[R-sig-ME] Reg. interpretation of parameter CIs from lmer()

Geetha Ramaswami geetha_r at ces.iisc.ernet.in
Wed Feb 15 04:39:17 CET 2012


Dear All,

 I am working with an ecological data set wherein I am trying to compare
 the growth rate of seedlings in plots where an invasive species is present
 or absent. Repeated measures on seedlings were made every two months
 across 40 plots of which 20 had the invasive species while the remaining
 20 did not. Seedling growth is measured as log(proportion increment in
 height) per month. I am also interested in looking at how rainfall
 received between two consecutive growth measurements and seedling habitat
 preferences affect growth. I came up with the following mixed effects
 model

 growth ~ invasive density (2 levels) + seedling habitat preference (3
 levels) + rainfall (mm) + all two-way interactions + random intercept on
 repeatedly measured plots

 The residuals on this model are highly overdispersed and do not meet the
 normality criteria, so i decided to use nonparametric bootstrapping
 (refitting the model with 10000 random subsets of data) to obtain 95% CI
 on all the fixed effects parameters estimated (I assumed that the CIs
 non-overlapping with zero indicated significant fixed effects). Apart from
 the 'intercept', the 'rain' term and the 'invasive density' term, 95% Cis
 of all other parameters included zero. I am interested in graphically
 representing only these effects. Since the normality assumption of
 residuals is not satisfied, is it appropriate to simplify the model using
 anova (with REML = F)? Or can I create a new, simpler model with just the
 terms of interest, generate 95% CI for these parameters and use these for
 graphical representation?

 Thank you in advance for your help.
 Geetha

 Geetha Ramaswami
 PhD Student
 Centre for Ecological Sciences,
 Indian Institute of Science,
 Bangalore 560012,
 India


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.




More information about the R-sig-mixed-models mailing list