[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: RE: LIML xtivreg2
Filipa de Castro <email@example.com>
Re: st: RE: LIML xtivreg2
Tue, 6 Oct 2009 13:34:02 -0500
Mark, many thanks for your email. I will follow your suggestion and
see if it works.
Again, many thanks
On Sat, Oct 3, 2009 at 6:20 PM, Schaffer, Mark E <M.E.Schaffer@hw.ac.uk> wrote:
>> -----Original Message-----
>> From: firstname.lastname@example.org
>> [mailto:email@example.com] On Behalf Of
>> Filipa de Castro
>> Sent: 03 October 2009 01:30
>> To: firstname.lastname@example.org
>> Subject: st: LIML xtivreg2
>> Dear all,
>> I am estimating a model with one endogenous variable
>> (hasmigr_US) and around 20 instruments using xtivreg2.
>> xtivreg2 goschool edad indchild oldest child45 child6 lowsixfam
>> edumam1 edumam2 edumam35 edumam68 edumam912 mommarried
>> madrejefefam ownhome survfam madremigfam smalltown
>> divorcesep agricolaheadspouse ctapropheadspouse oladummy
>> (hasmigr_US= instrl* instrpi* instrac2*) if sex==2 &
>> edad_10_13==1, fe i(ident ) liml
>> Having failed Cragg/Donald WALD F test I decided to use LIML
>> option which supposedly is robust for weak instruments.
>> After re-running the model with LIML option, Cragg/Donald
>> Wald F test and the other tests are ok, but I get some quite
>> odd and unrealistic coefficient for the endogenous variable
>> (right sign but wrong
>> (impossible) level).
>> What´s happening? Can you suggest what´s wrong or any other
>> option to deal with weak instruments.
> It's hard to tell exactly without seeing the output, but it's not very surprising. It's not quite right to say "LIML is robust to weak instruments". It's MORE robust than 2SLS/IV, but it's still susceptible to weak-instrument problems. Also, the LIML estimator has no moments, and so finding extreme (and extremely odd) coefficient estimates is going to happen more often than we'd like.
> The alternative approach to weak instruments is in the Anderson-Rubin tradition. Modern versions are due to Moreira, Kleibergen, Stock-Watson, Chernozhukov-Hansen, and others. This approach is genuinely robust, in the sense that as the instruments get weaker, the confidence interval around the parameter of interest gets wider. This is implemented in Stata by -condivreg- (which requires an iid assumption) and now, in the latest Stata Journal, -rivtest- by Finlay and Magnusson. These don't support fixed effects, but you could pehaps include dummies explicitly. -xtivreg2- reports a simple version of this approach in the first-stage output (see the help file for -ivreg2- for details).
>> many thanks
>> * 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/
> Heriot-Watt University is a Scottish charity
> registered under charity number SC000278.
> * 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: