I have checked the guide and previous issues on weighting for Covid-19 study but I am still unclear of the best way to use them. I am running a) a cross-sectional analysis and b)a longitudinal fixed effects analysis using the waves 6 to 8 (waves g,h,i), baseline pre-Covid (jk) and the 4 waves of Covid-19 survey (ca to cd, until July). Should I use the i_indinui_xw for each of the cross-sectional waves until wave (I cannot find weights for the wave jk, so I am not sure how I would solve this) and the *_betaindin_xw for the Covid-19 survey waves? For the longitudinal analysis, I read in a previous query that you are working on a longitudinal weight, do you have an tentative date for releasing it? In the meantime, what is the best approach?
Many thanks in advance,
Updated by Alita Nandi 2 months ago
- Private changed from Yes to No
- % Done changed from 0 to 80
- Assignee set to Laura Silva
- Status changed from New to Feedback
(a) For cross-sectional analysis use the weights cW_betaindin_xw for Covid waves 1-4 (W=a,b, c, d), W_indinui for main survey waves 6-8 (W = f,g, h). Note that in Wave 6, if any variable in your estimation model was not asked of IEMB sample, then you should use the weight f_indinub_xw. You should not analyse the baseline pre-Covid (jk) files by themselves, only with the Covid wave data, and so no weights have been included in these jk files.
(b) For the longitudinal analysis I am assuming you are doing a separate analysis using main annual survey waves 6-8 and Covid waves 1-4 separately? For longitudinal analysis you need to use the weight from the last wave and apply that to all waves, as this weight will account for the wave -on-wave nonrespose across all waves. So, for FE analysis using main annual survey waves 6-8 use h_indinui_lw, and for FE analysis using Covid waves 1-4 use d_betaindin_lw.
Understanding Society User Support Team
Updated by Laura Silva 2 months ago
Thanks so much for your detailed and prompt reply.
Just a note around (b). We are not doing a separate analysis, since we are doing a diff-in-diff analysis with a dummy variable 0-1 meaning pre and post covid (where we take wave 6-8 and baseline jk as "pre-covid" and Covid-19 waves 1-4 as "post-Covid").Would your answer change in this case?
I wish you a lovely evening,
Updated by Laura Silva about 2 months ago
Sorry for bothering again with this! I still have this doubt regarding the longitudinal weight.. we have two diff-in-diff models which take respsctively these waves. In order:
- diff in diff model 1: wave 7, 8, 9, jk, ca(covid1), cb(covid2), cc(covid3), cd(covid4).
- diff in diff model 2: wave 7, 8, 9, jk, cb(covid2), cd(covid4). --> (in this case our outcome is measured only in the cb and cd waves of covid, that's why we discard the others).
To estimate the DiD coefficient, we don't need to have all individuals responding at all waves. In fact, we just need that individuals have responded to at least one wave pre-Covid AND at least one wave post-Covid t.
So, considering this, what kind of weight should we use?
Many thanks and, again, sorry to bother again!
Updated by Alita Nandi about 2 months ago
- Assignee changed from Laura Silva to Alita Nandi
(b) In that case there is no appropriate weights. If you were to use Main survey Wave 9 + Covid waves for DID (pre and post covid), then covid longitudinal weights would be appropriate as these build on Main survey Wave 9 weight. See Section "16.6Respondent longitudinal weights" in the user guide.