NCSS Statistical Software NCSS.

com

Chapter 415

Multivariate Analysis of Variance

(MANOVA)

Introduction
Multivariate analysis of variance (MANOVA) is an extension of common analysis of variance (ANOVA). In
ANOVA, differences among various group means on a single-response variable are studied. In MANOVA, the
number of response variables is increased to two or more. The hypothesis concerns a comparison of
vectors of group means. When only two groups are being compared, the results are identical to Hotelling’s
T² procedure.
The multivariate extension of the F-test is not completely direct. Instead, several test statistics are available,
such as Wilks’ Lambda and Lawley’s trace. The actual distributions of these statistics are difficult to calculate,
so we rely on approximations based on the F-distribution.

Technical Details
A MANOVA has one or more factors (each with two or more levels) and two or more dependent variables.
The calculations are extensions of the general linear model approach used for ANOVA.
Unlike the univariate situation in which there is only one statistical test available (the F-ratio), the
multivariate situation provides several alternative statistical tests. We will describe these tests in terms of
two matrices, H and E. H is called the hypothesis matrix and E is the error matrix. These matrices may be
computed using a number of methods. In NCSS, we use the standard general linear models (GLM) approach
in which a sum of squares and cross-products matrix is computed. This matrix is based on the dependent
variables and independent variables generated for each degree of freedom in the model. It may be
partitioned according to the terms in the model.

415-1
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

MANOVA Test Statistics

For a particular p-variable multivariate test, assume that the matrices H and E have h and e degrees of
freedom, respectively. Four tests may be defined as follows. See Seber (1984) for details. Let 𝜃𝜃𝑖𝑖 , 𝜙𝜙𝑖𝑖 , and 𝜆𝜆𝑖𝑖
be the eigenvalues of H(E+H)-1, HE-1, and E(E+H)-1 respectively. Note that these eigenvalues are related as
follows:

𝜙𝜙𝑖𝑖
𝜃𝜃𝑖𝑖 = 1 − 𝜆𝜆𝑖𝑖 =
1 + 𝜙𝜙𝑖𝑖

𝜃𝜃𝑖𝑖 1 − 𝜆𝜆𝑖𝑖
𝜙𝜙𝑖𝑖 = =
1 − 𝜃𝜃𝑖𝑖 𝜆𝜆𝑖𝑖

1
𝜆𝜆𝑖𝑖 = 1 − 𝜃𝜃𝑖𝑖 =
1 + 𝜙𝜙𝑖𝑖

Wilks’ Lambda
Define Wilks’ Lambda as follows:
|𝐸𝐸|
Λ 𝑝𝑝,ℎ,𝑒𝑒 =
|𝐸𝐸 + 𝐻𝐻|
𝑝𝑝

= ��1 − 𝜃𝜃𝑗𝑗 �
𝑗𝑗=1

with e ≥ p.
The following approximation based on the F-distribution is used to determine significance levels:

(𝑓𝑓𝑓𝑓 − 𝑔𝑔)�1 − Λ1/𝑡𝑡 �

𝐹𝐹𝑝𝑝ℎ,𝑓𝑓𝑓𝑓−𝑔𝑔 =
𝑝𝑝ℎΛ1/𝑡𝑡
where

1
𝑓𝑓 = 𝑒𝑒 − (𝑝𝑝 − ℎ + 1)
2

𝑝𝑝ℎ − 2
𝑔𝑔 =
2

𝑝𝑝2 ℎ2 − 4 2 2
�
𝑡𝑡 = � 𝑝𝑝2 + ℎ2 − 5 if 𝑝𝑝 + ℎ − 5 > 0
1 otherwise

This approximation is exact if p or h ≥ 2.

415-2
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Lawley - Hotelling Trace

The trace statistic, 𝑇𝑇𝑔𝑔2 , is defined as follows:

𝑠𝑠

𝑇𝑇𝑔𝑔2 = 𝑒𝑒 � 𝜙𝜙𝑗𝑗
𝑗𝑗=1

where

𝑠𝑠 = min(𝑝𝑝, ℎ)

The following approximation based on the F-distribution is used to determine significance levels:
𝑇𝑇𝑔𝑔2
𝐹𝐹𝑎𝑎,𝑏𝑏 =
𝑐𝑐𝑐𝑐
where

𝑎𝑎 = 𝑝𝑝ℎ

𝑏𝑏 = 4 + (𝑎𝑎 + 2)/(𝐵𝐵 − 1)

𝑎𝑎(𝑏𝑏 − 2)
𝑐𝑐 =
𝑏𝑏(𝑒𝑒 − 𝑝𝑝 − 1)

(𝑒𝑒 + ℎ − 𝑝𝑝 − 1)(𝑒𝑒 − 1)
𝐵𝐵 =
(𝑒𝑒 − 𝑝𝑝 − 3)(𝑒𝑒 − 𝑝𝑝)

Pillai’s Trace
Pillai’s trace statistic, 𝑉𝑉 (𝑠𝑠) , is defined as follows:

𝑠𝑠
(𝑠𝑠)
𝑉𝑉 = � 𝜃𝜃𝑗𝑗 = tr(𝐻𝐻(𝐸𝐸 + 𝐻𝐻)−1 )
𝑗𝑗=1

where

𝑠𝑠 = min(𝑝𝑝, ℎ)

The following approximation based on the F-distribution is used to determine significance levels:

(2𝑛𝑛 + 𝑠𝑠 + 1)𝑉𝑉 (𝑠𝑠)

𝐹𝐹𝑠𝑠(2𝑚𝑚+𝑠𝑠+1),𝑠𝑠(2𝑛𝑛+𝑠𝑠+1) =
(2𝑚𝑚 + 𝑠𝑠 + 1)(𝑠𝑠 − 𝑉𝑉 (𝑠𝑠) )

415-3
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

where

𝑠𝑠 = min(𝑝𝑝, ℎ)

𝑚𝑚 = (|𝑝𝑝 − ℎ| − 1)/2

𝑛𝑛 = (𝑒𝑒 − 𝑝𝑝 − 1)/2

Roy’s Largest Root

Roy’s largest root, 𝜙𝜙𝑚𝑚𝑚𝑚𝑚𝑚 , is defined as the largest of the 𝜙𝜙𝑖𝑖 ’s. The following approximation based on the F-
distribution is used to determine significance levels:

2𝜈𝜈2 + 2
𝐹𝐹(2𝜈𝜈1 +2),(2𝜈𝜈2 +2) = 𝜙𝜙
2𝜈𝜈1 + 2 𝑚𝑚𝑚𝑚𝑚𝑚

where

𝑠𝑠 = min(𝑝𝑝, ℎ)

𝜈𝜈1 = (|𝑝𝑝 − ℎ| − 1)/2

𝜈𝜈2 = (𝑒𝑒 − 𝑝𝑝 − 1)/2

Which Test to Use

When the hypothesis degrees of freedom, h, is one, all four test statistics will lead to identical results. When
h>1, the four statistics will usually lead to the same result. When they do not, the following guidelines from
Tabachnick (1989) may be of some help.
Wilks’ Lambda, Lawley’s trace, and Roy’s largest root are often more powerful than Pillai’s trace if h>1 and
one dimension accounts for most of the separation among groups. Pillai’s trace is more robust to
departures from assumptions than the other three.
Tabachnick (1989) provides the following checklist for conducting a MANOVA. We suggest that you consider
these issues and guidelines carefully.

Assumptions and Limitations

The following assumptions are made when using a MANOVA.
1. The response variables are continuous.
2. The residuals follow the multivariate-normal probability distribution with means equal to zero.
3. The variance-covariance matrices of each group of residuals are equal.
4. The individuals are independent.

415-4
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Multivariate Normality and Outliers

MANOVA is robust to modest amount of skewness in the data. A sample size that produces 20 degrees of
freedom in the univariate F-test is adequate to ensure robustness. Non-normality caused by the presence of
outliers can cause severe problems that even the robustness of the test will not overcome. You should
screen your data for outliers and run it through various univariate and multivariate normality tests and plots
to determine if the normality assumption is reasonable.

Homogeneity of Covariance Matrices

MANOVA makes the assumption that the within-cell (group) covariance matrices are equal. If the design is
balanced so that there is an equal number of observations in each cell, the robustness of the MANOVA tests
is guaranteed. If the design is unbalanced, you should test the equality of covariance matrices using Box’s M
test. If this test is significant at less than 0.001, there may be severe distortion in the alpha levels of the
tests. You should only use Pillai’s trace criterion in this situation.

Linearity
MANOVA assumes linear relationships among the dependent variables within a particular cell. You should
study scatter plots of each pair of dependent variables using a different color for each level of a factor. Look
carefully for curvilinear patterns and for outliers. The occurrence of curvilinear relationships will reduce the
power of the MANOVA tests.

