Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Re: RE: Re: selmlog: question


From   Rasmus Jørgensen <rjo.cebr@cbs.dk>
To   statalist@hsphsun2.harvard.edu
Subject   st: RE: Re: RE: Re: selmlog: question
Date   Sun, 30 Apr 2006 22:15:17 +0200

Rafa,

Thanks once again. 

In the latest version of selmlog you can choose between several techniques,
eg. Dmf(0) and dmf(1). However, while the dmf(1) option includes a
correction term for each alternative, dmf(0) includes only M-1 correction
terms (outcome "1" is always excluded"). 

Do you know why? 

My problem is that when I "pool" outcome 1 and 3, I have missing values for
cprob11 but not for cprob31. Should these missing values be replaced with
zeroes?

My second question is regarding the economic interpretation of the
correction terms (having the Roy model in mind). When you pool the
correction terms like you suggested, then I guess you're implicitely
restricting the payoff to unobservables to be the same across sectors (here
sectors equals self-employment experience or not).

I suppose that if you add an interaction term, cprob_13 x self-employment
spell, then you would allow sector-specific payoff. 

Do you agree? And is it correct to do so when using the selmlog procedure?

I'm looking forward to your reply.

Best regards,

Rasmus

 

 


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of R.E. De Hoyos
Sent: 21 April 2006 20:20
To: statalist@hsphsun2.harvard.edu
Subject: st: Re: RE: Re: selmlog: question

Rasmus,

The mlogit model you are running is the same, in fact the probabilities 
predicted are also the same (the base category does not matter) however the 
way they are being parameterised to be included in the second-stage equation

is NOT.

In the second-stage equation for "w1" the predicted probabilities are being 
parameterised taking into account the information that outcome 1 has been 
chosen (cprob1). This will be different from the selection components 
(cprob3) for different selected outcomes. That's why you have to estimate 
the model two times.

Best,
Rafa


----- Original Message ----- 
From: "Rasmus Jørgensen" <rjo.cebr@cbs.dk>
To: <statalist@hsphsun2.harvard.edu>
Sent: Thursday, April 20, 2006 1:55 PM
Subject: st: RE: Re: selmlog: question


>
> Rafa,
>
> Thanks for your reply.
>
> There is, however, one thing that is puzzling me:
>
> Why do you have to run selmlog twice in order to get cprob1 and cprob3? As

> I
> see it, you're just running the same selection model once more and -- ex
> ante -- I would expect that it makes no difference in the estimation of
> correction terms.
>
> However, it does!
>
> _m3 from the first model with dep. var. w1 is much different from _m3 in 
> the
> second model. Do you have any ideas why this is the case?
>
> I magine that the difference may be caused by the fact that mlogit uses
> category 1 as base category, regardless of whether we model w1 or w3. Do 
> you
> agree?
>
> Once again any advice are gratefully appreciated.
>
> Regards,
>
> Rasmus
>
>
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of R.E. De Hoyos
> Sent: 12 April 2006 01:47
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Re: selmlog: question
>
> Rasmus,
>
> A way to do this is by generating the conditional probabilities (_m
> in -selmlog-) for your two outcomes of interest. You can then use them in 
> a
> single wage equation. Say "w1" are the observed wages under outcome 1
> (missing values otherwise) and "w3" are the observed wages under outcome 3
> (missing values otherwise) as you specified the problem. Then:
>
>    selmlog w1 x1 x2, sel(outcome x1 x2 z1) gen(cprob1)
>    selmlog w3 x1 x2, sel(outcome x1 x2 z1) gen(cprob3)
>
> The above model will allow for full wage parameter heterogeneity across
> outcomes 1 and 3. Depending on your particular problem this might be the
> best way to account for selection (allows for separate market equilibriums
> and different payments for the unobserved characteristics determining
> selection [cprob]). However if you want to impose the constraint of
> homogeneity in parameters across the wage equation for outcomes 1 and 3 
> but
> still treating them as different outcomes in your selection equation:
>
>    gen cpron_13=.
>    replace cprob_13 = cprob1 if outcome==1
>    replace cprob_13 = cprob3 if outcome==3
>
>    gen w_13=.
>    replace w_13 = w1 if outcome==1
>    replace w_13 = w3 if outcome==3
>
>    reg w_13 x1 x2 cprob_13
>
> This last model will estimate the wage equation for outcomes 1 and 3
> accounting for the unobserved characteristics that made the individuals
> "choose" those particular outcomes (although the market payment for those
> unobservables will be the same for both groups).
>
> Notice that you will have to bootstrap the standard errors to account for
> the heteroskedasticity present in the two-step procedure.
>
> I hope this helps,
> Rafa
> ________________________
> Rafael E. De Hoyos
> Faculty of Economics
> University of Cambridge
> CB3 9DE, UK
> www.econ.cam.ac.uk/phd/red29/
>
> ----- Original Message ----- 
> From: "Rasmus Joergensen" <rjo.cebr@cbs.dk>
> To: <statalist@hsphsun2.harvard.edu>
> Sent: Tuesday, April 11, 2006 8:26 PM
> Subject: st: selmlog: question
>
>
>> Dear Statalist,
>>
>> I'm trying to estimate the effect of self-employment experience. My
>> analysis considers the following selection rules:
>>
>> 1. Wage-employed in period t and period t+5
>>
>> 2. Self-employment spell between t and t+5.
>>
>> This selection model thus consider 4 possible outcomes as illustrated
>> below:
>>
>>                          WE,t and WE,t+5
>>                          YES                       NO
>>               YES       1                           2
>> SE spell
>>               NO         3                           4
>>
>>
>> One way to estimate this selection model is to use --selmlog--.
>>
>> However, selmlog can only estimate the wage equation (the equation of
>> interest) for one outcome of the selection process. But I'm interested in
>> running a wage regression for outcome 1 and 3 (see above). In other 
>> words,
>
>> I'm trying to estimate a model that accounts for both sample selection 
>> and
>
>> endogenous treatment (the SE spell).
>>
>> Does anyone have any advice how to correct --selmlog-- to estimate the
>> equation of interest for two outcomes of the selection process? Any
>> suggestions are very welcome.
>>
>> Thanks,
>>
>> Rasmus Jørgensen
>> Research Assistant
>> Centre for Economic and Business Research
>> E:< rjo.cebr@cbs.dk
>>
>>
>>
>> *
>> *   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/


*
*   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