Thread: rank correlation

why while calculating rank correlation coefficient for tied ranks we take the average of the tied ranks to rank the same observations

for eg.
X : 29 30 25 30 30 45
rank : 5 3 6 3 3 1

why should we give the rank 3 (2+3+4)/3 for all the three 30's instead of

2, 3 & 4. Can any one explain me the logic

Originally Posted by Zandra

why while calculating rank correlation coefficient for tied ranks we take the average of the tied ranks to rank the same observations

for eg.
X : 29 30 25 30 30 45
rank : 5 3 6 3 3 1

why should we give the rank 3 (2+3+4)/3 for all the three 30's instead of

2, 3 & 4. Can any one explain me the logic

If you have a sorted list you can connect each value with an index. If there are equal values they should have equal indices (=ranks) too. Therefore you take the average of those ranks which belong to equal values. Thus it could happen that you have a rational number as rank (which is, at first sight, not very plausible)

Btw there is a slight mistake in your calculation. (see attachment)