Multicollinearity and Singularity

Multicollinearity occurs when one dependent variable is almost a weighted average of the others. This
collinearity may only show up when the data are considered one cell at a time. The R²-Other Y’s in the
Within-Cell Correlations Analysis report lets you determine if multicollinearity is a problem. If this R² value is
greater than .99 for any variable, you should take corrective action (remove one of the variables). To correct
for multicollinearity, begin removing the variables one at a time until all of the R²’s are less than .99. Do not
remove them all at once! Singularity is the extreme form of multicollinearity in which the R² value is one.
Forms of multicollinearity may show up when you have very small cell sample sizes (when the number of
observations is less than the number of variables). In this case, you must reduce the number of dependent
variables.

415-5
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Data Structure
The data must be entered in a format that places the dependent variables and values of each factor side by
side. An example of the data for a MANOVA design is shown in the table below. In this example, WRATR and
WRATA are the two dependent variables. Treatment and Disability are two factor variables. This database is
stored in the file MANOVA1.

MANOVA1 Dataset (Subset)

WRATR WRATA Treatment Disability

115 108 1 1
98 105 1 1
107 98 1 1
90 92 2 1
85 95 2 1
80 81 2 1
100 105 1 2
105 95 1 2
95 98 1 2
70 80 2 2
. . . .
. . . .
. . . .

Unequal Sample Size and Missing Data

You should begin by screening your data. Pay particular attention to patterns of missing values. When using
MANOVA, you should have more observations per factor category than you have dependent variables so
that you can test the equality of covariance matrices using Box’s M test.
NCSS ignores rows with missing values. If it appears that most of the missing values occur in one or two
variables, you might want to leave these out of the analysis in order to obtain more data and hence more
power.
NCSS uses the GLM procedure for calculating the hypothesis and error matrices. Each matrix is calculated
as if it were fit last in the model. This is the recommended way of obtaining these matrices. This method is
valid even when the sample sizes for the various groups are unequal.

415-6
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Example 1 – Multivariate Analysis of Variance

This section presents an example of how to run an analysis of the data contained in the MANOVA1 dataset.

Setup
To run this example, complete the following steps:

1 Open the MANOVA1 example dataset

• From the File menu of the NCSS Data window, select Open Example Data.
• Select MANOVA1 and click OK.

2 Specify the Multivariate Analysis of Variance (MANOVA) procedure options

• Find and open the Multivariate Analysis of Variance (MANOVA) procedure using the menus or the
Procedure Navigator.
• The settings for this example are listed below and are stored in the Example 1 settings file. To load
these settings to the procedure window, click Open Example Settings File in the Help Center or File
menu.

Variables Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Response Variables......................................... WRATR-WRATA

Factor Variable 1 ............................................. Treatment
Factor Variable 2 ............................................. Disability

3 Run the procedure

• Click the Run button to perform the calculations and generate the output.

Expected Mean Squares

Expected Mean Squares

─────────────────────────────────────────────────────────────────────────
Source Term Denominator Expected
Term DF Fixed? Term Mean Square
────────────────────────────────────────────────────────────────────────────────────
A: Treatment 1 Yes S(AB) S+bsA
B: Disability 2 Yes S(AB) S+asB
AB 2 Yes S(AB) S+sAB
S(AB) 12 No S
─────────────────────────────────────────────────────────────────────────
Note: Expected Mean Squares are for the balanced cell-frequency case.

The Expected Mean Square expressions are provided to show the appropriate error term for each factor.
The correct error term for a factor is that term that is identical except for the factor being tested.

Source Term
The source of variation or term in the model.

415-7
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

DF
The degrees of freedom. The number of observations “used” by this term.

Term Fixed?
Indicates whether the term is “fixed” or “random.”

Denominator Term
Indicates the term used as the denominator in the F-ratio.

Expected Mean Square

This expression represents the expected value of the corresponding mean square if the design was
completely balanced. “S” represents the expected value of the mean square error (sigma). The uppercase
letters represent either the adjusted sum of squared treatment means if the factor is fixed, or the variance
component if the factor is random. The lowercase letter represents the number of levels for that factor, and
“s” represents the number of replications of the experimental layout.
These EMS expressions are provided to determine the appropriate error term for each factor. The correct
error term for a factor is that term whose EMS is identical except for the factor being tested.

