hausman test

[Mustafa Ozer]

Dear Statalists,

Could you please please give me an answer? I have been looking for it via google and your forums since yesterday but i am not still quite sure.

does the test used with stata endog option in ivreg29 is hausman test for endogeneity?

I am not sure because whenever i read the forums they say you can run simple hausman test or use stata endog option? So what kind of endogeneity test they use with ivreg29?


Please please please help me!

[Mustafa Ozer]

From the help file of ivreg29,

"Unlike the Durbin-Wu-Hausman tests reported by ivendog, the endog option of ivreg2 can report
test statistics that are robust to various violations of conditional homoskedasticity; the ivendog option
unavailable in ivreg2 is the Wu-Hausman F-test version of the endogeneity test. See help ivendog (if
installed)."

Could you please tell me which test statistics are used in ivreg29 to calculate endogeneity test?

Kind regards,

Mustafa

[Mustafa Ozer]

Is there any body going to tell me what it is? I need to report the name of the test in my regression results.

i know it looks stupid question but i really need this information.

thanks

[Jeff Wooldridge]

You seem to be using an old version of Stata. As per board rules, you are supposed to tell us which version of Stata and show us your output. The -ivregress- command now does everything you need. I think I can help if you provide more information. I was unaware of an ivreg29 command, but I suspect it is using the control function test that I describe in my books. JW

[Eric de Souza]

Jeff
From the ivreg2 help file: "The most-up-to-date implementation of ivreg2 requires Stata version 11 or later. If ivreg2 is called under earlier versions of Stata, it will automatically run a legacy version ivreg2x, where "x" denotes the required Stata version. These versions of ivreg2 - ivreg28, ivreg29 and ivreg210 - are self-contained and require a minimum of Stata version 8/9/10, respectively.
"Self-contained" means these legacy versions (unlike the main up-to-date ivreg2 code) do not require access to any external Mata library or user-written Stata routines. These legacy
versions are installed with the ivreg2 package, can also be called directly from the Stata command line or in do files, and come with their own help files.

[Jeff Wooldridge]

Thanks, Eric. I didn't know this. So ivreg29 should have the same format and offer the same options as ivreg2, correct?

[Eric de Souza]

I assume so. I've never used it. I've always used ivreg2.

[Mark Schaffer]

Mustafa - the robust version of the DWH endogeneity test implemented in ivreg2 (and ivreg29 et al.) goes under various names, but a good one that describes how the test is obtained is a "GMM distance test" (in this case, it's a test of whether or not the endogenous regressor can be added to the set of orthogonality conditions). I'm sure it's described in Jeff's books somewhere but I'm on the road and can't find an exact reference for you.

Jeff - ivreg210, ivreg29 and ivreg28 have basically the same syntax as ivreg2. When we promote ivreg2 to requiring a newer version of Stata, we freeze the existing version and give it a new name. Until recently these were just stand-alone versions that were on SSC and had to be installed separately, but about 6 months ago we started bundling them in with ivreg2 along with their help files. This means that if someone has, say, Stata version 10 and wants to run ivreg2, and we promote ivreg2 to requiring Stata 11, they don't have to separately install and start using ivreg210. Rather, the existing ivreg2 will check the calling version of Stata and fork to the highest version of ivreg2 that will run under it (in this hypothetical example, ivreg210). It also means that someone can use version control to run an older version ivreg2 if they want to.

[Mustafa Ozer]

Dear Statalists,

Thank you for your answers.

I use Stata 11.2.

"The endogeneity test implemented by ivreg29, is, like the C
statistic, defined as the difference of two Sargan-Hansen statistics: one for
the equation with the smaller set of instruments, where the suspect
regressor(s) are treated as endogenous, and one for the equation with the
larger set of instruments, where the suspect regressors are treated as
exogenous. Also like the C statistic, the estimated covariance matrix used
guarantees a non-negative test statistic. Under conditional homoskedasticity,
this endogeneity test statistic is numerically equal to a Hausman test
statistic; see Hayashi (2000, pp. 233-34). The endogeneity test statistic can
also be calculated after ivreg or ivreg29 by the command ivendog. Unlike the
Durbin-Wu-Hausman tests reported by ivendog, the endog option of ivreg29 can
report test statistics that are robust to various violations of conditional
homoskedasticity; the ivendog option unavailable in ivreg29 is the Wu-Hausman
F-test version of the endogeneity test."

It says in other documents and here that endogeneity test with endog option of ivreg29 is C test, which is the difference of two hansen j statistic.
Having considered that my data is clustered, I suppose the test statistics calculated is not Durbin-Wu-Hausman test.? Am i right? it is C test, right?

This really confuse me, the data set i am using is clustered so there is no conditional homoskedasticity,which means it is C test. This sentence,
"the ivendog option unavailable in ivreg29 is the Wu-Hausman F-test version of the endogeneity test." make me belive that
it is not hausman test?

Thank you very much,

Mustafa

[Mustafa Ozer]

Dear Statalists,

I have understood that name of the test for endogeneity is GMM distance test. I found an article, Baum et al. (2007), describing it in details.

