Lecture 6 : The Normal Distribution

Jonathan Marchini

November 12, 2004

1 Introduction
In previous lectures we have considered discrete datasets and discrete probability
distributions. In practice many datasets that we collect from experiments consist
of continuous measurements. For example, Figures 1, 2, 3 and 4 show histograms
of real datasets consisting of continuous measurements. From such samples of
continuous data we might want to test whether the data is consistent with a spe-
cific population mean value or whether there is a significant difference between
2 groups of data. To answer these question we need a probability model for the
data. The Normal distribution is one such model and is used extensivley through-
out statistics.
10
8
Frequency
6
4
2
0

1000 2000 3000 4000 5000 6000

Birth weight (g)

Figure 1: The birth weights of the babies in the Babyboom dataset

1
 10
8
Frequency
6
4
2
0

700000 800000 900000 1100000

Brain size

Figure 2: The brain sizes of 40 Psychology students

12
10
8
Frequency
6
4
2
0

1.0 1.2 1.4 1.6 1.8

Petal length

Figure 3: The petal length of a type of flower

2
 60
Frequency
40
20
0

0 100 200 300 400

Serum level

Figure 4: Serum level measurements from healthy volunteers

2 Continuous probability distributions

When we considered the Binomial and Poisson distributions we saw that the prob-
ability distributions were characterized by a formula for the probability of each
possible discrete value. All of the probabilities together sum up to 1. We can vi-
sualize the density by plotting the probabilities against the discrete values (Figure
5). For continuous data we don’t have equally spaced discrete values so instead
we use a curve or function that describes the probability density over the range of
the distribution (Figure 6). The curve is chosen so that the area under the curve is
equal to 1. If we observe a sample of data from such a distribution we should see
that the values occur in regions where the density is highest.

3
 0.00 0.02 0.04 0.06 0.08 0.10 0.12
P(X)

0 5 10 15 20
X

Figure 5: A discrete probability distribution

0.04
0.03
density
0.02
0.01
0.00

60 80 100 120 140

X

Figure 6: A continuous probability distribution

4
3 The Normal Distribution
There will be many, many possible probability density functions over a contin-
uous range of values. The Normal distribution describes a special class of such
distributions that are symmetric and can be described by the distribution mean µ
and the standard deviation σ (or variance σ 2 ). 4 different Normal distributions are
shown in Figure 7 together with the values of µ and σ. These plots illustrate how
changing the values of µ and σ alter the positions and shapes of the distributions.

If X is Normally distributed with mean µ and standard deviation σ, we write

X∼N(µ, σ 2 )
µ and σ are the parameters of the distribution.

The probability density of the Normal distribution is given by

1 2 2
f (x) = √ exp−(x−µ) /2σ
σ 2π
For the purposes of this course we do not need to use this expression. It is included
here for future reference.

µ = 100 σ = 10 µ = 100 σ = 5
0.08

0.08
density

density
0.04

0.04
0.00

0.00

50 100 150 50 100 150

X X

µ = 130 σ = 10 µ = 100 σ = 15
0.08

0.08
density

density
0.04

0.04
0.00

0.00

50 100 150 50 100 150

X X

Figure 7: 4 different Normal distributions

5
3.1 Calculating probabilities from the Normal distribution
For a discrete probability distribution we calculate the probability of being less
than some value x, i.e. P (X < x), by simply summing up the probabilities of the
values less than x.

For a continuous probability distribution we calculate the probability of being

less than some value x, i.e. P (X < x), by calculating the area under the curve to
the left of x.

For example, suppose X ∼ N(0, 1) and we want to calculate P (X < 0) ?

P(Z < 0)

0
For this example we can calculate the required area as we know the distribution is
symmetric and the total area under the curve is equal to 1, i.e. P (X < 0) = 0.5.

6
 What about P (X < 1.0)?

P(Z < 1)

0 1
Calculating this area is not easy1 and so we use probability tables. Probability
tables are tables of probabilities that have been calculated on a computer. All we
have to do is identify the right probability in the table and copy it down! Obvi-
ously it is impossible to tabulate all possible probabilities for all possible Normal
distributions so only one special Normal distribution, N(0, 1), has been tabulated.

The N(0, 1) distribution is called the standard Normal distribution.

The tables allow us to read off probabilities of the form P (Z < z). Most of
the table in the formula book has been reproduced in Table 3.1. From this table
we can identify that P (X < 1.0) = 0.8413 (this probability has been highlighted
with a box)

1
For those Mathematicians who recognize this area as a definite integral and try to do the
integral by hand please note that the integral cannot be evaluated analytically

7
 0 z

z 0.0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 0.5000 5040 5080 5120 5160 5199 5239 5279 5319 5359
0.1 0.5398 5438 5478 5517 5557 5596 5636 5675 5714 5753
0.2 0.5793 5832 5871 5910 5948 5987 6026 6064 6103 6141
0.3 0.6179 6217 6255 6293 6331 6368 6406 6443 6480 6517
0.4 0.6554 6591 6628 6664 6700 6736 6772 6808 6844 6879
0.5 0.6915 6950 6985 7019 7054 7088 7123 7157 7190 7224
0.6 0.7257 7291 7324 7357 7389 7422 7454 7486 7517 7549
0.7 0.7580 7611 7642 7673 7704 7734 7764 7794 7823 7852
0.8 0.7881 7910 7939 7967 7995 8023 8051 8078 8106 8133
0.9 0.8159 8186 8212 8238 8264 8289 8315 8340 8365 8389
1.0 0.8413 8438 8461 8485 8508 8531 8554 8577 8599 8621
1.1 0.8643 8665 8686 8708 8729 8749 8770 8790 8810 8830
Table 1: N(0, 1) probability table

8
 Once we can know how to read tables we can calculate lots of other probabil-
ities

Example 1 P (X > 0.92)

P(Z > 0.92) P(Z < 0.92)

0 0.92 0 0.92
We know that P (X > 0.92) = 1 − P (X < 0.92) and we can calculate P (X <
0.92) from the tables.

Thus, P (X > 0.92) = 1 − 0.8212 = 0.1788

Example 2 P (X > −0.5)?

P(X > −0.5) P(X < 0.5)

−0.5 0 0 0.5
The Normal distribution is symmetric so we know that P (X > −0.5) = P (X <
0.5) = 0.6915

Example 3 We can use the symmetry of the Normal distribution to calculate

P (X < −0.76) = P (X > 0.76) = 1 − P (X < 0.76) = 1 − 0.7764 = 0.2236

P(X < −0.76) P(X < 0.76)

−0.76 0 0 0.76
9
Example 4 P (−0.64 < X < 0.43)

P(−0.64 < X < 0.43)

−0.64 0 0.43
P(X < −0.64) P(X < 0.43)

−0.64 0 0 0.43
We can calculate this using

P (−0.64 < X < 0.43) = P (X < 0.43) − P (X < −0.64)

= 0.6664 − (1 − 0.7389)
= 0.4053

Example 5 Consider P (X < 0.567)?

From tables we know that P (X < 0.56) = 0.7123 and P (X < 0.56) = 0.7157
To calculate P (X < 0.567) we interpolate between these two values

P (X < 0.567) = 0.3 × 0.7123 + 0.7 × 0.7157 = 0.71468

10
3.2 Standardization
All of the probabilities above were calculated for the standard Normal distribution
N(0, 1). If we want to calculate probabilities from different Normal distributions
we convert the probability to one involving the standard Normal distribution. This
process is called standardization.

Suppose X ∼ N(3, 4) and we want to calculate P (X < 6.2). We convert this

probability to one involving the N(0, 1) distribution by

(i) Subtracting the mean µ

(ii) Dividing by the standard deviation σ

Subtracting the mean re-centers the distribution on zero. Dividing by the standard
deviation re-scales the distribution so it has standard deviation 1. If we also trans-
form the boundary point of the area we wish to calculate we obtain the equivalent
boundary point for the N(0, 1) distribution. This process is illustrated in the figure
below. In this example, P (X < 6.2) = P (Z < 1.6) = 0.9452 where Z ∼ N(0,1)

N(3, 4)

3 6.2
−3 => N(0, 4)

0 3.2
/ 2 => N(0, 1)

0 1.6
This process can be described by the following rule
X−µ
If X ∼ N(µ, σ 2 ) and Z = σ
then Z ∼ N(0, 1)

11
Example 6
Suppose we know that the birth weight of babies is Normally distributed with
mean 3500g and standard deviation 500g. What is the probability that a baby is
born that weighs less than 3100g?

That is X ∼ N(3500, 5002 ) and we want to calculate P (X < 3100)?

We can calculate the probability through the process of standardization.

Drawing a rough diagram of the process can help you to avoid any confusion
about which probability (area) you are trying to calculate.

X ~ N(3500, 5002 ) Z ~ N(0, 1)

P(X < 3100) P(Z < −0.8)

Z = X − 3500
500

3100 3500 3100 − 3500 0

500
= −0.8

!
X − 3500 3100 − 3500
P (X < 3100) = P < = P (Z < −0.8) where Z ∼ N(0, 1)
500 500
= 1 − P (Z < 0.8)
= 1 − 0.7881
= 0.2119

12
3.3 Linear combinations of Normal random variables
Suppose two rats A and B have been trained to navigate a large maze. The time it
takes rat A is normally distributed with mean 80 seconds and standard deviation
10 seconds. The time it takes rat B is normally distributed with mean 78 seconds
and standard deviation 13 seconds. On any given day what is the probability that
rat A runs the maze faster than rat B?

X = Time of run for rat A X ∼ N(80, 102 )

Y = Time of run for rat B Y ∼ N(78, 132 )

Let D = X − Y be the difference in times of rats A and B

If rat A is faster than rat B then D < 0 so we want P (D < 0)?

To calculate this probability we need to know the distribution of D. To do this

we use the following rule
If X and Y are two independent normal variable such that

X ∼ N(µ1 , σ12 ) and Y ∼ N(µ2 , σ22 )

then X − Y ∼ N(µ1 - µ2 , σ12 + σ22 )

In this example,
D = X − Y ∼ N(80 − 78, 102 + 132 ) = N (2, 269)
We can now calculate this probability through standardization

D ~ N(2, 269) Z ~ N(0, 1)

P(D < 0) P(Z < −0.122)

Z=D−2
16.40

0 2 0−2 0
16.40
= −0.122

!
D−2 0−2
P (D < 0) = P √ <√ = P (Z < −0.122) where Z ∼ N(0, 1)
269 269
= 1 − (0.8 × 0.5478 + 0.2 × 0.5517)
= 0.45142

13
Other rules that are often used are
If X and Y are two independent normal variable such that

X ∼ N(µ1 , σ12 ) and Y ∼ N(µ2 , σ22 )

then

X + Y ∼ N(µ1 + µ2 , σ12 + σ22 )

aX ∼ N(aµ1 , a2 σ12 )
aX + bY ∼ N(aµ1 + bµ2 , a2 σ12 + b2 σ22 )

Example 7 Suppose two rats A and B have been trained to navigate a large maze.
The time it takes rat A is normally distributed with mean 80 seconds and standard
deviation 10 seconds. The time it takes rat B is normally distributed with mean 78
seconds and standard deviation 13 seconds. On any given day what is the proba-
bility that the average time the rats take to run the maze is greater than 82 seconds?

X = Time of run for rat A X ∼ N(80, 102 )

Y = Time of run for rat B Y ∼ N(78, 132 )

X+Y
Let A = 2
= 21 X + 12 Y be the average time of rats A and B
 
1 1 1 2 2 1 2 2
Then A ∼ N 2 80 + 2 78, ( 2 ) 10 + ( 2 ) 13 = N(79, 67.25)

We want P (A > 82)

A ~ N(79, 67.25) Z ~ N(0, 1)

P(A > 82) P(Z > 0.366)

Z = A − 79
8.20

79 82 0 82 − 79
8.20
= 0.366
!
A − 79 82 − 79
P (A > 82) = P √ < √ = P (Z > 0.366) where Z ∼ N(0, 1)
67.25 67.25
= 1 − (0.4 × 0.6406 + 0.6 × 0.6443)
= 0.35718

14
3.4 Using the Normal tables backwards
Example 8
The marks of 500 candidates in an examination are normally distributed wit a
mean of 45 marks and a standard deviation of 20 marks.

If 20% of candidates obtain a distinction by scoring x marks or more, estimate

the value of x.

We have X ∼ N(45, 202 ) and we want x such that P (X > x) = 0.2

⇒ P (X < x) = 0.8

X ~ N(45, 400) Z ~ N(0, 1)

P(X < x) = 0.8 P(Z < 0.84) = 0.8

Z = X − 45
20

45 x 0 x − 45
20
= 0.84

Standardizing this probability we get

!
X − 45 x − 45
P < = 0.8
20 20
!
x − 45
⇒P Z< = 0.8
20

From the tables we know that P (Z < 0.84) ≈ 0.8 so

x − 45
≈ 0.84
20
⇒ x ≈ 45 + 20 × 0.84 = 61.8

15
4 The Normal approximation to the Binomial
Under certain conditions we can use the Normal distribution to approximate the
Binomial distribution. This can be very useful when we need to sum up a large
number of Binomial probabilities to calculate the probability that we want.

For example, Figure 8 compares a Bin(300, 0.5) and a N(150, 75) which both
have the same mean and variance. The figure shows that the distributions are very
similar.

Bin(300, 0.5) N(150, 75)

0.04

0.04
0.03

0.03
P(X = x)

density
0.02

0.02
0.01

0.01
0.00

0.00

100 120 140 160 180 200 100 120 140 160 180 200
X X

Figure 8: Comparison of a Bin(300, 0.5) and a N(150, 75) distribution

In general

If X ∼ Bin(n, p) then

µ = np
σ 2 = npq where q =1−p

For large n and p not too small or too large

X ∼ N(np, npq)
1 1
n > 10 and p ≈ 2
OR n > 30 and p moving away from 2

16
Example 8
Suppose X ∼Bin(12, 0.5) what is P (4 ≤ X ≤ 7)?

For this distribution we have

µ = np = 6
σ 2 = npq = 3

So we can use a N(6, 3) distribution as an approximation.

Unfortunately, it’s not quite so simple. We have to take into account the fact
that we are using a continuous distribution to approximate a discrete distribution.
This is done using a continuity correction. The continuity correction appropriate
for this example is illustrated in the figure below

In this example, P (4 ≤ X ≤ 7) transforms to P (3.5 < X < 7.5)

0 1 2 3 4 5 6 7 8 9 10 11 12
3.5 7.5

!
3.5 − 6 X −6 7.5 − 6
P (3.5 < X < 7.5) = P √ < √ < √
3 3 3
= P (−1.443 < Z < 0.866) where Z ∼ N(0, 1)
= 0.732

The exact answer is 0.733 so in this case the approximation is very good.

17
5 The Normal approximation to the Poisson
We can also use the Normal distribution to approximate a Poisson distribution un-
der certain conditions.

In general,
If X ∼ Po(λ) then

µ = λ
σ2 = λ

For large λ (say λ > 20)

X ∼ N(λ, λ)
Example 9 A radioactive source emits particles at an average rate of 25 particles
per second. What is the probability that in 1 second the count is less than 28 par-
ticles?

X = No. of particles emitted in 1 second X ∼ Po(25)

So, we can use a N(25, 25) as an approximate distribution.

Again, we need to make a continuity correction

So P (X < 27) transforms to P (X < 26.5)

X ~ N(25, 25) Z ~ N(0, 1)

P(X < 26.5) P(Z < 0.3)

Z = X − 25
5

25 26.5 0 26.5 − 25
5
= 0.3

!
X − 25 26.5 − 25
P (X < 26.5) = P <
5 5
= P (Z < 0.3) where Z ∼ N(0, 1)
= 0.6179

18
Lecture 6 : The Normal Distribution

Jonathan Marchini
 Continuous data

In previous lectures we have considered discrete

datasets and discrete probability distributions. In prac-
tice many datasets that we collect from experiments
consist of continuous measurements.
So we need to study probability models for continuous
data.
The birth weights of the babies in the Babyboom dataset

10
8
Frequency
6
4
2
0

1000 2000 3000 4000 5000 6000

Birth weight (g)
The brain sizes of 40 students

10
8
Frequency
6
4
2
0

700000 800000 900000 1100000

Brain size
The petal length of a type of flower

12
10
8
Frequency
6
4
2
0

1.0 1.2 1.4 1.6 1.8

Petal length
Serum level measurements from healthy volunteers

60
Frequency
40
20
0

0 100 200 300 400

Serum level
 Continuous probability distributions

When we considered the Binomial and Poisson distri-

butions we saw that the probability distributions were
characterized by a formula for the probability of each
possible discrete value.
All of the probabilities together sum up to 1.
We can visualize the density by plotting the probabili-
ties against the discrete values.
A discrete probability distribution

0.00 0.02 0.04 0.06 0.08 0.10 0.12

P(X)

0 5 10 15 20
X
For continuous data we don’t have equally spaced
discrete values so instead we use a curve or function
that describes the probability density over the range of
the distribution.
The curve is chosen so that the area under the curve is
equal to 1.
If we observe a sample of data from such a distribution
we should see that the values occur in regions where
the density is highest.
A continuous probability distribution

0.04
0.03
density
0.02
0.01
0.00

60 80 100 120 140

X
 The Normal Distribution

There will be many, many possible probability density

functions over a continuous range of values.
The Normal distribution describes a special class of
such distributions that are symmetric and can be de-
scribed by two parameters
(i) µ = The mean of the distribution
(ii) σ = The standard deviation of the distribution
Changing the values of µ and σ alter the positions and
shapes of the distributions.
 µ = 100 σ = 10 µ = 100 σ = 5

0.08

0.08
density

density
0.04

0.04
0.00

0.00
50 100 150 50 100 150
X X

µ = 130 σ = 10 µ = 100 σ = 15
0.08

0.08
density

density
0.04

0.04
0.00

0.00

50 100 150 50 100 150

X X
If X is Normally distributed with mean µ and standard
deviation σ, we write
X∼N(µ, σ 2)
µ and σ are the parameters of the distribution.

The probability density of the Normal distribution is

given by
1 −(x−µ) 2
/2σ 2
f (x) = √ exp
σ 2π
For the purposes of this course we do not need to use
this expression. It is included here for future reference.
Calculating probabilities from the Normal distribution

For a discrete probability distribution we calculate

the probability of being less than some value x, i.e.
P (X < x), by simply summing up the probabilities of
the values less than x.
For a continuous probability distribution we calculate
the probability of being less than some value x, i.e.
P (X < x), by calculating the area under the curve to
the left of x.
Suppose Z ∼ N(0, 1), what is P (Z < 0) ?
P(Z < 0)

0
Symmetry ⇒ P (Z < 0) = 0.5
What about P (Z < 1.0)?
P(Z < 1)

0 1
Calculating this area is not easy and so we use proba-
bility tables. Probability tables are tables of probabil-
ities that have been calculated on a computer. All we
have to do is identify the right probability in the table
and copy it down!
Only one special Normal distribution, N(0, 1), has
been tabulated.
The N(0, 1) distribution is called
the standard Normal distribution.
The tables allow us to read off probabilities of the form
P (Z < z).

0 z
 z 0.0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 0.5000 5040 5080 5120 5160 5199 5239 5279 5319 5359
0.1 0.5398 5438 5478 5517 5557 5596 5636 5675 5714 5753
0.2 0.5793 5832 5871 5910 5948 5987 6026 6064 6103 6141
0.3 0.6179 6217 6255 6293 6331 6368 6406 6443 6480 6517
0.4 0.6554 6591 6628 6664 6700 6736 6772 6808 6844 6879
0.5 0.6915 6950 6985 7019 7054 7088 7123 7157 7190 7224
0.6 0.7257 7291 7324 7357 7389 7422 7454 7486 7517 7549
0.7 0.7580 7611 7642 7673 7704 7734 7764 7794 7823 7852
0.8 0.7881 7910 7939 7967 7995 8023 8051 8078 8106 8133
0.9 0.8159 8186 8212 8238 8264 8289 8315 8340 8365 8389
1.0 0.8413 8438 8461 8485 8508 8531 8554 8577 8599 8621
1.1 0.8643 8665 8686 8708 8729 8749 8770 8790 8810 8830

From this table we can identify that P (Z < 1.0) = 0.8413

 Example 1

If Z ∼ N(0, 1) what is P (Z > 0.92)?

P(Z > 0.92) P(Z < 0.92)

0 0.92 0 0.92

We know that P (Z > 0.92) = 1 − P (Z < 0.92) and we

can calculate P (Z < 0.92) from the tables.

Thus, P (Z > 0.92) = 1 − 0.8212 = 0.1788

 Example 2

If Z ∼ N(0, 1) what is P (Z > −0.5)?

P(Z > −0.5) P(Z < 0.5)

−0.5 0 0 0.5
The Normal distribution is symmetric so we know that
P (Z > −0.5) = P (Z < 0.5) = 0.6915
 Example 3

If Z ∼ N(0, 1) what is P (Z < −0.76)?

P(Z < −0.76) P(Z < 0.76)

−0.76 0 0 0.76
By symmetry
P (Z < −0.76) = P (Z > 0.76) = 1 − P (Z < 0.76)
= 1 − 0.7764
= 0.2236
 Example 4

If Z ∼ N(0, 1) what is P (−0.64 < Z < 0.43)?

P(−0.64 < Z < 0.43)

−0.64 0 0.43
P(Z < −0.64) P(Z < 0.43)

−0.64 0 0 0.43
We can calculate this probability as
P (−0.64 < Z < 0.43) = P (Z < 0.43) − P (Z < −0.64)
= 0.6664 − (1 − 0.7389)
= 0.4053
 Example 5

Consider P (Z < 0.567)?

From tables we know that P (Z < 0.56) = 0.7123

and P (Z < 0.57) = 0.7157
To calculate P (Z < 0.567) we interpolate between these
two values
P (Z < 0.567) = 0.3 × 0.7123 + 0.7 × 0.7157 = 0.71468
 Standardization

All of the probabilities above were calculated for the

standard Normal distribution N(0, 1). If we want
to calculate probabilities from different Normal
distributions we convert the probability to one
involving the standard Normal distribution.

This process is called standardization.

Suppose X ∼ N(3, 4), what is P (X < 6.2)?
N(3, 4)

3 6.2
−3 => N(0, 4)

0 3.2
/ 2 => N(0, 1)

0 1.6
We convert this probability to one involving the
N(0, 1) distribution by
(i) Subtracting the mean µ
(ii) Dividing by the standard deviation σ
Subtracting the mean re-centers the distribution on
zero. Dividing by the standard deviation re-scales the
distribution so it has standard deviation 1. If we also
transform the boundary point of the area we wish to
calculate we obtain the equivalent boundary point for
the N(0, 1) distribution.
⇒ P (X < 6.2) = P (Z < 1.6) = 0.9452 where Z ∼ N(0, 1)
This process can be described by the following rule

X−µ
If X ∼ N(µ, σ 2) and Z = σ
then
Z ∼ N(0, 1)
 Example 6

Suppose we know that the birth weight of babies is

Normally distributed with mean 3500g and standard
deviation 500g. What is the probability that a baby is
born that weighs less than 3100g?

That is X ∼ N(3500, 5002) and we want to calculate

P (X < 3100)?

We can calculate the probability through the process

of standardization.
Drawing a rough diagram helps

X ~ N(3500, 5002 ) Z ~ N(0, 1)

P(X < 3100) P(Z < −0.8)

Z = X − 3500
500

3100 3500 3100 − 3500 0

500
= −0.8
 !
X − 3500 3100 − 3500
P (X < 3100) = P <
500 500
= P (Z < −0.8) where Z ∼ N(0, 1)
= 1 − P (Z < 0.8)
= 1 − 0.7881
= 0.2119
Linear combinations of Normal random variables

Suppose two rats A and B have been trained to navigate

a large maze.
X = Time of run for rat A X ∼ N(80, 102)
Y = Time of run for rat B Y ∼ N(78, 132)
On any given day what is the probability that rat A runs
the maze faster than rat B?
Let D = X − Y be the difference in times of rats A and
B
If rat A is faster than rat B then D < 0 so we want
P (D < 0)?
To calculate this probability we need to know the
distribution of D. To do this we use the following rule
If X and Y are two independent normal
variable such that

X ∼ N(µ1, σ12) and Y ∼ N(µ2, σ22)

then X − Y ∼ N(µ1 - µ2, σ12 + σ22)

In this example,
D = X − Y ∼ N(80 − 78, 102 + 132) = N (2, 269)
We can now calculate this probability through stan-
dardization
D ~ N(2, 269) Z ~ N(0, 1)

P(D < 0) P(Z < −0.122)

Z=D−2
16.40

0 2 0−2 0
16.40
= −0.122
 !
D−2 0−2
P (D < 0) = P √ <√
269 269
= P (Z < −0.122) Z ∼ N (0, 1)

= 1 − (0.2 × 0.5478 + 0.8 × 0.5517)

= 0.45142
Other rules that are often used are
If X and Y are two independent normal
variables such that

X ∼ N(µ1, σ12) and Y ∼ N(µ2, σ22)

then
X + Y ∼ N(µ1 + µ2, σ12 + σ22)
aX ∼ N(aµ1, a2σ12)
aX + bY ∼ N(aµ1 + bµ2, a2σ12 + b2σ22)
 Using the Normal tables backwards

The marks of 500 candidates in an examination are

normally distributed with a mean of 45 marks and a
standard deviation of 20 marks.

If 20% of candidates obtain a distinction by scoring x

marks or more, estimate the value of x.

We have X ∼ N(45, 202) and we want x such that

P (X > x) = 0.2

⇒ P (X < x) = 0.8
X ~ N(45, 400) Z ~ N(0, 1)

P(X < x) = 0.8 P(Z < 0.84) = 0.8

Z = X − 45
20

45 x 0 x − 45
20
= 0.84
Standardizing this probability we get
!
X − 45 x − 45
P < = 0.8
20 20
!
x − 45
⇒P Z< = 0.8
20

From the tables we know that P (Z < 0.84) ≈ 0.8 so

x − 45
≈ 0.84
20
⇒ x ≈ 45 + 20 × 0.84 = 61.8
 The Normal approximation to the Binomial

Under certain conditions we can use the Normal distri-

bution to approximate the Binomial distribution.

Bin(300, 0.5) N(150, 75)

0.04

0.04
0.03

0.03
P(X = x)

density
0.02

0.02
0.01

0.01
0.00

0.00

100 120 140 160 180 200 100 120 140 160 180 200
X X
In general

If X ∼ Bin(n, p) then
µ = np
σ 2 = npq where q = 1 − p
For large n and p not too small or too large
X ∼ N(np, npq)

n > 10 and p ≈ 12 OR n > 30 and p moving

away from 12
 Example

Suppose X ∼Bin(12, 0.5) what is P (4 ≤ X ≤ 7)?

For this distribution we have
µ = np = 6
σ 2 = npq = 3
So we can use a N(6, 3) distribution as an
approximation.

Unfortunately, it’s not quite so simple. We have to take

into account the fact that we are using a continuous
distribution to approximate a discrete distribution.
This is done using a continuity correction.
0 1 2 3 4 5 6 7 8 9 10 11 12
3.5 7.5
P (4 ≤ X ≤ 7) transforms to P (3.5 < X < 7.5)
!
3.5 − 6 X − 6 7.5 − 6
P (3.5 < X < 7.5) = P √ < √ < √
3 3 3
= P (−1.443 < Z < 0.866) where Z ∼ N(0, 1)
= 0.732

The exact answer is 0.733 so in this case the approxi-

mation is very good.
 The Normal approximation to the Poisson

We can also use the Normal distribution to approxi-

mate a Poisson distribution under certain conditions.
In general,
If X ∼ Po(λ) then
µ = λ
σ2 = λ

For large λ (say λ > 20)

X ∼ N(λ, λ)
 Example

A radioactive source emits particles at an average rate

of 25 particles per second. What is the probability that
in 1 second the count is less than 27 particles?

X = No. of particles emitted in 1s X ∼ Po(25)

So, we can use a N(25, 25) as an approximate distribu-
tion.
Again, we need to make a continuity correction
So P (X < 27) transforms to P (X < 26.5)
X ~ N(25, 25) Z ~ N(0, 1)

P(X < 26.5) P(Z < 0.3)

Z = X − 25
5

25 26.5 0 26.5 − 25
5
= 0.3
 !
X − 25 26.5 − 25
P (X < 26.5) = P <
5 5
= P (Z < 0.3) where Z ∼ N(0, 1)
= 0.6179