MANOVA Tests

─────────────────────────────────────────────────────────────────────────
Term(DF): Name Decision
Test Statistic Test Value DF1 DF2 F-Ratio P-Value (α = 0.05)
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

A(1): Treatment
Wilks' Lambda 0.137721 2 11 34.44 0.000018 Reject
Lawley - Hotelling Trace 6.261036 2 11 34.44 0.000018 Reject
Pillai's Trace 0.862279 2 11 34.44 0.000018 Reject
Roy's Largest Root 6.261036 2 11 34.44 0.000018 Reject
F (WRATR) 2090.888889 1 12 46.12 0.000019 Reject
F (WRATA) 1494.222222 1 12 33.25 0.000089 Reject

B(2): Disability
Wilks' Lambda 0.255263 4 22 5.39 0.003528 Reject
Lawley - Hotelling Trace 2.895034 4 20 7.24 0.000896 Reject
Pillai's Trace 0.750481 4 24 3.60 0.019460 Reject
Roy's Largest Root 2.887241 2 12 17.32 0.000290 Reject
F (WRATR) 260.388889 2 12 5.74 0.017784 Reject
F (WRATA) 563.388889 2 12 12.54 0.001151 Reject

AB(2)
Wilks' Lambda 0.908068 4 22 0.27 0.893037 Accept
Lawley - Hotelling Trace 0.100954 4 20 0.25 0.904790 Accept
Pillai's Trace 0.092192 4 24 0.29 0.881598 Accept
Roy's Largest Root 0.098039 2 12 0.59 0.570550 Accept
F (WRATR) 1.055556 2 12 0.02 0.977029 Accept
F (WRATA) 26.388889 2 12 0.59 0.571116 Accept
─────────────────────────────────────────────────────────────────────────

This report gives the results of the various significance tests. Usually, the four multivariate tests will lead to
the same conclusions. When they do not, refer to the discussion of these tests found earlier in this chapter.

415-8
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Once a multivariate test has found a term significant, use the univariate ANOVA to determine which of the
variables and factors are “causing” the significance.

Term(DF): Name
The term in the design model with the degrees of freedom of the term in parentheses. For main effects, the
variable name is also given.

Test Statistic
The name of the statistical test shown on this row of the report. The four multivariate tests are followed by
the univariate F-tests of each variable.

Test Value
The value of the test statistic.

DF1
The numerator degrees of freedom of the F-ratio corresponding to this test.

DF2
The denominator degrees of freedom of the F-ratio corresponding to this test.

F-Ratio
The value of the F-test corresponding to this test. In some cases, this is an exact test. In other cases, this is
an approximation to the exact test. See the discussion of each test to determine if it is exact or approximate.

P-Value
The significance level of the above F-ratio. The probability of an F-ratio larger than that obtained by this
analysis. For example, to test at an alpha of 0.05, this probability would have to be less than 0.05 to make
the F-ratio significant.

Decision (α = 0.05)
The decision to accept or reject the null hypothesis at the given level of significance. Note that you specify
the level of significance when you select Alpha.

415-9
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Correlations and Within-Group Covariances

Correlations and Within-Group Covariances

─────────────────────────────────────────────────────────────────────────
WRATR WRATA
───────────────────────────────────────────────
WRATR 45.33333 2.583333
WRATA 0.0572313 44.94444
─────────────────────────────────────────────────────────────────────────
The correlations are shown in the off-diagonals in the lower-left half of the matrix.
The within-group covariances are shown on the diagonal and in the off-diagonals in the upper-right half of the matrix.

This report displays the correlations and covariances formed by averaging across all of the individual group
covariance matrices. The correlations are shown in the lower-left half of the matrix. The within-group
covariances are shown on the diagonal and in the upper-right half of the matrix.

Within-Cell Correlations Analysis

Within-Cell Correlations Analysis

─────────────────────────────────────────────────────────────────────────
R-Squared Canonical Percent Cumulative
Variable Other Y's Variate Eigenvalue of Total Total
──────────────────────────────────────────────────────────────────────────────────────────────────────
WRATR 0.003275 1 1.057231 52.86 52.86
WRATA 0.003275 2 0.942769 47.14 100.00
─────────────────────────────────────────────────────────────────────────

This report analyzes the within-cell correlation matrix. It lets you diagnose multicollinearity problems as well
as determine the number of dimensions that are being used. This is useful in determining if Pillai’s trace
should be used.

