Bookmark and Share

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

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: xtivreg first stage questions

From   "Schaffer, Mark E" <>
To   "" <>
Subject   st: RE: xtivreg first stage questions
Date   Wed, 4 Dec 2013 14:31:21 +0000


Some responses below.  But an extra note - you've left out some important details (see the Statalist FAQ).  Besides things like which versions of Stata and xtivreg you are using, you also should say which kind of estimator you are asking xtivreg to report (FE, G2SLS, EC2SLS, whatever).

> -----Original Message-----
> From: [mailto:owner-
>] On Behalf Of David Torres
> Sent: 03 December 2013 22:37
> To:
> Subject: st: xtivreg first stage questions
> Hello, Statlist,
> I have a couple quick questions about Stata's xtivreg command.  I'm using the
> command for a growth curve model and the variable I'm instrumenting for is
> a time-varying binary covariate.
> 1.  Does anyone know whether the first stage is calculated using OLS or logit
> or probit?  I'm told it's conventional approach to use a linear probability
> model when the predictor being instrumented for is binary.  I would
> appreciate a link to any literature discussing what is done in xtivreg's first
> stage.

This should be in the manual (previously hard copy, now PDF).

Exactly how to do the first stage depends on which estimator you are asking for (FE, EC2SLS, FD, whatever).

But for all of them, the first stage is least-squares-based, not logit or probit.  Your characterisation of the first stage as a "linear probility model" is actually a bit misleading.  The estimators implemented by xtivreg are single-equation (limited information) estimators.  The first stage is NOT a model and a structural interpretation is not guaranteed or required.

If you want to add some structure to the first stage, in effect you're thinking about multiple equation estimation and therefore different estimators.  This leads to the standard trade off - possible efficiency gains if the additional structural assumptions are right, possible inconsistency/bias if the additional structural assumptions are wrong.

A good place to read about this is Wooldridge's panel data textbook (easily google-able).  Another good place is Baltagi's panel data textbook (also easily google-able).

> 2.  Is there a way to get the F-statistic from the first stage?  Or is there
> another statistic (the z for the instrument, for instance) that I can use to
> show that my instrument is a good one?

Is the -first- option not reporting what you want?  It's described in the -xtivreg- online help.  Or use -xtivreg2- if you are using the FE or FD estimators, and the first-stage F will be reported automatically.

> 3.  I interact the binary variable and its instrument with dummies for race and
> sex and then include them in their respective portion of the parentheses
> part of the xtivreg model.  This is okay, right?

Hard to tell exactly what you mean.  But this sort of question comes up a lot on Statalist, and a search of the Statalist archives might be a good idea.  Here's a link to the most recent mention of a similar issue:

> 4.  Related to 2 and 3, since I have multiple instruments in the cross-level
> interaction model, I suppose I can get my Fs from the xtoverid command,
> using the noisily option.  Am I right?

Can't remember offhand (and I co-authored xtoverid, so this is a bad sign!).  Try it and see.


> Thanks,
> --
> David Diego Torres, PhD
> NSF SBE Postdoctoral Research Fellow
> Houston Education Research Consortium
> Rice University
> 6100 Main Street, MS-28
> Houston, TX  77005
> Phone: 713-348-2984
> Email: ddtorres at rice dot edu
> *
> *   For searches and help try:
> *
> *
> *

Sunday Times Scottish University of the Year 2011-2013
Top in the UK for student experience
Fourth university in the UK and top in Scotland (National Student Survey 2012)

We invite research leaders and ambitious early career researchers to 
join us in leading and driving research in key inter-disciplinary themes. 
Please see for further information and how
to apply.

Heriot-Watt University is a Scottish charity
registered under charity number SC000278.

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index