[R-sig-ME] Predictor variable with quantifiable measurement error

[Mike Lawrence]

Hi folks,

I'm wondering if there's a way to specify an lme model in a way that
lets it take into account quantifiable measurement error in a
predictor variable. The context is that I am interested in how fast
people categorize faces (as male vs female) as a function of the
attractiveness of the faces. I've measured the decision time to each
face multiple across multiple repetitions, eg:

categorization_data = expand.grid(
	subject = factor(1:40)
	, face = factor(1:20)
	, trial = 1:10
)
categorization_data$time = rnorm(nrow(a))

In a separate task, I've also elicited each subject's rating of each
face's attractiveness multiple times, eg:

rating_data = expand.grid(
	subject = factor(1:40)
	, face = factor(1:20)
	, trial = 1:2
)
rating_data$rating = sample(1:10,nrow(rating_data),replace=T)

Now, I could of course get the mean rating per face and use that mean
rating to predict categorization time (something along the lines of
"lmer( time ~ mean_rating  + ( 1 | face ) + ( 1 | subject ) )" ), but
I suspect that it might be important to let the lme know that there's
quantifiable measurement error in the rating data. I'm simply not sure
how to combine the data then formulate the model to achieve this.
Suggestions?

Mike

-- 
Mike Lawrence
Graduate Student
Department of Psychology
Dalhousie University

Looking to arrange a meeting? Check my public calendar:
http://tr.im/mikes_public_calendar

~ Certainty is folly... I think. ~