With Respect
Thanks for all the replies I have received about the gamma distribution.
I am also posting this to the STAT-L list because I want to start a discussion
The question I find of interest is most squarely raised by Blaise F. Egan's reply that there is an infinite set of distributuions and an infinites subset matching my data, and that I need to select a few "for my purposes" to test. "Which?" is a form of the "frame problem" of the cognitve scientists or what philosophers know as Pierce's abduction. What has been written on this particular form of frame problem, and how one chooses distributions to test? Simplicity is an obvious criterion but how should it be measured? and applied? How does the history of how distributions have been described and used show in this light? How much weight should be given to, and how much is known about, what processes produce different distributions of data in the real world? I don't suppose Helen MacGillivray's work includes a survey of how often distributions are found to be different from the normal by different amounts.
Yours Sincerely,
Alan E. Dunne
|