0% found this document useful (0 votes)
28 views

Lesson 21: Normal Distributions: What If A Normal Isn't Standard?

1) The document discusses rolling one or two standard six-sided dice. Rolling one die produces outcomes from 1 to 6 with equal probabilities. Rolling two dice produces sums from 2 to 12 and sample means from 1 to 6. 2) When rolling two dice, the distribution of sample means approaches a normal distribution with a mean of 3.5, even though the original distribution for one die is uniform. 3) As the sample size increases, the sampling distribution of the sample mean approximates the normal distribution, according to the Central Limit Theorem.

Uploaded by

Md Ibrahim Molla
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
28 views

Lesson 21: Normal Distributions: What If A Normal Isn't Standard?

1) The document discusses rolling one or two standard six-sided dice. Rolling one die produces outcomes from 1 to 6 with equal probabilities. Rolling two dice produces sums from 2 to 12 and sample means from 1 to 6. 2) When rolling two dice, the distribution of sample means approaches a normal distribution with a mean of 3.5, even though the original distribution for one die is uniform. 3) As the sample size increases, the sampling distribution of the sample mean approximates the normal distribution, according to the Central Limit Theorem.

Uploaded by

Md Ibrahim Molla
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 25

(Lesson 21: Normal Distributions) 21.

01

LESSON 21: NORMAL DISTRIBUTIONS


What if a Normal Isn’t Standard?
PART A: STANDARDIZING NORMAL SCORES TO FIND PROBABILITIES

Assume that X has a normal distribution with mean µ and standard deviation
(
(SD) σ . That is, X ~ N µ , σ . )
• µ and σ could have any units: points, inches, feet, pounds, etc.
Make sure µ and σ are written using the same units, however.

• Z is assumed to have a standard normal distribution, so


( )
Z ~ N µ = 0, σ = 1 . z scores have no units.

To find probabilities for values of X, we need to transform (convert) x scores


into z scores using our usual formula for z scores:

Formula for z Scores


x − mean x−µ
z = = (We usually round off to two decimal places.)
SD σ

Then, we find probabilities for the corresponding values of Z.


(Lesson 21: Normal Distributions) 21.02

Example 1 (Standardizing Normal Scores)


Let X be the height of a random American woman.
(
Then, X ~ N µ = 65 inches, σ = 3.5 inches . )
(Source: http://www.usablestats.com/lessons/normal)

Use these hints regarding the Z distribution:

( ) ( )
• a) Find P X < 60 inches , which is the same as P X < 5 feet . First write
the corresponding probability expression for Z. Show work by using the
Formula for z Scores.

( ) ( )
• b) Find P X > 60 inches , which is the same as P X > 5 feet . First write
the corresponding probability expression for Z .

( )
• c) Find P 60 inches < X < 72 inches , which is the same as
P (5 feet < X < 6 feet ) . First write the corresponding probability
expression for Z . Use the Formula for z Scores when showing work.
§ Solution

( )
• a) Find P X < 60 inches . First take the boundary x score, 60 inches, and
transform it into a z score by using the Formula for z Scores:
x−µ 60 − 65
x = 60 inches ⇒ z = = ≈ − 1.43
σ 3.5
Write the corresponding probability expression for Z ; from the figure on
the left, we see that this probability is about 0.0764:

P ( X < 60 inches ) ≈ P ( Z < −1.43) ≈ 0.0764


(Lesson 21: Normal Distributions) 21.03

( )
• b) Find P X > 60 inches . We want the complementary probability;
remember that the total area under the density curve is 1:

P ( X > 60 inches ) ≈ P ( Z > −1.43)


≈ 1− P ( Z < −1.43)
≈ 1− 0.0764
≈ 0.9236

( )
• c) Find P 60 inches < X < 72 inches . First take the boundary x scores,
60 inches and 72 inches, and transform them into z scores by using the
Formula for z Scores:
x−µ 60 − 65
x = 60 inches ⇒ z =
σ
=
3.5
≈ − 1.43 ( found in a))
x−µ 72 − 65
x = 72 inches ⇒ z = = ≈ 2.00
σ 3.5
Write the corresponding probability expression for Z ; use the given
probabilities in the figures:

P ( 60 inches < X < 72 inches ) ≈ P ( −1.43 < Z < 2.00 )


≈ 0.9772 − 0.0764
≈ 0.9008

§
(Lesson 21: Normal Distributions) 21.04

PART B: PERCENTILES OF NORMAL DISTRIBUTIONS

In Part A, we transformed x scores into z scores (by expressing z as a function of


x). Here, we will transform z scores back into x scores (by expressing x as a
function of z).

Formula for Transforming z Scores Into x Scores


x = µ + zσ

Proof
Solve for x:
x−µ
z=
σ
Multiply both sides by σ :
zσ = x − µ
Add µ to both sides:
µ + zσ = x
x = µ + zσ

Example 2 (Percentiles of Normal Distributions)


Let X be the height of a random American woman.
(
Then, X ~ N µ = 65 inches, σ = 3.5 inches . )
Find the 85th percentile of the distribution of heights of American women
and round it off to the nearest tenth of an inch. Also write the corresponding
probability statement for X and interpret it.
Hint: In Lesson 20, Example 3, we saw that the 85th percentile of the Z
distribution is about 1.04.
(Lesson 21: Normal Distributions) 21.05.

§ Solution

Transform the given z score into an x score:


z ≈ 1.04 ⇒
x = µ + zσ
≈ 65 + (1.04 ) ( 3.5)
≈ 68.6 inches

( )
The desired probability statement for X is: P X < 68.6 inches ≈ 0.85 .
That is, about 85% of American women are shorter than 68.6 inches.
§
(Lesson 22: The Central Limit Theorem (CLT)) 22.01

LESSON 22: THE CENTRAL LIMIT THEOREM (CLT)


How Do We Analyze Averages (Means)?
PART A: ROLL ONE DIE

We will roll one standard six-sided die.

Let X = the result of the die.

(
X has the following probability distribution, which we will call D X ~ D ; )
we saw this in Lesson 15, Example 1.

Value Probability
(x) P (x)
1 1/6
2 1/6
3 1/6
4 1/6
5 1/6
6 1/6

This is a discrete uniform distribution, but the ideas of this lesson will apply to all
discrete and continuous distributions with a finite standard deviation (which we
will assume).

Using the formulas from Lesson 16, we can find that D has:

mean, µ = 3.5
SD, σ ≈ 1.7078

• When in doubt, round off to at least five significant figures.


(Lesson 22: The Central Limit Theorem (CLT)) 22.02

As an alternative to a histogram, we can use a spike plot:

For one die, it is convenient that the die result, the sum, and the mean are all
the same. For example, if the die comes up a “3,” then the sum is 3, and the mean
is 3.

PART B: ROLL TWO DICE

We roll two standard six-sided dice, one red and one green.

We consider this to be a sample of size n = 2 from the original distribution D for


one die.

Let X 1 = the result of the red die.


Let X 2 = the result of the green die.

We will consider:

∑X = X 1 + X 2 = the sample sum (or total) of the dice

X1 + X 2
X = = the sample mean (or average) of the dice
2

Table of Sums (∑ X ) ( )
Table of Means X
(Lesson 22: The Central Limit Theorem (CLT)) 22.03

The spike plot below describes two sampling distributions: the distribution of
sums (∑ X ) as well as the distribution of means ( X ) , although they use
different scales. Possible sums are between 2 and 12, while possible means are
still between 1 and 6, with a mean of 3.5 (as for one die).

3.5

We can say that “the mean of the means” (denoted by µ X ) is 3.5.


That is, the mean of the sampling distribution for the sample means is 3.5.

X~D

↙ ↘
Sample: X1 X2
↘ ↙
Sample Mean: X

X has the triangular distribution at the top


of the page, using the bottom (blue) scale.

Imagine 1000 people, each rolling a pair of dice. Each person takes a sample of
size n = 2 from D, the original, uniform distribution. Each person finds the sample
mean of the two dice; these are values of X . For example:

X1 X2 X
Sample #1 2 3 2.5
Sample #2 6 2 4.0
! ! ! !
Sample #1000 1 5 3.0
(Lesson 22: The Central Limit Theorem (CLT)) 22.04

The relative frequency histogram (or spike plot) for the values of X should
resemble the triangular distribution. This is because of the Law of Large
Numbers (LLN).

n=2

The sampling distribution for X has:

mean, µ X = µ = 3.5
σ 1.7078
SD or SE, σ X = ≈ ≈ 1.2076
n 2

σ X is also called the standard error (SE) of the mean.

• As the sample size n increases, σ X decreases. Notice that the probability


mass for the triangular distribution is more tightly clustered around x = 3.5
here for n = 2 than for n = 1. Here, the average of the dice is more likely to
be close to 3.5, the mean of the original distribution.

Think About It: As n, the number of dice in a sample, increases, what will the
shape of the sampling distribution for X begin to resemble? See the next page ….
(Lesson 22: The Central Limit Theorem (CLT)) 22.05
D = Uniform Standard Die Distribution; µ = 3.5 ; σ ≈ 1.7078
x P( x)
1 1/6 ≈ 0.16667
2 1/6 ≈ 0.16667
3 1/6 ≈ 0.16667
4 1/6 ≈ 0.16667
5 1/6 ≈ 0.16667
6 1/6 ≈ 0.16667
In the figures below, n dice are to be rolled. “N” means “Normal,” not population size.
n =1 n=2

n=4 n=8

n = 16 n = 32

CLT for Sums:


32 approx. ⎛ mean = nµ = 112, ⎞
∑ i X ∼ N ⎜ ⎟
⎝ SD = σ n ≈ 9.6608⎠
i=1

CLT for Means (or Averages):


approx.
⎛ mean = µ = 3.5, ⎞
X ∼ N⎜ σ ⎟
⎜ SD = ≈ 0.30190⎟
⎝ n ⎠
(Lesson 22: The Central Limit Theorem (CLT)) 22.06

PART C: ROLL 32 DICE

The Central Limit Theorem (CLT) applies when the sample size is large
enough; usually n > 30 is the standard.

It also applies when the original distribution D is approximately normal,


regardless of sample size.

( )
According to the CLT for Means, if a machine rolls 32 dice n = 32 , then the
sampling distribution for X , is approximately normal:

n = 32

The sampling distribution for X has:

mean, µ X = µ = 3.5
σ 1.7078
SD or SE, σ X = ≈ ≈ 0.30190
n 32
Therefore:

( )
approx.
X ∼ N µ X = 3.5, σ X ≈ 0.30190

We expect the sample mean to be very close to 3.5, since we expect high and low
numbers to have a strong tendency to balance each other out. The standard
error (SE) is about 0.3, much smaller than the 1.7 or so that we started with for σ .
(Lesson 22: The Central Limit Theorem (CLT)) 22.07

( )
In Lesson 9, we saw the “Two SD” 2σ Rule for Usual Values.

( )
We can extend this to the following “Two SE” 2σ X Rule for Usual Values of a
Sample Mean.

( )
The "Two SE" 2σ X Rule for Usual Values of a Sample Mean

An appropriate interval of usual values for a sample mean is given by


( µ − 2σ X )
, µ + 2σ X .

Example 1 (Interval of Usual Values for a Sample Mean: 32 Dice)


A machine rolls 32 standard six-sided dice. What is the interval of usual
values of the sample mean X , based on the “Two SE” Rule?

Use the following distribution for X ; round off the value of σ X to 0.3.

( )
approx.
X ∼ N µ X = 3.5, σ X ≈ 0.3

§ Solution

µ − 2σ X ≈ 3.5 − 2 ( 0.3) ≈ 3.5 − 0.6 ≈ 2.9


µ + 2σ X ≈ 3.5 + 2 ( 0.3) ≈ 3.5 + 0.6 ≈ 4.1

The interval of usual values of the sample mean is given by:


( µ − 2σ X )
, µ + 2σ X , which is ( 2.9, 4.1) here.

