Multivariate Analysis

Elliot M. Cramer; R. Darrell Bock

Review of Educational Research, Vol. 36, No. 5, Methodology of Educational Research. (Dec.,
1966), pp. 604-617.

Stable URL:
[Link]

Review of Educational Research is currently published by American Educational Research Association.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
[Link] JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
[Link]

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact support@[Link].

[Link]
Fri Aug 24 [Link] 2007
 CHAPTER VIII

Multivariate Analysis

ELLIOT M. CRAMER * and R. DARRELL BOCK

This review covers multivariate statistical methods other than factor

analysis. Previous reviewers have only touched on multivariate methods
in connection with other topics (Norton, 1963) or considered the special
case of discriminant analysis (Tatsuoka and Tiedeman, 1954). To bring
the reader up to date, the present review includes contributions to multi-
variate analysis back to the initial work in the 1930's.
The multivariate methods (exclusive of factor analysis) most important
for educational research are discriminant analysis, canonical correlation,
and multivariate analysis of variance (MANOVA). Of these, multivariate
analysis of variance is the most general and, in fact, can be formulated
comprehensively to include canonical correlation and discriminant analysis
as adjuncts of the analysis (Bock, 1 9 6 3 ~ ) .
MANOVA and canonical correlation are multivariate generalizations of
the familiar Fisherian analysis of variance and analysis of regression.
Although they are part of the statistical development which originated in
Student's and Fisher's involvement with agricultural field trials, general
multivariate procedures are not especially relevant in analysis of agricul-
tural experiments. Agricultural field trials usually have a clear-cut, one-
dimensional response variable-namely, yield of crop-and multivariate
methods are not needed (Finney, 1956).
The educational equivalent of a field trial-a methods study or curricu-
lum evaluation-is much more clearly in need of multivariate methods. In
educational experiments, the subjects' responses to the treatments may be
expressed through a multiplicity of interrelated variables. Often there is
no objective basis for giving priority to certain variables or defining a
single composite of variables. Merely to classify possible dimensions of
response may require an extensive taxonomy (Bloom, 1956; Krathwohl,
Bloom, and Masia, 1964). Multivariate analysis provides models and pro-
cedures for dealing separately with each of a number of variables in esti-
mation, while at the same time providing tests of hypotheses which lead
to a single probability statement referring to all variables jointly.
At the present time exploitation of multivariate methods in the educa-
tional research represented in the literature is irregular. Discriminant
* This work was partially supported by Public Health Service Research Grant No. MH7285,
National Institute of Mental Health, to the Biometric Laboratory, University of Miami (Dean
Clyde, principal investigator).
December 1966 MULTIVARIATE
ANALYSIS

analysis now seems well established and is in practical use in the field of
guidance and also appears with some regularity in multivariate compari-
sons of two or more treatment groups (e.g., Kelly, Veldman, and McGuire,
1964).
I n the analysis of two-way (or more) designs, or where control variables
are involved, however, the literature continues to include studies which
rely exclusively on univariate statistics calculated separately for each of
the response variables (e.g., Hopkins, Oldridge, and Williamson, 1965).
Such a procedure inevitably forces the investigator to pick out and interpret
the largest t or F statistic, correlation coefficient, or whatever, often from
rather large numbers of nonindependent statistics and, of course, increases
the probability of errors of the first kind to levels far above their nominal
values. Since the educational literature is undoubtedly already biased
toward studies in which chance was favorable to the outcome (i.e., results
of significance tests influence the acceptance of papers for publication),
uncritical selection of and emphasis on a few variables which show signifi-
cance in multivariate studies are bound to increase the reporting of find-
ings which will not replicate. For the most part, this problem may be
avoided by the use of multivariate statistical methods when the data are
multivariate. The investigator then has the protection of an exact test of
the treatment effect which is comprehensive of all variables simultaneously.
If the multivariate statistic is significant, he can look at the separate uni-
variate t's, F's, and r's with some assurance that there is a significant effect
in the data, an approach which is analogous to a univariate analysis in
which the investigator looks at t-statistics for contrasts of particular treat-
ment groups with the protection of a comprehensive F-test of between-
group differences.
Procedures for carrying out a wide variety of multivariate tests appli-
cable to multivariate normal populations can be found in the statistical
literature cited in this review. Since most of these procedures require con-
siderable calculation, computer programs are essential for their practical
implementation. Therefore, it was felt that programing specifications
and documents describing important programs should be included in the
review.
The standard reference in multivariate analysis is undoubtedly Ander-
son's (1958) book An Introduction to Multivariate Analysis, but its high
difficulty level and the paucity of examples make it an unsuitable reference
for the research worker. Rao's (1952) book remains an important refer-
ence for the research worker, with its emphasis on applications of dis-
criminant theory; his later work (1965a) has some overlap in the multi-
variate analysis area but is at a higher mathematical level. Other books
of general interest are those by Cooley and Lohnes (1962), Kendall
(1957), Roy (19571, and Seal (1964 1 . Unfortunately, no single reference
yet exists which treats the subject of multivariate methods in a compre-
hensive manner.
REVIEWOF EDUCATIONAL
RESEARCH Volume X X X V I , No. 5

Development of Multivariate Analysis of Variance

(MANOVA)
The basic method has been available for over 15 years and, looking back
to Tukey's little known presentation (1949), we find that the generalization
seems natural and obvious. The computational complexities for more than
two criteria were undoubtedly sufficient to retard its use; also of impor-
tance were the difficulties of obtaining equal numbers of observations in
cells with multiple criteria as required by the original formulation. The
test of significance using Wilks' criterion, identical to that for the multiple
discriminant function, would be found in Rao (1952) with a very good F
approximation. Rao (1955) made an important contribution in considering
the general multivariate factorial design with unequal numbers in cells
and was one of the first to consider the unequal N problem with more than
two factors, even for the univariate case. He gave and illustrated tests for
two and three factor designs testing interactions and main effects, elim-
inating effects of equal or lower order and ignoring those of higher order.
His computational ~ r o c e d u r e swere several orders of magnitude more
difficult than those of univariate ANOVA with equal N although the prob-
lem was a more realistic one. The methods, however, were not readily
adapted to digital computers. The treatment by Roy and Gnanadesikan
(1959) was very general. It covered both fixed and random effects models
with the multivariate extension for any type of analysis of variance and
possibly unequal numbers in cells. The matrix treatment provided a natural
framework for computer implementation.
Rao (1952) argued that Wilks' criterion does not provide a satisfactory
test for the multigroup discriminant function since it gives equal weight to
all dimensions of variation (where the number of dimensions is the lesser
of the number of criteria and the number of degrees of freedom). H e felt
that if the differences were concentrated in one or two dimensions they
might be obscured by the error dimensions, at least if the effects were small.
The development of the largest root test by Roy (1957) provided an alterna-
tive to the use of Wilks' criterion. Charts of 0.05, 0.025, and 0.01 signifi-
cance levels for tests involving up to five roots (up to five criteria or up to
five degrees of freedom) were provided by Heck (1960). Pillai (1960,
1964, 1965) tabled 0.05 and 0.01 significance levels for up to 1 0 roots.
A third criterion, called the "trace" criterion, which involves all dimen-
sions of variation, was proposed by Hotelling (1957) and tabled by Pillai
(1960,1964, 1965). The trace criterion is the sum of the roots and involves
all dimensions of departure from the null hypothesis.
I t is known that no one of these tests is uniformly more powerful than
either of the others; indeed, for the case of one degree of freedom tests,
they are equivalent. It was shown by Srivastava (1964) and by Das Gupta,
Anderson, and Mudholkar (1964) that all have the monotonicity property,
the power of the test increasing as the differences in any dimension in-
December 1966 MULTIVARIATE
ANALYSIS

