Re: st: RE: instruments in ivreg2

[Bidisha Mandal <mandal.7@osu.edu>]

Kit and Steve,

Thank you. I think I have finally figured out the appropriate 
instruments. I have 2 endogenous regressors, and managed 4 excluded 
instruments. The overidentification test gives a p-value of 0.15.
I do have another question though. Change in Marital status (the 
endogenous regressor) is categorical. So I am doing the 2SLS by brute 
force, using logistic regression in the first stage. Obviously 
correcting for the standard errors is slightly more complicated. Any 
suggestion?

Thanks again,

Bidisha Mandal

Dept. of AED Economics
The Ohio State University  


Steven Stillman wrote:

> Hi Bidisha.
>
> As Kit points out, it is fine to use a time-invariant variable as an
> excluded instrument for a time-variant endogenous variable, but you are
> quite likely to end up with a weak instrument problem.
>
> Depending on your data, you might be able to create slightly more
> informative time-variant instruments such as the number of cumulative years
> of each parent's current marriage or number of years since being married if
> not-married, cumulative number of marriages for each parent, average length
> of each parents marriages, etc.  If your data allows you to create more than
> one instrument of this flavour, you could then run an overidentification
> test.  You would still need to carefully test that none of these instruments
> was weak.
>
> Cheers,
> Steve
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu]On Behalf Of Kit Baum
> Sent: Sunday, October 22, 2006 1:59 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: instruments in ivreg2
>
>
> Bidisha said
>
> d(mental health) = a1 + d(marital status) + d(income) + e1
>
> You realize that the presence of the constant in this equation
> implies that there is a trend in mental health status? Although many
> of us may feel that way, I wonder if you really mean that. Same issue
> arises for the second equation.
>
> As to your question -- whether you can use parents' marital status
> (presumably still married v. divorced or widowed) as an instrument
> for the change in marital status observed in your sample... Surely
> you will observe some changes in your sample. For parents' status to
> be a valid instrument, it must be correlated with the endogenous
> measure ( d(mar.status)) and independently distributed of the error
> e1. The latter is reasonable, given that parental status is
> predetermined; but I wonder how well correlated it will be to the
> included measure. Look at the first stage regression results closely.
>
> As you have written them, both of these equations appear exactly
> identified, so that you cannot carry out any overid tests.
>
> Kit Baum, Boston College Economics
> http://ideas.repec.org/e/pba1.html
> An Introduction to Modern Econometrics Using Stata:
> http://www.stata-press.com/books/imeus.html
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
>   
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/