Saturday, January 12, 2008, 8:38:33 PM, Stephen Cole wrote:
SC> I have a data set that is infested by a plague of zeros that is causing me
SC> to violate all assumptions of classic parametric testing. ...

Stephen,

if you're measuring density, abundance or such and you end up with
many zeros and some non-zero values, then you should turn your
analysis into something that is aimed at testing the independence of
the "zero" response relative to some conditions, rather than testing
the quantitative differences including zeros in the computations.

For instance, assuming that you're interested in measuring changes in
density of a copepod species in the upper layer of the water column
between day and night, then the you can test whether differences
between mean values (or median, in case you go for a non-parametric
approch) during day and night are equal. That's more or less the
rationale of the experimental design you mentioned, just a tad
simpler.

Now, imagine you realize that very often that species is absent (you
might wonder whether you're collecting samples that are not large
enough, or whether it is only a night vs day issue, but that's another
story...): this means having too many zeros, that in turn means that
info about variability is not reliable, so you already know that the
methods you were planning to use (both parametric and non-parametric)
won't work.

In this case the real issue is changed: now the problem is "Are zero
densities more frequent during the night?", i.e. H0: non-zero
densities are independent of the time (day or night). Now you just
need to build a 2x2 contingency table and a chi-square test will
provide the answer to your question. Of course, you can also design
something more complex, like a n-way design, or test whether a given
frequency (e.g. zero density during the night) in your contingency
table is significantly larger than the expected value or not. That's a
sort of a posteriori test you can add to your basic procedure

I hope I've been able to explain what I'm suggesting. As you probably
guessed, English is not my first language...

All the best,

Michele


--------------------------------
Michele Scardi
Associate Professor of Ecology

Department of Biology
University of Rome "Tor Vergata"
Via della Ricerca Scientifica
00133 Roma
Italy

http://www.mare-net.com/mscardi
--------------------------------

Reply via email to