Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

Re: st: logit model with endogenous


From   John Antonakis <[email protected]>
To   [email protected]
Subject   Re: st: logit model with endogenous
Date   Tue, 25 Mar 2014 21:31:56 +0100

Hi Richard:

Actually, with MPLUS and when using the MLR (maximum likelihood robust) estimator, one has to use the same "trick" that I showed below, that is:

MODEL:
y on binary;
binary on x1 x2 x3 x4;
f BY y@1 binary;
f@1;
[f@0];

The covariance of the disturbance is the loading of binary (and this is the Hausman test).

However, in MPLUS if one uses the WLSMV (Muthén's robust WLS) estimator then one can simply correlate the disturbances of the mediator and the endogenous variable.

It would be be great if Stata had the WLSMV estimator one day.

Ref: Muthén, B. O., du Toit, S. H. C., & Spisic, D. (in press). Robust inference using weighted least squares and quadratic estimating equations in latent variable modeling with categorical and continuous outcomes. Psychometrika.

Best,
J.

__________________________________________

John Antonakis
Professor of Organizational Behavior
Director, Ph.D. Program in Management

Faculty of Business and Economics
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis

Associate Editor:
The Leadership Quarterly
Organizational Research Methods
__________________________________________

On 25.03.2014 15:56, Richard Williams wrote:
Thanks John. When I worked with LISREL, I would do things like zap most of the matrices, e.g. all the observed variables would be Ys (no Xs), all the latent variables would also be in a single set of matrices rather than be split between exogenous and endogenous. This gave me a great deal of flexibility, e.g. I culd have correlated measurement errors between observed Xs and observed Ys. I haven't played around with SEM too much; I did find that a trick I thought should work had a terrible time converging. That was a while back though. I should see if things are working better now. It would be nice if you could do some things without having to use "tricks" to get them; my impression is that MPLUS may be better in this respect.

At 03:42 AM 3/25/2014, John Antonakis wrote:
[I don't think that this went through on the first send--resending]

Hi Richard:

-gsem- does not allow an explicit correlation of disturbances; thus, if the mediator is endogenous, the estimator will not be consistent. Thus, one is better off using ivprobit or cmp (from SSC). With gsem one would have to do a couple of tricks to get going as an instrumental-variable estimator. Here is a work-around that Stata's tech support gave me when I pinged them on this question, which I used to do with MPlus but had trouble doing it when using Stata. The solution is quite straight forward and includes modeling an extra latent variable, "L":

  cmp setup
  webuse laborsup, clear
  replace fem_inc = fem_inc - 10
  cmp (fem_work =  fem_educ kids) ///
      (other_inc = fem_work fem_educ kids male_educ), ///
      ind($cmp_probit $cmp_cont)

  gsem (fem_work <- fem_educ kids L@1, probit) ///
       (other_inc <- fem_work fem_educ kids male_educ L), var(L@1)

  /* Coefficients in the other_inc equation match */
  /* Probit assumes an error variance of 1, but we have added
   *    and extra variance of 1 from L. */
  /* To obtain probit coefficients, we can divide by the
   *    square root of 2 */

  nlcom (fem_educ : _b[fem_work:fem_educ] / sqrt(2)) ///
        (kids : _b[fem_work:kids] / sqrt(2)) ///
        (cons : _b[fem_work:_cons] / sqrt(2))

/* We also need to back transform results in order to obtain sigma and
   *  rho that are reported by -cmp-. */

  nlcom (sig: sqrt(_b[var(e.other_inc):_cons] +_b[other_inc:L]^2)) ///
(rho: _b[other_inc:L]/(sqrt(2)*sqrt(_b[var(e.other_inc):_cons] ///
      + _b[other_inc:L]^2)))

Best,
J.

__________________________________________

John Antonakis
Professor of Organizational Behavior
Director, Ph.D. Program in Management

Faculty of Business and Economics
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis

Associate Editor:
The Leadership Quarterly
Organizational Research Methods
__________________________________________

On 24.03.2014 02:04, Richard Williams wrote:
> Unclear. Do you mean you tried ivprobit, and it would not run? Or are
> you saying you don't think ivprobit is appropriate? If the former,
> show us the code and error messages. If the latter, explain what you
> have against using probit instead. I think it is rare that some great
> (or even small) harm would come from running a probit rather than a
> logit.
>
> I haven't tried it, but i wonder if you could do something like an
> ivlogit model using -gsem-.
>
> At 06:28 PM 3/23/2014, maggie zhang wrote:
>> Hi,
>>
>> I need to run logit model with endogenous covariates. There is only
>> ivprobit/ivtobit in stata. However, I could not run the probit
>> regression for my model instead of the logit. What could I do?
>>
>> Thanks,
>> Best,/Maggie
>> *
>> *   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/
>
> -------------------------------------------
> Richard Williams, Notre Dame Dept of Sociology
> OFFICE: (574)631-6668, (574)631-6463
> HOME:   (574)289-5227
> EMAIL:  [email protected]
> WWW:    http://www.nd.edu/~rwilliam
>
> *
> *   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/


*
*   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/

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  [email protected]
WWW:    http://www.nd.edu/~rwilliam

*
*   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/

*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index