• Since X has an approximately normal distribution, the “68-95-99.7%”


Empirical Rule applies, and we can say that the probability that the sample
mean will be between 2.9 and 4.1 is about 95%, or 0.95.
§
(Lesson 22: The Central Limit Theorem (CLT)) 22.08

( )
When finding probabilities for the sample mean X , we need to adapt our
Formula for z Scores:

Formula for z Scores for Sample Means

x − mean x−µX x−µ


z = = , or (Round off to two decimal places.)
SE σX σ
n

Example 2 (Probabilities for a Sample Mean: 32 Dice)


A machine rolls 32 standard six-sided dice.
The distribution for X , the result on one die, has:
mean, µ = 3.5
SD, σ ≈ 1.7078

( )
Find the probability that the average of the 32 dice X will be between 3.0

( )
and 4.0. That is, find P 3.0 < X < 4.0 .

Use these hints regarding the Z distribution:


(Lesson 22: The Central Limit Theorem (CLT)) 22.09

§ Solution

The sampling distribution for X has:

mean, µ X = µ = 3.5
σ 1.7078
SD or SE, σ X = ≈ ≈ 0.30190
n 32

The sample size n = 32 . Since n > 30 , the CLT applies, so we use the
following distribution for X :

( )
approx.
X ∼ N µ X = 3.5, σ X ≈ 0.30190

( )
We want to find P 3.0 < X < 4.0 . Take the boundary x scores, 3.0 and
4.0, and transform them into z scores by using the Formula for z Scores for
Sample Means:
x−µX 3.0 − 3.5
x = 3.0 ⇒ z = ≈ ≈ − 1.66
σX 0.30190

x−µX 4.0 − 3.5


x = 4.0 ⇒ z = ≈ ≈ 1.66
σX 0.30190
(can use symmetry about the mean, 3.5)

Write the corresponding probability expression for Z ; use the given


probabilities in the figures:

P ( 3.0 < X < 4.0 ) ≈ P ( −1.66 < Z < 1.66 )


≈ 0.9515 − 0.0485
≈ 0.9030
(Lesson 22: The Central Limit Theorem (CLT)) 22.10

• Note: If we are more interested in the sum (or total) of the dice, observe
( ) ( )
that: P 3.0 < X < 4.0 = P 96 < ∑ X < 128 . Here, a sum is 32 times an
average. The CLT for Sums implies that:
32 ⎛ mean = nµ = 112,
approx. ⎞
∑ i X ∼ N ⎜ ⎟
⎝ SD = σ n ≈ 9.6608⎠
i=1

PART D: THE CENTRAL LIMIT THEOREM (CLT) FOR MEANS

The Central Limit Theorem (CLT) for Means


Assume X ~ D with mean µ and finite SD σ .

Let X be the sample mean of n iid (independent and identically


distributed) results from D.

( )
If the sample size n is large n > 30 , or if D is approximately normal, then
approx.
⎛ σ ⎞
X ∼ N ⎜ mean = µ X = µ , SD or SE = σ X = ⎟
⎝ n⎠

• Note: The CLT works best when D is symmetric and does not have thick
tails.

An asymmetric distribution D is analyzed in the table on the next page. For small
values of n, the asymmetry is still evident, but it becomes less of an issue by the
n = 32 case.
(Lesson 22: The Central Limit Theorem (CLT)) 22.11.

D = Biased Die Distribution; µ ≈ 1.6667 ; σ ≈ 0.74536


(Three faces are “1”s. Two faces are “2”s. One face is a “3.”)

x P x()
1 1/2 = 0.5
2 1/3 ≈ 0.33333
3 1/6 ≈ 0.16667
In the figures below, n dice are to be rolled. “N” means “Normal,” not population size.
n =1 n=2

n=4 n=8

n = 16 n = 32

CLT for Sums:


32 approx. ⎛ mean = nµ ≈ 53.333, ⎞
∑ i ∼ N ⎜⎝ SD = σ n ≈ 4.2164⎟⎠
X
i=1

CLT for Means (or Averages):


approx.
⎛ mean = µ ≈ 1.6667, ⎞
X ∼ N⎜ σ ⎟
⎜ SD = ≈ 0.13176⎟
⎝ n ⎠
(Lesson 23: Normal Approximations to Binomial Distributions) 23.01

LESSON 23: NORMAL APPROXIMATIONS TO BINOMIAL


DISTRIBUTIONS
How Can Normal Distributions Help Us Analyze Binomial
Distributions?
PART A: WHEN CAN A NORMAL DISTRIBUTION APPROXIMATE A
BINOMIAL DISTRIBUTION?

