Hi,

I am having a question regarding testing whether the difference of coefficients is significant.
I am running the following code:
Code:
svy: logit X ib4.age
margins ib4.age, atmeans
Age is a categorical variable with 6 different levels.

I get the following output:


Code:
. margins ib4.age

Adjusted predictions                            Number of obs     =  2,411,677
Model VCE    : OIM

Expression   : Pr(InEmp), predict()

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       age |
      16-24  |   .4930075   .0009591   514.03   0.000     .4911276    .4948873
      25-34  |   .7947116    .000704  1128.84   0.000     .7933317    .7960914
      35-44  |   .8203046    .000623  1316.77   0.000     .8190836    .8215255
      45-54  |   .8179822   .0005948  1375.19   0.000     .8168164     .819148
      55-64  |   .5946761    .000772   770.31   0.000      .593163    .5961892
      65-74  |   .1446094   .0005738   252.02   0.000     .1434847     .145734
------------------------------------------------------------------------------

What I would like to know if whether when I the difference in the coefficient for one level of Age to the base category (e.g. |0.4517| in the case of 16-24 year olds relative to 55-64 year olds) is statistically significant.

As both coefficients stem from the same model, and have been subtracted from each other, I am not sure what command to use? If it were two different regressions I would use a difference in means test, but that doesn't seem appropriate here.

Does anyone know what I should do?

Thank you.