crease. Little is known about power generally; some specialized results were
obtained for Wilks' test by Posten and Bargmann (1964).
Additional References: Bartlett (1939, 1947, 1954) ; Hopkins and Clay
(1963) ; Ito (1962) ; Mikhail (1965) ; Rao (1948).

Discriminant Function
Illustrations of the use and interpretation of multiple discriminant analy-
sis were given by Cooley (1959), Johnson (1955), and Jones and Bock
(1960). The use of dichotomous variables to form a multiple discriminant
was introduced by Claringbold (1958). Maxwell (1961) gave an example
with dichotomous variables in a psychological application. The method
does yield an "almost" continuous criterion which maximally discriminates
among groups and might be very useful for that reason, questions of
significance aside. While one would hope that the tests of significance in
multivariate analysis would be rather insensitive to nonmultivariate nor-
mality, this does seem to be a rather extreme application for the usual tests.
Cochran and Hopkins (1961) went back to the underlying theory in
forming a two-group discriminant function with qualitative data. The dis-
crete multivariate distribution was estimated, and observations could then
be classified in the group corresponding to the higher probability for an
observed state of the criteria. The authors suggested that this might be a
practical procedure where the number of criteria and states of each are
small. Adjustments were given for estimating probabilities of misclassifica-
tion. As a useful approximation to discriminant analysis in practice, Hall
(1963) proposed assignment based on similarity of profiles, a procedure
equivalent to linear discriminant classification if independent criteria were
used and similarity were judged by sum of squared deviations. The use-
fulness of the ipsative mean as suggested is not completely clear.
Collier (1963) observed that since there is a formal relationship be-
tween regression analysis and discriminant function analysis for the two-
group case, methods used for deleting variables in regression analysis could
be carried over to discriminant function analysis. He showed the equiva-
lence between the regression test and that given by Rao (1952) for testing
additional measures. The generalization of this to MANOVA involves a
multivariate analysis of covariance using the doubtful variables as criteria
and the others as covariates. This generalization is implicit in Rao's work.
Cochran (1964) was concerned with choosing the criteria to use in a
discriminant analysis. He gave some results indicating the circumstances
under which a new discriminator would increase the power of a discrim-
inator already chosen. Several examples from the literature indicated that
the discriminatory power of a group of criteria could be judged by the
squared normalized differences of the individual criteria. His examples,
however, were based on relatively small numbers of criteria, probably
already subjected to selection.
607
REVIEWOF EDUCATIONAL
RESEARCH Volume X X X V I , No. 5

Additional References: Bartlett (1951) ; Michael and Perry (1956) ;

Rao (1962) ; Williams (1955).

Relationship Between MANOVA and Discriminant

Analysis

In the chapter on discriminant analysis in the December 1954 REVIEW,,

Tatsuoka and Tiedeman extensively discussed the two-group discriminant
function and its obvious extension, the multigroup discriminant function.
As they pointed out, the two-group discriminant function may be ap-
proached by way of regression analysis using a criterion variable having
a value of 1 in one group and 0 in another group, but this does not
generalize directly to the multigroup case.
Horst (1956) did consider a formulation involving separate regression
equations contrasting each group in turn with all others, that is, defining
variable i to have the value 1 for subjects in that group and 0 otherwise.
He used the total covariance matrix of the predictors to estimate the co-
efficients; using the within-groups matrix in place of this does provide a
solution equivalent to the standard discriminant solution when the cutoff
points are appropriately chosen. Another way of generalization takes as its
starting point univariate analysis of variance on a number of criteria.
Tukey (1949) gave a very clear discussion of this type of generalization
in which the ordinary analysis of variance (ANOVA) calculations on sums
of squares are applied to cross-products of the criteria as well. This analysis
of variance for vectors answers the question, Are there any differences
among treatment groups on any of the criteria?-equivalent to the
question asked by discriminant analysis, Is there any linear combination of
the criteria which discriminates among the groups? As Rao (1955) indi-
cated, an ordinary analysis of variance could be performed on this optimum
composite score obtaining a higher F ratio than for any other composite.
Viewed in this way the multiple-group discriminant problem is a one-way
analysis of variance on several criteria simultaneously. All the logic and
computational methods of univariate ANOVA can then be directly applied
to the general multivariate problem.

Canonical Correlation

Hotelling (1935, 1936) gave the solution for determining the linear com-
posite of a set of criteria most predictable in the least squares sense from a
set of predictors: a generalization of multiple correlation where optimum
weights are chosen for both criteria and predictors, the resulting correla-
tion-the canonical correlation-being the highest possible between com-
posites from the two sets of variables. Other composites independent of the
first two can be obtained giving measures of relationship between criteria
608
December 1966 MULTIVARIATE
ANALYSIS

and predictors on dimensions independent of the first. Bartlett (1941)

considered tests of significance of overall relationship and of residual
relationship after the largest dimension corresponding to the first canonical
correlation had been eliminated.
There have been few applications of canonical correlation although it
provides an exact solution for a frequently asked question, Do two sets of
measures have factors in common? They have common factors only if the
canonical correlations between them are not all zero. Hotelling (1957)
observed that such questions formulated in terms of factor analvsis can be
better answered by multivariate methods. Canonical correlation ~ r o v i d e s
the method for testing independence of two groups of variables under the
multivariate normal hypothesis. Roy and Bargmann (1958) obtained a
generalization for testing independence among several sets of measures.
The step-down method developed for this procedure has found other
applications in Bock (1966) and Roy (1958). Horst (1961) was concerned
with similar generalizations, but his results dealt with linear transforma-
tions to maximize measures of correlation among groups rather than
with tests of significance.
Computational procedures appropriate to small computers were given
by Koons (1962), but the methods do not generalize well to today's larger,
faster machines. Uses of multivariate methods and particularly. partial.
canonical correlation were noted by Bargmann (1961). Computational
procedures may be found in Roy (1957).
The correction for attenuation in ordinary correlations was extended to
the general case by Meredith (1964). Striking increases in the correla-
tions may be obtained (from 0.68 to 0.97 in his example), but, as in the
two-variable case, the correlations could be greater than 1. It should be
remembered that except for very large samples, canonical correlations
have a very substantial upward bias which would carry over to the cor-
rected correlations.
Additional References: Anderson and Das Gupta (1964) ; Lawley (1956,
1959).

Multivariate Analysis in the Repeated Measures Design

Norton, in the December 1963 REVIEW,considered developments in the
use of the repeated measures design. Lana and Lubin (1963) pointed out
the frequent use and misuse of the repeated measures design in psychology
both for repeated measures, where it may sometimes be defensible, and for
multiple criteria, where it seems to be completely indefensible. They sug-
gested strategies for the investigator with regard to using the conservative
test of Greenhouse and Geisser (1959) and the exact, though nonspecific,
multivariate test.
Bock (196313) used orthogonal polynomials in the repeated measures as
multivariate criteria. He gave other tests for specific situations with as-
REVIEWOF EDUCATIONAL
RESEARCH Volume X X X V I , No. 5

sumptions about the structure of the repeated measures. Potthoff and Roy
(1964) generalized multivariate analysis of variance by appending a post-
matrix to the general linear hypothesis model. They indicated that the
obvious practical application of this involved orthogonal polynomials for
growth curves. In the case where the number of polynomials equals the
number of repeated measures, their analysis is equivalent to a multivariate
analysis on the original measures transformed by the orthogonal poly-
nomials. Where a smaller number of orthogonal polynomials is used, one
of their special cases reduces to this same analysis. Rao (196513) objected
to the arbitrariness of the Pothoff and Roy formulation because it involved
the choice of an arbitrary matrix which presumably affects the power of
the test. He suggested doing the conventional multivariate analysis on the
measures transformed by the orthogonal polynomials but using some or all
of the higher-order terms as covariates. He indicated that some progress
had been made toward the solution of the problem of choosing the best
covariates from the higher-order polynomials. There still remains the
problem of choosing the number of polynomials to use as criteria.
Additional References: Chan (1964) ; Elston and Grizzle (1962) ;
Geisser (1959) ; Rao (1958, 1959).

Analysis of Covariance Structures

A number of workers have proposed tests of a hypothesis that the popu-

lation covariances matrix has some special pattern or structure. Wilks
(1946) provided a sample criterion for testing the hypothesis that the
observations are from a multivariate normal population with a specified
covariance matrix. Votaw (1948) gave a test of the hypothesis of com-
pound symmetry, i.e., that the population matrix has equal variances and
equal covariances. Bargmann (1957) showed that compound symmetry is
one of several "reducible" covariance patterns which may be diagonalized
by a known linear transformation. Bock (1960) observed that a classifica-
tion of psychological test items in a 2" factorial design implies that the
covariance matrix of the test scores has the reducible pattern called "equi-
predicability" by Bargmann (1957).
Since this pattern is diagonalized by transformation with the 2" Hada-
mard matrix, Bock (1960) proposed using a criterion of no-association
in the transformed sample covariance matrix as a test of "equi-predica-
bility" in the population. He also showed the relationship of this transfor-
mation to the "mixed model" analysis of variance for this type of data.
Mukherjee (1966) derived the likelihood criterion for testing the so-
called "Guttman quasi-simplex" structure and gave a simple procedure for
its calculation. Mukherjee's solution is an advance over those solutions in
the literature which assume a perfect simplex and do not account for the
measurement error (Kaiser, 1962).
December 1966 MULTIVARIATE
ANALYSIS

Srivastava (1966) identified a class of covariance structures which arise

from any balanced or partially balanced cross-classification of test items.
He demonstrated that this class is reducible and derived a likelihood ratio
test for the structure. Strivastava found also that the likelihood ratio
test is not equivalent to testing the hypothesis that the reduced matrix
is diagonal.
Bock and Bargmsnn (1966) gave likelihood ratio criteria for testing
the hypothesis that the observed scores are composed of a specified linear
transformation of independent latent scores plus an independent measure-
ment error. They derived the criteria assuming both homogeneous and
nonhomogeneous error variances for the tests. In cases where the test
scores are not all in the same metric, they included in the hypothesis scale
factors accounting for the unit of measurement of each variable. Thus their
tests may be applied to the correlation matrix as well as covariance matrix.
They discussed the role of their tests in analyzing the m u l t i t r a i ~ u l t i -
method matrix (Campbell and Fiske, 1959).

Other Work
Although the problem of missing observations can be handled in the
general form of MANOVA, no allowance is made for partially missing
observations where some subjects have some measures missing. Methods
for estimating the variance-covariance matrix have been given by Ander-
son (1957), Buck (19601, and Nicholson (1957). These methods may be
used as Buck has done to estimate regression coefficients. The modification
of multivariate tests of significance is another matter. Rao (1956) did
- a test criterion based on Wilks' test for the special case where all the
give
missing scores are on one criterion. No generalization has apparently been
obtained. A somewhat different situation was treated by Trawinski and
Bargmann (19M) in which their model provided for obtaining subsets of
the criteria on several groups of subjects.
Another criterion for tests in MANOVA is the step-down method as
given by Roy (1958). This involves a univariate ANOVA on the most
important criterion followed by covariance analyses on the remaining
criteria eliminating the previous ones. It is, however, dependent on the
initial ordering and on the significance levels chosen for each test. Its use
as a test for the value of doubtful criteria in addition to surer ones was
advocated by Bock (1966).
Roy and Murthy (1960) tabled some significance levels for Wilks' test
of simultaneous equality of variances and of covariances and also the
test of equality of means, variances, and covariances.
Chung and Fraser (1958) and Dempster (1960) considered multivariate
tests for two groups where there are too few observations for Hotelling's T2.
Dempster's test rests upon normal theory while the method of Chung and
Fraser is a randomization test.
REVIEWOF EDUCATIONAL
RESEARCH Volume X X X V I . No. 5

Dempster (1963) developed a stepwise MANOVA related to the step-

wise procedure of Roy (1958), but based on an ordering determined by a
principal components analysis. As Dempster pointed out, there is no rea-
son why this ordering should be preferable to an a priori ordering. As
with the tests above, when the number of variables is large in comparison
to the number of observations, the method does provide a test involving a
combination of all criteria using only the largest component.
A very interesting graphical procedure was proposed by Wilk and
Gnanadesikan (1964) to give a meaningful summary of the real effects
present in a set of data analyzed by MANOVA. In the general case it con-
sists of choosing sets of orthogonal contrasts (in the main effects and inter-
actions) and obtaining scores analogous to the D' of discriminant analysis.
They suggested that plots of these scores provide "a useful summary of
indications and a stimulus for insight."
Additional Reference: Bennett (1962, 1965) .

Applications of Multivariate Analysis i n Research

Finney (1956) criticized the use of multivariate analysis, in agriculture
in particular, and in research more generally. His particular objections
were that optimum linear combinations of criteria were of no interest, that
significance tests were of slight importance in comparison with estimates
of effects, and that inexact or nonindependent tests of the right thing were
preferable to exact tests of the wrong thing. Rao (1960) vigorously upheld
multivariate methods as an "indispensable statistical aid in applied re-
search," giving numerous examples of the usefulness of these methods.
Finney's objections do not generalize very well to education where many
criteria may be involved and judgments based on many correlated tests are
hazardous. Although the interpretation may be difficult, the optimum linear
combinations of multivariate analvsis do ~ r o v i d einformation not avail-
able from the separate univariate analyses.
Recent papers have attempted to display the virtues of multivariate
methods and advocate their use. They are at present better references for
the research worker than is any available single text. Smith, Gnanadesikan,
and Hughes (1962) described the analysis of a one-way MANOVA with
covariates. The analysis was done on an IBM 709 computer, and the inter-
mediate steps involved were indicated. Bock and Haggard (in press) gave a
description of the use of multivariate methods generally, considering multi-
variate analysis of variance, covariance, and regression. Their presentation
consisted of the application of the various methods to the same real data,
giving some insight into the interrelationships of the methods. Interpreta-
tion of results was emphasized with the computational procedures outlined.
Three examples of factorial experiments with multiple criteria were used
by Jones (1966) as examples of experiments designed for MANOVA. The
value of discriminant functions and discriminant scores as an aid in the
December 1966 MULTIVARIATE
ANALYSIS

interpretation of results was emphasized. Bock (1966) considered appli-

cations in educational research and gave worked examples of multivariate
tests of equality of covariance matrices, homogeneity of regression, no-
association between sets of variables, equality of treatment effects, and
adjusted treatment effects.

Computer Procedures

Bock ( 1 9 6 3 ~ )gave a thorough discussion of the computational pro-

cedures of MANOVA for arbitrary multifactor designs with multiple co-
variates for the general unequal n case. This paper provided the basic
framework for handling almost any type of univariate or multivariate
analysis within a single computer program. It gave the logical basis for
several of the most widely used computer programs in this area. Bock
(1963a) discussed the implementation of these methods on a. computer
and provided detailed flow charts. Finn (1966) prepared a very large
capacity Fortran I1 program from these flow charts, but the program uses
special subroutines which are machine dependent (IBM 7090/7040).
Another such program based on the same techniques was provided by
Hall and Cramer (1963) for an IBM 7090 in Fortran 11. A newer program
(Clyde, Cramer, and Sherin, 1966) in Fortran IV for the IBM 7090 and
IBM 7040 gave a somewhat different implementation. Cooley and Jones
(1964) described an integrated package of univariate and multivariate
computer programs in Fortran 11. As described, it included the basic
multivariate procedures except for MANOVA.
Although canonical correlation and discriminant function can be con-
sidered special aspects of MANOVA, programs for these procedures are
useful in their own right since they frequently provide different types of
output such as discriminant scores and classification tables. Current sources
of these include Dixon's Biomed series (1964) ; Clyde, Cramer, and
Sherin's multivariate package (1966) ; and the Cooley and Lohnes text
(1962). Probably all of these sources will have updated programs for the
- ~

newer generation of computers.

The availability of computer programs such as these and the develop-
ment of newer ones should make possible the wider use of multivariate
methods. Only then will the usefulness of multivariate analysis become
apparent for research workers in education.

Bibliography
ANDERSON,
T. W. ''IVlaximum Likelihood Estimates for a Multivariate Normal Dis-
tribution When Some Observations Are Rlissing." Journal of the American Statisti-
cal Association 52: 200-203; June 1957.
ANDERSON, T. W. An Introduction to Multivariate
Analysis. New York: John Wiley
and Sons, 1958. 374 pp.

REVIEWOF EDUCATIONAL
RESEARCH
Volume X X X V I , No. 5

B A R G M A N NROLF.
, A S t u d y of Independence and Dependence i n Multivariate Normal
Analysis. Institute o f Statistics, hlimeograph Series No. 186. University o f North
Carolina, Chapel Hill, December 1957. 95 pp.
B A R G M A N NR., E. "Multivariate Statistical Analysis i n Psychology and Education."
Bulletin de L'Institut International de Statistique 38, Part I V : 79-86; 1961.
BARTLETT,M . S. " T h e Statistical Significance o f Canonical Correlations." Biometrika
32: 29-37; January 1941.
BLOOM,B E N J A M I NS., editor. Taxonomy of Educational Objectives: T h e Classifica-
tion of Educational Goals. flandbook I : Cognitive Domain. New Y o r k : Longmans,
Green, and Co., 1956. 207 pp.
B O C K , R . DARRELL."Components o f Variance Analysis as a Structural and Dis-
criminal Analysis for Psvchological Tests." British Journal o f Statistical Psrchologr -.
13: 151-63; ~ o v e m b e r1960.
BOCK,R. DARRELL." A Computer Program for Univariate and Multivariate Analysis of
Variance." Proceedings of the IBM Scientific Computing Symposium o n Statistics.
W h i t e Plains, N.Y.: IBM Data Processing Division, October 1963. pp. 69.111. ( a )
B O C K , R. DARRELL."Multivariate Analysis o f Variance o f Repeated Measurements."
Problems in Measuring Change. (Edited b y Chester W . Harris.) Madison:
University o f Wisconsin Press, 1963. Chapter 5, pp. 85-103. ( b )
BOCK,R . DARRELL."Programming Univariate and Multivariate Analysis o f Variance."
Technometrics 5 : 95-117 ; February 1963. ( c )
BOCK, R. DARRELL."Contributions o f Multivariate Experimental Designs t o Educa-
tional Research." Handbook of ,Plultivariate Experimental Psychology. (Edited
b y Raymond B. Cattell.) Chicago: Rand McNally and Co., 1966. pp. 820-40.
B O C K , R. DARRELL,and B A R G M A N NROLF , E. "Analysis o f Covariance Structures."
Psychometrika 31: 507-34; December 1966.
BOCK, R . DARRELL,and HAGGARD,ERNEST A. " T h e U s e o f Multivariate Analysis
o f Variance i n Behavioral Research." Handbook of Measurement i n Educational
Psychology and Education. Reading, Mass.: Addison-Wesley Publishing Co. ( I n
press )
BUCK, S. F. " A Method o f Estimation o f Missing Values i n Multivariate Data
Suitable for U s e with a n Electronic Computer." Journal o f the Royal Statistical
Society (Series B ) 2 2 : 302-306; No. 2, 1960.
CAMPBELL, DONALDT., and F I S K E ,DONALDW . "Convergent and Discriminant Valida-
tion b y t h e Multitrait-Multimethod Matrix." Psychological Bulletin 56: 81-105;
March 1959.
C H U N G J, . H., and FRASER,D. A. S. "Randomization Tests for a Multivariate T w o -
Sample Problem." Journal of the American Statistical Association 53: 729-35;
September 1958.
CLARINGBOLD, P. J . "Multivariate Quanta1 Analysis."
Journal o f the Royal Statistical
Society (Series B ) 20: 398-405; No. 2, 1958.

