[R-sig-ME] How to Model U-Shaped Distribution in LME4

[Poe, John]

Given that you've only got 8 response options and the data are multimodal
it's arguable that you're best off going with a discrete choice model like
multinomial or mixed logit. You can collapse adherence, partial adherence,
and no adherence into three or four categories and then model them that
way.


On Mar 16, 2017 8:41 PM, "David Jones" <david.tn.jones at gmail.com> wrote:

I am looking at medication adherence as a DV. I am having difficulty
considering how to model adherence in LME4, as the distribution is very
u-shaped.

In particular, the distribution is bimodal - participants usually
didn't adhere at all, or adhered completely (30% of responses did not
adhere at all, while 55% of responses had full adherence, despite it
being measured on a ratio scale). Q-Q and P-P plots also reflect
problems with modeling adherence as Gaussian.

I was wondering what thoughts would be regarding the best distribution
to apply to this variable - beta binomial has been mentioned, and I
was wondering if other options come to mind that are available in
LME4.

Of note, as it has turned out there have only been 8 response
categories despite adherence being a continuous measure (the measures
asked people how many pills they missed weekly; many people were
prescribed only 1 pill/day, thus there were only 8 options).

Many thanks!

_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

	[[alternative HTML version deleted]]