Re: Set Repeatability

ravenous.wolves_at_gmail.com
Date: 02/03/05 

Date: 3 Feb 2005 10:59:07 -0800

Any idea how to apply this same concept for correlation between two
tools? For example, suppose on T1 and T2 I get the following (column
totals / nRuns) for F1/F2.

F1:
1 = .9
2 = 1.0
3 = .1
4 = .2
5 = .2
6 = 1.0

F2:
1 = .9
2 = .9
3 = .9
4 = .2
5 = .8
7 = .1

I was planning to take the 1.0 - the absolute difference for each
discrete value and average them, but after doing repeatability wrong, I
thought I should check whether this is statistically sound.

A1 = 1.0 - ABS(0.9 - 0.9) = 1.0
A2 = 1.0 - ABS(1.0 - 0.9) = 0.9
A3 = 1.0 - ABS(0.1 - 0.9) = 0.2
A4 = 1.0 - ABS(0.2 - 0.2) = 1.0
A5 = 1.0 - ABS(0.2 - 0.8) = 0.4
A6 = 1.0 - ABS(1.0 - 0.0) = 0.0
A7 = 1.0 - ABS(0.0 - 0.1) = 0.9

A-Avg = 0.628

However, this doesn't look right, as the # of low-frequency values
which appear on only one tool increase (such as A7), the correlation
score trends upwards since A7 is judged as having high correlation
since both tools tended to not find it. This is the same problem my
earlier repeatibility formula had :(

Thanks,
-ken