# st: Re: Comparing means of two variables using -suest-

 From "Leonor Saravia" To statalist@hsphsun2.harvard.edu Subject st: Re: Comparing means of two variables using -suest- Date Wed, 8 Oct 2008 22:49:19 -0400

```Steve and Martin,

I´m incorporating them.  As soon as I have some results I will be in
touch with you again. Thank you very much for your help!

Leonor

Leonor-

1, Means of dummy variables are proportions. To further Martin's
question, I am not sure what proportions you want to estimate and
compare.

Consider the following 2 x 2 table  with n=50 for pension =1
Participate
1           0  Total

Males              5         15    20
Females           10         20    30
Total             15         35    50

This translate into the following table of proportions:
Participate
1           0   Total

Males              .1         .3   .4
Females            .2         .4   .6
Total              .3         .7   1.0

With your definitions, the mean of act_m is 0.10. and the mean of
act_f is 0.20. Are those the numbers you want? If so, show us a
similar table for pension =0 and show us what your hypothesis tests
are meant to compare.

2. Your -svyset- command means that "folio" was a primary sampling
unit-- the kind of unit selected in the first-stage of sampling, and
that there were other units selected later. Is this correct? Were
there any sampling strata? Show us the result of the -svydes- command
and please describe the sampling design.

3. You did not use any -svy- commands in your analysis. However, since
I am not sure what the proper analysis is, I won't try to correct your
syntax yet.

-Steve

On Oct 7, 2008, at 5:13 AM, Martin Weiss wrote:

Just out of curiosity. Why not: -proportion male female if
participating == 1 & pension == 1- and compare the resulting CIs? The
-if- statement records what is common to both groups, namely that they
participate and that they
draw a pension.

HTH
Martin

2008/10/7 Leonor Saravia <lmisaravia@gmail.com>:
> Hello,
>
> I´d like to compare two means of two different groups (dummy
> variables), where each one is affected by a grouping variable. I was
> looking in the Stata FAQS and found that this was posible with the -
> suest - command, but when I use it, it doesn´t work and I don´t know
> what´s wrong. I´appreciate if you could give me some advice.
>
> My dummy variables are like this:
>
> act_m = (participating == 1 & male == 1) ; 0 otherwise
> act_f =  (participating == 1 & female == 1) ; 0 otherwise
> pension = 1 if recieves a pension; 0 otherwise
>
> And I need to compare the 'act_m' mean when 'pension' == 1 with the
> mean of 'act_f' when 'pension' == 1.
>
> What I was doing was (and didn´t work):
>
> svyset folio [pw = factor]
> mean act_m if (hm3_hym == 1 & menor15h == 1) [fw = factorex], over(pension)
> est store act_m1
> mean act_f if (hm3_hym == 1 & menor15h == 1) [fw = factorex], over(pension)
> est sto act_f1
> suest act_m1 act_f1,svy
> test [act_m1]pensionh [act_h1]pensionh
>
> The error that Stata brings is:
>
> suest act_m1 act_h1,svy
> unable to generate scores for model act_m1
> suest requires that predict allow the score     option
> r(322);
>
>
> Well, I´d appreciate very much your help! :)
>
> Leonor
>
>
>
>
>
> Thank you so much Austin!..... This has been very helpful... I appreciate it
>
> Regards
>
> Renato
>
> On Feb 24 2008, Austin Nichols wrote:
>
>
> Renato <ravi0023@umn.edu>:
> Try -suest- e.g.
>
> sysuse auto
> svyset [pw=wei]
> qui svy: reg len for
> est sto length
> qui svy: reg turn for
> est sto turn
> qui svy: reg tru for
> est sto trunk
> suest length turn trunk, svy
> test [length]foreign [turn]foreign [trunk]foreign
>
>
> On 23 Feb 2008 16:34:29 -0600,  <ravi0023@umn.edu> wrote:
>
> I have a data base with 2 different test scores and two groups, a treatment
> group and a control group. I have managed to test whether the mean of each
> test scores is different among the two groups using svymean and then lincom
> in this way:
>
> svy: mean testA, over(treatment)
> lincom [testA]1-[testA]0
>
> svy: mean testB, over(treatment)
> lincom [testB]1-[testB]0
>
> What I want to do is jointly test whether both variables (testA and testB)
> are jointly statistically different among the two groups, but incorporating