Hello,
I'm analyzing bray-curtis values over time. I would like to quantify if 1) there is a rapid or gradual change in bray curtis values, and 2) how long this change occurs. As a hypothetical example, Fig. 1a, a rapid increase until sampling week four. Is there a trend analysis I could run that would allow me to use terminology such as 'gradual' or 'rapid' based on slope? Or something along those lines.
I primarily use R.
Thank you.
Josh
Did you already think about using regression analysis with time as explanatory variable? Than you could use the slope as an indicator of increase (or decrease). Of course normally you would transform your data in order to linearize your relationship (between time and BC). But then you would capture the general trend (over 19 weeks) in your model and not the different parts in your trend (as you describe in fig. 1a
Is it necessary to quantify the changes? Can't you just qualitatively describe the changes?
Koenraad Van Meerbeek I think that may be the best bet. Thanks for your insite. I think what I'm going to do is a segmented regression. Calculate the breakpoint, and get a best fit line for which ever weeks increase until the breakpoint. I found a function called 'segmented' in R that does this.
As another option you could also look at using the Mann Kendall trend test. This test does not assume that your data is normally distributed, so you won't need to transform them. Check out the 'trend' R package; it includes the Mann Kendall test as well as other trend tests and tests for change-point detection. From this package you can also use Sen's Slope to quantify the magnitude of change in your Bray-Curtis values. I have not used all of the tests from the 'trend' package, but you might find something in there that will allow you to perform a non-parametric alternative to a segmented regression. Here is the vignette for the 'trend' package: https://rdrr.io/cran/trend/f/inst/doc/trend.pdf
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling