[R-sig-ME] Linear mixed effect model
bbolker at gmail.com
Fri Mar 18 13:51:49 CET 2011
On 11-03-18 08:19 AM, Manuel Spínola wrote:
> Dear list members,
> I am trying to fit a linear mixed model using the following variables::
> Response variable:
> Swiftness.2 (This is the time it took for the otter to first approach
> the lure. The time ranges from 1 second (in which case the otter
> approached the lure almost immediately) to 600 seconds (10 minutes).
> Explanatory variables:
> 1) Subject (this is the individual otter -- each otter is measured for
> response to each lure, so it is a repeated measure on the individual);
> 2) sex;
> 3) facility size (small, med, large);
> 4) lure type (there were 6).
> I would like to see if the response variable is influenced by the
> explanatory variables including Subject like a "repeated measure" term
> (same animal expose to different lures).
> I am fitting the model:
> otter$Facility.Size = factor(otter$Facility.Size)
> otter$Sex = factor(otter$Sex)
> mod1 = lmer(Swiftness.1 ~ Lure + Sex + Facility.Size + (1|Subject), data
> = otter)
> > mod1 = lmer(Swiftness.1 ~ Lure + Sex + Facility.Size + (1|Subject),
> data = otter)
> > summary(mod1)
> Linear mixed model fit by REML
> Formula: Swiftness.1 ~ Lure + Sex + Facility.Size + (1 | Subject)
> Data: otter
> AIC BIC logLik deviance REMLdev
> 1277 1295 -631.3 1302 1263
> Random effects:
> Groups Name Variance Std.Dev.
> Subject (Intercept) 0 0.00
> Residual 21558 146.83
> Number of obs: 102, groups: Subject, 17
> Fixed effects:
> Estimate Std. Error t value
> (Intercept) 92.883 44.711 2.077
> Lure -6.286 8.513 -0.738
> Sex1 -3.266 29.199 -0.112
> Facility.Size2 24.174 37.628 0.642
> Facility.Size3 58.528 38.692 1.513
> Correlation of Fixed Effects:
> (Intr) Lure Sex1 Fcl.S2
> Lure -0.666
> Sex1 -0.327 0.000
> Facilty.Sz2 -0.516 0.000 -0.055
> Facilty.Sz3 -0.519 0.000 0.000 0.617
> Is the model a plausible model and is it well parameterized?
Plausible, yes, except that you have apparently failed to
transform Lure into a factor -- as it stands, lmer is treating
it as a continuous covariate.
Effects seem quite small.
I would worry a little about your distribution, because I would guess
that elapsed times are likely to be skewed. Have you looked at the
residuals/thought about log-transforming?
You are getting zero variance for the random effect (and a huge
residual variance), which suggests a general lack of power.
More information about the R-sig-mixed-models