I am working on analyzing survey data which was collected using a stratified random sample, with disproportionate allocation to strata (single stage, no clusters). The weight I am using is the product of the sampling weight and the nonresponse rate.
I used svyset to declare the survey design as follows:
Code:
svyset _n, strata(DistrictLocaleCollapsed_code) fpc(fpc) weight(TotalWeight) vce(linearized) singleunit(missing)
I wanted to compare estimates, standard errors, and confidence intervals for my data as unadjusted, using pweight, and using svy. To do so I ran the following code:
Code:
mean TotalNumberOperationalSchools
svy: mean TotalNumberOperationalSchools
mean TotalNumberOperationalSchools [pweight = TotalWeight]
For both options with the survey weights the mean estimates were what I expected them to be and were the same. However, there is a noticeable difference between the standard errors and confidence intervals for the svy v. pweight outputs, with the svy results being smaller. For example:
Method |
Mean |
SE |
CI-LL |
CI-UL |
svy:mean |
8.141049 |
.2834497 |
7.582634 |
8.699464 |
mean [pweight] |
8.141049 |
.7248509 |
6.713137 |
9.568961 |
I am hoping someone might be able to help me understand why this difference is occurring and how to troubleshoot it.
0 Response to Difference in SE and CI using svyset v. pweight - reason for difference?
Post a Comment