Hausman Test

[Ishetu Mengesha]

I am facing the following problem when running the following test and how can I correct it?

. hausman fe re

Note: the rank of the differenced variance matrix (2) does not equal the number of coefficients being tested (4); be sure this is what you expect, or there may be
        problems computing the test.  Examine the output of your estimators for anything unexpected and possibly consider scaling your variables so that the
        coefficients are on a similar scale.

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |       fe           re         Difference          S.E.
-------------+----------------------------------------------------------------
         fdi |    1.93e-06     2.57e-06       -6.44e-07        1.39e-07
         gdp |    .1889387      .200659       -.0117204        .0034824
       nbtot |    8.87e+07     8.15e+07         7231909         2633943
      exrate |    282453.4     103381.3        179072.1        192646.8
------------------------------------------------------------------------------
                           b = consistent under Ho and Ha; obtained from xtreg
            B = inconsistent under Ha, efficient under Ho; obtained from xtreg

    Test:  Ho:  difference in coefficients not systematic

                  chi2(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                          =        8.75
                Prob>chi2 =      0.0126

[Carlo Lazzaro]

Ishetu:
try:

hausman fe re, sigmamore

[Ishetu Mengesha]

Hello Carlo Thank you for your support. But still the problem is there after I tried by the following

. hausman fe re, sigmamore

Note: the rank of the differenced variance matrix (2) does not equal the number of coefficients being tested (4); be sure this is what you expect, or there may be
        problems computing the test.  Examine the output of your estimators for anything unexpected and possibly consider scaling your variables so that the
        coefficients are on a similar scale.

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |       fe           re         Difference          S.E.
-------------+----------------------------------------------------------------
         fdi |    1.93e-06     2.57e-06       -6.44e-07        1.56e-07
         gdp |    .1889387      .200659       -.0117204        .0036625
       nbtot |    8.87e+07     8.15e+07         7231909         2998864
      exrate |    282453.4     103381.3        179072.1        202271.2
------------------------------------------------------------------------------
                           b = consistent under Ho and Ha; obtained from xtreg
            B = inconsistent under Ha, efficient under Ho; obtained from xtreg

    Test:  Ho:  difference in coefficients not systematic

                  chi2(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                          =        6.79
                Prob>chi2 =      0.0335

.

[Carlo Lazzaro]

Ishetu:
you may want to give it one more try with the community-contributed command -xtoverid-.

[Ishetu Mengesha]

Carlo:
I have tried but it is unrecognized command.

[Carlo Lazzaro]

Ishetu:
you should first install -xtoverid-.
Type:

search xtoverid

and follow the instructions to install it.

[SABA GULNAZ]

When I run Fixed Effects (within) regression without robust command all variables are significant.

Fixed-effects (within) regression Number of obs = 210
Group variable: IDINDIAASEAN Number of groups = 10

R-sq: Obs per group:
within = 0.8041 min = 21
between = 0.3309 avg = 21.0
overall = 0.3474 max = 21

F(5,195) = 160.09
corr(u_i, Xb) = -0.8470 Prob > F = 0.0000

---------------------------------------------------------------------------------
logXijt | Coef. Std. Err. t P>|t| [95% Conf. Interval]
------------------+--------------------------------------------------------------
logGDPitGDPjt 2.395327 .6264141 3.82 0.000 1.15991 3.630743
logPCGDPitPCGDPjt | -2.828763 .784914 -3.60 0.000 -4.376774 -1.280752
logDistanceij | .4089278 .0314809 12.99 0.000 .3468412 .4710145
TRGDPit | 1.624585 .246403 6.59 0.000 1.138628 2.110542
TRGDPjt | .3707311 .0784522 4.73 0.000 .2160075 .5254548
Borderij | 0 (omitted)
_cons | -31.49102 9.28393 -3.39 0.001 -49.80083 -13.18122
------------------+----------------------------------------------------------------
sigma_u | 1.660084
sigma_e | .21798139
rho | .98305059 (fraction of variance due to u_i)
-----------------------------------------------------------------------------------
F test that all u_i=0: F(9, 195) = 31.03 Prob > F = 0.0000

. xtcsd, pesaran abs

Pesaran's test of cross sectional independence = 6.797, Pr = 0.0000

Average absolute value of the off-diagonal elements = 0.433

. xttest3

Modified Wald test for groupwise heteroskedasticity
in fixed effect regression model

