FAQ/ranksum - CBU statistics Wiki
location: FAQ / ranksum

What is the expected total discrepancy score in a R choice task?

Suppose we have R possible choices and each of these is equally likely to be the true one.

If we consider a discrepancy as the difference between the true choice and the one given by a subject then

The expected total discrepancy of the absolute value of discrepancies equals

$$\sum_{k=1}^R (k=1)k $$, $$1 \leq k \leq R $$

with the average sum of the absolute values of discrepancies per rating equal to

$$\frac{\sum_{k=1}^R (k=1)k}{R}$$.

For example the table below gives all the abs(discrepancies) for the case where R = 4.

k=1

k=2

k=3

k=4

True Rank

0

1

2

3

1

1

0

1

2

2

2

1

0

1

3

3

2

1

0

4

Expected total score assuming random guesses at true rank

= 2(1+2+3)+2(1+1+2)= 20

= 1x2 + 2x3 + 3x4

= $$\sum_{k=1}^4 (k=1)k $$.

The average sum of abs(discrepancies) per rating = 20/4 = 5.

None: FAQ/ranksum (last edited 2013-03-08 10:17:10 by localhost)