Kendall's W
Kendall's concordance analysis (after Maurice George Kendall ) is a non-parametric statistical method for quantifying the agreement between multiple raters ( raters ). This is Kendall's concordance coefficient W is an alternative to
- Kappa statistics - which are designed for nominal scaled data - and
- Rank correlation coefficients for ordinal data (such as Spearman's and Kendall's ) - which are mainly intended for two judge -
Dar.
The concordance coefficient W is similar to Cronbach's alpha to determine the reliability, for example, a test procedure. It takes values between 0 and 1.
Formula
If appraisers bring the cases ( = objects to observe people, characteristics) in a rank order, each case receives from each judge a rank order; the sum of all assigned rankings for a case is then:
If a judge a case no clear rank order (1,2,3, ... N) assigns, but, for example, several cases have to share a rank order, one speaks of "rank binding ". The number of cases that each share a specific rank place in a judge, called rank bond length.
Of course, multiple rank bonds can also occur with a judge when cases are judged equal. The total number of bonds is at a rank raters:
Kendall's W is it calculated as follows:
In which
And
W is the Friedman - coefficient (English article) as well as the rank correlation coefficient of Spearman directly related:
And
,
Where the average of all rank correlations between the possible combinations representing each of two raters.
Literature and sources
- Bortz, J., Lienert, GA & Boehnke, K. ( 1990): Distribution -free methods in biostatistics. Chap. 9 Berlin: Springer.
- MG Kendall, Babington Smith, B.: The problem of m rankings. In: The Annals of Mathematical Statistics. 10, No. 3 Sep 1939 p 275-287.
- Descriptive Statistics