-marginsplot- after -logit-

[Marco Tacchi]

Hello,

After running a logit model, -logit- command with -vce(cluster)-, with continuous-by-continuous interaction, c.x##c.m, I proceed by calculating marginal effects using the -margins- command:

. margins, at(x=(0(2)4) m=(0 65))
------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         _at |
          1  |   .0014409    .000167     8.63   0.000     .0011137    .0017681
          2  |     .00019    .000052     3.65   0.000     .0000881    .0002919
          3  |   1.84e-26   2.63e-25     0.07   0.944    -4.98e-25    5.35e-25
          4  |   .0388714   .0345513     1.13   0.261     -.028848    .1065908
          5  |   2.18e-49   6.25e-48     0.03   0.972    -1.20e-47    1.25e-47
          6  |   .6466306   .2712054     2.38   0.017     .1150778    1.178183
------------------------------------------------------------------------------

However, when I try to plot this with -marginsplot-, point 5 (x=4, m=0) does not appear on the graph for some reason.
Does anyone know why this could be the case and how the blue line can be continued?

[Joseph Coveney]

point 5 (x=4, m=0) does not appear on the graph for some reason.
Does anyone know why this could be the case . . ?

I don't know, but at O(10^-26) ± O(10^-25), it might be considered as missing-valued.

. . . and how the blue line can be continued?

You could try extending the x-axis using something like an xlabel(0(1.5)4.5) option. Or maybe the y-axis to just below zero (same method).

Otherwise, maybe plot the margins in the original estimation metric instead of probabilities.

[Marco Tacchi]

I don't know, but at O(10^-26) ± O(10^-25), it might be considered as missing-valued.

You could try extending the x-axis using something like an xlabel(0(1.5)4.5) option. Or maybe the y-axis to just below zero (same method).

Otherwise, maybe plot the margins in the original estimation metric instead of probabilities.

Thank you for the reply. I have tried adjusting xlabel and ylabel, but it did not work. Perhaps, 2.18e-49 and 6.25e-48 can indeed be too low values for Stata and it treats them as missing.