No chi2 value on xtreg, re vce(clustered)

[Rafael Moura]

Dear All,

My first post here, I hope I'm doing it right.

As stated in the title, I'm not getting any value for the chi2 on Random-effects GLS regression and am not sure why. I tried without the dummies (I have 4 in my equation) but still the same.

I appreciate any help to solve this.

Thank you so much.

Best regards,
Rafael Moura

[William Lisowski]

Welcome to Statalist, Rafael.

Please review the Statalist FAQ linked to from the top of the page, as well as from the Advice on Posting link on the page you used to create your post. Note especially sections 9-12 on how to best pose your question.

There are a lot things that can cause the problem, but without seeing your results, it's hard to guess which it might be, and unproductive to write up an exhaustive list of things to look for and what to do as a result. Please copy from your Stata Results window your xtreg command and all the output from it, and then paste that into Statalist using code delimiters, as the FAQ describes, and I expand on below.

The more you help others understand your problem, the more likely others are to be able to help you solve your problem.

Code blocks:

To assure maximum readability of results that you post, please copy them from the Results window or your log file into a code block in the Forum editor, as explained in section 12 of the Statalist FAQ linked to at the top of the page. For example, the following:

[code]
. sysuse auto, clear
(1978 Automobile Data)

. describe make price

storage display value
variable name type format label variable label
-----------------------------------------------------------------
make str18 %-18s Make and Model
price int %8.0gc Price
[/code]

will be presented in the post as the following:

. sysuse auto, clear
(1978 Automobile Data)

. describe make price

              storage   display    value
variable name   type    format     label      variable label
-----------------------------------------------------------------
make            str18   %-18s                 Make and Model
price           int     %8.0gc                Price

[Rafael Moura]

Thank you, William!

The results are pasted below. I'm using Stata/MP 13.0.

Before this, I ran the Hausman test in which I got Prob>chi2 = 0.5489, so I used RE after.

When I use default standard error I have no problem with the chi2 (0.0000, by the away), only when I use clustered robust.

. xtreg G logPOPGRO LFG RED RLFPR INV Trade CH U5MR, re vce(cluster Pais)

Random-effects GLS regression                   Number of obs      =       136
Group variable: paisnum                         Number of groups   =         5

R-sq:  within  = 0.3057                         Obs per group: min =        18
       between = 0.4727                                        avg =      27.2
       overall = 0.3058                                        max =        35

                                                Wald chi2(4)       =         .
corr(u_i, X)   = 0 (assumed)                    Prob > chi2        =         .

                                   (Std. Err. adjusted for 5 clusters in Pais)
------------------------------------------------------------------------------
             |               Robust
           G |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   logPOPGRO |  -.0033898   .0020639    -1.64   0.100    -.0074349    .0006553
         LFG |   1.212222    .748755     1.62   0.105    -.2553113    2.679754
         RED |   .1943591   .0404313     4.81   0.000     .1151152     .273603
       RLFPR |   .0310618   .0296789     1.05   0.295    -.0271077    .0892313
         INV |   .3390996   .1005654     3.37   0.001     .1419949    .5362042
       Trade |  -.0318276   .0237251    -1.34   0.180     -.078328    .0146728
          CH |   .0233269   .0047566     4.90   0.000     .0140041    .0326497
        U5MR |   .0030066   .0002531    11.88   0.000     .0025104    .0035027
       _cons |  -.3395725   .0551928    -6.15   0.000    -.4477484   -.2313967
-------------+----------------------------------------------------------------
     sigma_u |          0
     sigma_e |  .01982808
         rho |          0   (fraction of variance due to u_i)
------------------------------------------------------------------------------

The thing here is that I need to test for heteroskedasticity and autocorrelation and can't use rvfplot since I have panel data.

[Clyde Schechter]

Read -help j_robustsingular-, particularly the paragraph on the cluster robust estimator for a full explanation. In brief, the cluster robust estimator only supports a number of simultaneous parameter estimates less than the number of groups. Your model has 8 predictors but you have only 5 groups, so this cannot be done. The individual Wald tests in the main regression table are nevertheless not affected individually.

I should also point out that nearly all experts would agree that the vce(cluster robust) should not be used with only 5 clusters. There is no consensus on the minimum number of clusters needed, some say 10, some say 50, others say other numbers. But everyone would, I think, agree that at 5 clusters the vce() is not only no better than the standard vce(), it is actually worse.

[Rafael Moura]

Read -help j_robustsingular-, particularly the paragraph on the cluster robust estimator for a full explanation. In brief, the cluster robust estimator only supports a number of simultaneous parameter estimates less than the number of groups. Your model has 8 predictors but you have only 5 groups, so this cannot be done. The individual Wald tests in the main regression table are nevertheless not affected individually.

I should also point out that nearly all experts would agree that the vce(cluster robust) should not be used with only 5 clusters. There is no consensus on the minimum number of clusters needed, some say 10, some say 50, others say other numbers. But everyone would, I think, agree that at 5 clusters the vce() is not only no better than the standard vce(), it is actually worse.

Thank you for your explanation!

How would you suggest that I test for heteroskedasticity and autocorrelation? Should I use xtscc command?

. xtscc G logPOPGRO LFG RED RLFPR INV Trade CH U5MR  dumPT dumGR dumSU dumAL, re lag(4)
(119 missing values generated)

Regression with Driscoll-Kraay standard errors   Number of obs     =       136
Method: Random-effects GLS regression            Number of groups  =         5
Group variable (i): paisnum                      Wald chi2(8)      =    219.12
maximum lag: 4                                   Prob > chi2       =    0.0000
corr(u_i, Xb) = 0 (assumed)                      overall R-squared =    0.3422

------------------------------------------------------------------------------
             |             Drisc/Kraay
           G |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   logPOPGRO |  -.0025377   .0017002    -1.49   0.144    -.0059768    .0009014
         LFG |   1.087144   .7492214     1.45   0.155    -.4282997    2.602587
         RED |   .1901542   .0668886     2.84   0.007     .0548593    .3254491
       RLFPR |   .1732371   .0823767     2.10   0.042     .0066145    .3398597
         INV |   .3886117   .1140728     3.41   0.002     .1578777    .6193458
       Trade |  -.0059719   .0411082    -0.15   0.885     -.089121    .0771772
          CH |  -.0739275   .0853505    -0.87   0.392    -.2465651    .0987101
        U5MR |   .0019737   .0015282     1.29   0.204    -.0011173    .0050647
       dumPT |  -.0757584   .0682914    -1.11   0.274    -.2138907     .062374
       dumGR |   .0116881   .0105487     1.11   0.275    -.0096487    .0330249
       dumSU |   .0068724   .0157956     0.44   0.666    -.0250773    .0388221
       dumAL |   .0642993    .057324     1.12   0.269    -.0516495    .1802481
       _cons |  -.1865939   .1931039    -0.97   0.340    -.5771834    .2039956
-------------+----------------------------------------------------------------
     sigma_u |          0
     sigma_e |  .01982808
         rho |          0   (fraction of variance due to u_i)
------------------------------------------------------------------------------