[R-sig-ME] ENQUIRY: independent slopes of categorical effects in a random-effects

[Blything, Liam]

Dear Dr Bolker,

Enquiry: Independent slopes of categorical effects in a random-effects.

An important prior note is that I am dealing with IVs that are binomial (e.g., the influence of age: young vs old).
My advisor and I both cannot understand the output for a random effects structure that occurs when excluding the correlations between the slope and the intercept (i.e., without investing a parameter for their correlation).

Specifically, if I put in a simple syntax for any dataset, such as below, ....
NAMEOFMODEL= glmer(DV + (IVone + IVtwo) +
(1|subject) +(0+IVone|subject), data=ANYDATASET, family = "binomial")

.... The random effect structure is not what we would expect:

Random effects







        Variance
SD

Subject (intercept)







        0.10
0.32

Subject.1 (slope) young







        0.20
0.45

Subject.1 (slope) old







        0.42
0.65


Obviously, we would expect the following output (below) just as we get when we do not exclude correlations [e.g., (IVone+1|subject) ]

Random effects









Variance

SD

Subject (intercept)









0.53

0.72

Subject.1 (slope) IVone









0.29

0.54



I find the first random effects output extremely difficult to interpret, and I wonder whether it indicates a bug?


Your expertise would be very valuable here, and hoprfully it helps raise an issue for the future when people want to interpret a random effects structure that has a binomial IV slope that is not allowed to correlate with the intercept.

Thank you in advance,

Liam Blything

	[[alternative HTML version deleted]]