Dear all,
I am facing a problem that seems to me more tricky now that it did at first
sight.
I have a collection of data which consist in frequency distributions: 6
patches had been proposed to female insects (for oviposition), 3 of them
corresponding to one treatment (A), the other 3 to another treatment (B).
The results are the number of patches that have been chosen for oviposition
by each female (0 to 3 patches can be used for oviposition, for each
treatment). Thus, the experiment is replicated (each female's choice is a
replication).
I am interested in testing if females do prefer one treatment to the other
(if they oviposit preferently on one sort of patch), i.e. testing the
goodness of fit of my data to an equal ratio of oviposition on patches A and B.
I intended to use the replicated goodness-of-fit test (G-statistic)
described in Sokal & Rohlf (1981): this test seemed accurate for such
data...
But a first problem is that some frequencies in my data set are = 0... I
wonder if an x->( x + 1) transformation could be reasonnably performed to
permit the taking of ln(X) in calculations.
Moreover, low frequencies (less than 5) are usually reported as a problem
for G-tests, and because my females had only the choice to oviposit on 6
patches (3 for treatment A, 3 for treatment B), I have indeed low frequencies!
Do you know another way to perform such an analysis, more accurate in my
case (or do you thing a G-test could nevertheless be used)?
I hope my explanations were comprehensible, and if it isn't the case,
don't
hesitate to ask for more.
Thanks for your reply
Kind regards,
Karine
