Virata manual

svy: tabulate

Cross-tabulation with survey-correct percentages and test.

What we have

  • one-way and two-way tables with design-based errors
  • row, column, count, se and ci
  • the Rao-Scott corrected F test, which matches R's svychisq exactly (checked)

Coming later

  • standard errors inside the main table — here they follow it as separate tables

How you write it

svy: tabulate var1 [var2] [if] [, row column count se ci]

Square brackets mean optional. A vertical bar means choose one.

What it does

One variable gives population shares; two give a grid with a test of whether the two are related.

row, column and count choose what the cells hold; se and ci add the precision as tables underneath. The test of independence is the Rao-Scott corrected version of chi-squared — the ordinary one assumes independent observations and a clustered survey has none. Its degrees of freedom are not whole numbers, which is normal and not a bug.

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: tabulate sector sex, row
Survey: Two-way tabulation

Number of strata   =         12   Number of obs      =        6,000
Number of PSUs     =        400   Population size    =   10,612,910
                                  Design df          =          388

              sex
      sector |      Male    Female |     Total
-------------+---------------------+----------
       Rural |     .5037     .4963 |         1
       Urban |     .4893     .5107 |         1
-------------+---------------------+----------
       Total |     .4998     .5002 |         1
Key: row proportions

Pearson:
  Uncorrected   chi2(1)                 =     0.9812
  Design-based  F(1.0000, 388.00)       =     1.0683
                                       P =     0.3020
The same thing in pandas and in R

pandas

# survey table: ratio estimators, Taylor-linearised; Rao-Scott second-order corrected test of independence

R

library(survey)
tab <- svytable(~sector + sex, design)
prop.table(tab, 1)
svychisq(~sector + sex, design)

# This R was run in a real R (4.6.1) and gave the same numbers.
The full built-in help (help svy: tabulate)
svy: tabulate var1 [var2] [if] [, row column count se ci]

Survey tables with design-based errors, using the design from svyset.  One
variable gives a one-way table of proportions; two give a two-way table and a
test of independence.

    . svyset village [pw=multiplier], strata(stratum)
    . svy: tabulate sector sex, row se

  (nothing)   cell proportions: each cell over the whole table
  row         row proportions: each row adds to 1
  column      column proportions: each column adds to 1
  count       estimated population counts instead of proportions
  se          add a table of linearised standard errors
  ci          add a table of 95% confidence intervals (a t interval on the
              logit scale for proportions, so it stays between 0 and 1)

The test of independence is the Rao-Scott second-order corrected Pearson
statistic, which Stata calls the design-based F.  Its numerator degrees of
freedom are not whole numbers.  It matches R's survey::svychisq exactly
(checked); it has not been compared with a real Stata run.

Stata prints standard errors and intervals inside the main table; here they
follow it as separate tables.  The numbers are the same.

Nearby

svyset svy: mean svy: total svy: proportion svy: ratio svy: regress