WILP Advanced Statistical Methods

BITS Pilani
Semester 1 (2024-25)
Mid-Semester Exam (Makeup) Solutions

1. A particular brand of drinking water is sold in three sizes: 10L, 20L, and 30L. 20%
ofcustomers bought 10L water bottle, 40% of customers bought 20L water bottle and
remaining 40% of customers bought 30L water bottle. If X1 and X2 denote the water bottle
sizes selected independently by two customers (sample), then write down the sampling
distributions of the sample mean and the sample standard deviation. Don’t plot any
histogram.
Hint : The probability that the sample mean or the sample standard deviation takes a
particular value is the joint probability of the corresponding sample observations, that is,
the product of the individual probabilities (due to independence between X1 and X2 ).
Solution: Here X1 and X2 form a sample of size 2. Let xi denote the value of Xi so
xi ∈ {10, 20, 30} where i = 1, 2. The sample variance is s2 = (x1 − x)2 + (x2 − x)2 where
x = (x1 + x2 )/2 is the sample mean. Given P (Xi = 10) = 0.2, P (Xi = 20) = 0.4 and
P (Xi = 30) = 0.4. The joint distribution of X1 and X2 is p(x1 , x2 ) = P (X1 = x1 , X2 =
x2 ) = P (X1 = x1 )P (X2 = x2 ) since X1 and X2 are independent RVs. (1)
All possible pairs (x1 , x2 ) along with the joint distribution and the values of x, s2 and s
are listed below.

x1 x2 p(x1 , x2 ) x s2 s
10 10 0.04 10 0 0
10 20 0.08 15 50 7.0711
10 30 0.08 20 200 14.1421
20 10 0.08 15 50 7.0711
(2)
20 20 0.16 20 0 0
20 30 0.16 25 50 7.0711
30 10 0.08 20 200 14.1421
30 20 0.16 25 50 7.0711
30 30 0.16 30 0 0

Using the above table, the sampling distribution of X is given as follows.

x 10 15 20 25 30
(2)
P (X = x) 0.04 0.16 0.32 0.32 0.16

Similarly using the initial table, the sampling distribution of S is given as follows.

s 0 7.0711 14.1421
(1)
P (S = s) 0.36 0.48 0.16

2. A sample of test spot welds yields the following data of their shear strengths, which are
assumed to be normally distributed.

392 376 401 367 389 362 409 415 358 375

1
 (a) Obtain the point estimates of the true average and the true variance of shear strength.
(b) Discuss the bias and consistency properties of each of the above estimators with
justification.

Solution: Here the sample size is n = 10. Let x1 , . . . , x10 denote the shear strengths of
test spot welds.

(a) The point estimate of the true average shear strength is the sample mean (x) of shear
strengths whereas the point estimate of the variance of shear strength is the sample
variance (s2 ) of shear strengths. The sample mean and the sample variance are as
follows.
n n n
1X 3844 1 X 1X 3556.4
x= xi = = 384.4, s2 = (xi −x)2 = (xi −384.4)2 = = 395.1556.(2)
n 10 n−1 9 9
i=1 i=1 i=1

(b) Let X and S 2 denote the point estimators of the population mean µ and the popu-
lation variance σ 2 respectively. We know

E(X) = µ

so X is an unbiased estimator of µ. (1)

Also
σ2
Var(X) = → 0 as n → ∞
n
so X is a consistent estimator of µ. (1)

Now we know

E(S 2 ) = σ 2

so S 2 is an unbiased estimator of σ 2 . (1)

Finally note that
2σ 4
Var(S 2 ) = → 0 as n → ∞.
n−1
Hence S 2 is a consistent estimator of σ 2 . (1)

3. Assume that the following sample is drawn from a normal population.

1 6 2 4 3 6 1 4

(a) Test whether the population variance is different from 5 using α = 0.10.

(b) Calculate a 95% confidence interval for the population variance.

Solution: Given the sample size is n = 8. Let x1 , . . . , x8 denote the sample observations.
Then the sample mean and the sample variance are
n n n
1X 27 1 X 1X 27.875
x= xi = = 3.375, s2 = (xi −x)2 = (xi −3.375)2 = = 3.9821.(2)
n 8 n−1 7 7
i=1 i=1 i=1

2
 (a) Let σ 2 denote the population variance. We want to test the hypotheses
H0 : σ 2 = 5 vs Ha : σ 2 ̸= 5. (1)
The value of the test statistic under H0 is
(n − 1)s2 7 × 3.9821
χ2 = 2
= = 5.575.
σ 5
Note that χ2 ∼ χ2n−1 = χ27 under H0 . (1)
2 2
Given α = 0.10 so 1 − α/2 = 0.95, χ7 (0.95) = 2.1673 and χ7 (0.05) = 14.0671. Since
χ27 (0.95) < χ2 < χ27 (0.05), we can’t reject H0 at 10% level of significance and conclude
that the population variance is not different from 5. (1)
(b) Given the confidence level is 90% so 100(1 − α) = 0.90 or α = 0.10. (1)
Hence a 90% confidence interval for the population variance is
(n − 1)s2 (n − 1)s2 7s2 7s2
     
