Hello everyone :)
For my bachelor‘s thesis, I‘m conducting a linear mixed models analysis using R. Since LMMs aren't part of our statistics course until the master‘s program, I'm honestly a bit at loss about the intricacies of the lme4() command.
My missing data is MCAR since it‘s due to a mistake in programming the study. Put simply, due to missing one letter in the randomization, four out of eight experimental groups haven‘t been shown 5 items each for one of 12 Texts. Their other data is complete. According to https://rpsychologist.com/lmm-slope-missingness MCAR data is largely ignorable.
My supervisor suggested just using the data as is. But what exactly would lme4() do in that case?
Or would I be better advised reading into multiple imputation (for example using the mice package - which I‘ve seen recommended elsewhere)?
Many thanks to everyone who can give their input!
(I know it can be quite frustrating talking complete newbies. I‘m honestly looking for a sensible starting point where I can dig more deeply without wasting too much time trying to understand dead ends to my problem.)
It depends if the data missing are predictors or outcomes, but in your case it seems you lost both missing predictors and outcomes for some trials. In this case the issue is really imbalance. Multilevel models and by extension lme4 handles this well and will correctly estimate the model (though with some loss of precision for some effects because of the lost trials).
More generally if only outcomes are missing the estimates of multilevel models are OK under MCAR or MAR (the latter basically assumes the predictors in the model account for missingness). There can still be reasons to use multiple imputation - mainly for missing predictors - this is often more to preserve power by not dropping too many cases.
- Badges
- Science topic