Use Statistix software 8.1 (Analytical Software, USA).

Dear Simone,

The answer to your question largely depends on the scope of your work, often (but not always) stated as experimental hypothesis. Can you provide some further explanation about what do you want to do which those data?

Thank you both for your reply. The data is in the form of areas of different habitat types (n=7) inside 29 animals home ranges. I have three different sampling regimes so I would like to assess if a change in sampling regime creates differences in the amount of each habitat type. The raw data is measured in m2, contains zeros and is therefore not normally distributed. Otherwise I might have tried a nested ANOVA. Hope you can advise?

hi Simone,

you may want to try

1. categorical regression models for habitat selection analysis suggested by Kneib et al. (2007)

https://epub.ub.uni-muenchen.de/2052/1/tr001.pdf

and/or

2. ecological niche factor analysis suggested by Hirzel et al. (2002)

Hirzel, A. H., Hausser, J., Chessel, D. & Perrin, N. (2002) Ecological niche-factor analysis: How to compute habitat-suitability maps without absence data?. Ecology, 83, 2027–2036.

cheers, Kostas

Thank you Kostas.

Dear Simone,

I understood that your question is about the effect of sampling, let's stay the home ranges aside. Depending on how many zeros you have the following alternatives could work:

(1) Multivariate ANOVA on raw data or some measure of distance among samples. Input the 7 habitat types as 7 response variables and compile a 7 x n matrix with your n replicated scores (or the relative distances among pairs of samples, e.g. Anderson 2001). Consider sampling regime an experimental factor. Perform the analysis by randomization, so that you will get correct results even if data do not held the normality asumption (Manly 1997). Note that you will need to check for homogeneity of variances anyway.

(2) ANOVA (or regression) on some univariate index capturing the information that you want to focus on. For example: number of different habitat types, contagion index, etc...

(3) ANOVA (or regression) on the ranks of your original raw data. This is in practice a non parametric analysis and therefore you can also ignore non-normality in your data. Yet it is often advised that variances should be similar. This option would be close to option 1 by Kostas.

(4) One of the many non parametric methods, which in general are robust to large proportion of zero values in your data matrix. CAP would fit your needs (Anderson & Willis 2003). Available for free at

http://www.esapubs.org/archive/ecol/E084/011/suppl-1.htm

Hope this help. If home ranges and repeated measures are involved, other considerations should be made.

Many thanks Tomas for all your input, I will give your suggestions a try.

If you have false-color pictures of the habitat, you can use modeling and GIS to figure out the amounts of each subdivision of the habitat.

JMP (SAS)

CANOCO software is a great way to analyse multivariate data with illustrative visual output.  It basically uses Exploratory Data Analysis, but even provides capability for hypothesis testing.  You can input both categorical and continuous data, and there is no need for a normal distribution. 

Hello, Simone

Compositional analysis to study habitat selection has been used widely, but it has several important statistical snags: see Fattorini et al. (2014), which I have attached.

Fattorini et al. (2014) suggest an alternative method free of the important flaws of compositional analysis. I have used it on habitat selection of crested porcupines (Mori et al. 2014) which I have attached, too, just in case you wish to have a look to an example.

Have a good luck with your study!

SANDRO

Hello Sandro, Many thanks for those documents, they are very interesting and I will try applying your methods to my data.

Best wishes,

Simone

Sandro, did you carry out your sign tests in R? If so could you provide an example of the code you used please?

Simone,

Please, write directly to Prof Lorenzo Fattorini (he is on ResearchGate)

He will be able to answer (and guide) you properly, much better than me!

Cheers,

SANDRO

PS - You can write to him on my behalf, if you wish!

Many thanks Sandro, I have written to Prof. Fattorini.

best wishes,

Simone