Interpreting and reporting xtgee

[Robin Klomp]

Hello,

I recently switched from an xtreg, fe robust model to xtgee, vce(robust) model because the regression assumptions were not met to perform xtreg. I learned how to interpret xtreg, but not xtgee. I would do it similarly to the xtreg output, reporting the significance of Wald > chi2 and then the p-values and coefficients of the regression - is that possible? Also, is it right that there is no r2 in xtgee? Can I use anything similar?

Any help is much appreciated!

Best,
Robin

 xtgee ROA2 EnvironmentalPillarScore crt_InvestorPressureScore IndependentBoardScore CeoDuality FirmAge at FinancialLeverage2 Envi
> ronmentalInnovationScore i.GicSectorCod i.Year, family(gauss) link(ident) corr(exch) vce(robust)       

Iteration 1: tolerance = .060165
Iteration 2: tolerance = .00246858
Iteration 3: tolerance = .00008445
Iteration 4: tolerance = 2.883e-06
Iteration 5: tolerance = 9.838e-08

GEE population-averaged model                        Number of obs    =  3,683
Group variable: firm                                 Number of groups =    559
Family: Gaussian                                     Obs per group:  
Link:   Identity                                                  min =      1
Correlation: exchangeable                                         avg =    6.6
                                                                  max =     11
                                                     Wald chi2(28)    = 557.53
Scale parameter = .0058369                           Prob > chi2      = 0.0000

                                                   (Std. err. adjusted for clustering on firm)
----------------------------------------------------------------------------------------------
                             |               Robust
                        ROA2 | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-----------------------------+----------------------------------------------------------------
    EnvironmentalPillarScore |  -.0000617   .0001273    -0.48   0.628    -.0003112    .0001878
   crt_InvestorPressureScore |  -7.28e-06   9.43e-06    -0.77   0.440    -.0000258    .0000112
       IndependentBoardScore |  -.0000255   .0000654    -0.39   0.696    -.0001538    .0001027
                  CeoDuality |    .000137   .0032736     0.04   0.967    -.0062791    .0065532
                     FirmAge |  -.0000395   .0000852    -0.46   0.643    -.0002065    .0001276
                          at |  -3.63e-08   1.42e-08    -2.56   0.010    -6.41e-08   -8.51e-09
          FinancialLeverage2 |  -.0432841    .042199    -1.03   0.305    -.1259927    .0394244
EnvironmentalInnovationScore |   .0000565   .0000801     0.71   0.481    -.0001005    .0002135
                             |
               GicSectorCode |
                         15  |   .0251066    .011395     2.20   0.028     .0027729    .0474403
                         20  |   .0595791   .0102459     5.81   0.000     .0394975    .0796607
                         25  |    .073868    .012161     6.07   0.000     .0500329    .0977031
                         30  |    .082808   .0180956     4.58   0.000     .0473413    .1182746
                         35  |   .0473829   .0110256     4.30   0.000     .0257732    .0689927
                         40  |  -.0477227   .0130711    -3.65   0.000    -.0733417   -.0221038
                         45  |   .0525775   .0111712     4.71   0.000     .0306823    .0744726
                         50  |   .0187791    .012591     1.49   0.136    -.0058987    .0434569
                         55  |  -.0181651   .0093623    -1.94   0.052    -.0365148    .0001846
                         60  |  -.0306139   .0260309    -1.18   0.240    -.0816335    .0204057
                             |
                        Year |
                       2011  |   .0049791   .0022142     2.25   0.025     .0006394    .0093188
                       2012  |  -.0028938   .0029369    -0.99   0.324      -.00865    .0028624
                       2013  |  -.0034104   .0030602    -1.11   0.265    -.0094084    .0025875
                       2014  |   .0000771   .0036499     0.02   0.983    -.0070765    .0072306
                       2015  |  -.0138303   .0063112    -2.19   0.028    -.0261999   -.0014606
                       2016  |  -.0077075   .0057525    -1.34   0.180    -.0189822    .0035673
                       2017  |  -.0035088    .006504    -0.54   0.590    -.0162565    .0092388
                       2018  |   .0031203   .0049443     0.63   0.528    -.0065704     .012811
                       2019  |  -.0090347   .0062895    -1.44   0.151    -.0213619    .0032925
                       2020  |  -.0233336   .0075638    -3.08   0.002    -.0381584   -.0085088
                             |
                       _cons |   .1309138    .013712     9.55   0.000     .1040387    .1577889
----------------------------------------------------------------------------------------------

[Joro Kolev]

I think what you are fitting with -xtgee- is equivalent to -xtreg, re-, so I do not see what you are gaining from this. There was a time when -xtreg- could not compute robust variances and -xtgee- could, but this was fixed in Stata 9 I think.

And the interpretation in linear -xtgee, family(gauss) link(ident)
- is the same as in -xtreg-, both are linear models.

[Robin Klomp]

Joro Kolev thank you for your answer. My supervisor suggested using xtgee because my DV doesn't fit the normality assumptions and she said a GEE makes up for that.

Is there something similar to the r-squared though? Do you know the best way to report the results?

[Carlo Lazzaro]

Robin:
please note that normality is a (weak) requirement for residuals only.
I do not know how -xtgee- could take non-normality of the regressand into account.
In addition, -xtgee- returns -e(chi2)- instead of -Rsq-.