You are on page 1of 8

Multivariate Analysis of Variance (MANOVA) (copy this section to methodology)

Multivariate analysis of variance evaluates differences among centroids for a set of dependent
variables when there are two or more levels of independent variables (groups). This technique
provides a multivariate test to compare the mean vectors of k random samples for significant
difference when the levels of the grouping variable are two or more: in this study it was used
when the levels were more than two. For the case of two levels of the independent term (group)
in the model, the Hotellings T
2
described earlier would be used to test the equality of the mean
vectors between the two samples.
Consider k independent random samples of size n obtained from pvariate normal populations.
The model for each observation is:
(8) , , 2 , 1 ; , , 2 , 1 , p j k i
y
ij i
ij i ij
= = + =
+ + =
c
c o

In terms of the p variables in Y
ij
, (1) becomes
|
|
|
.
|

\
|
+
|
|
|
.
|

\
|
=
|
|
|
.
|

\
|
+
|
|
|
.
|

\
|
+
|
|
|
.
|

\
|
=
|
|
|
.
|

\
|
ijp
ij
ip
i
ijp
ij
ip
i
p ijp
ij
y
y
c
c

c
c
o
o


1 1 1 1 1 1

With respect to this study, our interest was to compare the mean vectors of the k samples for
significant differences. The multivariate model
c + = X Y
leads to multivariate hypothesis of the form
0 = ' A C

Where is a matrix of parameters, C specifies constraints on the design matrix X for a particular
hypothesis, and A provides a transformation of Y. An estimate of is provided by
( ) Y X X X ' ' =


The error sum of squares and cross products (SSCP) matrix is
( )
( )( )

- -
= =
- -
' ' =
'
=
' ' ' =
y y
n
y y
y y y y
A X X Y Y A W
i
ij
ij ij
k
i
n
j
i ij i ij
1

1 1
| |

and the SSCP matrix for the hypothesis is
( ) ( ) { } ( )
( )( )
- - - - - -
=
- - - - - -

' ' =
'
=
' ' '
'
=

y y
kn
y y
n
y y y y n
A C C X X C C A B
i i
k
i
i i
1 1

1
1
| |


The goal of this research using the MANOVA technique is to discover whether plant height,
plant girth and foliage (number of leaves of plants) is changed by treatment (or other action) of
the of the distance and presence of leguminous plants. The null hypothesis tested here is that a
given predictor has no effect on either of the outcome variables.
DATA ANALYSIS (CHAPTER FOUR)
Plant Height

Figure 1: Profile plot of plant height over time
As shown in Figure 1, the control group has significantly higher mean values as compared to the
treatment groups. With regards to the two treatment groups, mucuna tends to have significant
influence on the height of maize plants than cowpea.





4
0
6
0
8
0
1
0
0
M
e
a
n

H
e
i
g
h
t

o
f

P
l
a
n
t
s
hp1 hp2 hp3
Measurement Occasion
control cp
mc mean
Table 1: MANOVA table for plant height
Source Test Statistic Estimate df F p-value
Model Wilks lambda 0.7077 4, 12, 375.7 3.79 0.0000 a
Pillais trace 0.3156 12, 375.0 3.67 0.0000 a
Lawley-Hotelling trace 0.3802 12, 365.0 3.85 0.0000 a
Roys largest root 0.2599 4, 125.0 8.12 0.0000 u
Residual 125
Legume Wilks lambda 0.8785 2, 6, 246 2.74 0.0134 e
Pillais trace 0.1241 6, 248 2.73 0.0137 a
Lawley-Hotelling trace 0.1354 6,244 2.75 0.0132 a
Roys largest root 0.1081 3, 124 4.47 0.0051 u
Distance Wilks lambda 0.8457 2, 6, 246 3.58 0.0020 e
Pillais trace 0.1567 6, 248 3.51 0.0024 a
Lawley-Hotelling trace 0.1795 6, 244 3.65 0.0017 a
Roys largest root 0.1618 3, 124 6.69 0.0003 u
Residual 125
Total 129
e = exact, a = approximate, u = upper bound on F


The first part of Table 1 shows the test results for the overall model. At the 5% level of
significance, the result indicates that the distance and presence of the legumes has an effect on
the height of plants.
Controlling for distance, all the four multivariate tests rejects the null hypothesis of no
leguminous effect on plant height, suggesting some kind of difference between the three-
dimensional mean vectors of the three legumes (control group inclusive). Also, the distance of
maize plants to the legumes also produced significantly different results, indicating non-
parallelism of among plants planted at different distances from the leguminous plants.

Foliage (Number of Leaves)

Figure 2: Profile plot of mean number of plant leaves
According to Figure 2, the number of the mean number of leaves of maize plants grown by
cowpea tends to decrease with increasing number of days whilst maize plants grown by mucuna
produces more foliage as they plants grow.
6
6
.
5
7
7
.
5
8
M
e
a
n

N
u
m
b
e
r

o
f

L
e
a
v
e
s
lp1 lp2 lp3
Measurement Occasion
control cp
mc mean


. manova lp1 lp2 lp3 = legume distance

Number of obs = 130

W = Wilks' lambda L = Lawley-Hotelling trace
P = Pillai's trace R = Roy's largest root

Source | Statistic df F(df1, df2) = F Prob>F
-----------+--------------------------------------------------
Model | W 0.7349 4 12.0 325.7 3.35 0.0001 a
| P 0.2861 12.0 375.0 3.29 0.0001 a
| L 0.3326 12.0 365.0 3.37 0.0001 a
| R 0.2147 4.0 125.0 6.71 0.0001 u
|--------------------------------------------------
Residual | 125
-----------+--------------------------------------------------
legume | W 0.8726 2 6.0 246.0 2.89 0.0097 e
| P 0.1285 6.0 248.0 2.84 0.0109 a
| L 0.1447 6.0 244.0 2.94 0.0086 a
| R 0.1355 3.0 124.0 5.60 0.0012 u
|--------------------------------------------------
distance | W 0.8408 2 6.0 246.0 3.71 0.0015 e
| P 0.1625 6.0 248.0 3.66 0.0017 a
| L 0.1853 6.0 244.0 3.77 0.0013 a
| R 0.1604 3.0 124.0 6.63 0.0003 u
|--------------------------------------------------
Residual | 125
-----------+--------------------------------------------------
Total | 129
--------------------------------------------------------------
e = exact, a = approximate, u = upper bound on F
























Plant Girth

Figure 3: Profile plot of mean plant girth over time

. manova gp1 gp2 gp3 = legume distance
Number of obs = 129

W = Wilks' lambda L = Lawley-Hotelling trace
P = Pillai's trace R = Roy's largest root

Source | Statistic df F(df1, df2) = F Prob>F
-----------+--------------------------------------------------
Model | W 0.6287 4 12.0 323.1 5.16 0.0000 a
| P 0.4102 12.0 372.0 4.91 0.0000 a
| L 0.5289 12.0 362.0 5.32 0.0000 a
| R 0.3589 4.0 124.0 11.13 0.0000 u
|--------------------------------------------------
Residual | 124
-----------+--------------------------------------------------
legume | W 0.7544 2 6.0 244.0 6.16 0.0000 e
| P 0.2585 6.0 246.0 6.09 0.0000 a
| L 0.3086 6.0 242.0 6.22 0.0000 a
| R 0.2366 3.0 123.0 9.70 0.0000 u
|--------------------------------------------------
distance | W 0.8486 2 6.0 244.0 3.48 0.0026 e
3
4
5
6
M
e
a
n

P
l
a
n
t

G
i
r
t
h
gp1 gp2 gp3
Measurement Occasion
control cp
mc mean
| P 0.1519 6.0 246.0 3.37 0.0033 a
| L 0.1779 6.0 242.0 3.59 0.0020 a
| R 0.1747 3.0 123.0 7.16 0.0002 u
|--------------------------------------------------
Residual | 124
-----------+--------------------------------------------------
Total | 128
--------------------------------------------------------------
e = exact, a = approximate, u = upper bound on F



References
Tabachnick, B. G., and Fidell L. S. (2004). Using Multivariate Statistics. 5
th
ed. Boston: Allyn &
Bacon.
Rencher, A. C. (2002). Methods of Multivariate Analysis. 2
nd
ed. New York: Wiley.

You might also like