Skip to content

mixed effects model glmer

4 messages · Ben Bolker, Thierry Onkelinx, M West

Original
M West
# 
I am trying to fit a mixed effects model with repeated measures data.

Data are:

y variable = percentage (# females/total)
x variable = percentage

measured across multiple sites for 4 years.

here's the model:

y <- cbind(total females, (Total - total females)))

mod1 <- with(data, glmer(y ~ disease prevalence +  (1|Site) + (1|Year),
family = binomial,  data = data1))

1) This model runs, but the summary(mod1) just generates a series of the
following....which doesn't make any sense so something must be wrong with
the model specification...I'm just not sure what.

2) Also, what is the default AR correlation on these models (i.e., do I
need to specify it or is the temporal psuedoreplication taken care of)?

3) Finally, do you suggest another form of the model that's better etc.?

Fixed effects:
                                Estimate             Std. Error    z value
   Pr(>|z|)
(Intercept)                  -1.60267            0.11618    -13.794    <
2e-16 ***
disease prevalence    -0.40212           0.15557     -2.585     0.009745 **
disease prevalence    0.035088231    -1.46452    0.22860  -6.406 1.49e-10
***
disease prevalence    0.064935065     -0.36344   0.30810  -1.180 0.238157

disease prevalence    0.078507945    -2.57479    0.46537  -5.533 3.15e-08
***
disease prevalence    0.120039255    -3.30998    0.71915  -4.603 4.17e-06
***
disease prevalence    0.182623706     -0.14362   0.19899  -0.722 0.470438


Many thanks in advance,
M.
Ben Bolker
#
On Wed, Sep 23, 2015 at 2:36 PM, M West <westm490 at gmail.com> wrote:
Just to be clear, disease prevalence is a number in [0,1]?
AR models are not currently easy in lme4.  My suggestion (=hack) would
be to get the residuals and use nlme::gls(resid~1,correlation=corAR1()) (or
something like that) to see if you should worry about autoregressive structure.

  Four years is not very many, so you might need to treat Year as a
fixed effect (e.g. I would consider that option if the random effects variance
is estimated as zero)

  How many sites?  How many total observations?

  I have to admit that I'm stumped by your apparent model output (i.e.
that there are multiple parameters for disease prevalence when there
should only be one)

  Perhaps you could send the results of summary(data1) and/or
str(data1) and summary() of your whole model?
Thierry Onkelinx
# 
Adding a random effect is equivalent to a compound symmetry
correlation structure. Since you have only 4 years, it would be too
bad compared to an AR1 correlation structure.

If you really need correlated random effects, then you can have a look
at the INLA package. Not on CRAN but on www.r-inla.org.
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature
and Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium

To call in the statistician after the experiment is done may be no
more than asking him to perform a post-mortem examination: he may be
able to say what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does
not ensure that a reasonable answer can be extracted from a given body
of data. ~ John Tukey


2015-09-23 22:41 GMT+02:00 Ben Bolker <bbolker at gmail.com>:
M West
# 
Ah. that worked - no idea why that happened.
Thanks so much! - I'll know what it looks like if that ever happens again.

M.

On Thu, Sep 24, 2015 at 3:44 AM, Thierry Onkelinx <thierry.onkelinx at inbo.be>
wrote: