Statalist


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

RE: st: Cluster-robust estimate of the VCE, GMM, Mata


From   Rodolphe Desbordes <[email protected]>
To   "[email protected]" <[email protected]>
Subject   RE: st: Cluster-robust estimate of the VCE, GMM, Mata
Date   Fri, 24 Jul 2009 10:52:45 +0100

Dear Stas,

I apologise for my late answer. I have tried your solution and I have encountered two difficulties.

The first problem was the incompatibility of -optimize_result_V_robust()- with an evaluator of type d. I replaced - optimize_init_evaluatortype(S, "d2")- by  -optimize_init_evaluatortype(S, "v2"). The initial do file still worked fine after this modification.

The second problem is the output after inserting the code that you suggested:

"
 clear mata
 capture drop cons
 global xlist private medicaid age age2 educyr actlim totchr
 * Nonlinear 2SLS IV estimator for Poisson: computation using command optimize
 generate cons = 1
 local y docvis
 local xlist private medicaid age age2 educyr actlim totchr cons
 local zlist income ssiratio medicaid age age2 educyr actlim totchr cons

. mata
------------------------------------------------- mata (type end to exit) -------------------------------------------------------------------------------------------------
:   void pgmm(todo, b, y, X, Z, Qb, g, H)
>   {
>     Xb = X*b'
>     mu = exp(Xb)
>     h = Z'(y-mu)
>     W = cholinv(cross(Z,Z))
>     Qb = h'W*h
>     if (todo == 0) return
>     G = -(mu:*Z)'X
>     g = (G'W*h)'
>     if (todo == 1) return
>     H = G'W*G
>     _makesymmetric(H)
>    }

:   st_view(y=., ., "`y'")

:   st_view(X=., ., tokens("`xlist'"))

:   st_view(Z=., ., tokens("`zlist'"))

:   S = optimize_init()

:   optimize_init_which(S,"min")

:   optimize_init_evaluator(S, &pgmm())

:   optimize_init_evaluatortype(S, "v2")

:   optimize_init_argument(S, 1, y)

:   optimize_init_argument(S, 2, X)

:   optimize_init_argument(S, 3, Z)

:   optimize_init_params(S, J(1,cols(X),0))

:   optimize_init_technique(S,"nr")

:   b = optimize(S)
Iteration 0:  f(p) =  156836.04
Iteration 1:  f(p) =  21765.741
Iteration 2:  f(p) =  2087.4467
Iteration 3:  f(p) =  186.55764
Iteration 4:  f(p) =    182.298
Iteration 5:  f(p) =  182.29546
Iteration 6:  f(p) =  182.29541
Iteration 7:  f(p) =  182.29541

:
: optimize_result_V_robust( S )
[symmetric]
       1   2   3   4   5   6   7   8
    +---------------------------------+
  1 |  .                              |
  2 |  .   .                          |
  3 |  .   .   .                      |
  4 |  .   .   .   .                  |
  5 |  .   .   .   .   .              |
  6 |  .   .   .   .   .   .          |
  7 |  .   .   .   .   .   .   .      |
  8 |  .   .   .   .   .   .   .   .  |
    +---------------------------------+

:
:
: // some prep work in creating empty score_vars
: st_view( score_vars=., ., tokens("`score_vars'"))

: score_vars[,] = optimize_result_scores(S)
                 <istmt>:  3200  conformability error
r(3200);

: end
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------

. // of Mata
. _robust `score_vars' , cluster( cluster )
varlist required
r(100);

end of do-file

r(100);
"

Do you have any ideas on how to tweak the code?

