I'm working with large datasets that use complex sampling procedures (ie, clusters and strata). I can specify the sample design using the "svyset" commands when working in STATA. However, I have not yet succeeded in applying the same in the lme4 or nlme R. packages.
I'm slowly switching to R from Stata and some basic functions are still hard to find.
I suggest you look at the notes in this link. You want to build the RHS I believe.. Best wishes, David Booth
https://www4.stat.ncsu.edu/~davidian/st732/notes.html
I suggest you look at the notes in this link. You want to build the RHS I believe.. Best wishes, David Booth
https://www4.stat.ncsu.edu/~davidian/st732/notes.html
I also recommend the book from Gelman and Hill:
http://www.stat.columbia.edu/~gelman/arm/
There are three elements as I see it in complex designs
1 clusters - that is what multilevel (mixed) models do, so no special issues there.
2 Strata - these can be included as fixed effects in the mixed model so that you condition on them
3 Oversampling - good software allows you to include sampling weights and you can do this at each level.
In reality most multilevel models do not bother beyond using the clusters and do so as more than a technical fix.
Article Multilevel approaches to modelling contexuality: from nuisan...
You can account for clusters, strata and survey/sampling weights using the "survey"-package (https://cran.r-project.org/package=survey). There is an accompanying website and book (more info: http://r-survey.r-forge.r-project.org/survey/).
Currently, mixed models in R do not correctly account for sampling weights. However, Thomas Lumley, author of the survey-package, works on a survey-function for mixed models (https://github.com/tslumley/svylme), probably the GitHub version is quite stable (haven't tested yet).
An alternative would be the "scale_weights()" function from the sjstats-package ( https://strengejacke.github.io/sjstats/articles/mixedmodels-statistics.html#rescale-model-weights-for-complex-samples ), which rescales sampling weights so they can be used as "weights" for the mixed models function you have in R (lme4, lme, ...).
Thank you for your answers.
I believe that my question was not clear. But I was looking on how to include sample weights correctly in a complex survey. The answer from Daniel Lüdecke directly addressed it.
I also understand that we can specify the model according to the survey clusters and strata to obtain correct estimates. The answers from David Eugene Booth Kelvyn Jones and Jochen Wilhelm helped me with this. But for accounting for strata, I also need to include the strata as both fixed and random effect, right? By including only as fixed effect I would be assuming that the effect of strata does not vary between clusters.
I am reading the recommended book from Gelman and Hill and I am trying to specify models that can correctly represent the survey.
in practice I would be aware of the need of sampling weights at each level; were diffent clusters differentially sampled?how many strata are there
?
Kelvyn Jones
The study is a national survey aimed to evaluate cardiovascular risk factors and the sampling procedure considered participants (level1) nested within country regions (level2). The procedure also included age group and gender as strata.
I really do not understand why age group and gender should be included as strata, but as this is a secondary analysis, the procedure was already done and now I have a dataset with only one sample weight variable.
The inclusion of the sample weight as a unique variable is a problem because I want to fit a mixed model which includes gender and age as varying slopes. But I do not know how this should be correctly done as the sampling procedure included this variables as strata.
The evidence mentioned by Ricardo Orozco also denotes the problems of not considering level specific sample weights.
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Sampling Design