I have 75 archaeological sites (not identical in size or shape or number of species but all the same/similar vegetation type). An Excel spreadsheet contains presence and absence data for each site in the form of botanical species. Can I use Jaccard or Sorensen's index to compare each site with a Ghost site which is artificially populated with all the possible important taxa (arrived at by regression) - then arrive at a % similarity across the 75 sites? OR compare two groups from the 75 sites separately as above and then find the possible difference between the two groups from the % similarity for each group? OR is there a better way of doing this!?
You could calculate a Bray-Curtis similarity index matrix. That will give you the percent similarity between each pair of sites. That matrix can be used as a basis for a number of subsequent analyses. You can then visualize the data with a non-metric multidimensional scaling plot (n-MDS). You can also run an analysis of similarity (ANOSIM) or a permutational ANOVA (PERMANOVA), both of which are mutivariate analogs of ANOVAs.
A lot depends on what question you want to ask with this data and how it was collected. Did you do multiple quadrats and identify all the species in each? If that's the case (i.e., if you have multiple samples from each site) then you have a lot of options as far as comparing sites. If you did a complete survey of each site and IDed all species, that could be a bit of a problem as sites with a smaller area are likely to have fewer species just by chance. If that's the case you may need to account for area sampled.
Thanks William. I have used the Sorensen's similarity measure because I can use it to compare each site with one master "ghost" (or imaginary) site populated with the taxa that are the important ones in my study! If one uses other methods then each pair of comparisons do not necessarily compare the same set of taxa throughout, do they? I am afraid I am not a stats boffin, but I have struggled for some time to get my mind around them, with the result that this seems like the best method for my purposes. What I want to know is - has anyone else used the same method, and if so with good or bad results (and acceptance) and if not, why not! The question I am asking is quite simple, how much similarity is there between sites regarding a set of 50 important taxa hared by the sites. There are another 15 taxa of lesser importance which are sometimes shared as well and they are also put into the ghost site. The sites are widely scattered in about 6 Biomes and many different vegetation types, the fact that they are so similar tends to suggest a possible anthropogenic process. But I have to show statistically first that in fact they are similar!! I did a complete survey fo each site and Id'd all the species present - so that makes it difficult as you say! However the size of the site does not make much difference to the presence and absence data regarding the species which are most common - even the smallest site may have up to 15 important species present.
The Sorensen's and B-C similarities are actually the same if used on presence-absence data. B-C can be used on abundance/density data as well so it’s just a little more general. Anyway, as long as you have a complete dataset, i.e. you have a 0 or 1 for every species at every site you are comparing the same set of taxa. So no need of ghosts. It is perfectly valid to cull the rare species if your question just concerns the common ones. What you do, is get your similarity matrix and run an ANOSIM with Biome as the factor. If the Biomes are as similar as you hypothesize you should get a low Global R value and a high p value. p > 0.05 will allow you to say there is no statistical difference among the sites. You can follow up with a SIMPER (similarity percentage) analysis again with Biome as a factor. This will give you the average similarity within each of the Biomes, the average dissimilarity between each pair of Biomes and a list of the species that are most important in contributing to the similarity/dissimilarity. What you should see (if the Biome are all similar) is the average similarity within each Biome is about the same as 1-dissimlarity between the Biomes. That is, if the sites in Biome 1 are on average 60% similar and the average dissimilarity with Biome 2 is 40% then the sites in Biome 1 are as similar to the sites in Biome 2 as they are to each other. These analyses should allow you to demonstrate clearly how similar the different Biomes are.
I noticed you hail from South Africa and that was making me a bit homesick. I grew up in Botswana (and somehow ended up in Alaska of all places) and we took plenty of holidays and trips down to South Africa as well. I particularly miss Southern Africa this time of year when the days are so short and we don’t get much sun.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis