margins, contrast and margins, pwcompare with svy: which is the correct distribution?

[Lydia Palumbo]

Dear Stata Users,

I am using the following command to estimate a multinomial logit with survey data
:

svy, subpop(if marital==3): mlogit uniontype2 i.durationtype i.employ2  ,  level(95) base rrr
margins i(1 2 3 4)b1.employ2, level(95) pr(out(1)) pr(out(2))   vsquish  force  atmeans   post  subpop(if marital==3) noatleg vce(unconditional)

I am testing whether certain predictions differ signficantly from each other. For doing so, I used both pwcompare and contrast. Both the commands give identical results.
However, i noticed that the output of pwcompare reports the probability distribution is a t distribution, whereas the contrast an F one.

Does someone know whether both distributions are correct or whether there is a difference?

Thank you. ,
Lydia


pwcompare

                           Contrast    std. err.    t    P>t
            
(1#M/E,F/U)    vs    (1#E/E)    .0400832    .0203203    1.97    0.049
(1#M/U,F/E)    vs    (1#E/E)    .0953815    .038329    2.49    0.013
(1#M/U,F/U)    vs    (1#E/E)    .0605543    .0329893    1.84    0.067
(2#E/E)    vs    (1#E/E)         .066152    .0106399    6.22    0.000
(2#M/E,F/U)    vs    (1#E/E)    .041327    .0208385    1.98    0.048
(2#M/U,F/E)    vs    (1#E/E)    .0034187    .0269758    0.13    0.899
(2#M/U,F/U)    vs    (1#E/E)    .0044137    .0263974    0.17    0.867

(output omitted)

contrast

employ2@_predict    
(M/E,F/U vs E/E) 1    1    3.89    0.0488
(M/E,F/U vs E/E) 2    1    1.24    0.2653
(M/U,F/E vs E/E) 1    1    6.19    0.0130
(M/U,F/E vs E/E) 2    1    5.05    0.0249
(M/U,F/U vs E/E) 1    1    3.37    0.0667
(M/U,F/U vs E/E) 2    1    4.64    0.0315
Joint    6    3.18    0.0043