Best regards,

Rodolphe


________________________________________
From: [email protected] [[email protected]] On Behalf Of Stas Kolenikov [[email protected]]
Sent: 17 July 2009 16:59
To: [email protected]
Subject: Re: st: Cluster-robust estimate of the VCE, GMM, Mata

Without going into any of the substantive details -- you could get the
sandwich VCE with

optimize_result_V_robust( S )

and in all likelihood you could get cluster corrected standard errors
along the lines of

// some prep work in creating empty score_vars
st_view( score_vars=., ., tokens("`score_vars'"))
score_vars[,] = optimize_result_scores(S)
end
// of Mata
_robust `score_vars' , cluster( cluster )

As a side note, I personally find clustering on age to be a relatively
strange idea, to tell you the truth. If you think there's dependence
on age, you could just as well use it as a covariate in your model
somewhere. Clustering is a problem when your sample was collected
using clusters of units sampled together; imposing clustering based on
exogenous variables is infrequently sensible, I think.

On Thu, Jul 16, 2009 at 6:53 AM, Rodolphe
Desbordes<[email protected]> wrote:
> Dear all,
>
> In "Microeconometrics Using Stata" by Colin Cameron and Pravin Trivedi, they explain, pp. 381-383, how to calculate a GMM estimator for a Poisson model with an endogenous regressor. The Mata code can be found here http://cameron.econ.ucdavis.edu/racd/trcount2009.do and I reproduce it below:
>
> "
> * Nonlinear 2SLS IV estimator for Poisson
>
>  capture drop cons
> capture drop cluster
>
> gen cluster=group(age)
>
> clear mata
> generate cons = 1
> local y docvis
> local xlist private chronic female income cons
> local zlist private chronic female income cons
> local cluster cluster
> mata
>  void pgmm(todo, b, y, X, Z, Qb, g, H)
>  {
>    Xb = X*b'
>    mu = exp(Xb)
>    h = Z'(y-mu)
>    W = cholinv(cross(Z,Z))
>    Qb = h'W*h
>    if (todo == 0) return
>    G = -(mu:*Z)'X
>    g = (G'W*h)'
>    if (todo == 1) return
>    H = G'W*G
>    _makesymmetric(H)
>   }
>  st_view(y=., ., "`y'")
>  st_view(X=., ., tokens("`xlist'"))
>  st_view(Z=., ., tokens("`zlist'"))
>
> st_view(C=., ., "`cluster'")
>
>  S = optimize_init()
>  optimize_init_which(S,"min")
>  optimize_init_evaluator(S, &pgmm())
>  optimize_init_evaluatortype(S, "d2")
>  optimize_init_argument(S, 1, y)
>  optimize_init_argument(S, 2, X)
>  optimize_init_argument(S, 3, Z)
>  optimize_init_params(S, J(1,cols(X),0))
>  optimize_init_technique(S,"nr")
>  b = optimize(S)
>  // Compute robust estimate of VCE
>  Xb = X*b'
>  mu = exp(Xb)
>  h = Z'(y-mu)
>  W = cholinv(cross(Z,Z))
>  G = -(mu:*Z)'X
>  Shat = ((y-mu):*Z)'((y-mu):*Z)*rows(X)/(rows(X)-cols(X))
>  Vb = luinv(G'W*G)*G'W*Shat*W*G*luinv(G'W*G)
>  st_matrix("b",b)
>  st_matrix("Vb",Vb)
> end
> "
>
> I would like to obtain a cluster-robust estimate of the variance-covariance matrix of the estimator (VCE).  Hence, in the Mata code, I tried to replace the expression of Shat by:
>
> "
>
> V = J(cols(Z), cols(Z), 0)
>        for(i=1; i<=colmax(C[.,1]); i++) {
>                st_subview(x_j=., select(X, C:==i), ., .)
>            st_subview(z_j=., select(Z, C:==i), ., .)
>                st_subview(y_j=., select(y, C:==i), ., .)
> V = V +(cross(cross(z_j, (y_j- exp(x_j*b')))', cross(z_j, (y_j- exp(x_j* b')))'))
>        }
>
>
> Shat = V*(rows(X))/(rows(X)-cols(X))*(colmax(C[.,1])/(colmax(C[.,1])-1))
>
> "
>
> However, I do not obtain the same estimates as those given by "poisson docvis private chronic female income, cl(age)". I am not sure what I am doing wrong. Any help would be greatly appreciated.
>
> Best regards,
>
> Rodolphe
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>



--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

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

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



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index