Math 218, Mathematical Statistics we need to find the probability that a standard normal ran-

D Joyce, Spring 2016 dom variable Z has a value less than −0.9798. Look up
Φ(−0.9798) in the table to get 0.1635.
Comment. Even though about 31 of the cans are that light,
Assignment. Chap. 5, p. 189, exercises 4, 5, 8, 14, 16,
the probability that six cans will average out to be that light
17, 23 .
is only about 16 .

Selected answers. 5. A random sample X1 , . . . , X150 is drawn from a popu-

lation with mean µ = 40 and standard deviation σ = 15 but
4. A soft drink company uses a filling machine to fill cans. from an unknown distribution. Let U = (X1 + · · · + X50 )/50
Each 12 oz. can is to contain 355 milliliters of beverage. In represent the sample mean of the first 50 observations and
fact, the amount varies according to a normal distribution V = (X51 + · · · + X150 )/100 the sample mean of the last 100
with mean µ = 355.2 ml and standard deviation σ = 0.5 ml. observations.
a. What is the probability that an individual can contains a. What are the approximate distributions of U and V ?
less than 355 ml? Since they’re large samples, the central limit theorem says
Our random variable X is normal with mean µ = 355.2 they’ll be approximately normal. They’ll have the same
and standard deviation σ = 0.5. Here are three slightly mean as the population mean, that is, µU = µV = µ = 40.
different ways you can get the answer. You’ll probably find But their variances will be smaller than the population vari-
2
one of these methods a little more to your liking. ances by factors of their sizes, that is, σU = σ 2 /50 =
2 2 2 2
Method 1: The mean is 355.2, so the value 355 is 0.2 ml 15 /50 = 4.5 and σV = σ /100 = 15 /100 = 2.25
below the mean. Since a standard deviation is 0.5 ml, that (Alternatively, you could answer this question in terms
means 0.2 ml is 0.2/0.5 = 0.4 standard deviations. That of standard deviations. Their standard deviations will be
means that we need to find the probability that a standard smaller than the population standard deviations by √ factors
normal random variable Z has a value less than −0.4. In of the
√ square roots of their sizes,
√ that is, σ
√ U = σ/ 50 =
the notation of our text, that’s Φ(−0.4). Look that up in 15/ 50 = 2.121 and σV = σ/ 100 = 15/ 100 = 1.5.)
Table A.3 of standard normal curve areas Φ(z) = P (Z ≤ z). b. Which would you expect to be larger, P (35 ≤ U ≤ 45)
You get Φ(−0.4) = 0.3446. or P (35 ≤ V ≤ 45)? Why?
Comment: Thus about 13 of the cans are too light. The interval is symmetric about the mean 40, from 40 − 5
Method 2: Normalize X to get a standard normal dis- to 40 + 5, so the question is which of U and V has a smaller
tribution Z. Then Z = (X − µ)/σ. We want to know variation. Since V has the smaller variation, it’s more likely
P (X ≤ 355). But X ≤ 355 iff X − µ ≤ 355 − µ iff to be inside this interval. Therefore, P (35 ≤ V ≤ 45) is
Z = (X − µ)/σ ≤ (355 − µ)/σ iff Z ≤ (355 − 355.2)/0.5 larger.
iff Z ≤ −0.4. Therefore, we need P (Z ≤ −0.4). Proceed as c. Find P (35 ≤ U ≤ 45) and P (35 ≤ V ≤ 45) using the
in method 1 to get 0.3446. normal approximation.
Method 3: Normalize X. Then Z = (X − µ)/σ. We
need P (X ≤ 355). But X = σZ + µ, so that means we For U , a standard deviation is 2.121, so 5 is 5/2.121 =
need P (σZ + µ ≤ 355). That simplifies to P (Z ≤ −0.4), so 2.357 standard deviations. We need P (−2.357 ≤ Z ≤
proceed as before. 2.357). That’s the same as 1 − 2Φ(−2.357) = 1 − 2 · 0.0091 =
0.9818.
b. What is the probability that the mean content of a For V , a standard deviation is 1.5, so 5 is 5/1.5 = 3.333
six-pack of cans is less than 355 ml? standard deviations. We need P (−3.333 ≤ Z ≤ 3.333).
The last part was about the capacity X of 1 can, but in That’s the same as 1 − 2Φ(−3.333) = 1 − 2 · 0.0004 = 0.9992.
this part, we take the average X = 61 (X1 + · · · + X6 ) of 6
cans. The mean of X is the same as the mean of X, namely 8. The lifetime of disc brake pads varies according to a
µ = 355.2, but the standard deviation of X is different from normal distribution with mean µ = 50, 000 miles and stan-
that of X. In fact, dard deviation σ = 3000 miles. Suppose that a sample of
nine break pads is tested.
σX 0.5
σX = √ = √ = 0.2041. a. What is the distribution of the sample mean X? Give
n 6 the mean and standard deviation of this distribution.
Now, 355 is 0.2 below the mean of 355.2, and since one It’s a normal distribution. Its mean is µX = µ = 50,
√ 000
√
standard deviation for X is 0.2041, that means 0.2 is miles. Its standard deviation is σX = σ/ n = 3000/ 9 =
0.2/0.2041 = 0.9798 standard deviations. That means that 1000 miles.

