[R-sig-ME] Analysis of unbalanced design data from a before-during-after control-impact study of breeding birds and railway construction
Thierry.ONKELINX at inbo.be
Mon Jan 14 10:30:08 CET 2013
See my inline comments.
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
+ 32 2 525 02 51
+ 32 54 43 61 85
Thierry.Onkelinx op inbo.be
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
Van: r-sig-mixed-models-bounces op r-project.org [mailto:r-sig-mixed-models-bounces op r-project.org] Namens Adriaan De Jong
Verzonden: maandag 14 januari 2013 10:04
Aan: r-sig-mixed-models op r-project.org
CC: nils.bunnefeld op stir.ac.uk
Onderwerp: [R-sig-ME] Analysis of unbalanced design data from a before-during-after control-impact study of breeding birds and railway construction
· What are the effects of the transformation from numbers of territories to densities (per site) on the analyses and the conclusions? (the sites vary a lot in size!)
Use a generalised linear mixed model with poisson or negative binomial family on the number of territories and use the log acreage as an offset factor. This is equivalent to modeling the density but takes into account the fact that the number of territories is discrete.
· Did I choose the right model approach, or are there better ways to do it? (most likely there are, probably in the realm of medical statistics)
I would refrain from (stepwise) model selection. Just construct the models that you need to test your hypotheses. E.g.
M0 <- glmer(resp ~ treatment * year + (year|site), data=bot, family = poisson)
M1 <- glmer(resp~ year + (year|site), data=bot, family = poisson)
· Should sites where the species never occurred during the study period be excluded from the model selection process? (non-existing birds cannot be affected)
Yes. Including those sites will a) dampen trends and b) cause numerical problems in glmm models.
Still check for zero inflation after removing the sites.
· Did I use the right random effect alternatives?
You used a random slope along year per site. Not year nested in site.
· Should/can I include a treatment-duration trend in the random effect part (time nested within treatment within site)?
You probably lack data to do that. Note that adding only treatment as a random 'slope' requires 6 parameters to be estimated instead of 1 for only the random intercept. Adding both treatment and year as a random slope is problematic as they are collinear.
· Does the absence of treatment effect in the most parsimonious model mean that there was no significant effect of railway construction? (again: all impact sites were included)
See question two.
· Is there a multivariate (multi-species) alternative?
Thanks a lot in advance for giving this a thought.
Adriaan “Adjan” de Jong
Wildlife, Fish, and Environmental Studies Swedish University of Agricultural Sciences
PS. We will continue the studies in 2013-2015 to monitor the possible effects of train traffic. Train traffic started with a limited timetable on parts of the track in autumn 2010, but full scale traffic has only recently begun. After that we will have four treatment classes.
[[alternative HTML version deleted]]
* * * * * * * * * * * * * D I S C L A I M E R * * * * * * * * * * * * *
Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is door een geldig ondertekend document.
The views expressed in this message and any annex are purely those of the writer and may not be regarded as stating an official position of INBO, as long as the message is not confirmed by a duly signed document.
More information about the R-sig-mixed-models