[R-sig-ME] Matrix of predicted values using lme or lmer
Ben Bolker
bbolker at gmail.com
Sat Jun 23 21:30:43 CEST 2012
Matthew Smith <mts002 at ...> writes:
>
> I am currently trying to used a mixed model to perform a Procrustes
> Trajectory analysis on some shape data. I have 17 relative warps
> scores as my dependent matrix accompanied by individual, treatment,
> size, and day. Shape data was collected at 3 separate time points for
> each individual, so this is a repeated measures design. Since I do not
> want to analyze each relative warp separate, I have chosen to use
> individual as a random effect along with my fixed effects.
> The procedure calls for a residual randomization, in which residual
> values from a reduced model are added to the predicted values of the
> full model.
>
> The origin code use the lm function, but since I wanted to add random
> effects, I am using lme. My problem lies in when I attempt to get a
> matrix of predicted values rows=# of obs, cols=17 (each RW), it does
> not match up.
>
> I read several discussion forums online and other threads, but I
> cannot seem to come up with a clear answer. Is there a predict command
> for the lme function that will take into account both the fixed and
> random effects? If not, is there one that just takes into account the
> fixed effects?
Without a reproducible example, it's a little hard to say. The
predict method for lme definitely takes the random effects into account
(have you read ?predict.lme ?) -- whether the random effects are taken
into account or not depends on the 'level' argument: level=0 excludes
all random effects; the behavior defaults to including all random effects.
In order to do a multivariate analysis with lme, I believe you will
need to convert your data to 'long form', i.e. each row will represent
a measurement of a particular warp score on a particular individual
at a particular time. Someone here may be able to suggest a good reference
for doing this kind of multivariate analysis in lme (as I recall, it
is *not* covered thoroughly in Pinheiro and Bates 2000, which is otherwise
the standard reference text).
Ben Bolker
More information about the R-sig-mixed-models
mailing list