[R-sig-ME] Output glmer

[Ben Bolker]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 15-03-10 07:01 AM, Davide Bellone wrote:
> Good afternoon,
> 
> I have a little problem in my glmer output. In my model, before run
> the model I used
> 
> options(contrasts=c("contr.sum", "contr.poly"))
> 
> So after stepwise deletion I arrive at the finel output:

  First of all, I would caution against stepwise deletion (see e.g.
Frank Harrell's book _Regression Modeling Strategies_, or Google
"stepwise regression problems"


> Formula: y ~ Model$Manage + Model$age + Model$veg + Model$wood + 
> Model$under + Model$veg * Model$Manage + Model$age * Model$veg +
> Model$under * Model$wood + Model$veg * Model$under + (1 |
> Model$Site) + (1 | obs)

  Second, I would suggest that you leave the "Model$" out of your
formula, and that you recognize that * incorporates both main effects
and interactions: your model can be written more simply as

  y ~ Manage + age + veg + wood + under + veg:Manage + veg:age +
      under:wood + under:veg + (1|Site) + (1|obs)

or even

  y ~ veg*(Manage+age + under) + under*wood + (1|Site) + (1|obs)

(there is one redundant term here -- the main effect of under is
incorporated in both terms -- but R will take care of dropping it
automatically)

or better, retain all two-way interactions:

  y ~ (veg+Manage+age+under+wood)^2 + (1|Site) + (1|obs)

> 
> AIC      BIC   logLik deviance df.resid 568.0    604.1   -272.0
> 544.0      138
> 
> Scaled residuals: Min       1Q   Median       3Q      Max -0.76077
> -0.15913  0.00041  0.26754  0.70326
> 
> Random effects: Groups     Name        Variance Std.Dev. obs
> (Intercept) 11.954   3.458 Model$Site (Intercept)  7.405   2.721 
> Number of obs: 150, groups:  obs, 150; Model$Site, 10
> 
> Fixed effects: Estimate      d. Error z value Pr(>|z|) (Intercept)
> 16.67112     7.73426   2.155 0.0311 * Model$Manage1
> -5.72100    2.79907  -2.044   0.0410 * Model$age
> -2.10245     0.84062  -2.501 0.0124 * Model$veg
> -0.62276     0.29968  -2.078 0.0377 * Model$wood1
> 0.36500     0.41976   0.870 0.3846 Model$under1
> 3.69383     2.73372   1.351 0.1766 Model$Manage1:Model$veg
> 0.20478     0.09007   2.274   0.0230 * Model$age:Model$veg
> 0.08319    0.03326   2.502   0.0124 * Model$wood1:Model$under1
> 1.02751    0.44610   2.303   0.0213 * Model$veg:Model$under1
> -0.17567    0.08846  -1.986   0.0470 * --- Signif. codes:  0 ‘***’
> 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
> 
> Manage, wood and under are categorical with 2 levels each.
> 
> My question is: how can I find the real value of the estimates in
> the summary output (since I used the contrast)? Also, how it works
> with the interactions estimate?

   What do you mean by the "real value of the estimates"?  I think you
might want to take a look at the lsmeans or effects packages, or you
could use predict() to compute the expected outcome for some
particular combination of factors ...

  The question about contrasts/interpretation of parameters in linear
or generalized linear models is not really specific to mixed models.
Maybe take a look at Crawley's book, or Faraway's ...


> The books that I am reading don´t help much since they don´t show 
> interactions between variables. Usually, they show only one
> variable with more levels.  I Hope this is the right section to ask
> this question.
> 
> Thank you for who can help to understand this (maybe simple)
> problem.
> 
> Davide
> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iQEcBAEBAgAGBQJU/vVpAAoJEOCV5YRblxUHGMQH/3HMpXDO7Zouj4hBEPX9O9lm
1x6czZa+SQSCphKT01WevbT0wjLgFwaBhwfE/cFKbqr6YLsD5TsC3iG6FK5qzmob
Mmm1nWl1L04ZI7lPVkfH07UTDUNyQlToagPFuZYs/g+HlOx7vlQ44VgWIgTnGA3X
bmEtorIDF2n6PZvnMothJZrrri+0dOQBrFFDpdACpCR0zsGFQ4oJP5VJsW3mE40M
pqolU3CRMxO5Q9GzPO2I+P+IMb0rIj6Ke1ADdnkx1rt0QZcnndhFyA4wHk/rJwoQ
2NBXSgy+dBsX479CgCoS4aIJBTevUoHZ/S9SAAapoq8XAruEVwxPpYGxYqC1fMk=
=lwqn
-----END PGP SIGNATURE-----