MATH/STAT 394: MIDTERM II.

NAME and student ID:

There are 5 problems. Stop now and make sure you have all of the problems. If
you do not have them all, then request a new exam. You must show all of your
work. Just a numeric answer without arguments will not receive any points.
Now, take a deep breath and read the questions.

Problem Score

1 /10

2 /10

3 /15

4 /10

5 /15

Total /60

Date: November 19, 2018.

1
2

Problem 1. (10 points) A pollster would like to estimate the fraction p of people
in a population who intend to vote for the particular candidate. How large a sample
size does she have to take in order to be at least 95% certain that the fraction p̂ of
the positive answers in the sample is within 0.01 of the true p? You can use the fact
that, for a standard normal random variable Z, the following estimates are true:
P (|Z| ≤ 1) ≥ 68%, P (|Z| ≤ 2) ≥ 95%, P (|Z| ≤ 3) ≥ 99%.

Solution. By normal approximation, we know that, for large n,

√

P (|p̂ − p| ≤ ) ≥ P |Z| ≤ 2 n .
√
If I √
want the right hand side to be more than 95%, it is enough to take 2 n = 2,
or  n = 1. Taking  = .01, gives us n = 10, 000.
 3

Problem 2. (5+5 points) Suppose 20% of the households in a city earn over
$80, 000 a year, and 0.25% of households earn over $450, 000. A random sample
of 400 house holds has been chosen. In this sample, let X be the number of
households that earn over $80, 000 and let Y be the number of households that
earn over $450, 000. Estimate the probabilities using either the normal or the
Poisson approximation. In case of normal approximation leave the final answer in
terms of the standard normal c.d.f. Φ. Do not evaluate it.
(i) P (X ≥ 88). Argue why your approximation is appropriate.
(ii) P (Y ≥ 2). Argue why your approximation is appropriate.
Solution. (i) X is distributed as binomial(400, 0.2). Thus E(X) = 80 is large.
normal approximation is appropriate. σ 2 = Var(X) = 400 · 0.2 · 0.8 = 64,
Thus a √
or σ = 64 = 8. Hence
 
X − 80 88 − 80
P (X ≥ 88) = P ≥ ≈ P (Z ≥ 1) = 1 − Φ(1).
8 8
(ii) Y is distributed as binomial(400, 0.0025). Thus E(X) = 1 is small. Thus a
Poisson approximation is appropriate with λ = 1.
P (Y ≥ 2) = 1 − P (Y = 0) − P (Y = 1) = 1 − 2e−1 .
4

Problem 3. (10+3+2 points) Choose a point uniformly at random inside the

triangle with corners (0, 0), (1, 0) and (1, 1). Let S be the slope of the line through
the chosen point and the origin.
(i) Find the cumulative distribution function of S.
(ii) Find the probability density function of S.
(iii) Find the expected value of this slope.
Solution.

(i) Method 1. The slope S of the line through the m

point X is less than m if and only if X lies below
the line of slope m. In the figure, the slope of the
X h
line through X is less than m. Thus, for 0 ≤ m ≤ 1,
P (S ≤ m) is the area of the shaded triangle, divided
by the total area of the big triangle. 1

m = h/1 ⇒ h = m
1
·1·h
P (S ≤ m) = 21 =h=m for 0 ≤ m ≤ 1.
2 ·1·1
Thus, 
 0 for m < 0,
F (m) = m for 0 ≤ m ≤ 1,
1 for m > 1.

(Note that this means S ∼ Uniform[0, 1].)
(i) Method 2. Let θ be the angle that the line through X makes with the
positive x-axis. The possible value of θ is the interval [0, π/4] and of S = tan(θ) is
the interval [0, 1]. Now, pick any 0 ≤ a ≤ π/4. Then θ ≤ a if and only if X lies in
the shaded region in the figure. This shaded triangle is right angled and has base
length 1 and height tan(a). Thus its area is 12 · tan(a) · 1 = 12 tan(a). Since total
area is 21 · 1 · 1 = 1/2, thus
(tan(a))/2
P (θ ≤ a) = = tan(a).
1/2
Thus, for any 0 ≤ u ≤ 1,
F (u) = P (S ≤ u) = P (tan(θ) ≤ u) = P (θ ≤ arctan(u)) = tan(arctan(u)) = u.
For u ≤ 0, F (u) = 0, and for u ≥ 1, F (u) = 1.

d 1 for 0 ≤ m ≤ 1,
(ii) f (m) = F (m) =
dm 0 otherwise.
R1  1 2 1 1
(iii) E[S] = 0 m · 1dm = 2 m 0 − 2 . The expected value of Uniform[0, 1] is 12 .
 5

Problem 4. (10 points) The amount of loss in a car accident has an exponential
distribution with rate one. An insurance company applies a deductible of r, which
means the following. If the amount of loss is less than or equal to r, then the
customer pays out of his pocket and the company pays nothing. If the amount of
loss exceeds r, then the customer pays the first r and the company pays the rest.
Calculate the expected amount that the company has to pay to a customer in the
event of an accident as a function of the deductible r.
Solution. Let X be the amount of loss. The payment of the company is a function
0, x ≤ r,

g(x) =
x − r, x > r.
Hence, by integration by parts, expected payment is
Z ∞
Eg(X) = (x − r)e−x dx = e−r (1 + r) − re−r = e−r .
r
Alternatively, you can use the memoryless property of exponentials to claim that
Eg(X) = P (X > r)E(X) = e−r .
6

Problem 5. (10+2+3 points) Sample 3 points independently at random uniformly

from the unit interval [0, 1]. Let X be the maximum of these 3 points.
(i) Find the probability density function of X.
Hint: Relate the cdf of X to that of the 3 random points.
(ii) Find the expected value of X.
(iii) Find the variance of X.
Solution. (i) The possible values of X is [0, 1]. Let U1 , U2 , U3 by the three points.
Take 0 ≤ a ≤ 1 Then {X ≤ a} is the same as {U1 ≤ a} ∩ {U2 ≤ a} ∩ {U3 ≤ a}.
Thus
P (X ≤ a) = P ({U1 ≤ a} ∩ {U2 ≤ a} ∩ {U3 ≤ a}) = P (U1 ≤ a)P (U2 ≤ a)P (U3 ≤ a) = a3 .
Hence the density of X is f (a) = 3a2 , for 0 ≤ a ≤ 1, and zero elsewhere.
(ii)
Z 1 Z 1
2 3
E(X) = a3a da = 3 a3 da = .
0 0 4

(iii)
Z 1 Z 1
3
E(X 2 ) = a2 3a2 da = 3 a4 da = .
0 0 5
3 3 9 48−45 3
Hence Var(X)= 5 − (3/4)2 = 5 − 16 = 80 = 80 .
 7

Extra page.