Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

R: st: -cmp for mixed and nonrecursive process?


From   Jordana Rodrigues Cunha <jordana.rodrigues@unibo.it>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   R: st: -cmp for mixed and nonrecursive process?
Date   Tue, 7 Dec 2010 10:02:08 +0000

Dear Robert thank you very much for the response, I've taken a bit of time to understand better my problem and after answer the questions you've made, I rethink the model differently:

First of all: 
I ran an OLS regression for the 1st equation, to check the significance level of X and W in Y:

1)reg Y X W var1 var2 var3, where:

Y is a continuous variable no time-varying;
X  is binary dummy that varies from 0 to 1
W  is an ordinal variable that varies from 1 to 5
var* are controls

In managing the possibility of bias of the OLS regression; I decided to do a Hausmann test using a 2 stage regression, using Z1 and Z2 as IV to check for the endogeneity in this way:

2)probit X  Z1 var4 var5 var6
predict res_X

3) oprobit W Z2 var4 var5 var6 
predict res_W*


4) reg Y X resX W res_W1 res_W2 res_w3 res_W4 res_W5 var1 var2 var3

The coefficient of the residuals are not significant correlated to Y, meaning that neither X or W are correlated to Y error term and so I conclude that they are not endogenous in the fisrst equation.

Are my results valid? I tried to use 

cmp (Y = X W var1 var2 var3) (X = Z1 var4 var5 var6) (W = Z2 var4 var5 var6), ind ($cmp_cont $cmp_probit $cmp_oprobit) and I can get convergence even after 100 interactions even after setting the scaled gradient tolerance to a value larger than its default of 10(-5) to 10(-3). 

obs: your precedent questions:
1 -Are these variables time-varying or not?  No, they aren't.
2- Do you have strong instruments for any of them? If so, how many?  Yes, I have strong instruments for both of the variables which I would like to test the endogeneity;
3- How many control variables are you using? in the first equation I am using 6 control variables (plus YEAR controls, I have 37 different years, 10 industry sectors controls  and controls for 65 countries ) for the 2nd and the 3rd variables I am using 3 (more YEAR controls) 
4- Can you bifurcate the ordinal variable? I've tried dot do so, and I lost significance, I've built it in a really ordered scale, where at each level foward I increase complexity respect to the precedent level, so I prefere do not do it if it would be possible

Jordana Rodrigues Cunha
PhD. Candidate
University of Bologna
Department of Management
Via Capo di Lucca, 34, 1st floor
40126 ? Bologna, ITALY
Fixed line:  0039 (051) 20 98 073
Fax: 0039 (051) 20 98 074
jordana.rodrigues@unibo.it
www.sa.unibo.it
________________________________________
Inizio: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] per conto di Robert A Yaffee [bob.yaffee@nyu.edu]
Inviato: giovedì 25 novembre 2010 16.48
Fine: statalist@hsphsun2.harvard.edu
Oggetto: Re: st: -cmp for mixed and nonrecursive process?

Jordana,
   I have a few preliminary questions?  Are these variables time-varying or not?   Do you have strong instruments for any of them? If so, how many?   Can you impose constraints?   Have you tested for identification of your model with an order or rank test?  If the parameters are time-varying, you will also have to be concerned with the stability of the feedback loops?
   Assuming that you can impose enough constraints or have enough instruments, and that the resulting model is identified,  you might want to consider using a structured equation model approach with a polyserial-polychoric covariance matrix as inputs.  Check out Stas Kolenkov's work on confa for the application of such input.   How many control variables are you using?
    Can you bifurcate the ordinal variable?
             Robert

Robert A. Yaffee, Ph.D.
Research Professor
Silver School of Social Work
New York University

Biosketch: http://homepages.nyu.edu/~ray1/Biosketch2009.pdf

CV:  http://homepages.nyu.edu/~ray1/vita.pdf

----- Original Message -----
From: Jordana Rodrigues Cunha <jordana.rodrigues@unibo.it>
Date: Wednesday, November 24, 2010 10:16 am
Subject: st: -cmp for mixed and nonrecursive process?
To: "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>


> Dear statalisters, I really need your help, seems that this is an
> impossible mission. I have consulted all the precedent faq's to arrive
> until here, but without success.
>
> I am estimating the effects of X, Z  and W over Y, where they are (in
> sequence): a dummy, an ordinal, a dummy and a continuous variable. The
> three independent variables are linked by a nonrecursive looping,
> meaning that they are linked by reciprocal feedbacks: X determines Z
> and W, as well as Z determines X and W and W determines Z and Y.
>
> X = a1+ bZ + cW + λ1Controls + e1; (binary dummy that varies from 0 to
> 1)
> Z = a2 + dX + eW + λ2Controls + e2;(ordinal variable that varies form
> 1 to 5)
> W = a3 + fX + gZ + λ3Controls + e3 ; (binary dummy that varies from 0
> to 1)
>
> the full model would be:
>
> Y = a4 + hX + iZ + jW + λ4Controls + e4 (continuous variable)
>
>
> I have made simple probits and ordered probits to check the
> relationship among the independent variables and I executed -cmp
> models to check the correlations among their error terms (in pairs of
> variables), confirming that they were really non-independent of each
> other. The problem is that as -cmp doesn't allow for reciprocal
> interaction among the variables and I included the second variable in
> the first equation and omitted the first variable in the second
> equation and so on. I have run, a naive OLS where the coefficients of
> X and Z were significant but not the coefficient of  W. I think that
> the best would run a SEM to use the 3 equations of X, Z and W with the
> fourth Y in the full model structured to allow for endogeneity but I
> have two main problems:
>
> 1- I cannot execute a Hausmann test ( to check for endogeneity in the
> full model) because when I ask for the residuals prediction after
> running oprobit (for estimate the variable Z) this option is not
> allowed and so I cannot regress the full model with the residual of Y
> and check the significance of its coefficient.
>
> 2- I have four different functional distributions and so if I would
> like to do 2sls or 3sls in different stages I wouldn't know  how to
> indicate the nature of the distribution for each variable and 3sls
> assumes that all the variables are continuous, right?
>
> 3 - I am using the same controls in all the equations, this could be a
> problem?
>
>
> Please, somebody could give me an advice? I hope I had been clear in
> explaining, thank you all in advance,
>
> jordana
> *
> *   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/

*
*   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 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index