Jeehoon Kim posted on Sunday, February 06, 2011 - 2:51 pm
I'm running a path model with a national data set was designed in two stage stratified sampling.
I created balanced repeated replicate weights using Stata and have used them in Mplus.
My question is I am wondering how I can use a subset of the whole data set (G is 0 or 1, not 2). I tried to use subpopulation is G < 2, but it didn't work.
If I didn't use replicate weights, I can use subpopulation option. Could you please advise me how I can only use a subset of the whole data set with replicate weights? Should I use useobservation command?
No, USEOBERVATIONS uses only the observations specified.
Jeehoon Kim posted on Tuesday, February 08, 2011 - 1:30 pm
Thanks, Linda. My question is back how I can select subset of the whole data set when use replicate weights?
Because it was degined in complex survey, I think that I need to use the whole data set even though I'm only interested in part of it. But with replicate weights, I cannot use subpopulation option and don't know which command I can use. Could you please advise me? Thanks a lot.
With BRR, setting the weights to 0 for all observations outside of the subpopulation, will give you the same result as useobseravtions (that is because BRR SE are obtained from point estimates and those do not change from adding the 0 weight observations), i.e., useobseravtions is easiest way to get correct estimates.
If I want to use the Taylor Series Linearization method for variance estimation, how can I do subpopulation analysis with the type=twolevel option? I wanted to compare the results with what I get from the type=complex option?
p.s. I am new to the model-based approach to survey analysis and I want to compare it with the design-based approach considering the weights, stratification and clustering. This is the main reason why I bought MPLus with the combination add-on. My package arrived just yesterday.
Stata posted on Friday, September 14, 2012 - 5:00 pm
In one of the previous Mplus discussions talked about multiple group CFA with complex data. My understanding is that useobservation command is preferred over subpopulation (only need a subset of the population). However, when I used useobservation command, it showed that cluster and stratification should not be used. In fact,I really don't need the stratum and cluster variables in the analysis. Is it appropriate if I only use useobservation and sampling weight without stratification and cluster as well as type=complex? Will it be a problem without taking unproportional sampling into consideration? Any comments and suggestions are highly appreciated.
I think you are asking if you can use a weight variable only in the model estimation without any other complex survey options. The answer is yes. And because you are not using TYPE=COMPLEX, you would use USEOBSERVATIONS instead of SUBPOPULATION. However, you are taking only the unequal probability of selection into account not stratification and non-independence of observations.
Stata posted on Saturday, September 15, 2012 - 11:36 am
Than you very much.
Luo Wenshu posted on Wednesday, June 24, 2020 - 3:03 am
Dear Dr Muthen,
For two level data, students nested in classes, when I run analysis at student level using type=complex, and choose to work on some selected students (e.g., from Sec 3 classes), should I use SUBPOPULATION or USEOSERVATIONS to select students? Which will lead to correct error estimation?