See the gallery of 50 binomial histograms on the next page.

Think About It: Which of these histograms looks closest to normal?

⎛ 1 ⎞
In fact, Bin ⎜ n = 10, p = or 0.5⎟ is the most basic binomial distribution that can
⎝ 2 ⎠
be approximated by a normal distribution.

Normal approximations to binomial distributions are more appropriate if:

• The sample size n is large, and

• p and q are not close to 0 or 1.


1
p= is ideal.
2
Remember that p is the success probability on a trial and that
q is the complementary failure probability; q = 1− p .

Conditions for Using Normal Approximations to Binomial Distributions


We will use normal approximations to binomial distributions
if and only if the following conditions apply:
• np ≥ 5 , and
• nq ≥ 5
(Lesson 23: Normal Approximations to Binomial Distributions) 23.02

GALLERY OF BINOMIAL HISTOGRAMS

p = 1/8 p = 1/4 p = 1/2 p = 3/4 p = 7/8

n=1

n=2

n=3

n=4

n=5

n=6

n=7

n=8

n=9

n = 10
(Lesson 23: Normal Approximations to Binomial Distributions) 23.03

PART B: PARAMETERS FOR BINOMIAL DISTRIBUTIONS AND THEIR


NORMAL APPROXIMATIONS

Mean and SD of a Binomial Distribution

( )
If X ∼ Bin n, p , then:
mean, µ = np
SD, σ = npq

If a normal approximation is appropriate, then it makes sense to use these values


of µ and σ for the normal distribution that we use to approximate the binomial
distribution.

Using Normal Approximations to Binomial Distributions

( )
Let X ∼ Bin n, p . If np ≥ 5 and nq ≥ 5 , then:

( )
approx.
X ∼ N mean = µ = np, SD = σ = npq
(Lesson 23: Normal Approximations to Binomial Distributions) 23.04

PART C: CONTINUITY CORRECTIONS

We use continuity corrections to adjust for the fact that we are using a continuous
distribution (a normal distribution) to approximate a discrete distribution
(a binomial distribution).

⎛ 1⎞
( )
Let’s say we want to find P 45 ≤ X ≤ 55 , where X ∼ Bin ⎜ n = 100, p = ⎟ .
⎝ 2⎠
Since the distribution is discrete, it may matter whether we use “<” or “< .”

• We associate the integer value a in the binomial distribution with the interval
( )
a − 0.5, a + 0.5 in the approximating normal distribution. Think: “rounding.”

• We associate the integer value “45” in the binomial distribution with the interval
( )
44.5, 45.5 in the approximating normal distribution.

• We associate the integer value “55” in the binomial distribution with the interval
( )
54.5, 55.5 in the approximating normal distribution.

( )
To approximate P 45 ≤ X ≤ 55 for the binomial random variable X ,
(
we will use continuity corrections and find P 44.5 ≤ X c ≤ 55.5 )
for the normal random variable X c . Think: “corrected X.”

( )
• X c is continuous, so we may consider P 44.5 < X c < 55.5 instead.
(Lesson 23: Normal Approximations to Binomial Distributions) 23.05

PART D: COIN EXAMPLES

Example 1 (Normal Approximation to a Binomial Distribution: Coins)


We will flip a fair coin 100 times.
Let X = the number of heads in the 100 flips.

( )
Approximate P 45 ≤ X ≤ 55 by following these steps:

• a) Describe the distribution of X.


• b) Verify that a normal approximation to the distribution of X would be
appropriate.
• c) Describe the normal distribution that can be used to approximate the
distribution of X.

( )
• d) Apply continuity corrections and rewrite P 45 ≤ X ≤ 55 in terms of
Xc .

• e) Find the z scores for the boundary values of xc using the Formula for
z Scores.
• f) Write the corresponding probability expression for Z .

