Hello!

I have conducted a systems-GMM analysis aiming to determine whether a relationship exists between GDP per capita growth (agr) and income inequality (ginid), and I'm now looking to test whether this effect differs for low-income and high-income countries.

To do this, I have first conducted separate regressions for both high-income and low-income countries using the following codes:

High-Income:
Code:
xtabond2 agr gdpL1 ginid schl invrt pli if avggdpL1 > 9.594971, gmmstyle(gdpL1 ginid schl invrt pli, lag(2 2) eq(level) collapse) gmmstyle(gdpL1 ginid schl invrt pli, lag(3 3)eq(diff) collapse) twostep robust small
Low-Income:
Code:
xtabond2 agr gdpL1 ginid schl invrt pli if avggdpL1 < 6.849174, gmmstyle(gdpL1 ginid schl invrt pli, lag(2 2) eq(level) collapse) gmmstyle(gdpL1 ginid schl invrt pli, lag(3 3)eq(diff) collapse) twostep robust small
This yielded the following results:

High-Income:

Code:
Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: cid                             Number of obs      =       147
Time variable : period                          Number of groups   =        26
Number of instruments = 11                      Obs per group: min =         2
F(5, 25)      =      5.08                                      avg =      5.65
Prob > F      =     0.002                                      max =         6
------------------------------------------------------------------------------
             |              Corrected
         agr |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       gdpL1 |  -74.23678   105.4379    -0.70   0.488    -291.3903    142.9167
       ginid |    2.42002    7.60249     0.32   0.753     -13.2376    18.07764
        schl |   48.99133   24.56882     1.99   0.057    -1.609101    99.59177
       invrt |   2.935979   8.312672     0.35   0.727    -14.18429    20.05625
         pli |  -.9992304   .9320143    -1.07   0.294     -2.91875    .9202891
       _cons |   519.9719   919.0785     0.57   0.577    -1372.906     2412.85
------------------------------------------------------------------------------
Low-Income:

Code:
Dynamic panel-data estimation, two-step system GMM
------------------------------------------------------------------------------
Group variable: cid                             Number of obs      =       149
Time variable : period                          Number of groups   =        34
Number of instruments = 11                      Obs per group: min =         2
F(5, 33)      =     13.96                                      avg =      4.38
Prob > F      =     0.000                                      max =         6
------------------------------------------------------------------------------
             |              Corrected
         agr |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       gdpL1 |   6.896571   17.06682     0.40   0.689    -27.82614    41.61929
       ginid |   .0412858    3.93958     0.01   0.992     -7.97385    8.056422
        schl |   10.49205   4.164243     2.52   0.017     2.019832    18.96426
       invrt |  -.1814473    1.65804    -0.11   0.914    -3.554754     3.19186
         pli |   -1.22711    .279851    -4.38   0.000    -1.796471   -.6577486
       _cons |   23.14023   129.7007     0.18   0.859    -240.7379    287.0184
------------------------------------------------------------------------------
My ultimate goal is then to conduct a test to determine whether or not there is equality in the coefficients on "ginid" in these two separate regressions.

The first step I took was to store the results for the two-regressions and then use the "suest" command to test for coefficient equality, but this yielded the following error:

Code:
suest high low
high was estimated with a nonstandard vce (Corrected)
r(322);
I then tried re-estimating the models without utilising corrected standard errors to see if this would solve my issue, but I was again met with an error:

Code:
suest low high
unable to generate scores for model low
suest requires that predict allow the score option
r(322);
At this stage I must confess to being somewhat stuck as to how I might accomplish my goal of comparing the two coefficients on "ginid", and testing for equality. Any help in this matter would therefore be greatly appreciated.

Many thanks in advance!