Measurement Error

[pzivich]

Measurement error is a problem in analyses. For a single mis-measured variable, two options include multiple imputation for measurement error (MIME) and re-parameterized imputation for measurement error (RIME).

While easy to apply by 'hand', it would be nice to include frameworks for these as tools. I will have to think how to work these in the current structure (the multiple imputation part makes somewhat complicated when used with other methods). Also RIME has its own unique likelihood function

What this adds:

A new functionality to correct for measurement error of a single variable. This would allow support for internal and external validation samples. The generated data sets would then be applied through the remaining estimation procedure via a for loop. Imputed data can then be merged via rubins_rules.

References:

• MIME: relies on internal validation subsample. https://pubmed.ncbi.nlm.nih.gov/16709616-multiple-imputation-for-measurement-error-correction/

• RIME: allows for internal and external validation subsample. https://academic.oup.com/aje/advance-article/doi/10.1093/aje/kwaa011/5714914