martin ★★ Austria, 2017-03-28 17:25 (2578 d 11:35 ago) Posting: # 17194 Views: 2,521 |
|
Dear all, I have a question regarding estimating intra-subject variability in a fixed sequence 2 period design (i.e. A followed by B). This could be analyzed by a paired t-test or mixed model whereas the latter has some benefit in case of missing values in one period assuming that missing values are missing at random and I will focus therefore on the mixed model approach (although probably challenged by some regulatory agencies for that reason). With the mixed model approach you model the subject as random factor and treatment as fixed and you get an estimate for the between subject variability and the residual error. My question: on which premises can the residual error obtained from the mixed effects model to be interpreted as within subject variability? Best regards & looking forward to an interesting discussion Martin |
DavidManteigas ★ Portugal, 2017-03-29 12:59 (2577 d 16:01 ago) @ martin Posting: # 17199 Views: 2,008 |
|
Hi Martin, A mixed model is only useful if you have more than 2 periods (or repeated measures). In the case you are explaining, if data is missing for one period it can't be used in any way regardless of the statistical method used. Also, I can't understand why would you use a single sequence design in bioequivalence, and also why you use a mixed effects model in a "pre-post" design... Anyway, I guess that as long as you use a random slope and intercept model, residual variability will be your "within subject variability"... but not sure how that applies to a mixed effects model with two periods only. Regards, David |