haascn at dunx1.ocs.drexel.edu (Charles N Haas) wrote:
... text removed...
>If plate count data are in control, then the counts should be Poisson
>distributed. To compute the density, sum the total colonies by the total
>volumes plated.
>The log transformation is one way to approximately transform Poisson
>variates into normal variates. There is a long literature on this, although
>some of it is fairly old. You do not indicate what your overall objective
>is, so it is hard to go further. But if this thread continues, I will put
>in a few references of mine that I have written regarding statistical
>analysis of microbial data (mostly in the context of disinfection
>experiments, however it is equally applicable to growth curve
>experiments).
>
Many workers lose sight of the fact that as soon as colonies are
sufficiently dilute to be countable on plates, samples should be
statistically described by the Poisson distribution. If I remember
correctly, the Poisson distribution is actually the general case, and
at large numbers, it devolves into the Gaussian distribution.
In microbial quality control in the food and beverage industries, the
lack of understanding of these facts leads many to conclude that there
is a real difference between results of 2 and 10 colonies on a plate,
for example. Whereas the reality is that you can't reliably replicate
such counts at all. In such a case, all you can reasonably say is
that the colony density is less than some value. But I guess this is
another topic.
I would appreciate the references which you mention. Thanks.