a. Draw the scatterplot for the six points on the rotated axis given at Chart B on
the following page and rain the points onto the u axis. Verify if we affected
a separation of variables onto the u-axis. (If this is so, then all the information
on the points contained in the two axes in the xy system, is now contained in
one axis, the u-axis in the uv system. This is called reduction of axes.)
[15]
142
141.8
141.6
141.4
141.2
141
122 122.5 123 123.5 124 124.5 125 125.5
B
3
2. Consider the bivariate dataset below. This dataset for a Multiple Sclerosis
experiment, consists of two variable: X1, the left arm response to a given stimulus,
and X2 the right leg response to the same stimulus for two MS patients.
X1 X 2
X 2 6
6 20
Find:
a. The mean vector [4,13]
[3]
b. The covariance matrix [6]
Solution:
Using the H matrix:
1 1
1 1 0 1
1 1 2 2
H In Jn
2 1 1 1
n 0 1 1
2 2
Then
4 14
AT HA
14 49
c. The correlation matrix [6]
Solution:
ij
ij
i2 j2
For example:
12
12
12 22
Thereby,
1 1
1 1
4 49 14 14 196 196 0
Generalized Variance = det =
f. The Mahalanobis Transformation of the data matrix X (if it exists). [20]
Solution:
2 6
A
6 20
Eigenvalues:
A I 0
2 6
6 0
20
2 20 36 0
2 22 4 0
1 21.82, 2 0.18
Using 1 21.82, we have
1
v1
3.3
Using 2 0.18 , we have
3.3
v2
1
Normalizing the eigenvectors:
Ni vi viT
N 3.4
Let the normalized vector for vi be ei . Thus,
vi
ei
N
and
1/3.4 0.3
e1 ~1
3.3 /3.4
3.3 /3.4 1
e2
1/3.4 0.3
The Mahalanobis Transformation:
5
1
Z X 2
1 1
2
T 2
1
0
0.3 1 21.82 0.3 1
=
1 0.3 1
1 0.3
0
0.18
2.37 0.65
=
0.64 0.42
And
2 6
X A 4 13
6 20
2 7
2 7
Therefore,
1
2.37 0.65 2 7
Z 2 X
0.64 0.42 2 7
6 21.1
=
2.13 7.46
2 6
A
6 20
26 6 20
X1 4 and X 2 13
2 2
2 1 2
d1 X1 X 1 J1 4
6 1 2
6 1 7
d2 X2 X 2 J1 13 1 7
20
2
2 11 d1T d1 2 2 8
2
11 8 /2 4
7
2 22 d2T d2 7 7 98
7
22 98 /2 49
7
2 12 d1T d2 2 2 28
7
12 21 28 /2 14
Thus,
11 12 4 14
14 49
21 22
Section 2
Each question in this section is worth 3 points.
1 , 2 ,..., p
1. The mean vector of a p-variate random variable is written
True False
The matrix is
a. Positive definite
b. Positive semi-definite
5. We can find the Mahalanobis Transfomation of the data matrix A that generated
the covariance matrix given in Question 4 above.
True False