In the paper it also describes how to conduct orthog test for excluded instrument. I used the following regression and recived three test statistics regarding orthog condition.
Only second one fails to reject H null hypothesis. This made me confuse. Could you please help me with this? Which test statistics should i take into account?

May i say my instrument, instrumentr26d3n is valid?

Thank you

(By the way, i use stata 11.2. )

xi: ivreg29  H7  (V133=instrumentr26d3n wealths parity) yinvest26r26d3n i.cocuk26 yobin2-yobin5  r26d3  GE26ag_yob52- GE26ag_yob55 i.agedummies [aw=v005], nocon cluster  (cocuk26) first partial(i.cocuk26 yinvest26r26d3n yobin2-yobin5 GE26ag_yob52-GE26ag_yob55 r26d3 i.agedummies) orthog(instrumentr26d3n)


IV (2SLS) estimation


Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on cocuk26

Number of clusters (cocuk26) = 26                     Number of obs =     3684
F(  2,    25) =    41.86
Prob > F      =   0.0000
Total (centered) SS     =  776.1752258                Centered R2   =  -0.0443
Total (uncentered) SS   =  776.1822668                Uncentered R2 =  -0.0442
Residual SS             =  810.5264151                Root MSE      =    .4691


Robust
Shepa3_A       Coef.   Std. Err.      z    P>z     [95% Conf. Interval]

V133    .0605835   .0065392     9.26   0.000      .047767       .0734
r26d3   -.0015579   .0008939    -1.74   0.081    -.0033098    .0001941

Underidentification test (Kleibergen-Paap rk LM statistic):             23.017
Chi-sq(3) P-val =    0.0000

Weak identification test (Cragg-Donald Wald F statistic):              471.409
(Kleibergen-Paap rk Wald F statistic):        138.198
Stock-Yogo weak ID test critical values:  5% maximal IV relative bias    13.91
10% maximal IV relative bias     9.08
20% maximal IV relative bias     6.46
30% maximal IV relative bias     5.39
10% maximal IV size             22.30
15% maximal IV size             12.83
20% maximal IV size              9.54
25% maximal IV size              7.80
Source: Stock-Yogo (2005).  Reproduced by permission.
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.

Hansen J statistic (overidentification test of all instruments):        10.379
Chi-sq(2) P-val =    0.0056
-orthog- option:
Hansen J statistic (eqn. excluding suspect orthog. conditions):          0.123
Chi-sq(1) P-val =    0.7256
C statistic (exogeneity/orthogonality of suspect instruments):          10.256
Chi-sq(1) P-val =    0.0014
Instruments tested:   instrumentr26d3n

Instrumented:         V133
Included instruments: r26d3
Excluded instruments: instrumentr26d3n wealths parity
Partialled-out:       _Icocuk26_2 _Icocuk26_3 _Icocuk26_4 _Icocuk26_5
_Icocuk26_6 _Icocuk26_7 _Icocuk26_8 _Icocuk26_9
_Icocuk26_10 _Icocuk26_11 _Icocuk26_12 _Icocuk26_13
_Icocuk26_14 _Icocuk26_15 _Icocuk26_16 _Icocuk26_17
_Icocuk26_18 _Icocuk26_19 _Icocuk26_20 _Icocuk26_21
_Icocuk26_22 _Icocuk26_23 _Icocuk26_24 _Icocuk26_25
_Icocuk26_26 yinvest26r26d3n yobin2 yobin3 yobin4 yobin5
GE26ag_yob52 GE26ag_yob53 GE26ag_yob54 GE26ag_yob55
_Iagedummie_2 _Iagedummie_3
nb: small-sample adjustments account for
partialled-out variables


.
end of do-file

Three tests results to test the orthog cond. for instrumentr26d3n were reported. Among the tests given below, copied from the above results, only
second one fail to reject H null hypothesis. May i still say the excluded intstrument, instrumentr26d3n is valid?

Hansen J statistic (overidentification test of all instruments):        10.379
Chi-sq(2) P-val =    0.0056
-orthog- option:
Hansen J statistic (eqn. excluding suspect orthog. conditions):          0.123
Chi-sq(1) P-val =    0.7256
C statistic (exogeneity/orthogonality of suspect instruments):          10.256
Chi-sq(1) P-val =    0.0014
Instruments tested:   instrumentr26d3n

Regards,

Mustafa

[Mark Schaffer]

Mustafa:

1. You have 3 instruments, instrumentr26d3n, wealths and parity, and one endogenous regressor.

2. When you use two of the instruments, wealths and parity, the J stat is very small (0.123). This suggests that these two instruments are exogenous and the equation using just these two instruments is well specified.

3. When you use all three instruments, the J stat is very large (10.379). This suggests that the three instruments are not all exogenous.

4. The difference between the two J stats is very large (10.256). This is the C or GMM Distance statistic. It suggests that the additional instrument, instrumentr26d3n, is not exogenous. That is, it is not "valid".

Basically, adding instrumentr26d3n to the instrument set causes the J stat to go from pretty small (suggesting the estimation is well specified) to very large (suggesting misspecification).

HTH,
Mark