1
$\begingroup$

I have a model something like this:

lmer(y ~ factor(month) + x + (1|subject))

Where month represents repeated measures of y. The x variable is fixed-per subject.

However x is an aggregate of multiple measurements taken before the main experiment started, and these are unbalanced. For some subjects x is based on 2 observations, but for others it's based on as many as 30.

Is there any way to include uncertainty in the measurement of x in the main model without re-running this as some kind of SEM?

EDIT: I've just discovered this is called an errors-in-measurement model, and is possible using the brms package, which in turn relies on Stan (see https://github.com/paul-buerkner/brms/issues/114). If there is a simpler way to run these models without resorting to MCMC I'd still be interested though.

$\endgroup$
1
$\begingroup$

It sounds like what you want is not to model measurement error but to weight x according to the number of trials the variable is based on. lme4 has a weight parameter, but I believe that's for weighting entire observations. The only thing I can think of that lets you stick with lmer is to transform x in some fashion so that it has a bigger absolute value when the number of trials is bigger. Since I have no idea what x represents, I couldn't give you a concrete suggestion of a transformation.

$\endgroup$

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.