As we intend to calculate the standardized prevalence of symptoms to model associations overtime in a prospective cohort study by using survey instruments, how could we employ the Generalized Estimating Equations functions by using SPSS, STATA or SAS? Is there a possibility for weighting respondents for the likelihood of attrition over time of the study period? Much importantly, how do we interpret the model?
In Stata you can use 'xtmixed' with the 'noconstant' option (to omit the random intercept otherwise included by default), the 'mle' option (to get ML parameter estimates) and then use the 'residuals()' option to specify the covariance matrix between occasions (within-subject) and the 'vce(robust)' to get the correct standard errors.
If you want your model to accommodate different timing of occasions, number of occasions or spacing between them I am afraid you need to allow residuals' covariance to differ across subjects. This means that, at least in Stata, you need to use random coefficient models (cannot use marginal models like GEEs). As a consequence, the interpetation of the betas will be no longer 'marginal' (i.e. averaged out across all subjects) but subject-specific.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS