The multivariate model has the form
where is the N
p vector of responses, for p > 1;
is the N
k design matrix;
is the k
p matrix of model parameters that correspond to the columns of
and
; and
is an N
p vector of errors, where
In PROC GLMPOWER, the model parameters are not specified directly, but rather indirectly as
, which represents either conjectured response means or typical response values for each design profile. The
values are manifested as the collection of dependent variables in the MODEL
statement. The matrix
is obtained from
according to the least squares equation,
Note that, in general, there is not a one-to-one mapping between and
. Many different scenarios for
might lead to the same
. If you specify
with the intention of representing cell means, keep in mind that PROC GLMPOWER allows scenarios that are not valid cell means according to the model that is specified in the MODEL
statement. For example, if
exhibits an interaction effect but the corresponding interaction term is left out of the model, then the cell means (
) that are derived from
differ from
. In particular, the cell means that are derived in this way are the projection of
onto the model space.
It is convenient in power analysis to parameterize the design matrix in three parts,
, defined as follows:
The q k essence design matrix
is the collection of unique rows of
. Its rows are sometimes referred to as "design profiles." Here, q
N is defined simply as the number of unique rows of
.
The q 1 weight vector
reveals the relative proportions of design profiles, and
. Row i of
is to be included in the design
times for every
times that row j is included. The weights are assumed to be standardized (that is, they sum up to 1).
The total sample size is N. This is the number of rows in . If you gather
copies of the ith row of
, for i =
, then you end up with
.
The preceding quantities are derived from PROC GLMPOWER syntax as follows:
It is useful to express the crossproduct matrix in terms of these three parts,
because this expression factors out the portion (N) that depends on sample size and the portion () that depends only on the design structure.
A general linear hypothesis for the univariate model has the form
where is an l
k between-subject contrast matrix with rank
,
is a p
m within-subject contrast matrix with rank
, and
is an l
m null contrast matrix (usually just a matrix of zeros).
Note that model effect tests are just between-subject contrasts that use special forms of , combined with an
that is the p
1 mean transformation vector of the dependent variables (a vector of values all equal to
). Thus, this scheme covers both effect tests (which are specified in the MODEL
statement and the EFFECTS=
option in the POWER
statement) and custom between-subject contrasts (which are specified in the CONTRAST
statement).
The matrix is often referred to as the dependent variable transformation and is specified in the MANOVA
or REPEATED
statement.
The model degrees of freedom are equal to the rank of
, denoted
. The error degrees of freedom
are equal to N –
.
The hypothesis sum of squares in the univariate model generalizes to the hypothesis SSCP matrix in the multivariate model,
The error sum of squares in the univariate model generalizes to the error SSCP matrix in the multivariate model,
where
and
The population counterpart of is
and the population counterpart of is
The elements of are specified in the MATRIX=
and STDDEV=
options and identified in the CORRMAT=
, CORRS=
, COVMAT=
, and SQRTVAR=
options in the POWER
statement.
The power and sample size computations for all the tests that are supported in the MTEST=
option in the POWER
statement are based on and
. The following two subsections cover the computational methods and formulas for the multivariate and univariate tests that
are supported in the MTEST=
and UEPSDEF=
options in the POWER
statement.
Power computations for multivariate tests are based on O’Brien and Shieh (1992) (for METHOD= OBRIENSHIEH) and Muller and Peterson (1984) (for METHOD= MULLERPETERSON).
Let s = , the smaller of the between-subject and within-subject contrast degrees of freedom. Critical value computations assume that
under
, the test statistic F is distributed as
, where
if s = 1 but depends on the choice of test if s > 1. Power computations assume that under
, F is distributed as
, where the noncentrality
depends on
,
, the choice of test, and the power computation method.
Formulas for the test statistic F, denominator degrees of freedom , and noncentrality
for all combinations of dimensions, tests, and methods are given in the following subsections.
The power in each case is computed as
Computed power is exact for some cases and approximate for others. Sample size is computed by inverting the power equation.
Let , and define
as the
vector of ordered positive eigenvalues of
,
, where
. The population equivalent is
where is the s
1 vector of ordered positive eigenvalues of
,
for
.
Case 1: s = 1
When s = 1, all three multivariate tests (MTEST=
HLT, MTEST=
PT, and MTEST=
WILKS) are equivalent. The test statistic is F = , where
.
When the dependent variable transformation has a single degree of freedom (), METHOD=
OBRIENSHIEH and METHOD=
MULLERPETERSON are the same, computing exact power by using noncentrality
. The sample size must satisfy
.
When the dependent variable transformation has more than one degree of freedom but the between-subject contrast has a single
degree of freedom (), METHOD=
OBRIENSHIEH computes exact power by using noncentrality
, and METHOD=
MULLERPETERSON computes approximate power by using
The sample size must satisfy .
Case 2: s > 1
When both the dependent variable transformation and the between-subject contrast have more than one degree of freedom (s > 1), METHOD=
OBRIENSHIEH computes the noncentrality as , where
is the primary noncentrality. The form of
depends on the choice of test statistic.
METHOD=
MULLERPETERSON computes the noncentrality as , where
has the same form as
except that
is replaced by
Computed power is approximate for both methods when s > 1.
Hotelling-Lawley Trace (MTEST= HLT) When s > 1
If , then the denominator degrees of freedom for the Hotelling-Lawley trace are
,
where
which is the same as in O’Brien and Shieh (1992) and is due to McKeon (1974).
If , then
,
which is the same as both in O’Brien and Shieh (1992) and
in Muller and Peterson (1984) and is due to Pillai and Samson (1959).
The primary noncentrality is
The sample size must satisfy
If , then the test statistic is
where
and
If , then the test statistic is
Pillai’s Trace (MTEST= PT) When s > 1
The denominator degrees of freedom for Pillai’s trace are
The primary noncentrality is
The sample size must satisfy
The test statistic is
where
Wilks’ Lambda (MTEST= WILKS) When s > 1
The denominator degrees of freedom for Wilks’ lambda are
where
The primary noncentrality is
The sample size must satisfy
The test statistic is
where
Power computations for univariate tests are based on Muller et al. (2007) and Muller and Barton (1989).
The test statistic is
Critical value computations assume that under , F is distributed as
, where
and
depend on the choice of test.
The four tests for the univariate approach to repeated measures differ in their assumptions about the sphericity of
,
Power computations assume that under , F is distributed as
.
Formulas for and
for each test and formulas for
,
, and
are given in the following subsections.
The power in each case is approximated as
Sample size is computed by inverting the power equation.
The sample size must be large enough to yield ,
,
, and
.
Because these univariate tests are biased, the achieved significance level might differ from the nominal significance level.
The actual alpha is computed in the same way as the power, except that the noncentrality parameter is set to 0.
Define as the vector of ordered eigenvalues of
,
, where
, and define
as the jth eigenvector of
. Critical values and power computations are based on the following intermediate parameters:
The degrees of freedom and noncentrality in the noncentral F approximation of the test statistic are computed as follows:
Uncorrected Test
The uncorrected test assumes sphericity , in which case the null F distribution is exact, with the following degrees of freedom:
Greenhouse-Geisser Adjustment (MTEST= UNCORR)
The Greenhouse-Geisser adjustment to the uncorrected test reduces degrees of freedom by the MLE of the sphericity,
An approximation for the expected value of is used to compute the degrees of freedom for the null F distribution,
where
Huynh-Feldt Adjustments (MTEST= HF)
The Huynh-Feldt adjustment reduces degrees of freedom by a nearly unbiased estimate of the sphericity,
where
The value of is truncated if necessary to be at least
and at most 1.
An approximation for the expected value of is used to compute the degrees of freedom for the null F distribution,
where
and
Box Conservative Test (MTEST= BOX)
The Box conservative test assumes the worst case for sphericity, , leading to the following degrees of freedom for the null F distribution: