Dear Stata Users,

I am analyzing the probability that cohabiting couples will marry or dissolve according to certain characteristics.
I am using a multinomial logit, where marriage and cohabitation are considered competing risks.

I show the results that I get according to couple employment (employ2) on marriage (omitting dissolution)
by using the output of the multinomial.
I controlled for a series of other characteristics that I omitted.

uniontype2 = occurrence of marriage (or dissolution)

svy: mlogit uniontype2 i.durationtype i.agesm2_f##i.mage_gap_cat i.wave5 i.employ2 ib0.evunionpr i.shared i.educ2_f i.homog i.tercilehh if marital==3&duratyrs3>0&duratyrs3<6&educ2_m<4&educ2 _f<4&sampst==1&ivfioall==1&(minagentry>18&minagent ry<37)&employ2<5, level(90) rrr base


employ2 | Coef SE t P -value 90% CIs
Both employed | 1 (base)
Male unemployed | .4910494 .2654903 -1.32 0.189 .2013042 1.197837
Female unemployed | .7248931 .2145792 -1.09 0.278 .4448765 1.181159
Both unemployed | .7624074 .4288538 -0.48 0.630 .3014883 1.927985


All the results are not statistically significant.

If I compute the margins and test differences, such that

margins r.employ2, level(90) pr(out(1)) pr(out(2)) post atmeans*

I get that the following couple type (Male not employed/female employed) has a significant lower risk of marriage than those
who are both employed (I omitted dissolution)


Contrast Std. Err. [90% Conf. Interval]

(Male unemployed vs Both employed) -.0751986 .0345901 -.1322493 -.0181479

(Female unemployed vs Both employed) -.037832 .0281888 -.0843248 .0086607

(Both unemployed vs Both employed) 2 -.0317464 .0541869 -.1211187 .057626


I am wondering why there is this difference. I thought that both should have the same level of significant.
Is there a reason why these results diverge?
Which output should I trust if want to understand the relationship between being in a certain couple type
and the probability of entering a marriage or going through a dissolution?


Thank you.
Best,
Lydia


*The result remains also by using as observed