0
$\begingroup$

I have data from a between-subject experiment (Repeated measures, two conditions). Condition A has 15 participants while condition B is with 19. The difference is due to missing data.

My aim is to use this data to compare the frequentist( lme4) approach to the Bayesian approach (brms).

What can I do about the difference in sample size between A & B?

$\endgroup$
  • $\begingroup$ It might be informative to remove the missing data from both groups and regress them together as a single data set. There is a possibility that something interesting could be found from this, and such a test is certainly easy to perform.. $\endgroup$ – James Phillips Aug 31 '19 at 11:08
  • $\begingroup$ @JamesPhillips Why & how to combine two different datasets(two conditions) into a single data set to compare the difference between the two conditions? $\endgroup$ – amarykya_ishtmella Sep 1 '19 at 13:48
  • $\begingroup$ I was discussing this portion of your question: "The difference is due to missing data". $\endgroup$ – James Phillips Sep 1 '19 at 13:59
1
$\begingroup$

Both the frequentist and Bayesian approaches are likelihood-based approaches in this case. And likelihood-based approaches give you valid results under both the missing completely at random and missing at random missing data mechanisms. This is under the proviso that the model is correctly/flexibly specified. This includes also the variance-covariance structure for the repeated measurements. That is, you should model the correlations in the repeated measurements adequately.

In both approaches you can/should work with all available data. You should not do a complete cases analysis (i.e., only consider the 15 subjects who provide all measurements) because this will be less efficient and also be valid only under missing completely at random.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ For bayesian I was not skeptical about including all data. But for doing ANOVA, isn't different sample sizes a problem considering the difference between the conditions w.r.t the data is around 20%. $\endgroup$ – amarykya_ishtmella Sep 1 '19 at 14:02
  • $\begingroup$ Package lme4 does not do ANOVA. It fits mixed-effects models using maximum likelihood or restricted maximum likelihood (the latter being only applicable for continuous outcomes and normal error terms). $\endgroup$ – Dimitris Rizopoulos Sep 1 '19 at 18:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.