( )
• g) Approximate P 45 ≤ X ≤ 55 . Use these hints regarding the Z
distribution:
(Lesson 23: Normal Approximations to Binomial Distributions) 23.06

§ Solution
• a) Describe the distribution of X.

⎛ 1⎞
X ∼ Bin ⎜ n = 100, p = ⎟
⎝ 2⎠

• b) Verify that a normal approximation to the distribution of X would be


appropriate.
⎛ 1⎞
( )
np = 100 ⎜ ⎟ = 50 ≥ 5
⎝ 2⎠
⎛ 1⎞ ⎡ 1 1⎤
nq = (100 ) ⎜ ⎟ = 50 ≥ 5 ⎢ q = 1− p = 1− =
⎝ 2⎠ ⎣ 2 2 ⎥⎦

µ or E ( X ) = np = 50 , the “expected number of successes (heads).”


nq = 50 , the “expected number of failures (tails).”

• c) Describe the normal distribution that can be used to approximate the


distribution of X.
The parameters for both binomial and normal distributions are:
⎛ 1⎞
mean, µ = np = (100 ) ⎜ ⎟ = 50 ⎡⎣ heads ⎤⎦
⎝ 2⎠

SD, σ = npq = (100) ⎛⎜⎝ 12 ⎞⎟⎠ ⎛⎜⎝ 12 ⎞⎟⎠ = 5 ⎡⎣ heads ⎤⎦

approx.
(
Therefore, X ∼ N µ = 50, σ = 5 )
(
• d) Apply continuity corrections and rewrite P 45 ≤ X ≤ 55 in terms of )
Xc .
( )
From Part C, we found that: P 45 ≤ X ≤ 55 ≈ P 44.5 < X c < 55.5 . ( )
(Lesson 23: Normal Approximations to Binomial Distributions) 23.07

• e) Find the z scores for the boundary values of xc using the Formula for
z Scores.
x−µ 44.5 − 50
xc = x = 44.5 ⇒ z = = ≈ − 1.10
σ 5
x−µ 55.5 − 50
xc = x = 55.5 ⇒ z = = ≈ 1.10
σ 5
• f) Write the corresponding probability expression for Z .

P ( 44.5 < X c < 55.5) ≈ P ( −1.10 < Z < 1.10 )

(
• g) Approximate P 45 ≤ X ≤ 55 . )
P ( −1.10 < Z < 1.10 ) ≈ 0.8643− 0.1357
≈ 0.7286

• Software may be able to use the Binomial Probability Formula


⎛ n ⎞ x n−x
P ( x ) = n Cx p x q n−x , or ⎜ ⎟pq
⎝ x ⎠

( ) ( ) ( )
and calculate P 45 + P 46 + ...+ P 55 to get a more precise result.

• The actual probability to six decimal places is 0.728747, so our answer of


0.7286 was quite good!

• If we had not applied the continuity corrections, our answer would have
been about 0.6826.
§
(Lesson 23: Normal Approximations to Binomial Distributions) 23.08

Example 2 (Continuity Corrections: Coins)

Apply continuity corrections and rewrite the following in terms of X c .

( )
• a) The probability of at most 45 heads, P X ≤ 45 .

(
• b) The probability of more than 55 heads, P X > 55 . )
§ Solution
• a)

P ( X ≤ 45) ≈ P ( X c < 45.5)

• b)

P ( X > 55) ≈ P ( X c > 55.5)


§
(Lesson 23: Normal Approximations to Binomial Distributions) 23.09.

PART E: WHY ARE NORMAL DISTRIBUTIONS USEFUL?

• They are often seen in practice. This is why the “68-95-97%” Rule for
normal distributions is called the Empirical Rule.

• According to the Central Limit Theorem (CLT), averages of iid


(independent and identically distributed) random variables have
approximately normal distributions under robust conditions (the CLT applies
in a great many situations).

• Binomial distributions can often be approximated by normal


distributions.

PART F: WHY DO THESE NORMAL APPROXIMATIONS TO BINOMIAL


DISTRIBUTIONS WORK? (OPTIONAL)

The Central Limit Theorem (CLT) can be applied to a sum of many indicator
variables. In our coin examples, we consider the sum of 100 indicator variables:

You might also like