I'm not finding it easy to find a simple way of interpreting/explaining
zero counts.
The simplest approach would be to use Confidence intervals. For example,
you could use Wald confidence intervals for analysing bats flying along
a linear corridor. Such as...
No records were made over five nights suggesting that the proportion of
times you could record bats over a longer sampling period would be
between 0 and 0.4. (90% Wald CIs)
This in contrast to another linear corridor where bats were recorded on
all 5 nights out of 5, giving a 90% Wald CI of between 0.60 and 1.
"very loosely" interpreting the CIs, this gives the impression that with
a longer term study, it is unlikely that bats would be recorded using
linear corridor 1 more than 40% of the nights surveyed (but it could be
0% of the nights), and with Linear corridor 2 it would be unlikely that
bats would be recorded less than 60% of the nights surveyed. Which does
at least give some idea of what a zero count over 5 sample nights might
mean, and allows a simple comparison of relative bat use between the two
corridors.
I know this can be done in a more satisfying way with a bayesian
approach, but I am looking for something simple and easy to apply,
primarily to avoid the often used statement " no records were made of
species X, but this doesn't mean they weren't present."
The important thing is to try and give a measure of what a zero count
might mean, which this on the face of it, this does, but I just feel
uncomfortable with it, because the proper interpretation of the CIs
seems difficult to make practical sense.
It seems such a common problem that I am amazed I cannot find any
"standard" approaches.
Can anyone give me any pointers to where I can follow this up and see
what others are doing.
Many thanks,
Graham
