- or —
ANALYSIS( REPEATED  (CONDITIONAL] ) /
(UNCONDITIONAL]

PARTITION subdivides the degrees of freedom of a factor.
PARTITION( factorname) [- (dfl, df2,..)]. /

CONTRAST indicates the type of contrast desired for a factor.

DEVIATION [(refcat)]
DIFFERENCE
HELMERT
CONTRAST( factorname) < SIMPLE ((refcat)]
REPEATED
| POLYNOMIAL ((metric)]
SPECIAL (matrix)

PREN ENERENE NOW ONENINIH
—

ERROR specifies the error term to be used in the model.

(1 WITHIN or W
( RESIDUAL or R
ERROR < | WITHIN 4 RESIDUAL or WR
n

DESIGN specifies the design model to be analyzed.
DESIGN - effectl, effect2,../

The DESIGN specification should be the last subcommand of a complete MANOVA run. All the
computational and output options are applied to the subseguent DESIGN models unless
overridden.

As an example of specifications for MANOVA, consider the following:

MANOVA Y BY A(1,3) B(1,4) WITH X/
PRINT-CELLINFO ( MEANS) /
METHOD<ESTIMATION ( BALANCED) /
DESIGN<A,B/
METHOD<ESTIMATION(OR) /
DESIGN-A,B,A BY B/

—

An analysis of covariance model is specified with Y as the dependent variable, X as the covariate,
and A and B as factor variables with three and four levels respectively. The PRINT subcommand
reguests cell information. The METHOD subcommand indicates that a special balanced
processing method be used for parameter estimation. These two options apply to the first DESIGN
specification, which reguests a main effects model. The second METHOD subcommand reguests
the (default) OR method for estimating the parameters in the second DESIGN specification (a full
model). The PRINT subcommand applied to the first DESIGN will also apply to the second
DESIGN.

Note that if the last command is not a DESIGN specification, MANOVA will generate a full
model design specification for the problem.

1.2 UNIVARIATE ANALYSIS OF VARIANCE

The basic features of MANOVA useful for univariate analysis of variance are illustrated in the
following example taken from Winer (1971, p. 436). An experiment was conducted to evaluate the
relative effectiveness of three drugs (Factor DRUG) in bringing about behavioral changes in two
categories of patients (Factor CAT). Three patients of each category were assigned at random to

one of three drugs, and criterion ratings (Y) were made for each patient. The data are given in
Table 1.2.

Table 1.2

CAT

MANOVA 3