CLYDE,DEAN J.; CRAMER,ELLIOT M.; and S H E R I N ,R I C H A R DJ . Multivariate Sta-

tistical Programs. Coral Gables, Fla.: Biometric Laboratory, University o f Miami,
1966. 61 pp. ( O f f s e t )
C O C H R A NW, I L L I A MG. " O n the Performance o f the Linear Discriminant Function."
Technometrics 6 : 179-90; May 1964.
C O C H R A NW, I L L I A MG., and H O P K I N S C , ARL E. "Some Classification Problems with
Multivariate Qualitative Data." Biometrics 17: 10-32; March 1961.
COLLIER,R A Y M O N DO., JR. " A Note on the Multiple Regression Technique for
Deleting Variables i n the Discriminant Function." Journal o f Experimental Educa-
tion 3 1 : 351-53; Summer 1963.
COOLEY,W I L L I A MW . "Identifying Potential Scientists: A Multivariate Approach."
School Science and Mathematics 59: 381-96; hlay 1959.
COOLEY,W I L L I A MW . , and J O N E S K , E N N E T HJ . "Computer Systems for Multivariate
Statistical Analysis." Educational and Psychological Measurement 24: 645.53;
Fall 1964.
COOLEY,W I L L I A MW . , and L O H N E S ,PAUL R. Multivariate Procedures for the Be-
havioral Sciences. New Y o r k : John W i l e y and Sons, 1962. 211 pp.
DAS GUPTA, S.; A N D E R S O NT, . W . ; and MUDHOLKAR,G. S. "Monotonicity o f the
Power Functions o f Some Tests o f the Multivariate Linear Hypothesis." Annals o f
Mathematical Statistics 35: 200-205; March 1964.
614
December 1966 MULTIVARIATE
ANALYSIS
DEMPSTER,A. P. " A Significance Test for the Separation o f T w o Highly Multi-
variate Small Samples." Biometrics 1 6 : 41-50; March 1960.
DEMPSTER,A. P. "Stepwise Multivariate Analysis o f Variance Based o n Principal
Variables." Biometrics 19: 478-90; September 1963.
D I X O N ,W . J . Biomedical Computer Programs. Los Angeles: Health Sciences Com-
puting Facility, University o f California, 1964. 585 pp. ( O f f s e t )
F I N N ,J E R E M YD. M E S A 95-Univariate and Multivariate Analysis o f Variance and
Couarzance: A Program for the IBM 7094 Computer. Department o f Education
Statistical Laboratory Research Memorandum No. 3. Chicago: University o f
Chicago, 1966. 68 pp. ( O f f s e t )
F I N N E Y ,D. J. "Multivariate Analysis o f Agricultural Experiments." Biometrics 12:
67-71; March 1956.
G R E E N H O U SSEA, M U E LW . , and G E I S S E RSEYMOUR.
, " O n Methods i n the Analysis o f
Profile Data." Psychometrika 24: 95-112; J u n e 1959.
HALL, C H A R L E SE. "Discriminant Analysis, Classification and Ipsative Scoring."
Journal of Experimental Education 32 : 159-65 ; W i n t e r 1963.
HALL, CHARLESE., and C R A M E RELLIOT , M. " A General Purpose Pro5;am T o C o m -
pute Multivariate Analysis o f Variance on an IBM 7090 Computer. Washington,
D.C.: Biometric Laboratory, T h e George Washington University, 1963. 34 pp. ( O f f s e t )
H E C K , D. L. "Charts o f Some Upper Percentage Points o f the Distribution o f the
Largest Characteristic Root." Annals o f Mathematical Statistics 31: 625-42;
September 1960.
H O P K I N SK, E N N E T HD.; OLDRIDGE, V . A.; and W I L L I A M S O N
MALCOLM
, L. " A n Em-
pirical Comparison o f Pupil Achievement and Other Variables i n Graded and
Ungraded Classes." American Educational Research Journal 2 : 207-15; November
1965.
HORST,PAUL. "Multiple Classification b y t h e Method o f Least Squares." Journal of
Clinical Psychology 12: 3-16; January 1956.
HORST, PAUL. "Relations Among m Sets o f Measures." Psychometrika 26: 129-49;
June 1961.
HOTELLING,H AROLD. " T h e Most Predictable Criterion." Journal of Educational
Psychology 26: 139-42; February 1935.
HOTELLING,HAROLD. "Relations Between T w o Sets o f Variates." Biometrika 28:
321-77; December 1936.
HOTELLING, HAROLD." T h e Relations o f the Newer Multivariate Statistical Methods t o
Factor Analysis." British Journal o f Statistical Psychology 10: 69-79; November
1957.
J O H N S O NM. , C L E M E N S"Classification
. b y Multivariate Analysis with Objectives o f
Minimizing Risk, Minimizing Maximum Risk, and Minimizing Probability o f
Misclassification." Journal of Experimental Education 2 3 : 259-64; March 1955.
J O N E SLYLE
, V . "Analysis o f Variance i n Its Multivariate Developments." Handbook of
Multivariate Experimental Psychology. (Edited b y Raymond B. Cattell.) Chicago:
Rand McNally and Co., 1966. pp. 244-66.
J O N E S , LYLE V . , and BOCK, R. DARRELL."Multiple Discriminant Analysis Applied t o
' W a y s T o Live' Rating from Six Cultural Groups." Sociometry 23: 162-76;
June 1960.
K A I S E RH, E N R YF. "Scaling a Simplex." Psychometrika 27: 155-62; J u n e 1962.
KELLY, FRANCISJ . ; V E L D M A NDONALD , J . ; and MCGUIRE,CARSON."Multiple Dis-
criminant Prediction o f Delinquency and School Dropouts." Educational and
Psychological Measurement 24: 535-44; Fall 1964.
KENDALL,MAURICEG. A Course i n Multivariate Analysis. New Y o r k : Hafner
Publishing Co., 1957. 185 pp.
K O O N S ,PAUL B., JR. "Canonical Analysis." Computer Applications i n t h e Behavioral
Sciences. (Edited b y Harold Borko.) Englewood C l i f f s , N.J.: Prentice.Hal1, 1962.
Chapter 12, pp. 266-79.
K R A T H W O H LDAVID
, R . ; BLOOM,B E N J A M I NS.; and MASIA, BERTRAMB. Taxonomy
of Educational Objectives, Handbook 11: Affective Domain. New Y o r k : David
McKay Co., 1964. 196 pp.
LANA, ROBERT E., and L U B I N ,A R D I E ." T h e Effect o f Correlation o n the Repeated
Measures Design." Educational and Psychological Measurement 23: 729-39; W i n t e r
1963.
REVIEWOF EDUCATIONAL
RESEARCH V o l u m e X X X V I , No. 5

M~XU'ELL. ,4. E. "Canonical Variate Analvsis When the Variables Are Dichotomous."
~ d u c a t i o n a land Psychological ,~easur;ment 21 : 259-71 ; Summer 1961.
[Link]."Canonical Correlations with Fallible Data." Psychometrika
29 : 55-65 ; hfarch 1964.
MCKIIERJEE,BISHWA NATII. "Derivation of Likelihood-Ratio Tests for Guttman
Quasi-Simplex Covariance Structures." Psychometrika 31: 97-123; March 1966.
NICHOLSON,GEORGEE., JR. "Estimation of Parameters from Incomplete Multi-
variate Samples." Journal of the American Statistical Association 52: 523-26;
December 1957.
NORTON, DEE W. "Developments in Analysis of Variance and Design of Experiments."
Review of Educational Research 33: 490-500; December 1963.
PILLAI,K. C. SREEDIIARAN. Statistical Tables for Tests of Multivariate Hypotheses.
hlanila: University of the Philippines Statistical Center, 1960. 46 pp. (Offset)
PILLAI,K. C. SREEDIIARAN. "On the Distribution of the Largest of Seven Roots of a
Matrix in Multivariate Analysis." Biometrika 51: 270-75; June 1964.
PILLAI,K. C. SREEDHAHAN. "On the Distribution of the Largest Characteristic Root
of a Matrix in Multivariate Analysis." Biometrika 52: 405-14; December 1965.
POSTEN,H. O., and BARGMANN, R. E. "Power of the Likelihood-Ratio Test of the
General Linear Hypothesis in Multivariate Analysis." Biometrika 51: 467-80;
December 1964.
POTTHOFF,RICHARDF., and ROY, S. N. "A Generalized Multivariate Analysis of
Variance Model Useful Especially for Growth Curve Problems." Biometrika 51:
313-26; December 1964.
RAO, C. RADHAKRISIINA. [Link] Statistical Methods in Biometric Research. New
York: John Wiley and Sons, 1952. 390 pp.
Rao, C. KADIIAKRISHNA. "Analysis of Dispersion for Multiple Classified Data with
Unequal Numbers in Cells." Sankhya 15: 253-80; July 1955.
RAO. C. RADHAKRISHNA. "Analvsis of Dis~ersion with I n c o m ~ l e t e Observations on
o n e of the Characters." ~our'nal of the = ~ o y aStatistical
l society (Series B ) 18:
259-64; No. 2, 1956.
Rao, C. RADHAKRISIINA. "Multivariate Analysis: An Indispensable Statistical Aid
in Applied Research." Sankhya 22: 317-38; June 1960.
RAO, C. RADIIAKRISHNA. Linear Statistical Inference and Its Applications. New
York: John Wiley and Sons, 1965. 522 pp. ( a )
RAO, C. RADIIAKRISHNA. "The Theory of Least Squares When the Parameters Are
Stochastic and Its Application to the Analysis of Growth Curves." Biometrika 52:
447-58; December 1965. ( b )
ROY, J. "[Link] Procedure in Multivariate Analysis." Annals of Mathematical
Statistics 29: 1177-87 : December 1958.
ROY, J., and MURTHY,V. K. "Percentage Points of Wilks' L,,, and LC, Criteria."
Psychometrika 25: 243-50; September 1960.
ROY, S. N. Some Aspects of Multiz'ariate Analysis. New York: John Wiley and Sons,
1957. 214 pp.
ROY, S. N., and BARGMANN, R. E. "Tests of Multiple Independence and the Associ-
ated Confidence Bounds." Annals of .[Link] Statistics 29: 491-503; June
1958.
ROY, S. N., and GNANADESIKAN, R. "Some Contributions to ANOVA in One or More
Dimensions, I and 11." Annals of Mathematical Statistics 30: 304-17 and 318-40;
June 1959.
SEAL, HILARYL. Multivariate Statistical Analysis for Biologists. New York: John
Wiley and Sons, 1964. 207 pp.
SMITH,H. ; GNANADESIKAN, R. ; and HUGIIES,J. B. "Multivariate Analysis of Variance
(MANOVA) ." Biornetrics 18: 22-41; March 1962.
S R I V . ~ S T . ~J.V . ~N., "On the hlonotonicity Property of the Three Main Tests for
Multivariate Analysis of Variance." Journal of the Royal Statistical Society
(Series B ) 26: 77-81; No. 1, 1964.
SR~VASTAVA, J. N. "On Testing Hypotheses Regarding a Class of Covariance Struc-
tures." Psychometrika 31: 147-64; June 1966.
TATSUOKA, ~ I A U R I CM.,
E and TIEDEMAN, DAVIDV.
"Discriminant Analysis." Review
of Educational Research 24: 402-20; December 1954.

616
Decemher 1966 MULTIVARIATE
ANALYSIS

TRAWINSKI, IRENE~ O N ~ I Iand ~ NBSRGMANN,

, R. E. "Maximum Likelihood Estima-
tion with Incomplete Multivariate Data." Annals of .+lathematical Statistics 35:
647-57; June 1964.
TUKEY,J O H N W. "Dyadic ANOVA, an Analtsis of Variance for Vectors." Human
Biology 21: 65-110; May 1949.
VOTAW,DAVIDF., JR. "Testing Compound Symmetry in a Normal Multivariate Dis-
tribution." Annals of Wathematical Statistics 19: 447-73; December 1948.
WILK, M. B., and GNANADESIKAN, R. "Graphical Methods for Internal Comparison in
Multiresponse Experiments." Annals of Mathematical Statistics 35: 613-31; June
1964.
WILKS, S. S. "Sample Criteria for Testing Equality of hleans, Equality of Variances,
and Equality of Covariances in a Normal Multivariate Distribution." Annals of
Mathematical Statistics 17 : 257-81 ; September 1946.

