topic Replicate Confidence intervals of mean from Stata in New SAS User

Replicate Confidence intervals of mean from Stata

nelngams — Sun, 22 Dec 2019 03:07:42 GMT

Dear Community.

I am new in SAS and loosing my mind with this.

DATA TEST;
INPUT ID UB CONS WGT;
CARDS;
1 0 23 12
2 0 13 10
3 0 21 20
4 0 20 13
5 1 19 12
6 1 24 14
7 0 26 16
8 0 17 10
9 0 19 19
10 1 22 17
11 0 27 10
12 0 18 12
13 1 22 13
14 1 21 20
15 1 20 16
RUN;

Where:

UB: is a dummy for whether one consumes a given beer or not
CONS: is the reported number of bottle drank for a given period of time
WGT is my sampling adjustment weight.

My data is from an online survey (no design), yet each respondent has a sampling adjustment weight to match the general population.

I want to estimate the mean consumption of beer and the corresponding 95% C.I for the two groups of UB (0,1) and test whether consumption is equal between the two groups.

Initially I ran this analysis in Stata and now I am tried to replicate it in SAS. I got the same weighted mean but different 95% C.I.

STATA Codes: Since STATA does not allow for weighted t-test, I ran a mean estimation (accounting for weight) and used test after mean estimation to compare the two means.

. mean CONS [pweight=WGT], over(UB)

. test [CONS]0=[CONS]1

. test	[CONS]0=[CONS]1
( 1)	[CONS]0 - [CONS]1	= 0
	F( 1, 14) =	0.26
Prob > F =	0.6202

In SAS, I am getting the same means but different 95% Confidence intervals.

Analysis Variable : CONS
UB	N Obs	Mean	Lower 95% CL for Mean	Upper 95% CL for Mean
0	9	20.6475410	17.5247509	23.7703311
1	6	21.3478261	19.6156577	23.0799944

Can some body help with this?

Re: Replicate Confidence intervals of mean from Stata

Rick_SAS — Sun, 22 Dec 2019 19:10:12 GMT

What SAS code did you use? That looks like PROC MEANS output. I suggest you use PROC SURVEYMEANS (if you want to correct for the survey nature) or PROC TTEST.

PROC TTEST with a WEIGHT WGT statement gives the same output, but also provides statistical tests for the difference of the means between the two groups. From the data, I would not assume that the variances are equal, which suggests using the Satterthwaite test results. The test statistic and p-value are close to the stata estimate you constructed.

Re: Replicate Confidence intervals of mean from Stata

PaigeMiller — Sun, 22 Dec 2019 20:29:16 GMT

proc means data=have;
    class ub;
    var cons;
    weight wgt;
run;

By the way, there's no reason to expect SAS and STATA will produce the exact same results, although it certainly is possible in this example.

Cheers to UB (University at Buffalo) for winning their first ever NCAA Football Bowl Game!