Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: SV: st: matched employer-employee panel data, IV-estimation, first stage: employer level, second stage: employee level


From   "Mark Schaffer" <M.E.Schaffer@hw.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: SV: st: matched employer-employee panel data, IV-estimation, first stage: employer level, second stage: employee level
Date   Mon, 13 Sep 2004 16:04:56 +0100

Jens,

Subject:        	SV: st: matched employer-employee panel data, IV-estimation, first stage: employer level, second stage: employee level
Date sent:      	Mon, 13 Sep 2004 16:09:22 +0200
From:           	"Jens Therkelsen" <JET@asb.dk>
To:             	<statalist@hsphsun2.harvard.edu>
Send reply to:  	statalist@hsphsun2.harvard.edu

> Mark,
> 
> Thanks for your response.  Maybe I should have made it clear, that
> each firm does not have the same number of employees. Actually, not at
> all. The largest firms will be represented by more observations in the
> first stage regression if it is performed on 300000 obs (individual
> level). When the largest firms "dominate" the first stage regression,
> they will also dominate the coefficients and therefore also the
> predictions. I don't think it is smart to let the largest firms
> dominate, but I'm not sure about this.
> 
> What do you think?

This is an interesting question.  One way to think about it is to ask 
whether you are concerned about letting the largest firms "dominate" 
the second stage, i.e., the structural equation.  It sounds like the 
answer is no, i.e., the coeffs will be determined mostly by lots of 
workers from a small number of large firms and this isn't a problem.  
If so, then maybe you shouldn't be concerned about the predicted 
values from the first stage either?  Or if not, then maybe you should 
be concerned in general?  There might be reasons to be concerned, but 
I guess you need to be explicit about why, and about why they matter 
for the profit-per-employee equation in one way and for the wage 
equation in another way.

Sorry for the wishy-washy answer, but maybe it will help anyway.  Or 
provoke one of the Statalist survey-data experts into a more informed 
response than mine.

Cheers,
Mark 

> 
> Jens
> 
> ________________________________
> 
> Fra: owner-statalist@hsphsun2.harvard.edu på vegne af Mark Schaffer
> Sendt: ma 13-09-2004 12:59 Til: statalist@hsphsun2.harvard.edu Emne:
> Re: st: matched employer-employee panel data, IV-estimation, first
> stage: employer level, second stage: employee level
> 
> 
> 
> Jens,
> 
> Subject:                st: matched employer-employee panel data,
> IV-estimation, first stage: employer level, second stage: employee
> level Date sent:              Mon, 13 Sep 2004 11:57:08 +0200 From:   
>                "Jens Therkelsen" <JET@asb.dk> To:                    
> <statalist@hsphsun2.harvard.edu> Send reply to:         
> statalist@hsphsun2.harvard.edu
> 
> > I have a matched panel with 500 firms (j) and 300000 employees (i) .
> >
> > I want to do a regression like this
> >
> > on the left:
> > wage(it)
> >
> > on the right:
> > individual variables such as age(it), education(it) and gender(i),
> > firm variables such as profits_per_employee(jt), firmsize(jt) and
> > fixed_assets_per_employee(jt) and industri-dummies(jt).
> >
> > But profits_per_employee is endogenous is the model as wages are
> > costs. And firm variables (j) are clustered.
> >
> > If I use "ivreg" or "xtivreg" the first stage regression seems to be
> > performed on 300000 observations and that must be wrong.(?)
> 
> I don't think this is "wrong", at least not in the sense you suggest.
> There are at least two ways to make this point.  One is to think of IV
> as a one-step estimator and the requirements to make it consistent. 
> The issue you've raised isn't one that violates these requirements.
> 
> Another way way to think about it is to ask what would be wrong with
> your first stage regression.  I think the answer is that the standard
> errors would be too small; but you don't need the first-stage SEs when
> you do the second-stage of IV.
> 
> You do have a problem, though, with the clustering by firm.  This will
> affect your first stage regression diagnostics (unless you adjust for
> clustering by firm or something like that) as well as your main
> regression.
> 
> Hope this helps.
> 
> --Mark
> 
> > An other option is to do the two steps seperately
> >
> > profits_per_employee = instruments........ on 500 oberservation and
> > save predictions
> >
> > wage = age education gender predicted_profits_per_employee
> > firmsize...,cluster(firm) on 300000 observations
> >
> > But am I getting it right this way?
> >
> > Any suggestions?
> >
> > Thanks for you time!
> >
> > Jens
> >
> >
> > *
> > *   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/
> 
> Prof. Mark E. Schaffer
> Director
> Centre for Economic Reform and Transformation
> Department of Economics
> School of Management & Languages
> Heriot-Watt University, Edinburgh EH14 4AS  UK
> 44-131-451-3494 direct
> 44-131-451-3008 fax
> 44-131-451-3485 CERT administrator
> http://www.som.hw.ac.uk/cert
> 
> *
> *   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/
> 
> 
> 

Prof. Mark E. Schaffer
Director
Centre for Economic Reform and Transformation
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS  UK
44-131-451-3494 direct
44-131-451-3008 fax
44-131-451-3485 CERT administrator
http://www.som.hw.ac.uk/cert


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index