Additional References
ANDERSON, T. W., and DAS GUPTA,S. "The >lonotonicity of the Power Functions of
Some Tests of Independence Between Two Sets of Variates." Annals of Mathe-
matical Statistics 35: 206-208; March 1964.
BARTLETT,% S.I."A Note on Tests of Significance in hlultivariate Analysis."
Proceedings 01 the Cambridge Philosophical Society 35: 180-85; April 1939.
BARTLETT,$1. S. "hlultivariate Analysis." Journal of the Royal Statistical Society,
Supplement 9: 176-97; No. 2, 1947.
BARTLETT, )I. S. "The Goodness of Fit of a Single Hypothetical Discriminant Func-
tion in the Case of Several Groups." Annals of Eugenics 16: 199-214; No. 3, 1951.
BARTLETT, $1. S. "A Note on the Multiplying Factors for Various X2 Approximations."
Journal of the Royal Statistical Society (Series B ) 16: 296-98; No. 2, 1954.
BENNETT,B. RI. "On Multivariate Sign Tests." Journal o j the Royal Statistical Society
(Series B ) 24: 159-61; No. 1, 1962.
BENNETT,B. 31. "On Multivariate Signed Rank Tests." Annals of the Institute o j
Statistical Mathematics 17: 55-61; No. 3, 1965.
CIIAN, MAN-WAI."A Multivariate Analysis of a n Experimental Design Involving a
Complete Set of Latin Squares." Psychometrika 29: 233-40; September 1964.
ELSTON,R. C., and GRIZZLE,J. E. "Estimation of Time-Response Curves and Their
Confidence Bands." Biometrics 18: 148-59; June 1962.
GEISSER,SEYMOUR."A Method for Testing Treatment Effects in the Presence of
Learning." Biometrics 15: 389-95; September 1959.
HOPKIKS,J. W., and CLAY, P. P. F. "Some Empirical Distributions of Bivariate
TZ and Homoscedasticity Criterion .lf Under Unequal Variance and Leptokurtosis."
Journal of the American Statistical Association 58: 1048-53; December 1963.
ITO, KOICHI."A Comparison of the Powers of Two Multivariate Analysis of Variance
Tests." Biometrika 49: 455-62; December 1962.
LAWLEY,D. N. "Tests of Significance for the Latent Roots of Covariance and Cor-
relation hlatrices." Biometrika 43: 128-36; June 1956.
LAWLEY,D. N. "Tests of Significance in Canonical Analysis." Biometrika 46: 59-66;
June 1959.
MICIIAEL,WILLIAMB., and PERRY,NORMANC. "The Comparability of the Simple
Discriminant Function and Multiple Regression Techniques." Journal of Experi-
mental Education 24: 297-301; June 1956.
MIKIIAIL, N. N. "A Comparison of Tests of the Wilks-Lawley Hypothesis in Multi.
variate Analysis." Biometrika 52: 149-56; June 1965.
RAO, C. RADHAKRISIINA. "Tests of Significance in Multivariate Analysis." Bionetrika
35: 58-79; March 1948.
RAO. C. RADIIAKRISHNA. "Statistical hlethods for Com~arison of Growth Curves."
Biometrics 14: 1-17; March 1958.
RAO, C. RADHAKRISHKA. "Some Problems Inlolving Linear Hlpotheses in h l u l t i ~ a r i -
ate Analtsis." Biometrika 46: 49-58; June 1959.
RAO, C. RADIIAKRISHNA. "Use of Discriminant and Allied Functions in Multivariate
Analysis." Sankhya (Series A ) 24: 149-54; May 1962.
WILLIAMS,E. J. "Significance Tests for Discriminant Functions and Linear Functional
Relationships." Biometrika 42: 360-81; December 1955.
[Link]

LINKED CITATIONS
- Page 1 of 9 -

You have printed the following article:

Multivariate Analysis
Elliot M. Cramer; R. Darrell Bock
Review of Educational Research, Vol. 36, No. 5, Methodology of Educational Research. (Dec.,
1966), pp. 604-617.
Stable URL:
[Link]

This article references the following linked citations. If you are trying to access articles from an
off-campus location, you may be required to first logon via your library web site to access JSTOR. Please
visit your library's website or contact a librarian to learn about options for remote access to JSTOR.

Bibliography

Maximum Likelihood Estimates for a Multivariate Normal Distribution when some

Observations are Missing
T. W. Anderson
Journal of the American Statistical Association, Vol. 52, No. 278. (Jun., 1957), pp. 200-203.
Stable URL:
[Link]

The Statistical Significance of Canonical Correlations

M. S. Bartlett
Biometrika, Vol. 32, No. 1. (Jan., 1941), pp. 29-37.
Stable URL:
[Link]

Programming Univariate and Multivariate Analysis of Variance

R. Darrell Bock
Technometrics, Vol. 5, No. 1. (Feb., 1963), pp. 95-117.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 2 of 9 -

A Method of Estimation of Missing Values in Multivariate Data Suitable for use with an
Electronic Computer
S. F. Buck
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 22, No. 2. (1960), pp.
302-306.
Stable URL:
[Link]

Randomization Tests for a Multivariate Two-Sample Problem

J. H. Chung; D. A. S. Fraser
Journal of the American Statistical Association, Vol. 53, No. 283. (Sep., 1958), pp. 729-735.
Stable URL:
[Link]

Multivariate Quantal Analysis

P. J. Claringbold
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 20, No. 2. (1958), pp.
398-405.
Stable URL:
[Link]

On the Performance of the Linear Discriminant Function

William G. Cochran
Technometrics, Vol. 6, No. 2. (May, 1964), pp. 179-190.
Stable URL:
[Link]

Some Classification Problems with Multivariate Qualitative Data

William G. Cochran; Carl E. Hopkins
Biometrics, Vol. 17, No. 1. (Mar., 1961), pp. 10-32.
Stable URL:
[Link]

Monotonicity of the Power Functions of Some Tests of the Multivariate Linear Hypothesis
S. Das Gupta; T. W. Anderson; G. S. Mudholkar
The Annals of Mathematical Statistics, Vol. 35, No. 1. (Mar., 1964), pp. 200-205.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 3 of 9 -

A Significance Test for the Separation of Two Highly Multivariate Small Samples
A. P. Dempster
Biometrics, Vol. 16, No. 1. (Mar., 1960), pp. 41-50.
Stable URL:
[Link]

Stepwise Multivariate Analysis of Variance Based on Principal Variables

A. P. Dempster
Biometrics, Vol. 19, No. 3. (Sep., 1963), pp. 478-490.
Stable URL:
[Link]

Multivariate Analysis and Agricultural Experiments

D. J. Finney
Biometrics, Vol. 12, No. 1. (Mar., 1956), pp. 67-71.
Stable URL:
[Link]

Charts of Some Upper Percentage Points of the Distribution of the Largest Characteristic
Root
D. L. Heck
The Annals of Mathematical Statistics, Vol. 31, No. 3. (Sep., 1960), pp. 625-642.
Stable URL:
[Link]

An Empirical Comparison of Pupil Achievement and Other Variables in Graded and

Ungraded Classes
Kenneth D. Hopkins; O. A. Oldridge; Malcolm L. Williamson
American Educational Research Journal, Vol. 2, No. 4. (Nov., 1965), pp. 207-215.
Stable URL:
[Link]

Relations Between Two Sets of Variates

Harold Hotelling
Biometrika, Vol. 28, No. 3/4. (Dec., 1936), pp. 321-377.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 4 of 9 -

Multiple Discriminant Analysis Applied to "Ways to Live" Ratings from Six Cultural Groups
Lyle V. Jones; R. Darrell Bock
Sociometry, Vol. 23, No. 2. (Jun., 1960), pp. 162-176.
Stable URL:
[Link]

Estimation of Parameters From Incomplete Multivariate Samples

George E. Nicholson, Jr.
Journal of the American Statistical Association, Vol. 52, No. 280. (Dec., 1957), pp. 523-526.
Stable URL:
[Link]

Developments in Analysis of Variance and Design of Experiments

Dee W. Norton
Review of Educational Research, Vol. 33, No. 5, Statistical Methodology. (Dec., 1963), pp.
490-500.
Stable URL:
[Link]

On the Distribution of the Largest of Seven Roots of a Matrix in Multivariate Analysis

K. C. Sreedharan Pillai
Biometrika, Vol. 51, No. 1/2. (Jun., 1964), pp. 270-275.
Stable URL:
[Link]

On the Distribution of the Largest Characteristic Root of a Matrix in Multivariate Analysis

