svy: mean
The average for the whole population, from a survey sample.
What we have
- the population mean with a linearised standard error
over()for domains, and anifas a subpopulation- the interval uses the design degrees of freedom, not the number of people
Coming later
- replicate weights
- pooling plausible values
How you write it
svy: mean | total | proportion | ratio | regress | tabulate ...
Square brackets mean optional. A vertical bar means choose one.
What it does
svy: mean mpce gives the population average with a standard error that accounts for the weights, the strata and the clusters recorded by svyset. The interval uses the design's degrees of freedom — the number of clusters minus the number of strata — rather than the number of people.
over(sector) reports it separately by group, still using the whole design.
Watch out. An
if condition here means a subpopulation, not a smaller dataset: the rest of the sample stays in the design. Deleting those rows first would lose clusters and shrink the standard errors incorrectly.Try it
Every example below was run after sysuse vira.sample02, clear.
. svyset village [pw=multiplier], strata(stratum)
pweight: multiplier
VCE: linearized
Single unit: missing
Strata 1: stratum
Sampling unit 1: village
FPC 1: <zero>
The same thing in pandas and in R
pandas
design = dict(psu='village', strata='stratum', weight='multiplier')
R
library(survey)
design <- svydesign(ids = ~village, strata = ~stratum,
weights = ~multiplier, data = df, nest = TRUE)
# This R was run in a real R (4.6.1) and gave the same numbers.
. svy: mean mpce, over(sector)
Survey: Mean estimation
Number of strata = 12 Number of PSUs = 400
Number of obs = 6,000
Design df = 388
| Mean Std. err. [95% conf. interval]
-----------------+--------------------------------------------------------
Rural | 3394.003 21.863 3351.018 3436.987
Urban | 4339.799 35.40776 4270.184 4409.414
The same thing in pandas and in R
pandas
# mean of ['mpce'] with linearised standard errors
R
library(survey) svyby(~mpce, ~sector, design, svymean) # This R was run in a real R (4.6.1) and gave the same numbers.
. svy: total hhsize
Survey: Total estimation
Number of strata = 12 Number of PSUs = 400
Number of obs = 6,000
Design df = 388
| Total Std. err. [95% conf. interval]
-----------------+--------------------------------------------------------
hhsize | 4.80755e+07 713048.5 4.66736e+07 4.94774e+07
The same thing in pandas and in R
pandas
# total of ['hhsize'] with linearised standard errors
R
library(survey) svytotal(~hhsize, design, na.rm = TRUE) # This R was run in a real R (4.6.1) and gave the same numbers.
. svy: proportion sector
Survey: Proportion estimation
Number of strata = 12 Number of PSUs = 400
Number of obs = 6,000
Design df = 388
| Proportion Std. err. [95% conf. interval]
-----------------+--------------------------------------------------------
sector=Rural | .7302614 .005520317 .7194079 .7411149
sector=Urban | .2697386 .005520317 .2588851 .2805921
The same thing in pandas and in R
pandas
# proportion of ['sector'] with linearised standard errors
R
library(survey) svymean(~factor(sector), design, na.rm = TRUE) # This R was run in a real R (4.6.1) and gave the same numbers.
. svy: ratio mpce/hhsize
Survey: Ratio estimation
Number of obs = 6,000
Design df = 388
| Ratio Std. err. [95% conf. interval]
-----------------+--------------------------------------------------------
mpce/hhsize | 805.5617 6.908603 791.9787 819.1447
The same thing in pandas and in R
R
# no R form for this one: a ratio needs a numerator and a denominator
Nearby
svyset svy: total svy: proportion svy: ratio svy: regress svy: tabulate