st: R: Implementing mmsel

["Francesco Pastore" <francesco.pastore@unina2.it>]

Dear Sami,

I am implementing your routine mmsel in Stata 11. I have a small sample of
young men and women (about 1300 observations) from Kosovo and wish to study
the gender wage gap among them. I found no gwg at the mean of the
distribution, but suspect that it might exist at different quantiles of the
wage distribution. The participation rate is very low for both gender, but
especially women, and I suspect then that sample selection bias might be an
important issue when estimating the gender gap. Only 350 circa of the
overall sample work. For all these reasons, I think that the MM
decomposition with sample selection bias is exactly what I need for the type
of data I have.

You posted in the statalist a call for users to report possible problems,
issues and bugs with the routine. I have some doubts regarding the
implementation and interpretation of results in your routine. I will now ask
these question, hoping in your kindness to help me using the routine for the
better:

a) In my view, the helpfile is a bit sketchy and many points are left
unclear. I give some examples below;

b) Is there available some application of the routine to real data, with an
explanation of how to read the output?

c) I have run the routine with a very basic specification, just log hours
and educational variables, and I got the results (after some time, more than
an hour). I used this command: mmsel wage lhours $educ $hc $ethnic if
wemployee1, group(women) filename(foo) reps(100) pooled. The results look
clear, although not all the output is clear and you would like to check that
what you have got is ok. 

d) I have tried adding the educational levels of parents only and the
routine get stack, with the expression: ?invalid 'decomposition.log'?. I
have used the following command: mmsel wage lhours $educ $hc $edufm $ethnic
if wemployee1, group(women) filename(foo) reps(100) pooled. I cannot
understand what it means and how to proceed. Should I give up including
these variables? probably there are not enough observations? Is there any
way to control for other variables in the estimates?

e) I cannot run the sample selection variant of the routine. I get the error
message that the Ps1 variable is missing. I use this command: mmsel wage
lhours $educ $hc $edufm $ethnic, group(women) single($educ $hc $edufm
$ethnic) filename(foo) reps(100) pooled adjust(var)

f) In order to run the sample selection variant, is it necessary to add
options and which ones are the minimum ones? I think this is important
information, considering the time that is necessary to get the results;

g) Is it possible, for instance to run the selection correction version of
the command without adding any option (not either the single option one), to
obtain only the classical probit estimate of the participation (selection)
equation? And how?

h) Should the single index generation be run beforehand and separately to
obtain the variables Ps1?

i) Is it possible to provide the specification of the selection equation in
the mmsel command? and how? 

j) Are instruments necessary or just possible? 

Best regards
Francesco 


--
Dr. Francesco Pastore
Assistant Professor of Economics, Seconda Università di Napoli 
Palazzo Melzi, Piazza Matteotti, I-81055, Santa Maria Capua Vetere
(Caserta), Italy
Tel./Fax.: +39.0823.275530 (Office); and +39.081.199.72951 (Home1);
+39.081.5495276 (Home2); 
Mobile: +39.3498720406
Email: francesco.pastore@unina2.it; fpastore@unina.it
www.giurisprudenza.unina2.it/news.php?nid=185&pagina=2

Research fellow, IZA (Institute for the Study of Labour)
www.iza.org/profile?key=692




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/