Performing Vuong's (1989) likelihood ratio test for equivalence of explanatory power in nonnested models

[Ali Eid]

Hello everyone, does anyone know how to perform Vuong's (1989) likelihood ratio test for equivalence of explanatory power in nonnested models. More specifically, i want to compare the explanatory power of two regression models. If you know, please give me the Stata commands to estimate Vuong (1989) test. Thank you very much.

[Jeff Wooldridge]

What kind of model are you using? I have Stata code comparing probit and logit, but it's easy to apply to any case where you can compute the log likelihoods for each observation. It is not enough to compute the total log likelihood. JW

[Jeff Wooldridge]

If you are using two linear regression models, then the approach is especially easy. Let me know, and I'll supply the Stata code here.

[Ali Eid]

Thank you dear Dr.prof Jeff Wooldridge. Actually, i am using fixed effects regression

[Ali Eid]

Dear Prof Jeff Wooldridge, could you please supply the Stata codes to compute Vuong (1989) in two linear regression models?

[Jeff Wooldridge]

Here is the Stata code. The following Vuong test for panel data can be reasoned from Section 13.11.2 in my 2010 MIT Press book, Econometric Analysis of Cross Section and Panel Data. In effect, you are comparing the sums of squared residuals from the two FE estimations. It is important that the dependent variable and the two samples of data used are the same.

The first model contains x1 ... xK and the second z1 ... zK. The cross sectional identifier is csid. The intercept in the final regression -- no other explanatory variables -- is the estimate of SSR(1)/NT - SSR(2)/NT. The clustering gives the proper standard error, and so the t statistics is the cluster-robust Vuong test.

xtreg y x1 x2 ... xK, fe
predict u1h, e
gen uh1sq = uh1^2
xtreg y z1 ... zM, fe
predict uh2, e
gen uh2sq = uh2^2
gen diff = uh1sq - uh2sq
reg diff, cluster(csid)

JW

[Zhigang Qin]

Dear Prof Jeff Wooldridge, could the code work for panel data re model?

[Canh Dang]

xtreg y x1 x2 ... xK, fe
predict u1h, e
gen uh1sq = uh1^2
xtreg y z1 ... zM, fe
predict uh2, e
gen uh2sq = uh2^2
gen diff = uh1sq - uh2sq
reg diff, cluster(csid)

JW

Dear Professor Wooldridge,

I am interested in how to perform Vuong (1989) test to "choose the best-fitting model when the models are nonnested". In your 2010 book (page 537), you mention that the test will see whether on log-likelihood value is significantly greater than another. I am perplexed on what log-likelihood values that you are referring to. I found this note from your course but still very confused on how you calculate the test statistics.

I want to do the test for a Cragg's model (following -craggit- command) and a Heckit model (following -heckman). Could you please provide more information?

Thank you.

[fernanda mazzotta]

Dear All,
I am interested in how to perform vuong test to choose the best model, between a IVtobit and graggmodels implemented with cmp, Somebody could help me?

[fernanda mazzotta]

Dear all,
about the previous question, Is it enough a LR test?

[fernanda mazzotta]

Dear Professor Wooldrige, could you send to me the code for comparing probit and logit, in this way I can try to apply to my cmp(ivtobit) and cmp (cragg model) becouse I understand that the total log likelihood is not enough

[sabrina mines]

Dear Professor Wooldridge,
I am interested in how to perform Vuong (1989) test to choose between test for a Cragg's model (following -craggit- command) and a Heckit model (following -heckman). Could you please provide more information?
thaank you

[Neophytos Lambertides]

Dear Professor Wooldridge, it is easy please to provide also the code for the Vuong test, for non-nested models, for Logit mainly models? Many thanks

[sufyan khan]

Dear Professor. I am working on impact of distance decay on household's willingness to pay for environmental quality applying choice modeling using mixed logit models, i run two models the 1st one don't include demographic factors and in the 2nd model i included demographic factors. now i wan to compare both the models, which test should i apply for the comparison of two models? also please send me the syntax. Regards

[Santosh Pathak]

Here is the Stata code. The following Vuong test for panel data can be reasoned from Section 13.11.2 in my 2010 MIT Press book, Econometric Analysis of Cross Section and Panel Data. In effect, you are comparing the sums of squared residuals from the two FE estimations. It is important that the dependent variable and the two samples of data used are the same.

The first model contains x1 ... xK and the second z1 ... zK. The cross sectional identifier is csid. The intercept in the final regression -- no other explanatory variables -- is the estimate of SSR(1)/NT - SSR(2)/NT. The clustering gives the proper standard error, and so the t statistics is the cluster-robust Vuong test.

xtreg y x1 x2 ... xK, fe
predict u1h, e
gen uh1sq = uh1^2
xtreg y z1 ... zM, fe
predict uh2, e
gen uh2sq = uh2^2
gen diff = uh1sq - uh2sq
reg diff, cluster(csid)

JW

Hi Professor Woolridge,

I am working on snp and sml models to estimate determinants of adaptation using univariate binary choice model. I wish to compare these two models using Vuong test? How can we implement this test under snp and sml framework?

Thanks in advance.

Regards,
Santosh