PROVIDE Project Technical Paper 2005:1
February 2005
PSU: <observations>
Number of PSUs
26177
Population size = 11221840
|
----------- Mean | |
Estimate |
Std. Err. |
[95% Conf. |
Interval] |
------------------- Deff |
|
totinc | |
42793.12 |
725.4612 |
41371.18 |
44215.06 |
------------------- 1.232494 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ |
The ci command with the aweight option gives the exact same point estimates as svymean
with pweight. However, the standard errors are different since aweight uses a different
formula for the standard error. The standard error of the point estimate, which should not be
confused with the standard deviation of a variable, as well as the 95% confidence interval of
the estimate, is slightly larger when pweight is used.
As argued before it is also important to specify clustering and stratification if applicable.
The IES 2000 data used in this example made use of clusters (PSUs) and stratification along
provincial and rural/urban lines (variable provloc). In the two examples that follow svyset psu
psuno is specified, and thereafter svyset strata provloc is added.12 In each instance the point
estimates are shown. Notice the effect on the standard error, confidence interval and deff (see
section 2.2.3).
. svyset psu psuno
. svymean totinc
Survey mean estimation
pweight: wgtselect
Strata: <one>
PSU: psuno
Number of obs = 26177
Number of strata = 1
Number of PSUs = 2956
Population size = 11221840
|
----------- Mean | |
Estimate |
Std. Err. |
[95% Conf. |
Interval] |
------------------- Deff |
|
totinc | |
42793.12 |
1042.246 |
40749.52 |
44836.72 |
------------------- 2.543879 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ |
. svyset strata provloc
. svymean totinc
Survey mean estimation
pweight: wgtselect
Strata: provloc
PSU: psuno
Number of obs = 26177
Number of strata = 18
Number of PSUs = 3327
Population size = 11221840
|
----------- Mean | |
Estimate |
Std. Err. |
[95% Conf. |
Interval] |
------------------- Deff |
|
totinc | |
42793.12 |
975.1044 |
40881.25 |
44704.99 |
------------------- 2.226684 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ |
12 Variable psuno is supplied by Statistics South Africa. Variable provloc, which represents the strata in the
survey, was created by grouping variables for province (prov) and location.
9
© PROVIDE Project