Hello, I am a political science graduate student, new to the Stata Forum and a first time poster. I am working on a project where I want to run a multilevel model using weighted survey data, but I am running into some issues with which I would greatly appreciate some help.
The data I am using is a national survey with final sample weights (weight) available that take into account the survey’s design effect. For each respondent, I created an “ideological distance” variable (ideo), which is how far away ideologically the respondent is from their member of Congress. For example, if a respondent is very liberal, and her Congressperson is very conservative, ideo would be a very big number. Ideo would be a very small number if they are both very liberal. The model I want to run would, in addition to having other variables, assess the impact of ideo on the likelihood that the respondent approves of their Congressperson (approve – a binary variable). Further, I want to look at only subsets of respondents – specifically, looking at one model including only Democrats, and one model including only Republicans.
The data is hierarchical - with the ideo variable, multiple respondents often come from the same Congressional district (district) and have the same member of Congress, so they are not completely independent). Because of this, I’d like to use a multilevel mixed effects model. However, when I run a survey-weighted model that tries to incorporate this hierarchy, this is the output I get:
svyset [pweight=weight]
svy, subpop(democrats): melogit approve ideo || :
(running melogit on estimation sample)
survey final weights not allowed with multilevel models;
a final weight variable was svyset using the [pw=exp] syntax, but multilevel models require that each stage-level
weight variable is svyset using the stage's corresponding weight() option
I can't provide a district-level weight, though, because a) the survey did not do any stage of sampling at the district level, and b) all I have is final weights. One solution I thought about was to create a weight variable equal to one for Congressional districts, and incorporate that into the svyset command, as such:
gen x = 1
svyset district, weight(x) || _n, weight(weight)
svy, subpop(democrat): melogit approve ideo || district:
Here, I get output that makes some sense, but does running a model like this incorporate the final sample weights correctly? To me, it logically seems like it would, but it's not an assumption I would want to make without asking people smarter than me
Is this a correct way to account for district-level effects while still applying survey weights correctly? If not, how could I account for the hierarchical nature of the data while still applying these final, respondent-level weights correctly? Any advice or thoughts would be helpful; thanks!
- Ryan Strickler
The data I am using is a national survey with final sample weights (weight) available that take into account the survey’s design effect. For each respondent, I created an “ideological distance” variable (ideo), which is how far away ideologically the respondent is from their member of Congress. For example, if a respondent is very liberal, and her Congressperson is very conservative, ideo would be a very big number. Ideo would be a very small number if they are both very liberal. The model I want to run would, in addition to having other variables, assess the impact of ideo on the likelihood that the respondent approves of their Congressperson (approve – a binary variable). Further, I want to look at only subsets of respondents – specifically, looking at one model including only Democrats, and one model including only Republicans.
The data is hierarchical - with the ideo variable, multiple respondents often come from the same Congressional district (district) and have the same member of Congress, so they are not completely independent). Because of this, I’d like to use a multilevel mixed effects model. However, when I run a survey-weighted model that tries to incorporate this hierarchy, this is the output I get:
svyset [pweight=weight]
svy, subpop(democrats): melogit approve ideo || :
(running melogit on estimation sample)
survey final weights not allowed with multilevel models;
a final weight variable was svyset using the [pw=exp] syntax, but multilevel models require that each stage-level
weight variable is svyset using the stage's corresponding weight() option
I can't provide a district-level weight, though, because a) the survey did not do any stage of sampling at the district level, and b) all I have is final weights. One solution I thought about was to create a weight variable equal to one for Congressional districts, and incorporate that into the svyset command, as such:
gen x = 1
svyset district, weight(x) || _n, weight(weight)
svy, subpop(democrat): melogit approve ideo || district:
Here, I get output that makes some sense, but does running a model like this incorporate the final sample weights correctly? To me, it logically seems like it would, but it's not an assumption I would want to make without asking people smarter than me
Is this a correct way to account for district-level effects while still applying survey weights correctly? If not, how could I account for the hierarchical nature of the data while still applying these final, respondent-level weights correctly? Any advice or thoughts would be helpful; thanks!- Ryan Strickler
Comment