Weighting for longitudinal youth analysis
I'm a running an analysis which involves data from wave 1 and wave 3 of the youth self-completion questionnaire. Bth wave 3 and wave 1 variables contributes to the dependent variable I am deriving; while my predictor variables and co-variables are from wave 1. What would be the appropriate weight variable for this analysis, as I can't seem to locate a longitudinal weighting in the W3 youth dataset.
#1 Updated by Alita Nandi about 5 years ago
You are essentially using response from waves 1 and 3, so the most appropriate weights for you will be longitudinal weights from wave 3. But note that these weights are non-zero for those who responded in waves 1, 2 and 3 and so your effective sample size will reduce as it will be zero for all those who missed at least one interview.
#2 Updated by Kareena McAloney-Kocaman about 5 years ago
Sorry I have been unclear - I was aware that I needed to use a longitudinal weight, my query was actually which weight that was in the wave 3 youth data, as I cannot see one in that particular dataset. Can you give me the variable label of the weight for longitudinal analysis of youth data using wave 3 and (if it is not held in the youth data set direct me to which dataset it is in)?
#3 Updated by Alita Nandi about 5 years ago
Sorry I misread your question and my advice was more general - about when to use longitudinal weights. You did not find youth longitudinal weights because we do not provide those. The longitudinal weights we provide for all other instruments are designed for those who respond continuously up until that wave. This method is not optimal for the youth questionnaire as there are very questions that are repeated annually and part of the sample becomes ineligible at each subsequent wave (as they move out of the age range).
#4 Updated by Kareena McAloney-Kocaman about 5 years ago
Ah, yes I see the problem with the youth data set. In that case I might proceed with a cross-sectional weight appropriate to the sweep of the outcome variable, and just note it in any write-up. My analysis requires only the participants who responded at both waves anyway, so that seems like a best compromise. Thanks!