The Normal Random Variable
By: Andrew Magyar
The Normal Random Variable
A normal rv is continuous rv with the following pdf,
2
( x−µ )
1 1 −
e 2σ 2
for − ∞ < x < ∞
2π σ
Where π = 3.141592645 and e = 2.718281828.
The support of a normal is the entire real number line.
The pdf of a normal is a unimodal, symmetric bell-shaped curve. There
are two parameters (numerical quantities) that dictate its form, its
expected value and variance/standard deviation.
E[X] = µ: This indicates where the bell-shaped curve is centered (its point
of symmetry). µ can be any real number, that is -∞ < µ < ∞.
Var[X] = σ 2. This determines the width of the bell-shaped curve. σ 2
must be a positive real number, that is σ 2 > 0.
Examples of Normal pdfs
Below are 4 different normals with varying values of µ and σ 2.
Note, just because the pdf of a rv has a bell-shape does not necessarily mean
it is that of a normal rv. There are other distributions that have pdfs with
bell-shapes.
Popularity of the Normal RV
The normal rv is the most commonly encountered continuous rv. Its
ubiquity stems from two main reasons:
i) For many real-life problems, the distribution of the data is
approximately normal (i.e. a histogram of the data is bell-shaped like the
pdf of a normal).
ii) Sample averages arise in many statistical analyses. Under certain
conditions, for large enough sample sizes, the sampling distribution of
the sample average closely follows the distribution of a normal rv, no
matter what the underlying distribution of the data being analyzed!!!
The normal rv is also called a Gaussian rv, named after its discoverer,
the famous German mathematician Carl Friedrich Gauss.
Notation for the Normal
If X is a normal rv with mean µ and variance σ 2, then this is denoted
X ~ Norm(µ, σ 2), read “X is distributed as a normal rv (or X
follows/has a normal distribution) with mean µ and variance σ 2.”
A special case of a normal rv is when µ = 0 and σ 2 = 1. This is called a
standard normal distribution. Since it commonly arises, a standard
normal rv is usually denoted Z and its pdf is given a special symbol, ϕ.
1 − z2 2
ϕ ( z) = e
2π
Obviously, E[Z] = 0 and Var[Z] = 1.
Finding Probabilities for Z
Let Z ~ Norm(0, 12), since the standard normal arises frequently, a
special notation is used to represent its cdf.
Pr ( Z ≤ z ) =
Φ( z)
With the cdf, one can obtain the probability a standard normal takes on
any set of values in an interval. For instance,
By the complement rule it follows
Pr ( Z > z ) =−
1 Pr ( Z ≤ z ) =−
1 Φ( z)
Additionally, for any real numbers a < b, it follows
Pr ( a < Z ≤ b ) = Pr ( Z ≤ b ) − Pr ( Z ≤ a ) = Φ ( b ) − Φ ( a )
Standard Normal
Table
Finding Probabilities for Z - Examples
To get values of the cdf function, one must either resort to tables, or
statistical software packages. They cannot be obtained by hand .
Pr(Z ≤ 0) = Φ(0) = 0.5 Pr(Z ≤ 1.64) = Φ(1.64) = 0.9495
Pr(Z > 1.96) = 1 – Pr(Z ≤ 1.96)
= 1 – Φ(1.96) = 1 – 0.975 = 0.025
Finding Probabilities for Z - Examples
Pr(1.28 < Z ≤ 1.5) = Pr(Z ≤ 1.5) – Pr(Z ≤ 1.28) = Φ(1.5) – Φ(1.28)
= 0.9332 – 0.8997 = 0.0335
Pr(|Z| > 1.35) = Pr((Z < -1.35) ∪ (Z > 1.35))
= Pr(Z < -1.35) + Pr(Z > 1.35)
= Pr(Z < -1.35) + (1 – Pr(Z < 1.35))
= 0.0885 + (1 – 0.9115)) = 0.177
Facts about Normals
If X ~ Norm(µ, σ 2) then it follows the rv, Z, defined as
𝑋𝑋−𝜇𝜇
𝑍𝑍 =
𝜎𝜎
follows a standard normal distribution. That is Z ~ Norm(0, 12).
If Z ~ Norm(0, 12) then it follows the rv, X, defined as
X = σZ + µ
has a normal distribution with mean µ and variance σ 2, i.e. X ~ Norm(µ, σ 2).
The above facts allow one to convert any normal into a standard normal and
vice-versa. This property of the normal is critical for the purposes of doing
statistical inference.
Facts about Probabilities
Let X be any rv, it follows
That is the set of outcomes that satisfy the inequality in the middle are the
same outcomes that satisfy the inequality on the right.
Using the above result, let X ~ Norm(µ, σ 2), utilizing the relationship
between the standard normal and general normal one has,
X −µ x−µ x−µ x−µ
Pr ( X ≤ x ) =
Pr ≤ =Pr Z ≤ =Φ
σ σ σ σ
The cdf for any normal can be written using the cdf of the standard normal!
Finding Probabilities for any Normal
Similarly, one can obtain the probability any normal takes on any set of
values. For instance,
By the complement rule it follows
x−µ
Pr ( X > x ) =−
1 Pr ( X ≤ x ) =−
1 Φ
σ
Additionally, for any real numbers a < b, it follows
b−µ a−µ
Pr ( a < X ≤ b ) = Pr ( X ≤ b ) − Pr ( X ≤ a ) = Φ −Φ
σ σ
Example – Getting
Normal
Probabilities in
Excel
Finding Probabilities for Normals - I
Suppose X ~ Norm(5, 22), find Pr(X ≤ 10).
X − 5 10 − 5 5
Pr ( X ≤ 10 ) =
Pr ