I want to do a poisson multilevel analysis of item nonresponse over 27 countries. The model holds 3 levels: respondent, interviewer, and country.

I am using data from the European social survey, and these data have two weights:

Design weight: Several of the sample designs used by countries participating in the ESS were not able to give all individuals in the population aged 15+ precisely the same chance of selection.The design weight corrects for these slightly different probabilities of selection, thereby making the sample more representative of a ‘true’ sample of individuals aged 15+ in each country.

Population weights: This weight corrects for the fact that most countries taking part in the ESS have very similar sample sizes, no matter how large or small their population. The population size weight makes an adjustment to ensure that each country is represented in proportion to its population size.

Now I was going to use the product of these two weights as the weighting at the first level. (Second and third level I'll use a weight of 1) But I got the advice to only use the design weight, since country is a variable in my model. And when weighting for population the country differences are going to dominate the data, masking other significant effects.

However, the website of the ESS advices: "when combining countries into a group, such as ’accession countries’ or ‘EU member states’, both design and population size weights should be applied.". And I am combining all the 27 countries into one group.

Does anyone have any good explanation as to why I should not use the population weights, when country is accounted for in the model. Is this a 'golden rule'?


1 Answer 1


I think the biggest issue here is sample-design, I would use the design weight, and depending on the values it takes you may have to normalize it, ie wt/mean(wt) Here is a good reference on using design weights in multilevel models:

A. C. Carle, “Fitting multilevel models in complex survey data with design weights: recommendations,” BMC Medical Research Methodology, vol. 9, no. 1, article 49, 2009.

  • $\begingroup$ Thanks, the reference is very helpful. The values range from 0 to 4 with a mean of 1 and SD of 0.5. So I think normalising is a good idea. $\endgroup$
    – Marloes
    Commented Mar 23, 2013 at 18:46
  • $\begingroup$ actually that's not so bad, sometimes weights are in the thousands, but those are typically population weights. $\endgroup$ Commented Mar 25, 2013 at 2:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.