I guess it is a quite simple task, but somehow I can't find the correct test. I am looking for a test/model to show whether or not the relative abundance of a species changed significantly over time.
The situation:
I have an area in which three categories/species/factors are present. The abundance sums up to 100%. I want to know if the share of e.g. one of the categories (better: the composition of the three factors) changed significantly over a time-period of seven years. I want to keep it simple, so I only have the time and the relative abundance included, other factors (like salinity or temperature) are in this step not needed.
Any ideas? A mutlinomial linear regression was suggested to me, but I am not sure if this will fit.
Dear Sven,
For species relative abundance, i suggest you to use PAST software, which is easy to use and has variety of functions. It will helps you better to calculate the statistical data in all ways. Best regards
Sven, regression is a trend determining tool, also a tool for forecasting, it is descriptive and is not a test.
Change, in simple terms can be determined by getting the difference of the values divided by the difference in time, (x2-x1)/y2-y1).
As to testing the difference, I would suggest that you review first your design to find out if you have probability samples, or if your data are fit for inferential analysis.
You could use multinomial ('logistic') regression, modelling the likelihood that each randomly drawn specimen is from species 'A', 'B', or 'C'. Then, you could extend to the model to see whether the likelihood of either changed over time.
Hope this is helpful,
Rense
Alternatively, I realise now, you could calculate an index of diversity - such as the Herfindahl index: http://en.wikipedia.org/wiki/Herfindahl_index
Not sure, though, if you could do so while controlling for other variables, but at the very least it could provide nice descriptives of change in composition over time.
Hi Sven, can you say more about the data you have? What units of measurement are you using for your variables? What software do you have the data saved onto?
Jackie
Hi Sven, In my opinion the most intuitive and informative method to assess changes in population trends in the Power Analysis. This analysis will take into account the uncertainty in your trend to find out if you can have a significant trend or not. See one of the paper a co-authored(Jewell et al 2012, MEPS) for further details. Hope that may help.
As I understand, you want to assess species composition changes over time, rather than (absolute) changes in abundance. Therefore, I agree with the view that multinomial regression is a logical way to go. You need to decide, however, whether time is going to be continuous or discrete. If an abrupt shift occured along the seven years period, a continuous tratment of time is not making much sense. Under the later scenario, you can formulate and compare multiple hipotheses (models) about the year when the shift occured and then compare these competing models using some information criterion, such as Akaike's. If you are an R user, check the multinom() function in the nnet package.
Thanks for the helpfull replies (and sorry for the indeed very late answer, time really flies by).
I tried several methods and I will do as Edwin suggested, it seem the ost logical way to go. Also, the PAST program looks like something I have to dive in deeper :)
Hi Sven, Have you tried Principal Component Analysis? If there were changes in abundance over time it may show up in the groups.
Good Luck,
Pete
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R Statistical Package