Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
American Sociological Association is collaborating with JSTOR to digitize, preserve and extend
access to Sociometry
Evaluating the extent of agreement among a set of judges each of whom has
ranked all of the members of a group except for himself is a common problem
in sociometric studies. Although Kendall's extension of the Coefficient of Con-
cordance, W, to Youden arrays is available as a measure of agreement in suck
a case, critical values of W have been given only by approximation and only
for large n. The current paper reports critical values of S for small n (4-10).
In addition, the paper demonstrates how Kendall's Coefficient of Concordance
can be used to test more specific hypotheses about agreement on the criteria
of choice within a group. This suggested extension of W applies not only to
the special case of square data matrices with zero major diagonals but to data
matrices in general.
* Revision of a paper presented at the annual meeting of the Pacific Sociological As-
sociation, Honolulu, Hawaii, April 1971.
496
W = 12S/m2n (n2 - 1)
where S is the sum of the squared deviations of the column sums from the
mean column sum,
S E(rj - r /M) 2
In the case where the objects to be ranked are the judges themselves, n = m
and
W= 12S/n3 (n2- 1)
W= 12S/X2n (n2 - 1)
where X is the number of times that a given comparison occurs. When each
member of a group ranks all of the members except himself, each pair of
members is ranked n - 2 times. This is obvious from the fact that the (ij)
pair is omitted by person i and person j but by no one else. In this case,
A = n - 2 and
Critical Values of S
1 A FORTRAN program was written that did the following: a random generator filled
the n(n-1) off-diagonal cells of a given data matrix, in each row the off-diagonal entries
(based on 50,000 data matrices per group size), together with maximum
values of S are shown in Table 1. Thus, under the null hypothesis that the
judges have ranked the objects at random, with seven judges Pr(S t 396)
9 .001.
TABLE 1
Pr(5SX)C 4 5 6 7 8 9 10
.990 2 4 10 22 44 88 146
.980 4 4 12 26 54 104 178
.900 4 10 22 50 96 176 292
.500 8 24 56 114 208 358 566
.250 14 36 82 164 288 488 758
.100 16 48 108 214 372 624 956
.050 .. 56 126 246 428 712 1094
.025 20 62 142 276 480 796 1214
.010 .. 68 158 312 542 900 1363
.005 .. 74 170 338 586 966 1462
.001 .. 80 194 396 686 1106 1710
Smax 20 90 280 700 1512 2940 5280
X2_A (n2 _1 ) W/ (k + 1 )
Comments
The purpose of the present paper is both to call attention to the extension
of the Coefficient of Concordance to data matrices with missing entries, and
were replaced by their rank, the column sums were calculated, S was computed, and
finally the frequency distribution for S was compiled. The program was run on the
Univac 1108 at Carnegie-Mellon University.
TABLE 2
to present the critical values for that statistic in the case where the major
diagonal is zero. Since it might appear that the extended form of the statistic
is severely limited in application, some comments are perhaps in order.
First, although the data collected may yield a complete data matrix, it is
not always the case that one wants to apply Kendall's Coefficient of Con-
cordance in its familiar form, that is, in the form appropriate to complete
data matrices. In many situations it may be more appropriate to eliminate the
judge's evaluation of himself in attempting to assess the extent to which
judges agree in their evaluations of the other members of a group. In most
situations, the judge's information about himself probably differs significantly
in both quantity and quality from the information he holds about others.
Jones (1959) has discussed this case in some detail.
Second, typical uses of Kendall's Coefficient of Concordance take the items
being judged as fixed and test the null hypothesis that the similarities in the
rankings of the objects could have been produced by chance. The alternative
hypothesis is that the judges are applying a common or similar standard of
judgment. There is an important case, however, where researchers would fail
to reject the null hypothesis even though the alternative hypothesis is true,
unless the researcher is careful in the test he performs. This can easily occur
if the criterion of evaluation is a function not simply of the items being
judged but is a function of the characteristics of the items and the judges
combined. Perhaps the point can be made most quickly by a short example.
Homans (1961) has suggested that if a group is characterized by a
prestige hierarchy (which we shall assume for present purposes is a strong
ordering) then, if the judges are asked to rank the other members of the
group according to their "preference" for them, the i-th judge will rank the
TABLE 3
Expected Rankings by Judges Where the Judges Are Ordered by Prestige in the Group-
Homans' Hypothesis
Person Ranked
1 2 3 4 5
1l-l [ 1 2 3 4
2 12 3 4
Judge 3 3 1 2 4
4 4 3 1 2
5 4 3 2 1
ing to the hypothesized preferences of the i-th judge, then the resulting table,
if Homans' hypothesis is correct, will result in a table in which the data of
the i-th row will contain the vector (1, 2, . . . , n - 1), with 0 as the i-th
element. Application of Kendall's Coefficient of Concordance to the trans-
formed table is appropriate. One would reject Homans' hypothesis either if
the results of the statistical test were consistently non-significant or if the
results were significant but not in the "direction" of the hypothesis. Direction
in this case refers to a monotonic increase of the column sums.
Table 4 shows the application of W to a hypothetical matrix and to the
same matrix after it has been transformed to test Homans' hypothesis.2
Whereas the untransformed data do not show substantial departure from
results which might be obtained in the absence of a common criterion of
judgment, the transformed matrix does. Secondly, and equally important, the
column sums increase in the manner suggested by the hypothesis.
2 To obtain the transformed matrix, rearrange the observed data matrix as follows: for
j <i, place in the (i,j) cell the observed rank of the person expected to be ranked j-th
by the i-th judge; for j = i, enter zero in the (i,j) cell; and for j > i, place in the (i,j)
cell the observed rank of the person expected to be ranked j - 1.
I 1?-0 2 1 4 3 1 0
2 1 0 3 2 4 2 1 0
3 3 1 0 2 4 3 1 2
4 3 1 0 2 4 1 2
4 3 1 2 0 5 2 1
12 9 6 10 13 5 7 1
S=30 S
W = .33 W =
p> .25 p=.0
* Transfor
REFERENCES
Bales, Robert F.
1953 "The equilibrium problem in small groups." Pp. 111-161 in T. Parsons,
R. F. Bales and E. A. Shils, Working Papers in the Theory of Action. New
York: Free Press.
Bock, R. Darrell
1955 Unpublished study of a ninth grade science class, Hyde Park High School,
Chicago.
Bock, R. Darrell and S. Z. Husain
1952 "Factors of the tele: A preliminary report." Sociometry 15:206-219.
Bonacich, Phillip
1968 Specialization and Differentiation in Small Laboratory Groups. Ph.D. Dis-
sertation: Harvard University.
Burke, Peter J.
1967 "The development of task and social-emotional role differentiation."
Sociometry 30 (December) :379-392.
Goldhammer, D.
n.d. Unpublished study of discussion groups in the Great Lakes Naval Training
Station.
Homans, George C.
1961 Social Behavior: Its Elementary Forms. New York: Harcourt, Brace and
World.
Jones, James A.
1959 "An index of consensus on rankings in small groups." American Sociological
Review 24:533-537.
Kendall, Maurice G.
1955 Rank Correlation Methods. New York: Hafner.
Morton, Anton S.
1959 Similarity as a Determinant of Friendship: A Multi-Dimensional Study.
Princeton University and Educational Testing Service.
Newcomb, Theodore M.
1961 The Acquaintance Process. New York: Holt, Rinehart and Winston.
Rehagen, Kenneth J.
1948 A Comparison of Pupil-Teacher Planning and Teacher-Directed Procedures
in Eighth Grade Social Studies Classes. Ph.D. Dissertation, University of
California, Department of Education.
Slater, Philip E.
1955 "Role differentiation in small groups." American Sociological Review 20:
300-310.
Theodorson, G. A.
1957 "The relationship between leadership and popularity roles in small groups."
American Sociological Review 22:58-67.
Weisman, Thomas
1968 "Liking and participating in small task-oriented problem-solving groups."
A.B. honors thesis, Yale University.