Understanding Society User Support

Ante,
Yes, you should use the longitudinal weight for wave 4, and the choice of d_indscus_lw is correct.
Olena

Dear Olena,

Thank you very much for your fast clarification.

Best regards,
Ante

Dear Olena,

I have a follow-up question.
If I choose d_indscus_lw to compare the sub-sample with the original sample (population), only matched IDs between Wave 4 and 1 (or 3) will be accounted for to calculate the population's figures for Wave 1 variables (which are not available in Wave 4). For the unmatched IDs from Wave 1, there is no d_indscus_lw weight so that these are excluded. The population size can thus be considerably reduced. Does d_indscus_lw adjust for these unmatched IDs, or is the use of Wave 1 weights (i.e. a_indscus_xw) then nevertheless the right approach to most accurately describe the population since it will cover all IDs from Wave 1 (the matched and unmatched ones)?

Thank you for your help.

Best regards,
Ante

Ante,

Your problem is not weights but as far as I understand your suggested analysis violates iid assumption - because the subgroups is part of the population. It would be easier to compare the subgroup to everyone else (population excluding the subgroup). You could do this in many ways one of which would be to use data from w1-w4 with the longitudinal w4 weight.

I hope this helps,
Olena

Dear Olena,

Thanks for your comment.

I agree that the subpopulation should be compared with population excluding the subpopulation.

There are indeed many different ways of which weight to use for which wave. The difficulty is to pick the combination that most accurately describes the population: is the accuracy of the population figures distorted more if

1) only matched IDs are used (between wave 1 and 4) with consistently using one weight for all variables (d_indscus_lw) or
2) all IDs are used with inconsistent choice of weights (for w1 variables a_indscus_xw and for w4 variables d_indscus_lw).

Does for this matter exist a clear criteria according to which one can decide what the better approach is?

Best regards,
Ante

Ante,

Yes, the easiest way is a) - the weight will automatically select the correct IDs - just by using the weight in your analysis you will take care of 'the selection'. b) is not impossible, but you really have to have theoretical reasons to use it, and the statistical software will not do the comparison automatically for you - so you will have to do some calculation by hand which I don't recommend.

Hope this helps,
Olena

Dear Olena,

I appreciate your patience in this matter. Thank you very much for your help.

Best regards,
Ante