[R] survey package rake results in very large weights

Imran Akbar skunkwerk at gmail.com
Mon Jan 19 21:27:16 CET 2015

    I'm trying to use both the survey package and the anesrake package to
perform raking on my sample dataset, to get the proportions to match up
with population data from the census.

I'm trying the following, but the resulting weights are very large:

> svy.unweighted <- svydesign(ids=~1, data=cleanPanelistData)
> svy.rake <- rake(design = svy.unweighted,
                       sample.margins = list(~age, ~education, ~ethnicity,
~gender, ~income, ~location, ~size),
                       population.margins = list(censusAge,
censusEducation, censusEthnicity, censusGender, censusIncome,
censusLocation, censusSize))

> summary(weights(svy.rake))
    Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
   133.9    567.6   1062.0   2606.0   2305.0 396800.0

When I use the anesrake package, the weights are capped at 5 and I'm able
to get a design effect of around 3.

My data set in .Rdata format is available here <http://we.tl/bUAuW3ajVs>
Am I doing something wrong?  Should I be using calibrate with
calfun='raking' instead of rake?

This was run on:
RStudio on Mac Mavericks, version 3.02
with survey_3.30-3


	[[alternative HTML version deleted]]

More information about the R-help mailing list