I am comparing habitat availability vs. use by a bat species using the R compana() function (package adehabitat) which uses Aebischer et al. (1993) method. It is a 3rd order habitat selection analysis, so the availability of habitats is different for each bat.
The compana randomisation test resulted in a nice habitat ranking plus a table with the significance of paiwise comparisons. The only thing I need more, are log-ratio differences or Manly selection ratio's (considering that these are given in almost every paper on habitat selection). Thus, I tried the Wi function that computes Manly's selection ratios. The problem is that the resulting ranking of habitats is very different from the one produced by compana, and the compana result seems to fit much better to the raw data. Apparently the results of these methods cannot be combined.
So now I am a bit lost. I hope someone is willing to explain
- Why these two methods (compana and Manly selection ratio's) produce such different habitat rankings and how they relate to each other.
- If I can obtain log-ratio differences within the compana function.
...Or any other advice on how to proceed.
Confusion like this is very common in my experience Simone – it usually occurs because the underlying criteria for the different statistical analyses are different. Although it's good to confirm results with multiple analyses, sometimes it's good to just use one reliable test and be confident of your results! I would use compana and be happy with that so there is no confusion. Or if you use both, interpret the results in light of the criteria :-)
You could also use a simple analysis, like Neu et al/Cherry Bailey or Jacobs.Indices just to get an impression which method does fit best.
I´ve build a simple template in EXCEL which you could use for that, if you would like to try.
Olli
The differences could be caused by zeros in the availability matrix in the compositional analysis. That critical issue is why I've switched to resource selection functions with a GLMM to allow for the individual random slopes of each individual.
http://www.bioone.org/doi/abs/10.2193/0022-541X(2004)068%5B0206:COTIER%5D2.0.CO%3B2
http://www.bioone.org/doi/abs/10.2193/2006-072?journalCode=wild
Thank you Andrew and thank you for citing those two papers, could you cite any examples of resource selection functions using a GLMM please?
Hi Simone
please consider also this alternative to compositional analysis, which overcomes several methodological problems of CA and it is quite easy to use:
Fattorini, L., Pisani, C., Riga, F., & Zaccaroni, M. (2016). The R package “phuassess” for assessing habitat selection using permutation-based combination of sign tests. Mammalian Biology-Zeitschrift für Säugetierkunde.
http://www.sciencedirect.com/science/article/pii/S161650471630204X
In my opinion, by using a GLMM you will overcome many problems originated by other methods. Additionally, GLMMs are much more flexible than other methods.
Example with radio-tagged individuals:
Weisshaupt, N., R. Arlettaz, T.S. Reichlin, A. Tagmann-Ioset & M. Schaub. 2011. Habitat selection by foraging Wrynecks Jynx torquilla during the breeding season: identifying the optimal habitat profile. Bird Study 58: 111-119.
Of course GLMM is a very nice oppurtunity!
In some cases this will not work. Consider also using GAM!
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology