Multilevel regression analysis with plausible values as dependent variable (PISA)

[Karin Freiberg]

Hello everyone

I am currently working with PISA data from 2018 and face a problem combining the “repest” command used to analyse PISA data correctly, taking account of the complex survey design, with multilevel regression analysis.

My dependent variable is reading test scores which is imputed as 10 plausible values. As the PISA data is clustered, I find it necessary to include these levels in the analysis. However, when I add these levels in the command, I get the following error message:

“REML not supported with weights”

The command I use is:
repest PISA, estimate(stata: mixed PV@READ ESCS || schoolid: , cov(unstructured))

Does anyone know whether repest is compatible with multilevel models or do you have ideas for an alternative procedure for analysing plausible values as the DV in multilevel models?

Thank you in advance,

best regards,
Karin

[Phil Bromiley]

You will increase your chances of useful answer by following the FAQ on asking questions – provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

With a user written procedure, help often depends on someone active on the list using that specific procedure. I have no idea what repest does. I'm also not sure what plausible values means – are you using imputed values or are you creating scales? There are certainly tools to create scales that can be used with other estimators.

[Victoria Haver]

Try using the pv module instead (install ssc pv).

Then you can run mixed using the following structure - pv, pv (pv1read pv2read pv3read pv4read pv5read pv6read pv7read pv8read pv9read pv10read): mixed @pv ESCS || schoolid: , cov(unstructured)