Kendall's Tau Distribution

 

Menu location: Analysis_Distributions_Kendall's tau.

 

Given a value for the test statistic (S) associated with Kendall's tau (t) this function calculates the probability of obtaining a value greater than or equal to S for a given sample size.

 

Consider two samples, x and y, each of size n. The total number of possible pairings of x with y observations is n(n-1)/2. Now consider ordering the pairs by the x values and then by the y values. If x3 > y3 when ordered on both x and y then the third pair is concordant, otherwise the third pair is discordant. S is the difference between the number of concordant (ordered in the same way, nc) and discordant (ordered differently, nd) pairs.

 

Tau (τ) is related to S by:

 

If there are tied (same value) observations then τb is used:

- where ti is the number of observations tied at a particular rank of x and u is the number tied at a rank of y. When there are no ties τb = τ.

 

This function does not calculate probabilities for τb.

 

Technical Validation

Probabilities are calculated by summation across all permutations when n ≤ 50 and by an Edgeworth series approximation when n > 50 (Best and Gipps, 1974). The two samples are assumed to have been ranked without ties.

 

The inverse is calculated by finding the largest value of S that gives a calculated upper tail probability (using the method above) closest to but not less than the P value entered. Please note that the results may differ slightly from tables in textbooks because StatsDirect calculates the inverse of Kendall's statistic more accurately than the routines used to calculate Best's widely quoted 1974 table.

 

Illustration

Consider the ten pairs of rankings in the Kendall rank correlation example, a tutor's rankings of ten students for suitability to their career and for knowledge of psychology. There are 34 concordant and 11 discordant pairs, so S = 23 and τ = 23/45 = 0.511111. To calculate the probability of an S at least this large when the two rankings are independent, select Kendall's tau from the Distributions section of the Analysis menu, enter 10 as the sample size and 23 as S (or enter 0.511111 as tau, from which StatsDirect calculates S) and click Calculate. StatsDirect shows the upper tail probability and reports it as:

 

P(Kendall's T 23, n 10) = 0.023311287477954 upper tail

 

Of the 10! = 3,628,800 equally likely orderings of one ranking against the other, 84,592 give an S of 23 or more. This is the upper side P of the exact test in the Kendall rank correlation report (P = 0.0233); the distribution of S is symmetrical about zero, so the two sided P is twice it (P = 0.0466).

 

For the inverse, enter 10 as the sample size and 0.05 as the upper tail P and click Invert. StatsDirect finds the largest S whose upper tail probability is not less than 0.05, shows it with its tau (S = 19, τ = 0.422222 to six places) and reports the probability that it cuts off:

 

Kendall's T (upper tail P 0.05, n 10) = 0.054156746031746

 

So with ten pairs an S of 19 does not quite reach the one sided 5% level. The next possible value, S = 21 (τ = 0.466667), is the smallest that does, with an upper tail probability of 0.036275077160494: with so few pairs the attainable significance levels are coarse.

 

StatsDirect displays the probabilities in this illustration to 15 decimal places; the R code below reproduces them.