27.875 27.875
, = , = , = (1.9816, 12.8616).(1)
χ2n−1 (α/2) χ2n−1 (1 − α/2) χ27 (0.05) χ27 (0.95) 14.0671 2.1673

4. During the first 13 weeks of a television season, the audience proportions for a programme
were recorded as ABC - 29%, CBS - 28%, NBC - 25%, and independents - 18%. Two
weeks after a revision of the programme schedule, the following viewing audience data
was recorded: ABC - 95 homes, CBS - 70 homes, NBC - 89 homes, and independents -
46 homes. Test with α = 0.05 to determine whether the viewing audience proportions
changed after the revision.
Solution: Denote the channels ABC, CBS, NBC and independents by A, B, C and D
respectively so there are k = 4 categories of viewers. Further let pA , pB , pC and pD
denote the respective audience proportions for a programme. To determine if the viewing
audience proportions changed after a revision in the programme schedule, we need to test
the following hypotheses.

H0 : pA = 0.29, pB = 0.28, pC = 0.25, pD = 0.18,

Ha : pA , pB , pC , pD are different from their values under H0 . (1)

Total number of homes surveyed surveyed after a change in the programme schedule =
n = 95 + 70 + 89 + 46 = 300. Let fA , fB , fC and fD denote the observed frequencies
whereas eA , eB , eC and eD denote the expected frequencies under H0 . The observed
viewing frequencies are given as follows.

fA = 95, fB = 70, fC = 89, fD = 46.

Since the observed data follows a multinomial distribution with parameters n, pA , pB , pC

and pD under H0 , the expected frequencies under H0 can be obtained as follows.

eA = npA = 300 × 0.29 = 87, eB = npB = 300 × 0.28 = 84,

eC = npC = 300 × 0.25 = 75, eD = npD = 300 × 0.18 = 54. (2)

The value of the goodness of fit test statistic is

(fA − eA )2 (fB − eB )2 (fC − eC )2 (fD − eD )2
χ2 = + + + = 6.8675.
eA eA eC eD

3
 Note that χ2 ∼ χ2k−1 = χ23 under H0 . (2)
2 2 2
Given α = 0.05 and χ3 (0.05) = 7.8147. Since χ < χ3 (0.05), we can’t reject H0 at 5%
level of significance and conclude that the viewing audience proportions for the channels
haven’t changed after the revision in the programme schedule. (1)
5. The following data are from a completely randomized design.

Treatment 1 Treatment 2 Treatment 3

20 28 20
26 26 19
24 31 23
22 27 22

Use analysis of variance (ANOVA) to test for a significant difference among the means of
the three treatments using α = 0.05.
Solution: Let x1 , x2 and x3 denote the sample means of Treatments 1, 2 and 3 respectively.
Also let s21 , s22 and s23 be the sample variances of Treatments 1, 2 and 3 respectively. From
the given data, we have
x1 = 23, x2 = 28, x3 = 21, s21 = 6.6667, s22 = 4.6667, s23 = 3.3333.
The overall sample mean for the three treatments is x = (x1 + x2 + x3 )/3 = (23 + 28 +
21)/3 = 24. The sample sizes for the three treatments are n1 = n2 = n3 = 4. The sum of
squares due to treatments is
3
X
SSTR = ni (xi − x)2 = 4[(23 − 24)2 + (28 − 24)2 + (21 − 24)2 ] = 104.
i=1

Let the jth observation in treatment j be xij . The sum of squares due to error is
3
X 3
X
2
SSE = (xij − xi ) = (ni − 1)s2i = (4 − 1)[6.6667 + 4.6667 + 3.3333] = 44.
i=1 i=1

The total sample size is n = 3 × 4 = 12. The mean sum of squares due to treatments and
error are
SSTR 104 SSE 44
MSTR = = = 52, MSE = = = 4.8889. (4)
3−1 2 12 − 3 9
Let µ1 , µ2 and µ3 denote the population means of Treatments 1, 2 and 3 respectively. We
want to test the following hypotheses.
H0 : µ1 = µ2 = µ3 vs Ha : µi ̸= µj for at least one pair (i, j).
The F-statistic for testing the above hypotheses is
MSTR 52
F = = = 10.6364.
MSE 4.8889
Note that F ∼ F3−1,12−3 = F2,9 under H0 . (1)
Since 10.6364 > F2,9 (0.05) = 4.2565, we reject H0 at 5% level of significance and conclude
that all the three treatment means are not equal. (1)