I have a problem understanding -margin results, even after reading Clyde Schechter's clarifications.
I run a regression of a continuous variable on two dummy variables and one continuous variable, including interactions.
Code:
reg relch_rd_led i.incumbent##c.share_rd##i.kabinetspartij
This is actually a simplyfied version of the model I use, in order to keep things manageable.
I am interested in the marginal effect of one of the dummies, incumbent, and how this marginal effect is influenced by the other dummy, kabinetspartij. So I run
Code:
margins kabinetspartij, dydx(incumbent) grand
Array


The result puzzles me. The grand margin is negative, while the margins at the only two possible values of the covariate kabinetspartij are both positive. How is this possible? Shouldn't the grand margin be somewhere in that range? I don't know how to interpret the results. Could anyone please help me?

I provide a data example to give an impression of the variables I use. However, -dataex truncated the sample and the above result is no longer obtained.

Code:
* Example generated by -dataex-. To install: ssc install dataex
clear
input float(incumbent kabinetspartij relch_rd_led share_rd)
0 0          0 26.666666
1 0          0 26.666666
0 1          0 26.666666
1 1        -25 26.666666
1 0          0        20
1 1 -23.076925        20
0 0          0 13.333333
0 0          0 13.333333
0 0          0 13.333333
1 1          0 13.333333
1 0          0 13.333333
1 1  15.384615 13.333333
0 1       -100  6.666667
0 0          0  6.666667
0 0          0  6.666667
0 0        100  6.666667
0 0          0 13.333333
0 0          0 13.333333
1 0  15.384615 13.333333
0 0          0 13.333333
0 0         50 13.333333
1 0 -33.333336        20
1 1         25 26.666666
1 1        -40 33.333332
1 0   33.33333        20
1 1        -50 26.666666
1 1          0 13.333333
0 0   73.07693 13.333333
0 1          0  6.666667
0 1        100  6.666667
1 1        -50 13.333333
0 0          0  6.666667
0 1          0  6.666667
0 1  15.384615  6.666667
1 1          0  30.76923
1 0          0  30.76923
1 0   8.333332  30.76923
1 1        -20 33.333332
1 1          0 26.666666
0 0        -25 26.666666
0 0          0  7.692307
0 0          0  7.692307
0 0 -13.333333  7.692307
0 0          0  6.666667
0 1        100  6.666667
0 0          0 13.333333
1 1          0  38.46154
1 1          0  38.46154
1 1  -13.33334  38.46154
1 0  20.000004 33.333332
1 1  -16.66667        40
1 1        -20 33.333332
0 0          0 23.076923
0 1          0 23.076923
0 1  15.555552 23.076923
0 1          0 26.666666
0 0          0 26.666666
1 1        -25 26.666666
1 1 -33.108105 12.121212
0 0  -5.128212 8.1081085
0 0  33.333344  7.692307
1 1 -25.000004  10.25641
1 1  -33.33333  7.692307
1 0          0  5.128205
1 1 -13.333332  5.128205
0 0        100  2.564103
0 1        -50  5.128205
0 0          0  2.564103
0 1  73.333336  2.564103
0 0  27.413124  21.21212
1 1   -52.5641 27.027027
0 1        -60 12.820513
0 1        -50  5.128205
0 0        200  2.564103
0 0        100  7.692307
1 1  -27.77778 15.384615
0 0  18.918922  9.090909
1 0  -5.128204  10.81081
1 0          0  10.25641
1 0 -25.000004  10.25641
0 0          0  7.692307
0 0  -33.33333  7.692307
0 0  73.333336  5.128205
1 1  -40.54054  27.27273
1 1   42.30769 16.216217
1 1 -22.222227 23.076923
1 0   71.42857 17.948717
1 1  -33.33333  30.76923
1 1 -37.500004  20.51282
1 0   21.33333 12.820513
0 0        -50  10.25641
0 0        100  5.128205
0 0 -35.000004  10.25641
1 0  27.413124  21.21212
1 1   13.84615 27.027027
1 1  -33.33333  30.76923
0 1 -12.500007  20.51282
1 0          0 17.948717
1 1 -28.571426 17.948717
1 1   38.66668 12.820513
end

reg       relch_rd_led i.incumbent##c.share_rd##i.kabinetspartij
margins kabinetspartij, dydx(incumbent) grand