bipartite (version 2.16)

null.t.test: Compares observed pattern to random webs.

Description

A little null-model function to check, if the observed values actually are much different to what one would expect under random numbers given the observed row and column totals (i.e.~information in the structure of the web, not only in its species' abundances). Random matrices are based on the function r2dtable. The test itself is a t-test (with all its assumptions).

Usage

null.t.test(web, N = 30, ...)

Arguments

web

A matrix representing the interactions observed between higher trophic level species (columns) and lower trophic level species (rows).

N

Number of null models to be produced; see ‘Note’ below!

…

Optional parameters to be passed on to the functions networklevel and t.test.

Value

Returns a table with one row per index, and columns giving

obs

observed value

null mean

mean null model value

lower CI

lower 95% confidence interval (or whatever level is specified in the function's call)

upper CI

upper 95% confidence interval (or whatever level is specified in the function's call)

t

t-statistic

P

P-value of t statistic

Details

This is only a very rough null-model test. There are various reasons why one may consider r2dtable as an incorrect way to construct null models (e.g.~because it yields very different connectance values compared to the original). It is merely used here to indicate into which direction a proper development of null models may start off. Also, if the distribution of null models is very skewed, a t-test is obviously not the test of choice.

Finally, not all indices will be reasonably testable (e.g.~number of species is fixed), or are returned by the function networklevel in a form that null.t.test can make use of (e.g.~degree distribution fits).

Examples

# NOT RUN {
data(mosquin1967)
null.t.test(mosquin1967, index=c("generality", "vulnerability",
    "cluster coefficient", "H2", "ISA", "SA"), nrep=2, N=10)
# }