ECLS-K data and weights

[zq8890]

When using ECLS-K data with EdSurvey package:

1. should child level weights be scaled in anyway for use with "long" data and/or mixed models?
2. when pulling in data, is it acceptable to only run analyses on a portion of data relevant for analyses? (i.e. only those with complete BMI data at all waves). In other software packages, normally the full dataset is used, but the PSU/Strata and weight variables only include the relevant observations.

[pdbailey0]

Thank you for your question. Since this is a question specific to ECLS data, we recommend contacting Jill McCarroll, the ECLS study director, at Jill.McCarroll@ed.gov. The ECLS team will be able to respond directly to your question about weights.

[zq8890]

Hello - after following up with Jill McCarroll, she suggested reaching out to the EdSurvey team about the appropriate methods for producing estimates on subpopulations using the EdSurvey package. For example, in SUDAAN and other software, it is recommended to use a subpopulation statement so that the full sample is used to estimate standard errors. Is there an equivalent in the EdSurvey package?

Thank you! Zerleen

Zerleen Quader, MPH Emory University Department of Epidemiology

[edited by pdbailey0 to remove verbose GH email headers.]

[pdbailey0]

@zq8890 there is no need to do this in EdSurvey. If you subset data to a subpopulation the mean and variance should account for the sampling design, using the full sample and replicate weights provided by NCES.

[pdbailey0]

I'm going to close this because this discussion seems to be complete.