Hi Christine.
Yes, from what I can tell, your 3-steps outlined below give you exactly the results you are expecting.
Cheers,
Steve
-----Original Message-----
From: [email protected]
[mailto:[email protected]]On Behalf Of Christer
Thrane
Sent: Wednesday, August 30, 2006 8:02 AM
To: [email protected]
Subject: st: Re: RE: Regression and "what-if" predictions
Thanks for your advice, Steven.
The problem with the automatic procedures is that they are designed for
(repeated) cross-sectional models, and not for panel regression models.
As regards the manual sollution, do you mean somthing like:
1.
use wagedata ...
drop if male==0
drop male
collapse (mean) edyears exp expsq nrchild (max) married year
save wagedata_men ...
This generates the married male in year 2002 with average values on the
remaining variables.
2.
use wagedata ...
xtreg logwage edyears exp expsq married if male==0, re cluster(lpnr)
This is the wage model for women.
3.
use wagedata_men ...
predict lwhat
list lwhat
This gets me the logwage for a married male in year 2002 with average values
on the remaining variables given the women's regression coefficients???
Am I on to something here?
BR
Christer
************************************************
Professor Christer Thrane
Department of Social Science
Lillehammer University College
2626 Lillehammer, Norway
+ 47 61 28 81 70 (fax)
+ 47 61 28 82 47 (phone, work)
+ 47 61 25 53 04 (phone, home)
E-mail, work: [email protected]
E-mail, home: [email protected]
************************************************
----- Original Message -----
From: "Steve Stillman" <[email protected]>
To: <[email protected]>
Sent: Tuesday, August 29, 2006 9:05 AM
Subject: st: RE: Regression and "what-if" predictions
> Hi Christine.
>
> The problem you are running into is that adjust isn't really designed to
> do something as complicated as you are trying to do.
>
> There are a number of user written commands that automate the process of
> doing wage decompositions. It is likely that one of these does what you
> are looking for - try findit "smithwelch", "jmp" , "oaxaca", "decompose",
> "jmp2".
>
> The other option is to use predict to do this by hand. one great feature
> of predict is that you can change datasets and still make predictions
> based on the last estimated model. So, all you need to do is set up a
> fake dataset (or add these as extra observations to your current dataset)
> with the values of the X variables that you want to use in your
> predictions and then run the appropriate regression model, change
> datasets, and use predict to create the appropriate fitted values.
>
> Hope this helps.
>
> Cheers,
> Steve
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of Christer
> Thrane
> Sent: Saturday, August 26, 2006 4:02 AM
> To: [email protected]
> Subject: st: Regression and "what-if" predictions
>
>
> Hi,
>
> The variables in my panel regression model are:
>
> logwage (log of annual wage)
> male (male=1, female=0)
> edlevel (5 levels of education)
> exp (working experience)
> expsq (working experience squared)
> year (1994=0, 1995=1, ... 2002=8)
> married (married=1, unmarried=0)
> nrchild (number of children)
>
> The first prediction I'm interested in is the:
>
> (i) log wage for a married male with edlevel = 3 in 2002 with average
> number
> of children and working experience.
>
> To obtain this, I use the code:
>
> xi: xtreg logwage i.edlevel exp expsq year /*
>> */ married nrchild if male==1, re cluster (lpnr)
>
> and
>
> adjust _Iedlevel_2=0 _Iedlevel_3=1 _Iedlevel_4=0 _Iedlevel_5=0 /*
>> */ exp expsq year=8 married=1 nrchild if male==1
>
> This seems to work well.
>
> The second prediction I'm interested is analogues to (i), but now I want
> to
> answer the question, What would be "the same" male's logwage if he had
> female averages on the various independent variables in the model?
>
> To obtain this second prediction (ii), I simply replace the "male==1" with
> "male==0" in the -adjust- command, as in:
>
> adjust _Iedlevel_2=0 _Iedlevel_3=1 _Iedlevel_4=0 _Iedlevel_5=0 /*
>> */ exp expsq year=8 married=1 nrchild if male==0
>
> This also seem to work well. At least exp, expsq and nrchild are given the
> female's averages, but I'm not sure of whether the effect of married and
> year are evaluated at the female's averages here.
>
> Now comes the tricky part, i.e. the two predictions I don't know how to
> get:
>
> (iii) What would "the same" male's logwage if he had male averages on the
> various independent variables, but the female's xt regression
> coefficients?
>
> (iv) What would "the same" male's logwage if he had female averages on the
> various independent variables, but the female's xt regression
> coefficients?
>
> At some point this, I think, involves the regression:
>
> xi: xtreg logwage i.edlevel exp expsq year /*
>> */ married nrchild if male==0, re cluster (lpnr)
>
> to obtain the female's xt regression coefficients, but I can't seem to
> figure out how to "force" the female's coefficients onto the male sample.
> Any help is appreciated!
>
> Best regards,
>
> Christer Thrane
> Lillehammer University College
>
>
> *
> * 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/
*
* 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/