H0: sigma(i)^2 = sigma^2 for all i

chi2 (10) = 884.53
Prob>chi2 = 0.0000

Because of presence of heteroskedasticity I run fe robust command, but now the two variables which are important to my model have become insignificant
Kindly suggest solution to this problem

. xtreg logXijt logGDPitGDPjt logPCGDPitPCGDPjt logDistanceij TRGDPit TRGDPjt Borderij , fe robust
note: Borderij omitted because of collinearity

Fixed-effects (within) regression Number of obs = 210
Group variable: IDINDIAASEAN Number of groups = 10

R-sq: Obs per group:
within = 0.8041 min = 21
between = 0.3309 avg = 21.0
overall = 0.3474 max = 21

F(5,9) = 37.57
corr(u_i, Xb) = -0.8470 Prob > F = 0.0000

(Std. Err. adjusted for 10 clusters in IDINDIAASEAN)
-----------------------------------------------------------------------------------
| Robust
logXijt | Coef. Std. Err. t P>|t| [95% Conf. Interval]
------------------+----------------------------------------------------------------
logGDPitGDPjt | 2.395327 1.590739 1.51 0.166 -1.203175 5.993829
logPCGDPitPCGDPjt | -2.828763 1.988721 -1.42 0.189 -7.327561 1.670036
logDistanceij | .4089278 .0810488 5.05 0.001 .2255828 .5922729
TRGDPit | 1.624585 .4255777 3.82 0.004 .6618609 2.587308
TRGDPjt | .3707311 .1472481 2.52 0.033 .0376327 .7038295
Borderij | 0 (omitted)
_cons | -31.49102 23.55634 -1.34 0.214 -84.77917 21.79712
------------------+----------------------------------------------------------------
sigma_u | 1.660084
sigma_e | .21798139
rho | .98305059 (fraction of variance due to u_i)
-----------------------------------------------------------------------------------

. xtreg logXijt logGDPitGDPjt logPCGDPitPCGDPjt logDistanceij logPCGDPDijt TRGDPit TRGDPjt , fe robust

Fixed-effects (within) regression Number of obs = 210
Group variable: IDINDIAASEAN Number of groups = 10

R-sq: Obs per group:
within = 0.8056 min = 21
between = 0.2801 avg = 21.0
overall = 0.2944 max = 21

F(6,9) = 30.43
corr(u_i, Xb) = -0.8796 Prob > F = 0.0000

(Std. Err. adjusted for 10 clusters in IDINDIAASEAN)
-----------------------------------------------------------------------------------
| Robust
logXijt | Coef. Std. Err. t P>|t| [95% Conf. Interval]
------------------+----------------------------------------------------------------
logGDPitGDPjt | 2.590503 1.591277 1.63 0.138 -1.009215 6.190221
logPCGDPitPCGDPjt | -3.023769 1.967147 -1.54 0.159 -7.473765 1.426227
logDistanceij | .4070484 .0808357 5.04 0.001 .2241855 .5899114
logPCGDPDijt | -.1787298 .1919635 -0.93 0.376 -.6129814 .2555218
TRGDPit | 1.527413 .4286397 3.56 0.006 .5577628 2.497063
TRGDPjt | .3896496 .1457469 2.67 0.025 .0599472 .719352
_cons | -34.10911 23.56937 -1.45 0.182 -87.42673 19.2085
------------------+----------------------------------------------------------------
sigma_u | 1.9337988
sigma_e | .21771865
rho | .98748305 (fraction of variance due to u_i)
--------------------------------------------------------

[Carlo Lazzaro]

Saba:
welcome to this forum.
For the future, please use CODE delimiters to share what you typed and what Stata gave you back (see the FAQ on this and other posting-related topics) open a new thread as you actually queued up to a previous one that has nothing to do with yours. Thanks.
That said:
1) if you have detected heteroskedasticity, you should invoke non-default standard errors (as you did): if the coefficients of the predictors you are intrested in turn out insignificant, you've got to live with that. The significance you got with default standard errors is biased by heteroskedasticity.
2) on the other side, 10 clusters are probably few to invoke non-default standard errors (and so they can be misleading).
) Finally I would check whether heteroskedastcity may depend on logging too many predictors.

[SABA GULNAZ]

Carlo Lazzaro Thank you so much for your reply and my apology for posting question like this. I will keep in my mind the caution while posting question next time.

With regards,
Saba