1
 b. Does your answer in part a require the use of the Thus, about 1% of people without ESP will guess 30 or
central limit theorem? Why or why not? more questions correctly.
No. Since the population distribution is normal, a sample
mean X is normal, too. Not approximately normal as in the 16. Using table A.5, find the following values: χ25,0.01 ,
conclusion to the central limit theorem, but exactly normal. χ210,0.05 , χ210,0.95 , χ210,0.75 .
c. Suppose that a sample mean less than 47,000 miles The value χ25,0.01 is the value of x such that a χ2 random
is considered evidence that µ < 50, 000 miles. What is the variable X with 5 degrees of freedom has the probability 0.01
probability that this will happen even when the true mean of being greater than x, that is, P (X ≥ x) = 0.01. That’s
µ = 50, 000, leading to an incorrect conclusion? found in the fifth row of the table, in the next to the last
We need to compute P (X < 47, 000). Now, 47,000 miles column. χ25,0.01 = 15.085.
is 3,000 miles below the mean, and 3,000 miles is 3 standard Likewise, χ210,0.05 = 18.307 and χ210,0.95 = 3.940.
deviations for X. Therefore, P (X < 47, 000) = P (Z < There is no column for α = 0.75 in the table, so χ210,0.75
−3) = Φ(−3) = 0.0013. So the probability of an incorrect can’t be determined from it.
conclusion under the given assumptions is 0.13%, almost 0.
17. Consider the χ2 distribution with 8 degrees of freedom.
14. In an ESP experiment, five choices are offered for each a. What are the mean and variance of a χ28 random vari-
question. Assume that a person without ESP guesses ran- able?
domly and thus correctly answers with probability 51 . Fur- The mean of a χ2 distribution with ν degrees of freedom
ther assume that the responses are independent. Suppose is ν, which is 8 in this exercise, and the variance is 2ν, which
that 100 questions are asked. is 16 in this exercise.
a. What are the mean and standard deviation of the
b. Using table A.5, find constants a, b, c, d, and e so that
number of correct answers?
P (χ28 > a) = 0.05,
Each Bernoulli trial Xi has mean µ = p = 15 and variance P (χ28 > b) = 0.99,
σ 2 = pq = 15 · 45 = 25 4
, so the standard
Pn deviation is σ = P (χ28 < c) = 0.90, and
2
5 . Therefore the sample sum, Sn = i=1 Xi , which is a P (d < χ28 < e) = 0.95.
binomial distribution, has mean µSn = nµ = npq = 20 and
See part c.
variance σS2 n = npq = 16, and so its standard deviation is
σS n = 4. c. Express the constants from part b in terms of the
notation χ2ν,α .
b. What are the mean and standard deviation of the
proportion of correct answers? P (χ28 > a) = 0.05 when a = χ28,0.05 = 15.507.
The proportion of correct answer is the same √
as the sample P (χ28 > b) = 0.99 when b = χ28,0.99 = 1.646.
mean X. Therefore, µX = µ = 15 and σX = σ/ n = 25 · 10 1
= P (χ28 < c) = 0.90 when P (χ28 > c) = 0.10, and that’s
0.04. when
c = χ28,0.10 = 13.362.
c. What is the probability that a person without ESP will
We can make P (d < χ28 < e) = 0.95 by making the two
correctly answer at least 30 of the 100 questions? Check if
tails the same, that is, P (χ28 > e) = 0.025 and P (χ28 >
the normal approximation will give accurate results. If so
d) = 0.975. That happens when d = χ28,0.975 = 2.180 and
use it with a continuity correction.
e = χ28,0.025 = 17.534. (You may have reason to make the
We need the probability P (Sn ≥ 30) with n = 100. That’s tails different sizes, in which case there are other values for
the sum d and e.
n   100   d. Sketch each of the probabilities from part b as an area
X n k n−k X 100 1
k 4
100−k
p q = 5 5 . under the χ28 probability density curve.
k k
k=30 k=30

There is a table to find cumulative sums like this, Table A.1 23. An engineer suspects that the temperature inside an
page 669, but it only goes up to n = 25. Beyond that the oven is not as uniform as when it was new, at which time
normal approximation is pretty good. the temperature varied ±10◦ F around its setting. (Taking
Now, Sn is approximately normal with mean 20 and stan- the range of a normal distribution to be roughly ±2σ, this
dard deviation 4. With the continuity correction, we need translates into σ = 5◦ F.) To verify his suspicion, he takes
to compute P (Sn ≥ 29.5). That’s 9.5 units above the 20 measurements in different parts of the oven. He wants a
mean, which is 9.5/4 = 2.375 standard deviations above rule which decides that the true σ > 5 if the sample standard
the mean. We need the probability P (Z ≥ 2.375) which is deviation of the measurements exceeds 5c where c > 0 is a
1 − Φ(2.375) = 1 − 0.9912 = 0.0088. suitably chosen constant. The rule must not have more than

2
a 10% chance of making a wrong decision, i.e., deciding that
σ > 5 when, in fact, σ = 5.
a. Find the value of c.
In this exercise, n = 20 and σ = 5.
We need to find a value c so that P (S > σc) = 0.10.
(n − 1)S 2
That’s the same as P (S 2 > c2 σ 2 ) = 0.10. Now,
σ2
is a χ2 random variable with n − 1 = 19 degrees of free-
dom. Convert the condition S 2 > c2 σ 2 so the left hand side
becomes that variable. The condition becomes
(n − 1)S 2
> (n − 1)c2 .
σ2
Thus, we need P (χ219 > 19c2 ) = 0.10. That happens when
19c2 = χ219,0.10 = 27.203. So c2 = 27.203/19 = 1.4317, and
therefore c = 1.197. Since the original specification of that
σ was 5◦ F was really only accurate to one or two digits, c
should only have that accuracy, c = 1.2.
b. Based on this value of c, does the rule decide that
σ > 5 if the sample standard deviation of the engineer’s
measurements is s = 7.5◦ F?
Is 7.5 > 5c? Since 5c = 6, yes it is. That is, if the
measured standard deviation of 20 measurements is higher
than 6◦ F, then this test concludes that the temperature
standard deviation is not 5◦ F.

Math 218 Home Page at

http://math.clarku.edu/~djoyce/ma218/