Can someone explain to me how to calculate the discordant and condordant pairs? I don't know how the C(++) or D(-+) etc are really derived. Sweeting doesn't show it. page 132
Don't you guys think there is an error in ActEd notes on the empirical calculation of C and D? page 15 of chapter 11. Example is ok until I get to the calculation of (64 - 72) - this gives me a negative sign (-) (85 - 45) - this gives me a positve sign (+) so the answer should be D(-,+) instead of C(+,+).
Let X and Y be random variables. Then for a sample x, y of X and Y the paid are concordant is x-E(X) >=0 and y-E(Y)>=0 or x-E(X) <=0 and y-E(Y)<=0 and discordant otherwise.
Thanks, I understand the definition, but can you refer to my example in the post? I think the ActEd contains an error....
Actually, it would be helpful if someone could show a numerical example, as I can't work it out from Sweeting page 132.
The pairs you are quoting are t=4 vs t=3 hence the bottom right cell in the table which is indeed D(-,+)?
A concordant pair is one where the pairs are in the same order relative to each other, so if we consider the ranks: X Y 1 4 2 6 Then for X: 1-2=-1 and for Y: 4-6=-2 so we have (-,-) and a concordant pair If we have: X Y 3 4 2 6 Then for X: 3-2=+1 and for Y: 4-6=-2 so we have (+,-) and a discordant pair.
This example is very simple. I just don't agree with the ActEd calculation in ActEd notes. With the example in sweeting, I don't know how he is getting his C and D. But I googled another method of calculation where Kendall's Tau = (C-D)/(C+D) and the calculation is by re-ranking one variable in ascending or descending order. AcdEd excercises also show this method. Is that acceptable for an exam?
It would be fine to sort first (as this simplifies the process somewhat!) I hope I dealt with that one in an earlier post? Let me know if you still disagree. (p132) He is using the same method. eg t=3 vs t=2 t X Y 2 95 25 3 15 10 15-95 = neg 10-25 = neg hence C(-,-) eg t=4 vs t=1 t X Y 1 10 20 4 35 15 35-10 = pos 15-20 = neg hence D(+,-)