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

Re: st: RE: 'movestay' command

From   Austin Nichols <>
Subject   Re: st: RE: 'movestay' command
Date   Thu, 9 Jul 2009 10:50:11 -0400

Clemence Berson<> :
It is not the responsibility of user programmers to support their
programs; contributions to SJ and SSC are provided "as is" with no
warranty, but I have several guesses as to why your questions did not
elicit a complete answer (e.g. claiming "I am not really able to read
the code" is not likely to increase your probability of good answers).
 It is clear from the -movestay- help file and implied by the SJ
article at
that the selection equation and the main outcome equation(s) can have
different sets of explanatory variables, and simply trying an example
illustrates that it is possible:

ssc inst movestay, replace
ren lmo_wage y
ren private s
movestay y edu*, select(s=age*)

But your "delta" is not a parameter of the -movestay- model, i.e. the
difference in expected wage rates conditional on entry to a given
sector does not influence the probability of entry except through the
correlation of error terms u_i and e_ji (this effect is not a constant
delta multiplying a difference in wages, and the selection model you
have written down is a linear probability model, so it is far from
clear what you intend).  You can predict u and either e1 and e2 given
the estimates, but not both e terms--though you could draw an e for
the unobserved sector from the estimated jointly normal distribution
of errors.  If you draw ten errors in this way and impute them to each
obs, you could estimate the partial correlation of the difference in
errors with various observed variables using a regression under
multiple imputation.  I have no idea if this kind of estimate would be
consistent or sensible, but you are free to derive the asymptotic
results and program such an estimator yourself.

On Thu, Jul 9, 2009 at 4:47 AM, Clemence
Berson<> wrote:
> I tried all these alternatives before borrowing the statalist, but I am not
> able to program in Stata then I am not really able to read the code.
> I sent an e-mail to M.Lokshin and he said he added this option in the last
> version of the command. I updated my version but I cannot find this option
> in the help file. Moreover, he did not answer me when I asked for some
> precisions. Does Anyone could help me ?
> Thanks for your attention,
> --
> Clémence Berson
> Nick Cox <> a écrit :
>> You can answer this kind of question yourself:
>> 1. Look at the help to see what options are documented.
>> 2. Look at the code using -viewsource- to see if any options are
>> undocumented.
>> Nick
>> Clemence Berson
>> My current working paper compares the discrimination between the private
>> and public sectors. I am using the Stata command 'movestay' developed
>> by Lokshin and Sajaia. I would like to know whether it is possible to
>> use control variables which are not present in the probit equation of
>> sector choice with this stata command.
>> ex:
>> lnw_1i = beta_1 (age age2 reg contract nbemployees) + u_1i
>> lnw_2i = beta_2 (age age2 reg contract nbemployees) + u_2i
>> I= delta(lnw_1i - lnw2i) + gamma(age age2 reg maritalstatus
>> csp_parents) + mu_i
>> where the type of contract and the number of employees in the firm do
>> not enter in the probability of working in a particular sector. Is
>> there an option to avoid adding these variables to the estimation of I
>> and getting the value of delta ?

*   For searches and help try:

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