K. C. Sreedharan Pillai
Biometrika, Vol. 52, No. 3/4. (Dec., 1965), pp. 405-414.
Stable URL:
[Link]

Power of the Likelihood-Ratio Test of the General Linear Hypothesis in Multivariate Analysis
H. O. Posten; R. E. Bargmann
Biometrika, Vol. 51, No. 3/4. (Dec., 1964), pp. 467-480.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 5 of 9 -

A Generalized Multivariate Analysis of Variance Model Useful Especially for Growth Curve
Problems
Richard F. Potthoff; S. N. Roy
Biometrika, Vol. 51, No. 3/4. (Dec., 1964), pp. 313-326.
Stable URL:
[Link]

Analysis of Dispersion with Incomplete Observations on One of the Characters

C. Radhakrishna Rao
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 18, No. 2. (1956), pp.
259-264.
Stable URL:
[Link]

The Theory of Least Squares When the Parameters are Stochastic and Its Application to the
Analysis of Growth Curves
C. Radhakrishna Rao
Biometrika, Vol. 52, No. 3/4. (Dec., 1965), pp. 447-458.
Stable URL:
[Link]

Step-Down Procedure in Multivariate Analysis

J. Roy
The Annals of Mathematical Statistics, Vol. 29, No. 4. (Dec., 1958), pp. 1177-1187.
Stable URL:
[Link]

Tests of Multiple Independence and the Associated Confidence Bounds

S. N. Roy; R. E. Bargmann
The Annals of Mathematical Statistics, Vol. 29, No. 2. (Jun., 1958), pp. 491-503.
Stable URL:
[Link]

Some Contributions to Anova in One or More Dimensions: I

S. N. Roy; R. Gnanadesikan
The Annals of Mathematical Statistics, Vol. 30, No. 2. (Jun., 1959), pp. 304-317.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 6 of 9 -

Multivariate Analysis of Variance (MANOVA)

H. Smith; R. Gnanadesikan; J. B. Hughes
Biometrics, Vol. 18, No. 1. (Mar., 1962), pp. 22-41.
Stable URL:
[Link]

On the Monotonicity Property of the Three Main Tests for Multivariate Analysis of Variance
J. N. Srivastava
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 26, No. 1. (1964), pp.
77-81.
Stable URL:
[Link]

Discriminant Analysis
Maurice M. Tatsuoka; David V. Tiedeman
Review of Educational Research, Vol. 24, No. 5, Statistical Methodology in Educational Research.
(Dec., 1954), pp. 402-420.
Stable URL:
[Link]

Maximum Likelihood Estimation with Incomplete Multivariate Data

Irene Monahan Trawinski; R. E. Bargmann
The Annals of Mathematical Statistics, Vol. 35, No. 2. (Jun., 1964), pp. 647-657.
Stable URL:
[Link]

Testing Compound Symmetry in a Normal Multivariate Distribution

David F. Votaw, Jr.
The Annals of Mathematical Statistics, Vol. 19, No. 4. (Dec., 1948), pp. 447-473.
Stable URL:
[Link]

Graphical Methods for Internal Comparisons in Multiresponse Experiments

M. B. Wilk; R. Gnanadesikan
The Annals of Mathematical Statistics, Vol. 35, No. 2. (Jun., 1964), pp. 613-631.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 7 of 9 -

Sample Criteria for Testing Equality of Means, Equality of Variances, and Equality of
Covariances in a Normal Multivariate Distribution
S. S. Wilks
The Annals of Mathematical Statistics, Vol. 17, No. 3. (Sep., 1946), pp. 257-281.
Stable URL:
[Link]

Additional References

Monotonicity of the Power Functions of Some Tests of Independence between Two Sets of
Variates
T. W. Anderson; S. Das Gupta
The Annals of Mathematical Statistics, Vol. 35, No. 1. (Mar., 1964), pp. 206-208.
Stable URL:
[Link]

Multivariate Analysis
M. S. Bartlett
Supplement to the Journal of the Royal Statistical Society, Vol. 9, No. 2. (1947), pp. 176-197.
Stable URL:
[Link]

A Note on the Multiplying Factors for Various #2 Approximations

M. S. Bartlett
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 16, No. 2. (1954), pp.
296-298.
Stable URL:
[Link]

On Multivariate Sign Tests

B. M. Bennett
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 24, No. 1. (1962), pp.
159-161.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 8 of 9 -

Estimation of Time-Response Curves and Their Confidence Bands

R. C. Elston; J. E. Grizzle
Biometrics, Vol. 18, No. 2. (Jun., 1962), pp. 148-159.
Stable URL:
[Link]

A Method for Testing Treatment Effects in the Presence of Learning

Seymour Geisser
Biometrics, Vol. 15, No. 3. (Sep., 1959), pp. 389-395.
Stable URL:
[Link]

Some Empirical Distributions of Bivariate T2 and Homoscedasticity Criterion M Under

Unequal Variance and Leptokurtosis
J. W. Hopkins; P. P. F. Clay
Journal of the American Statistical Association, Vol. 58, No. 304. (Dec., 1963), pp. 1048-1053.
Stable URL:
[Link]

A Comparison of the Powers of Two Multivariate Analysis of Variance Tests

Koichi Ito
Biometrika, Vol. 49, No. 3/4. (Dec., 1962), pp. 455-462.
Stable URL:
[Link]

Tests of Significance for the Latent Roots of Covariance and Correlation Matrices
D. N. Lawley
Biometrika, Vol. 43, No. 1/2. (Jun., 1956), pp. 128-136.
Stable URL:
[Link]

Tests of Significance in Canonical Analysis

D. N. Lawley
Biometrika, Vol. 46, No. 1/2. (Jun., 1959), pp. 59-66.
Stable URL:
[Link]
[Link]

LINKED CITATIONS
- Page 9 of 9 -

A Comparison of Tests of the Wilks-Lawley Hypothesis in Multivariate Analysis

N. N. Mikhail
Biometrika, Vol. 52, No. 1/2. (Jun., 1965), pp. 149-156.
Stable URL:
[Link]

Tests of Significance in Multivariate Analysis

C. Radhakrishna Rao
Biometrika, Vol. 35, No. 1/2. (May, 1948), pp. 58-79.
Stable URL:
[Link]

Some Statistical Methods for Comparison of Growth Curves

C. Radhakrishna Rao
Biometrics, Vol. 14, No. 1. (Mar., 1958), pp. 1-17.
Stable URL:
[Link]

Some Problems Involving Linear Hypotheses in Multivariate Analysis

C. Radhakrishna Rao
Biometrika, Vol. 46, No. 1/2. (Jun., 1959), pp. 49-58.
Stable URL:
[Link]

Significance Tests for Discriminant Functions and Linear Functional Relationships

E. J. Williams
Biometrika, Vol. 42, No. 3/4. (Dec., 1955), pp. 360-381.
Stable URL:
[Link]