Statistics question--a real hypothetical.

[_MCB]

Suppose you are asking yourself if two methods of authorship attribution arrive at reasonably consistent results. You decided to use word count, since one uses chapters, and the other uses pages from an old edition. You have sorted the selections into four files, and run a chi-square. The results are very satisfying. However, you want to run a fischer's exact test, but dealing with astronomically large numbers just isn't your cup of tea. Would it be legit to use the cell proportions for this? Seems logical to me.

I am sure that probability will be like a grain of sand in a ton, so I am not really worried about the results. However, I expect some will attack anything they imagine could damage the case.

[_MCB]

I am reviewing my four files, done the not spalding file and the yes jockers file, results are consistent with first running of data.

I will post numbers when I am done.

C'mon, Tarsky, fess up, you can tell me.

[_MCB]

Both say yes file, word- count is 51,802, both say no is 171,948. b says yes, j says no is 17,899, b says no, j says yes is 52,158. I won't give my value for chisquare, just so I know that any further stats analyist knows what he/she is doing.

Now, I did it three times, if there are any errors, they are random, and the errors of a mere human being. T'was a beastly job.

Dale, you are much too conservative. But that is OK. At least we know that the 51,802 are most probably the work of Spalding.

[_MCB]

Here is a discussion of chi-square:
http://www.uwsp.edu/psych/stat/14/nonparm.htm

It is probably the simplest statistical method for testing hypotheses.

If anyone could offer a different word-count, I would be interested.

[_MCB]

bump