Different results in Stata and Eviews fixed effects regression

[Ella Li]

I’m running a panel regression in both Stata and EViews, but I’m getting very different R² values and coefficient estimates despite using the same dataset and specifications (cross section fixed effects, cross section clustered SE). The panel is unbalanced with 409 cross sections and 14 periods, and Eviews auto-adjusted the periods to be 2012-2019 and 2023 because of lagged variables and missing data in 2020-2022.

R² is extremely low in Stata (<0.05) but high in EViews (>0.85). Some coefficient signs and significance levels are similar but not identical. eviews skipped 2020 and 2021; I didn't manually set that in Stata but the observation number matches
here is the results from Eviews. sorry I couldn't copy-paste the text or upload the file because I'm using the student lite version

. xtreg ln_homeless_vet_per10000_1 vet_black_rate vet_income median_rent_coc L1.own_vacancy_rate_coc L
> 1.rent_vacancy_rate_coc vet_pov_rate L1.vet_ue_rate ssi_coc own_burden_rate_coc rent_burden_rate_coc
>  L2.own_hpc L2.rent_hpc, fe vce(cluster coc_num)

Fixed-effects (within) regression               Number of obs     =      3,206
Group variable: coc_num                         Number of groups  =        362

R-squared:                                      Obs per group:
     Within  = 0.0495                                         min =          1
     Between = 0.0206                                         avg =        8.9
     Overall = 0.0255                                         max =          9

                                                F(12, 361)        =       4.94
corr(u_i, Xb) = 0.0442                          Prob > F          =     0.0000

                                       (Std. err. adjusted for 362 clusters in coc_num)
---------------------------------------------------------------------------------------
                      |               Robust
ln_homeless_vet_per~1 | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
----------------------+----------------------------------------------------------------
       vet_black_rate |  -.7471108   .3898904    -1.92   0.056    -1.513852    .0196308
           vet_income |  -4.72e-06   2.49e-06    -1.90   0.059    -9.61e-06    1.75e-07
      median_rent_coc |   1.03e-06   1.02e-06     1.01   0.312    -9.68e-07    3.02e-06
                      |
 own_vacancy_rate_coc |
                  L1. |   1.511589   3.545192     0.43   0.670    -5.460232    8.483411
                      |
rent_vacancy_rate_coc |
                  L1. |   .2485341   .4868649     0.51   0.610    -.7089136    1.205982
                      |
         vet_pov_rate |   .3256463   .4907035     0.66   0.507      -.63935    1.290643
                      |
          vet_ue_rate |
                  L1. |   1.098307   1.011106     1.09   0.278    -.8900894    3.086704
                      |
              ssi_coc |  -1.50e-07   5.36e-06    -0.03   0.978    -.0000107    .0000104
  own_burden_rate_coc |  -1.027336   .5897519    -1.74   0.082    -2.187117    .1324444
 rent_burden_rate_coc |   .7339676   .2552522     2.88   0.004     .2319997    1.235936
                      |
              own_hpc |
                  L2. |  -.2331625   .2056714    -1.13   0.258     -.637627    .1713021
                      |
             rent_hpc |
                  L2. |  -.1039091   .2698138    -0.39   0.700    -.6345134    .4266952
                      |
                _cons |   3.534651   .1609618    21.96   0.000     3.218111    3.851192
----------------------+----------------------------------------------------------------
              sigma_u |   1.140344
              sigma_e |  .45616874
                  rho |  .86205273   (fraction of variance due to u_i)
---------------------------------------------------------------------------------------

Stata’s diagnostic tests show presence of heteroskedasticity, serial correlation, and cross-sectional dependence, as shown below, but I’m unsure if I can use these results if the regression is so different from Eviews.

. xttest3

Modified Wald test for groupwise heteroskedasticity
in fixed effect regression model

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

chi2 (362)  =       4624.51
Prob > chi2 =          0.0000

. xtserial homeless_vet_per10000

Wooldridge test for autocorrelation in panel data
H0: no first-order autocorrelation
    F(  1,     358) =     12.948
           Prob > F =      0.0004

. xtcdf homeless_vet_per10000

xtcd test on variables homeless_vet_per10000
Panelvar: coc_num
Timevar: year
------------------------------------------------------------------------------+
    Variable    |  CD-test   p-value   average joint T | mean ρ   mean abs(ρ) |
----------------+--------------------------------------+----------------------|
 homeless_vet~0 +  65.874     0.000         10.73      +  0.06       0.27     | 18822 combinations o
> f panel units ignored (insufficient joint observations).
------------------------------------------------------------------------------+
 Notes: Under the null hypothesis of cross-section independence, CD ~ N(0,1)
        P-values close to zero indicate data are correlated across panel groups.

What else should I check to ensure both software are handling fixed effects and clustering the same way? Can I use robustness test results from Stata and regression results from Eviews?

Thank you in advance!

[Nick Cox]

Cross-posted at https://stats.stackexchange.com/ques...cts-regression and on https://www.reddit.com/r/stata/comme..._eviews_fixed/

It's an explicit request here, and a good idea everywhere, to tell people about cross-posting on other forums.

It's a rule on that subReddit -- as is explicit on the website.

[Carlo Lazzaro]

Ella:
the first detail that hit my eyes is the difference in the number of observations reported by the two statistical packages (3207 vs 3206).