Chapter 3: Correlation

CNCRD

Calculates and test the significance of the Kendall coefficient of concordance.

Required Arguments

XNOBS by K matrix containing the data.   (Input)
Each column of X is a set of observations (which can be converted to ranks) or a set of ranks.

FUZZ — Value to be used for determining ties.   (Input)
If within a column of X, the difference between two elements is less than or equal to FUZZ in absolute value, then the elements are said to be tied.

SUMS — Vector of length NOBS containing the sums of the K ranks in the corresponding row of X.   (Output)

STAT — Vector of length 4 containing the output statistics.   (Output)

i           STAT(i)

1          W, the coefficient of concordance

2          Chi-squared statistic corresponding to W with NOBS 1 degrees of freedom

3          Asymptotic probability of exceeding STAT(2) under the null hypothesis of independence

4          Kendall S statistic. This is the sum of the squared deviations from the expected sum of the ranks

Optional Arguments

NOBS — Number of observations per set of rankings.   (Input)
Default: NOBS = size (SUMS,1).

K — Number of sets of rankings.   (Input)
K must be greater than or equal to two.
Default: K = size (X,2).

LDX — Leading dimension of X exactly as specified in the dimension statement in the calling program.   (Input)
Default: LDX = size (X,1).

FORTRAN 90 Interface

Generic:          CALL CNCRD (X, FUZZ, SUMS, STAT [,…])

Specific:                             The specific interface names are S_CNCRD and D_CNCRD.

FORTRAN 77 Interface

Single:                                CALL CNCRD (NOBS, K, X, LDX, FUZZ, SUMS, STAT)

Double:                              The double precision name is DCNCRD.

Description

Routine CNCRD computes and tests the significance of the Kendall coefficient of concordance.

The coefficient of concordance is computed as follows: Within each of the k sets the n = NOBS observations are ranked. Tied ranks are used for tied observations where two observations are tied if they are within FUZZ of each other. Let xi denote the sum of the ranks for the i-th observation over the k sets. The mean of the xi is

Using this mean, compute the sums of squares of the xi about their mean as

This is the Kendall S statistic (STAT(4)). If there are tied ranks within a set i, compute the adjustment

where tj is the number of ties in the j-th group of ties, and the summation is over all tie groups for the set. Kendall’s coefficient of concordance, W, is computed as

Kendall’s coefficient of concordance is related to the Friedman one-way analysis of variance on ranks chi-squared test statistic T (see IMSL routine FRDMN.p<.STCH6.DOC!FRDMN;620;), as

When n or k is small, tables of the exact distribution of W exist. See Owen (1962, pages 396397). The probability reported in STAT(3) is asymptotic. It is only approximate when k and n are small.

Comments

1.         Workspace may be explicitly provided, if desired, by use of C2CRD/DC2CRD. The reference is:

CALL C2CRD (NOBS, K, X, LDX, FUZZ, SUMS, STAT, IWK, XWK)

The additional arguments are as follows:

IWK — Work vector of length NOBS.

XWK — Work vector of length NOBS * K.

2.         Informational errors

Type Code

3         6                  Within each of the K sets of rankings all observations are tied. STAT(1) STAT(3) cannot be computed and are set to NaN (not a number).

3         7                  The chi-squared degrees of freedom is less than 7. STAT(3) should be regarded with suspicion.

Example

The example is taken from Kendall (1962, pages 9798). It involves ten observations in three sets. The resulting coefficient of concordance, 0.828, is quite large, indicating a strong relationship.

 

      USE WRRRN_INT

      USE CNCRD_INT

      USE WRRRL_INT

 

      IMPLICIT   NONE

      INTEGER    K, LDX, NOBS

      REAL       FUZZ

      PARAMETER  (FUZZ=0.0001, K=3, LDX=10, NOBS=10)

!

      REAL       STAT(4), SUMS(NOBS), X(LDX,K)

      CHARACTER  CLABEL(2)*11, RLABEL(4)*11

!

      DATA RLABEL/'W', 'Chi-squared', 'p-value', 'S'/

      DATA CLABEL/'Statistic', '  '/

      DATA X/1, 4.5, 2, 4.5, 3, 7.5, 6, 9, 7.5, 10, 2.5, 1, 2.5, 4.5, &

          4.5, 8, 9, 6.5, 10, 6.5, 2, 1, 4.5, 4.5, 4.5, 4.5, 8, 8, 8, &

          10/

!

      CALL WRRRN ('X', X)

!

      CALL CNCRD (X, FUZZ, SUMS, STAT)

!

      CALL WRRRN ('SUMS', SUMS, 1, NOBS, 1, 0)

      CALL WRRRL ('  %/%/', STAT, RLABEL, CLABEL, FMT='(W10.6)')

      END

Output

 

               X
         1       2       3
 1    1.00    2.50    2.00
 2    4.50    1.00    1.00
 3    2.00    2.50    4.50
 4    4.50    4.50    4.50
 5    3.00    4.50    4.50
 6    7.50    8.00    4.50
 7    6.00    9.00    8.00
 8    9.00    6.50    8.00
 9    7.50   10.00    8.00
10   10.00    6.50   10.00


                                    SUMS

   1      2       3       4       5       6       7       8       9      10

5.50   6.50    9.00   13.50   12.00   20.00   23.00   23.50   25.50   26.50


Statistic

W                 0.828

Chi-squared      22.349

p-value           0.008

S               591.000



http://www.vni.com/
PHONE: 713.784.3131
FAX:713.781.9260