search

amices / mice

Multivariate Imputation by Chained Equations
https://amices.org/mice/
GNU General Public License v2.0
433 stars 107 forks source link

midastouch: use neff option #458

Closed sff06 closed 2 years ago

sff06 commented 2 years ago

I use midastouch to impute missing values. I further would like to apply the bias correction factor presented by Gaffert et al. (2016, eq. 4) that originates from Parzen et al. (2005). However, I am not sure how to do this in mice. In line with your R code instructions I created a new environment (PFLenv) and added to mice(..., neff = "PLFenv"). Would that suffice to use the correction factor?

I am having doubts because I ran mice(..., method = 'midastouch') with and without neff = "PLFenv", but the results were identical. Sample size is N = 227. Missingness in key variables is 16%.

Thank you for any support!

Best, Stanley

stefvanbuuren commented 2 years ago

This is best answered by Philipp Gaffert or Florian Meinfelder. Both are mice contributors.

sff06 commented 2 years ago

Thank you for your response! Both are not mentioned under constributors. So I guess they won't be able to notice my issue. I will try to contact them via other means and hopefully can post a clarification of the issue.

stefvanbuuren commented 2 years ago

They are package contributors. See DESCRIPTION.

PhilippGaffert commented 2 years ago

With every execution, the correction factor is calculated and added to your specified environment. The correction factor is not considered in the pooling that is done by mice. If you wanted to use these correction factors, you would need to write your own pooling. Be aware that the correction factor that is derived in Gaffert et al. 2018 is only applicable to means. Correction factors for more complex analysis models are not yet developed to the best of my knowledge. Whatever you discover in this process, I'd be keen to learn about.

sff06 commented 2 years ago

Thank you for your response and the note regarding the mean! Currently, I focus on robust correlations and differences of trimmed means, so I won't be using the PLF correction anymore and just rely on the midastouch without this correction.