Chennai Mathematical Institute
Multivariate Statistical Analysis
Mid Term Test
Time : 60 Minutes Max marks: 20
Choose the Best possible answer
1. An advantage of using an experimental multivariate design over separate univariate designs is
that using the multivariate analysis is:
A. Allows you to look at more complex relationships than does univariate strategy
B. Provides a more powerful test of hypotheses
a) Both A and B holds
b) A holds but not B
c) B holds but not A
d) Both A and B do not hold
2. If X has a multivariate Normal distribution, then
A. Linear combinations of X are normally distributed
B. All subsets of the components of X have a MVN distribution
C. All the conditional distributions of the components are not MVN
a) A and B are true
b) A and C are true
c) B and C are true
d) A,B, and C are true
3. Let X and Y be independent standard normal random variables. Define the random variable Z
by Z = X if XY > 0 ; Z = −X if XY < 0
A. Z is a standard normal random variable
B. Random Vector (Y, Z) is not jointly normal
a) A is true but not B
b) B is true but not A
c) Both A and B are true
d) Both A and B are not true
4. Examine the following statements
A. If the components of MVN are not independent, then the vector may not be jointly
Normal
B. Uncorrelated jointly normal random variables are independent
a) A holds but not B
b) B holds but not A
c) A and B do not hold
d) A and B holds
5. If Y follows MVN with (µ,Σ), then Y’AY ~ χ2 (r,γ) where γ = µ’Aµ and r(A) = r iff
A. AΣA = Σ and AΣ is idempotent
B. AΣA = Σ and A-1 Σ is idempotent
a) A holds
b) B holds
c) A and B holds
d) A and B does not hold
6. Examine the following properties on p variate MVN
A. Invariant under linear transformation
B. Uncorrelated sub-vectors are independent
a) A and B holds
b) A holds but not B
c) B holds but not A
d) A and B hold under certain other conditions
7. The MD between Y and μ in the metric of Σ is D(Y,μ) = [(Y − μ) Σ-1 (Y − μ)]1/2
A. Invariant under appropriate linear transformations
B. Adjusted Euclidean distance in the metric of reciprocal variance
a) A holds and not B
b) B holds and not A
c) A and B holds together
d) A and B do not hold
8. Q-Q plot can be used to
A. order Mahalanobis distances versus estimated quantiles for a sample of size n from a
chi-squared distribution with p degrees of freedom for evaluating multivariate normality
B. points on the right side of the plot for which the Mahalanobis distance is notably greater
than the chi-square quantile value exhibit the presence of outliers.
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
9. Examine the following statements in PCA
A. The construction of principal components does not require that the variables in Y
have a normal multivariate distribution.
B. Construction of PCA can be carried out through SVD of the data matrix
C. PCA minimizes the perpendicular distance between a data point and the principal
component
a) A and B are true
b) A and C are true
c) B and C are true
d) A, B, and C are true
10. Examine the following statements in PC
A. PC’s are invariant to linear transformation to the original variables
B. PC’s are not sensitive to the presence of outliers
a) Both A and B are true
b) A alone is true
c) B alone is true
d) A and B are untrue
11. Examine the following statements on Mahalanobis distance
A. MD is effectively a multivariate equivalent of Euclidean distance
B. MD could be used for classification and Outlier detection
C. MD is not possible for highly correlated variables
a) A,B and C are true
b) A,B is true but not C
c) A,C is true but not B
d) B,C is true but not A
12. Suppose X1, . . . , Xn are i.i.d. Np(µ, Σ); then the sample mean and sample variance are
A. √ n(X̅ − µ) ∼ Np(0, Σ),
independent, with
B. (n − 1)S ∼ Wp(n − 1, Σ).
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
13. let X ∼ Np(µ, Σ), mS ∼ Wp(m, Σ). then T2 p (m) = (X̅ − µ) ‘S -1 (X̅ − µ) and
[(m - p + 1)/(mp)] T2 p (m) follows Fp,m-p+1:
A. If X and S are independent
B. If X̅ and S are independent
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
14. If M ~ Wp(n; Σ) and a Є Rp is such that a’Σa ≠ 0 and n > p-1 then
A. a’Ma/a’Σa follows χ 2n
B. a’, Σ-1a/a’M-1 a follows χ 2n-p+1
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
15. If A is idempotent of rank r then
A. If y is Np(O, I), then y′Ay is χ2(r)
B. y is Np(µ, σ 2I), then y′Ay/σ 2 is χ2(r, µ′Aµ/σ 2)
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
16. Examine the following
A. Let B is a k × p matrix of constants, A is a p × p symmetric matrix of constants, and y is
distributed as Np(µ, Σ). Then By and y′Ay are independent if and only if BΣA = O
B. Let A and B be symmetric matrices of constants. If y is Np ( µ, Σ), then y ′Ay and y ′B y
are independent if and only if A Σ B = O.
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
17. If y us Np(µ, Σ), then
A. E(Y’AY) = tr(AΣ) + μ’Aμ
B. var(y′Ay) = 2tr[(AΣ) 2] + 4µ′AΣAµ.
C. cov(y, y′Ay) = 2ΣAµ.
a) A and B are true
b) A and C are true
c) B and C are true
d) A, B and C are true
18. Examine the following for Mahalanobis distance:
A. a measure that quantifies the dissimilarity between two data points in a multidimensional
space, considering the covariance structure of the data.
B. It takes into account the correlations between variables, making it suitable for datasets
where variables are interrelated.
C. It is not scale-invariant, meaning it is affected by the scaling of variables
a) A and B are true
b) A and C are true
c) B and C are true
d) A, B, and C are true
19. Let X follows Np(µ, Σ), and Y = Σ-1/2 (X - µ) then
A. Y follows Np(0, I p)
B. Y follows Np(µ, Σ)
a) A is true
b) B is true
c) A and B are true
d) A and B are not true
20. If X ~ N3(µ, Σ) with µ = [ -2 3 1]’ and
Then marginal distribution of X1 and X3 and [X1 X3 ]’ are
A. X1 ~ N ( 2, 2)
B. X2 ~ N ( 1,4)
C. X3 ~
a) A and B are but not C
b) B and C are true but not A
c) A and C are true but not B
d) A,B, and C are true
1–a
2–a
3–c
5-d
6-d
7-b
9-d
10-c
15-a
16-d
18- c
19 – a
20 – c
21-a
22-c
23-b
24 – a
25-a
