How to use weights?

[sbacelar]

ESS uses three types of weights, for ex. dweight. How to use it with essurvey?

[BernStZi]

@sbacelar

There are actually only two survey weights that should be used for analysis:

• the design weights dweight
• the post-stratification weights pspwght.

The design weights account only for unequal inclusion probabilities (i.e. the probability for a unit to end up in the gross sample.)
The post-stratification weights are calculated based on the design weights. Their purpose is to adjust the distributions of certain variables (age group, gender, education, and region) to known population values. Using them might help to counter selective response processes (from gross to net sample) in order to avoid or reduce response bias. But they might also help to reduce the sampling error.
The so called population size weights are only needed if you want to analyses data from multiple countries and/or rounds together (i.e. a regression involving data from multiple countries).
For more information see the ESS website.

For ESS data from a single country you could simple use:

ess_design <- 
svydesign(
  ids = ~idno,
  weights = ~ dweight,
  data = ess,
)

If you care for more accurate sampling error estimation you should check what sampling design the countries used. Especially if it involved multi-stage sampling and/or stratification. If so you should use the SDDF the get the information on the Primary Sampling Units (PSU) and/or the strata.
Your survey design object might then look like this if you merge the SDDF with the main data set:

ess_design <- 
svydesign(
  ids = ~ psu+idno,
  strata = ~ stratify,
  weights = ~ dweight,
  nest=TRUE,
  data = ess,
)

[cimentadaj]

Hi @sbacelar, essurvey only downloads the integrated files of the ESS or the separate country rounds which I think only have the analytics weights (not SDDF files). We're currently developing tools to download the complete SDDF files for complex survey analysis which you can follow in #40. Hopefully, in the near future, we'll develop a package to analyze the ESS survey data.