Bio Techniques Laboratories <cobactin at halcyon.com> wrote:
>> In recent years I have found two editorials on the "correct" method to
> statistically analyze plate count data. The first one said that the mean
> and the std. dev. should be determined from the arithmetic values of the
> data. The second one said the data should be transformed to logs and
> then the mean and std. dev. should be calculated and then retransformed
> to scientific notation. Are there any more opinions on how to deal with
> plate count data? If anyone could suggest a text or article that deals
> with statistics for the analysis of plate count data, I would appreciate it.
> Thank you!
>> Richard
You must use transformed data, usually log transformed. I learned
this the hard way. I once sent a very stupid letter to the Editor
of ASM NEWS on this subject and did not know what I was talking about.
Others responded to my letter and set me straight. I had egg all over
my face. But I checked with a real statistician and found out
that I was indeed wrong.
Here is the deal. Plate count data is usually not normally distributed,
especially if some of it comes from different dilution levels.
Parametric statistical tests require that the data be normally
distributed. Therefore it is customary to use the log transformation.
For Discussion, see my uninformed ignorant foot-in-mouth letter to
ASM News Ziprin, R.L. (1985). "Data log jam. ASM News 51(6):262-263."
And also look at the letters to the editor in the next two issues
where I am publically torn apart.
We learn from our mistakes. If I can be of further help,
e-mail to ziprin at usda.tamu.edu or tel. 409-260-9302.
