bbolker at gmail.com
Tue Mar 7 15:41:32 CET 2017
* this is a general mailing list (I asked you to redirect an off-line
query to r-sig-mixed-models, but I think I suggested that you change the
* a more informative subject would help (everyone who writes to the
list wants "Help" ...)
Are you sure you want all independent variables to be random effects?
That would suggest that you don't care about the effects of any
particular variables (e.g. streams), only about decomposing variance,
which I would find surprising. Also, for practical mixed modeling you
need to have a reasonable of levels (e.g. >5); do you have more than 5
I'm going to take a guess that you really want to assess effects of
A reasonable model for this would be
score ~ S + (1|A) + (1|Q) + (1|D)
This assumes that each applicant has a unique ID number (e.g. there is
not an applicant #24 in Stream 1 and in Stream 2, but these applicants
would be coded as 24-1 and 24-2). Also assumes S is a factor.
This is not a maximal model in the sense of Barr et al 2013: in
principle you could estimate among-applicant variability in Q, but that
would generally take a great deal of data.
For future reference it would help to know more about your design; how
many Streams, how many Days, how many Questions, how many total
observations? It would also help to know what your actual subject-area
question is: what are you trying to find out from this analysis?
You might also want to read the "nested or crossed?" section in
On 17-03-04 03:02 PM, Rose Rosei wrote:
> Dear Professor Bolker
> I have 100 applicants (A). They are nested in Streams (S). Applicants are
> nested in days (D). Applicants nested in sessions (S) and Applicants
> crossed in Question (Q), the dependent variable is a score. All
> independent variables are random
> Would you please advise me how to address the codes in lme4.
> Many thanks and very much appreciated.
> [[alternative HTML version deleted]]
> R-sig-mixed-models at r-project.org mailing list
More information about the R-sig-mixed-models