differing results for proportions using subpop vs 'over' with mi estimate using survey data

Hi Statalisters -

I am using multiple imputation for HIV status and have just identified some strange findings across what I had anticipated to be two different ways of calculating the same statistics.

The first way I evaluated % HIV positive across gender was as follows:

mi estimate: svy,subpop(if gender==1): proportion hiv_status
mi estimate: svy,subpop(if gender==2): proportion hiv_status

The second way was using proportion with option 'over' gender, as follows:

mi estimate: proportion hiv_status, over(gender)

With these commands, I found very different distributions of the variable, and am wondering whether anyone is able to help me understand why this may have occurred?

Thank you in advance.

Best,
Alison

Here is the output from my analysis:

.mi estimate: svy,subpop(if gender==1): proportion hiv_status

Multiple-imputation estimates Imputations = 50
Survey: Proportion estimation Number of obs = 1,044

Number of strata = 3 Population size = 92,506.999
Number of PSUs = 43 Subpop. no. obs = 401
Subpop. size = 43,445.999
Average RVI = 0.4196
Largest FMI = 0.3128
Complete DF = 40
DF adjustment: Small sample DF: min = 25.64
avg = 25.64
Within VCE type: Linearized max = 25.64

--------------------------------------------------------------
| Proportion Std. Err. [95% Conf. Interval]
-------------+------------------------------------------------
0 | .8004723 .0302365 .7382776 .8626669
1 | .1995277 .0302365 .1373331 .2617224
--------------------------------------------------------------


.mi estimate: svy,subpop(if gender==2): proportion hiv_status

Multiple-imputation estimates Imputations = 50
Survey: Proportion estimation Number of obs = 1,044

Number of strata = 3 Population size = 92,506.999
Number of PSUs = 43 Subpop. no. obs = 643
Subpop. size = 49,061
Average RVI = 0.3241
Largest FMI = 0.2593
Complete DF = 40
DF adjustment: Small sample DF: min = 27.82
avg = 27.82
Within VCE type: Linearized max = 27.82

--------------------------------------------------------------
| Proportion Std. Err. [95% Conf. Interval]
-------------+------------------------------------------------
0 | .7327915 .0227594 .6861577 .7794254
1 | .2672085 .0227594 .2205746 .3138423
--------------------------------------------------------------


.mi estimate: proportion hiv_status, over(gender)

Multiple-imputation estimates Imputations = 50
Proportion estimation Number of obs = 1,044
Average RVI = 0.3957
Largest FMI = 0.2938
Complete DF = 1043
DF adjustment: Small sample DF: min = 324.63
avg = 335.76
Within VCE type: Analytic max = 346.88

_prop_1: hiv_status = 0
_prop_2: hiv_status = 1

Male: gender = Male
Female: gender = Female

--------------------------------------------------------------
Over | Proportion Std. Err. [95% Conf. Interval]
-------------+------------------------------------------------
_prop_1 |
Male | .7633915 .0249646 .7142906 .8124925
Female | .6936858 .0215964 .6511993 .7361724
-------------+------------------------------------------------
_prop_2 |
Male | .2366085 .0249646 .1875075 .2857094
Female | .3063142 .0215964 .2638276 .3488007
--------------------------------------------------------------