R-Squared Other Y’s

This is the R-Squared index of this variable with the other variables. When this value is larger than 0.99,
severe multicollinearity problems exist. If this happens, you should remove the variable with the largest R-
Squared and re-run your analysis.

Canonical Variate
The identification numbers of the canonical variates that are generated during the analysis. The total
number of variates is the smaller of the number of variables and the number of degrees of freedom in the
model.

Eigenvalue
The eigenvalues of the within correlation matrix. Note that this value is not associated with the variable at
the beginning of the row, but rather with the canonical variate number directly to the left.

415-10
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Percent of Total
The percent that the eigenvalue is of the total. Note that the sum of the eigenvalues will equal the number
of variates. If the percentage accounted for by the first eigenvalue is relatively large (70 or 80 percent),
Pillai's trace will be less powerful than the other three multivariate tests.

Cumulative Total
The cumulative total of the Percent of Total column.

Univariate Analysis of Variance (ANOVA) Tables

Analysis of Variance (ANOVA) Table for WRATR

─────────────────────────────────────────────────────────────────────────
Source Sum of Mean Power
Term DF Squares Mean Square F-Ratio P-Value (α = 0.05)
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
A: Treatment 1 2090.889 2090.889 46.12 0.000019* 0.999988
B: Disability 2 520.7778 260.3889 5.74 0.017784* 0.763859
AB 2 2.111111 1.055556 0.02 0.977029 0.052757
S 12 544 45.33333
Total (Adjusted) 17 3157.778
Total 18
─────────────────────────────────────────────────────────────────────────
* Term significant at alpha = 0.05

Analysis of Variance (ANOVA) Table for WRATA

─────────────────────────────────────────────────────────────────────────
Source Sum of Mean Power
Term DF Squares Mean Square F-Ratio P-Value (α = 0.05)
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
A: Treatment 1 1494.222 1494.222 33.25 0.000089* 0.999519
B: Disability 2 1126.778 563.3889 12.54 0.001151* 0.981144
AB 2 52.77778 26.38889 0.59 0.571116 0.125682
S 12 539.3333 44.94444
Total (Adjusted) 17 3213.111
Total 18
─────────────────────────────────────────────────────────────────────────
* Term significant at alpha = 0.05

This is the standard ANOVA report as documented in the General Linear Models chapter. A separate report
is displayed for each of the response variables.

415-11
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Means and Standard Errors

Means and Standard Errors of WRATR

─────────────────────────────────────────────────────────────────────────
Standard
Term Count Mean Error*
───────────────────────────────────────────────────────
All 18 89.11111

A: Treatment
1 9 99.88889 2.234687
2 9 78.33334 2.234687

B: Disability
1 6 95.83334 2.736922
2 6 88.83334 2.736922
3 6 82.66666 2.736922

AB: Treatment, Disability

1, 1 3 106.6667 3.870592
1, 2 3 100 3.870592
1, 3 3 93 3.870592
2, 1 3 85 3.870592
2, 2 3 77.66666 3.870592
2, 3 3 72.33334 3.870592
─────────────────────────────────────────────────────────────────────────
* The standard errors are calculated using the mean square error from the corresponding Analysis of Variance (ANOVA) table.

Means and Standard Errors of WRATA

─────────────────────────────────────────────────────────────────────────
Standard
Term Count Mean Error*
───────────────────────────────────────────────────────
All 18 87.22222

A: Treatment
1 9 96.33334 2.234687
2 9 78.11111 2.234687

B: Disability
1 6 96.5 2.736922
2 6 88 2.736922
3 6 77.16666 2.736922

AB: Treatment, Disability

1, 1 3 103.6667 3.870592
1, 2 3 99.33334 3.870592
1, 3 3 86 3.870592
2, 1 3 89.33334 3.870592
2, 2 3 76.66666 3.870592
2, 3 3 68.33334 3.870592
─────────────────────────────────────────────────────────────────────────
* The standard errors are calculated using the mean square error from the corresponding Analysis of Variance (ANOVA) table.

This report provides the least-squares means and standard errors for each variable. Note that the standard
errors are calculated using the mean square error from the ANOVA table. They are not the standard errors
that would be calculated from the individual cells.

415-12
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

Means Plots

Means Plots
─────────────────────────────────────────────────────────────────────────

415-13
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Multivariate Analysis of Variance (MANOVA)

This section provides the means plots for main effects and interactions.

415-14
© NCSS, LLC. All Rights Reserved.