# st: Finite sample problem while using -Hausman-, and *suest- not working with *ivreg-

 From "Sophie AUDOUSSET" To Subject st: Finite sample problem while using -Hausman-, and *suest- not working with *ivreg- Date Thu, 03 Aug 2006 17:54:55 +0200

```.
Dear statalist,
I'm currently trying to compare the estimates obtained via two models (first model is IVREG and second model is OLS regression) because one of my independant variable proved to suffer from endogeneity (I've tested the endogeneity by showing that this variable is correlated with the residual).
So I run my *ivreg- model (, robust) (and store IVREG estimates), then I run my OLS model (-regress- *, robust) and store OLS estimates.
I finally run Hausman command :
. hausman IVREG OLS, constant sigmamore
and I end up with a positive ChiČ stat of 53.95 (prob. 0.000), which means that I have to reject the null hypothesis and consider the IVREG model to be better than the OLS. My problem is that I have the message "V_b-V_B is not positive definite", which indicates that I'm certainly facing a finite (small) sample problem (n=144*). As suggested in the help file (hausman), I've tried to use the command *suest- instead of *hausman-, but it seems that *suest- is not suitable for IVREG models.
My question is: is there another command that I can use to solve my problem and manage to compare the estimates of the IVREG and the OLS model, in my case of a small sample ?

(Hereunder are the relevant stata outputs)

Thank you very much for your help
Best Regards
Sophie Audousset
PhD student. HEC Paris, School of Management

----------------------------------------------
. hausman IVREG ORDINARY , constant sigmamore

---- Coefficients ----
|      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
|     IVREG       ORDINARY      Difference          S.E.
-------------+----------------------------------------------------------------
logNAF |    .2767367     .0486494        .2280873        .0367947
size |    .3372361     .4842723       -.1470362        .0272735
receiv |    .1583165     .5930488       -.4347323        .1748345
sqsubsint |    .0274407     .0416121       -.0141714               .
diversif |    .0851476     .0807569        .0043907        .0113417
crosslis |    .3002547     .4174471       -.1171924               .
block |    .2212551    -.3124698        .5337249        .0796082
indep |    .7656288     .4138254        .3518033        .0802583
onebig4 |     .527467     .4738404        .0536266               .
twobig4 |    .5217803     .5443661       -.0225859               .
mand12_2 |    .4531177     .3186652        .1344524        .0543505
_cons |   -3.246519    -4.606056        1.359537        .2265545
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from ivreg
B = inconsistent under Ha, efficient under Ho; obtained from regress

Test:  Ho:  difference in coefficients not systematic

chi2(12) = (b-B)'[(V_b-V_B)^(-1)](b-B)
=       53.95
Prob>chi2 =      0.0000
(V_b-V_B is not positive definite)

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```