Test Differences After Margins dydx

[Nik Honig]

Dear all, I am estimating the following regression

reghdfe SPREAD_Haupttest_mean i.mgr_exclude##c.NUMEST FRQ_MJM_V1 VOLA_Haupttest BTM SIZE LOSS TURNOVER_Haupttest_mean IMR6, absorb (FYQ INDUSTRY) vce (cluster gvkey)

The variable mgr_exclude is a binary one that takes the value 0 or 1. NUMEST is continous.

After running the regression, i want to see the marginal effects of mgr_exclude=1 in the cases that the variable NUMEST takes the values 25th, 50th and 75th percentile. moreover, and most importantly, i want check if the difference between the value in the 25th and the 75th percentile is significant. so i did the following

margins, dydx(mgr_exclude) at(NUMEST=(1.386294 1.94591 2.564949)) post

that part worked well. however, if i try to test the differences, i use

test _b[1._at] = _b[3._at]

but the result is


( 1) [0b.mgr_exclude]1bno._at - [0b.mgr_exclude]3o._at = 0
Constraint 1 dropped

chi2( 0) = .
Prob > chi2 = .


can you help me out? thanks!

[Nik Honig]

Maybe this clarifies my problem:

 margins, dydx(mgr_exclude) at(NUMEST=(1.386294 1.94591 2.564949)) post

Average marginal effects                                Number of obs = 55,176
Model VCE: Robust

Expression: Linear prediction, predict()
dy/dx wrt:  1.mgr_exclude
1._at: NUMEST = 1.386294
2._at: NUMEST =  1.94591
3._at: NUMEST = 2.564949

--------------------------------------------------------------------------------
               |            Delta-method
               |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
---------------+----------------------------------------------------------------
0.mgr_exclude  |  (base outcome)
---------------+----------------------------------------------------------------
1.mgr_exclude  |
           _at |
            1  |  -.1288038   .0172916    -7.45   0.000    -.1626947    -.094913
            2  |  -.0820267   .0151147    -5.43   0.000     -.111651   -.0524025
            3  |  -.0302826   .0188077    -1.61   0.107    -.0671449    .0065798
--------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

. test _b[1._at] = _b[3._at]

 ( 1)  [0b.mgr_exclude]1bno._at - [0b.mgr_exclude]3o._at = 0
       Constraint 1 dropped

           chi2(  0) =       .
         Prob > chi2 =         .

.

[Andrew Musau]

You are testing differences across the base level coefficients, which all take the values zero. Use the -coeflabel- option to see how to refer to the non-base level equation.

test [1.mgr_exclude]1._at = [1.mgr_exclude]3._at

[Andrew Musau]

Use the -coeflabel- option to see how to refer to the non-base level equation.

Read -coeflegend- for "coeflabel".