-xtabond2- with interaction term

[Connie Gao]

Hi Stata Experts:

I am using xtabond2 to estimate the impact of interaction term on outcome variable. The interaction term is based on two categorical variables.


My code is

xtabond2 lfstfy L.lfstfy i.A i.A#i.B i.housect i.Rstate i.year, gmmstyle(lfstfy i.A i.A#i.B, laglimits(2 3) collapse) ivstyle(i.Rstate i.year, equation(level)) noleveleq robust orthogonal

i.A#i.B is my main variable. I did not add B in the model, because the results omit B as well.

The results show some of cells of interaction terms were omitted. I assume it may be because collinearity. As interaction term is my interest variable, can someone tell me how to recover the omitted categories?

                           |               Robust
                 lfstfy |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
------------------------+----------------------------------------------------------------
                 lfstfy |
                    L1. |   .0452652   .0227559     1.99   0.047     .0006645     .089866
                        |
                      A |
                     1  |          0  (empty)
                     2  |   -.003393   .0722695    -0.05   0.963    -.1450385    .1382526
                     3  |  -.1802281   .0732506    -2.46   0.014    -.3237966   -.0366597
                     4  |  -.0337546   .0654204    -0.52   0.606    -.1619763    .0944671
                     5  |   .0199551   .0949678     0.21   0.834    -.1661784    .2060885
                     6  |  -.0485592   .0858893    -0.57   0.572    -.2168992    .1197807
                     7  |  -.0333336   .0667647    -0.50   0.618    -.1641901    .0975228
                     8  |  -.0004754     .10503    -0.00   0.996    -.2063304    .2053796
                     9  |  -.0693442   .0917054    -0.76   0.450    -.2490834    .1103951
                        |
                    A#B |
                   1 0  |          0  (empty)
                   1 1  |    .200792   .3310918     0.61   0.544    -.4481361    .8497201
                   1 2  |  -.4242789   .2050095    -2.07   0.038    -.8260902   -.0224676
                   2 0  |          0  (empty)
                   2 1  |   .3346134   .3740984     0.89   0.371     -.398606    1.067833
                   2 2  |  -.5909363   .2094511    -2.82   0.005    -1.001453   -.1804196
                   3 0  |          0  (empty)
                   3 1  |   .7852543   .3483723     2.25   0.024     .1024572    1.468051
                   3 2  |  -.2557902   .1799161    -1.42   0.155    -.6084192    .0968388
                   4 0  |          0  (empty)
                   4 1  |   .3312339   .3220198     1.03   0.304    -.2999133    .9623811
                   4 2  |  -.4006439    .196792    -2.04   0.042    -.7863491   -.0149387
                   5 0  |          0  (empty)
                   5 1  |          0  (omitted)
                   5 2  |          0  (omitted)
                   6 0  |          0  (empty)
                   6 1  |   .3071944   .4038064     0.76   0.447    -.4842516    1.098641
                   6 2  |  -.4528794   .2312391    -1.96   0.050    -.9060997    .0003409
                   7 0  |          0  (empty)
                   7 1  |    .101094   .3735869     0.27   0.787    -.6311228    .8333108
                   7 2  |  -.3408805   .2023023    -1.69   0.092    -.7373858    .0556247
                   8 0  |          0  (empty)
                   8 1  |   .7406616   .4301529     1.72   0.085    -.1024225    1.583746
                   8 2  |  -.8448848   .3371103    -2.51   0.012    -1.505609   -.1841608
                   9 0  |          0  (empty)
                   9 1  |   .6990125    .370923     1.88   0.059    -.0279832    1.426008
                   9 2  |  -.3290207    .215716    -1.53   0.127    -.7518164    .0937749

Thank you!

Connie

[Sven-Kristjan Bormann]

You could modify the code for -xtabond2- and remove the calls to -_rmcoll- which removes collinear variables. You would then modify the rest of the code accordingly. However, I have not tested these changes.
Without further information about your dataset and your estimation aim, it is difficult to provide further advice.

[Connie Gao]

Thank you Sven-Kristjan. I will try _rmcoll_to see whether it works.