At count rates near the instruments MDA, a normal distribution will usually not be present. In addition, the effect of background counts can have a large influence on peak shape for low-count ROI's. Software driven background subtraction functions may help, but can also produce false negatives. As the detector's efficiency will vary with energy, each ROI will have a different characteristic background, and therefore a different acceptable number of counts to pass the t-test. In practice, I would avoid trying to apply a normal distribution test to any peak that is less than 3 or 5 times the MDA.
Doug Taylor
WPRAP Laboratory Manager
Shaw Environmental & Infrastructure, Inc.
(513)-648-4355
-----Original Message-----
From: William Prestwich
Sent: Tuesday, October 22, 2002 9:05 AM
Subject: Re: Acceptance of Data Results
Since no one else has replied I thought I would stick my neck out. The
t-test is, I believe based upon the data being normally distributed.
Counting data, which usually follow Poisson statistics, may not cooperate
if the number of counts involved is not large enough to make the normal
distribution approximation valid.