com> wrote:
>> Dear Stata list -
>>
>> I am using Stata 10.1 on Mac and am seeking advice about the best way
>> to fit the following model:
>>
>> Yij = bo + b1 * Tj  + b2 * Ci + b3 * Cj + eij + ej
>>
>> Tj = v0 + v1 * Zj + epsilon j, where...
>>
>>
>> Yij = the dependent variable is a proportion (0<Yij <1), for person i
>> from group j.
>>
>> Zj = encouragement to take up one of two types of treatment - applied
>> at the group level j . Zj is binary: groups are either encouraged to
>> take up Zj=0 or Zj=1.
>>
>> Tj = treatment take up - takes place at the group level j. Tj is
>> binary (Tj=0 or Tj =1). Take up rates are about 80% for both
>> treatments.
>>
>> Ci = control variables at the individual level i
>>
>> Cj = control variables at the group level j
>>
>> The idea is to apply an encouragement research design, using Zj as an
>> instrumental variable for Tj.
>>
>> Given the nature of the dependent variable (proportion), my question
>> is how is it best to fit the model in Stata. In the past I have used
>> the glm command with link(logic) and family (bin) to fit a model with
>> a dependent variable that was a proportion, but the independent
>> variables were all exogenous. Is there a way to fit a glm with an
>> endogenous independent variable? Is it possible to use treatreg with
>> DV which is a proportion?
>>
>> I look forward for your astute recommendations.
>> Thanks!
>> Guy
>
> *
> *   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/
```