Skip to content

Thresholds of Synchrony in R

2 messages · Tania Bird, Botta-Dukát Zoltán

Original
Tania Bird
# 
Hi all,
Apologies if this is more a stats question than an R question but I do need
help with R for the solution.

I'm calculating the degree of synchrony of population fluctuations through
time in a community, based on Loreau & Mazencourt 2008 paper.
Loreau, Michel, and Claire de Mazancourt. (2008) "Species synchrony and its
drivers: Neutral and nonneutral community dynamics in fluctuating
environments." The American Naturalist 172, no. 2: E48-66.
doi:10.1086/589746.

I am using this code:

dat = cbind(sp1 = rnorm(100, 10, 2), sp2 = rnorm(100, 10, 2))
    #Two species with random independent abundance sampled 100 times.
V = var(dat)   # variance-covariance matrix for all species
    # calculate synchrony index from covariance matrix
synchrony = function(V) {
d = sqrt(diag(V))
sum(V) /sum(d%*%t(d))
}

S = synchrony(V)

S runs from 0 (total Asynchrony) to 1 (total Synchrony) with 0.5 = random
or no synchrony as described by the authors.

I am wondering if there is a to calculate a threshold or value of S at
which I can say that there is "significant Asynchrony in this community" or
 significant Synchrony in this community". For example is 0.3 asynchronous
or not really ?

Should this threshold based on the data range or the variation itself?
In which case how would I code this calculation?

Or should I just say that anything <0.25 is significantly/very
Asynchronous?
In which case how can I generate a column to say
 if S <0.25 its asynchronous or if its >0.75 its synchronous?

Thanks for your feedback
Tania


Tania Bird MSc
PhD Student,
Geo Ecology Lab
Ben Gurion University
Botta-Dukát Zoltán
# 
Dear Tania,

I suspect that there is no general threshold.
You can check significance of departure from 0.5 by randomization test:
1)  randomize your data,
2) calculate synchrony for the randomized data
3) repeat step 1 and 2 many times (at least 999)
4) p-value = min(# cases where observed S is higher than random; # cases 
where observed S is lower than random)/(number of randoms +1)

There are (at least) two ways of randomization:
- simply shuffling values within one species' time series - it is 
simple, but destroys the temporal autocorrelation within time series
- shifting values within one one species' time series (as it would be 
circular data) - it preserves the autocorrelation, so I would use it.

Best wishes,

Zoltan

2017.08.02. 18:58 keltez?ssel, Tania Bird ?rta: