Hello,
I would like to run a multilevel model (random-effects linear regression) on European Social Survey (ESS) data. The ESS data were collected in 18 countries. In each country, about 1000 individuals were interviewed. The data are therefore individual-level data, clustered in countries. Here is the kind of model I want to run:
xtset country
xtreg depvar countryindepvars individualindepvars [iweight=weight], mle
country is the variable identifying each of the 18 countries, depvar is the dependent variable, countryindepvars are the country-level independent variables, individualindepvars are the individual-level independent variables.
I am interested in the effects the country-level and individual-level variables have on the depedent variable. My question is, should I use weights in this analysis? If yes, what kind of weights are appropriate? Post-stratification weights? Or a combination of population size weights and post-stratification weights?
The ESS team recommends that users always use appropriate weights with the data. The ESS data have post-stratification weights, which correct bias introduced by sampling design. In addition, there are population size weights. The population size weights are described in the ESS documentation as weights that are "used when examining data for two or more countries combined. The population size weights are the same for all persons within a country but differ across countries. These weights correct for the fact that most countries taking part in the ESS have different population sizes but similar sample sizes." According to the ESS documentation, population size weights should not be used alone, they should always be combined with post-stratification weights.
Thanks a lot!.
Zuzana Ringlerova
I would like to run a multilevel model (random-effects linear regression) on European Social Survey (ESS) data. The ESS data were collected in 18 countries. In each country, about 1000 individuals were interviewed. The data are therefore individual-level data, clustered in countries. Here is the kind of model I want to run:
xtset country
xtreg depvar countryindepvars individualindepvars [iweight=weight], mle
country is the variable identifying each of the 18 countries, depvar is the dependent variable, countryindepvars are the country-level independent variables, individualindepvars are the individual-level independent variables.
I am interested in the effects the country-level and individual-level variables have on the depedent variable. My question is, should I use weights in this analysis? If yes, what kind of weights are appropriate? Post-stratification weights? Or a combination of population size weights and post-stratification weights?
The ESS team recommends that users always use appropriate weights with the data. The ESS data have post-stratification weights, which correct bias introduced by sampling design. In addition, there are population size weights. The population size weights are described in the ESS documentation as weights that are "used when examining data for two or more countries combined. The population size weights are the same for all persons within a country but differ across countries. These weights correct for the fact that most countries taking part in the ESS have different population sizes but similar sample sizes." According to the ESS documentation, population size weights should not be used alone, they should always be combined with post-stratification weights.
Thanks a lot!.
Zuzana Ringlerova
