Dear all,

I have been analyzing survey data in Stata, and I encounter a problem regarding value of confidence interval.

I am using the national survey data to evaluate satisfaction of medical care that was received in Japan.
I used multi-stage random sampling, and pweight was attached to the data.
By using the weight, I would like to calculate point estimate and confidence interval of patient satisfaction in each prefecture (Japan has 47 prefectures).

Originally, our survey had 5 possible answers for each question.
I categorized these answers to two groups: positive and negative answer.

I tried two different ways to calculate the confidence interval, and these two method gave me completely different values.
I am wondering the reason, and which method should I be using for the analysis.

Code:
use data.dta, clear

svyset facid [pw=pweight], strata(pref) fpc(Nst) || patid, strata(group) fpc(Nhi) singleunit(certainty)

gen Q20 = .
replace Q20 = 1 if Answer = 5 |
Answer = 4 | Answer = 3
replace Q20 = 0 if
Answer = 2 | Answer = 1
Code:
* Method 1
keep if pref == 1
svy: tab Q20, ci
Output:
Mean
Linearized Std Err.
 95% CI Lower 95% CI Upper
.9139879
.0323847
.5025009
 1.325475
Code:
* Method 2
svy: mean Q20, over(pref)
Code:




Output:









Mean




Linearized Std Err.


95% CI Lower
95% CI Upper






.9140944


.0060102
.9021265
.9260624











Could anyone explain why I got the different value depending on using "svy:mean" or "svy:tab"?

Also, why the 95CI of svy:tab had much wider range compared to svy:mean?



Sincerely,

Yuichi Ichinose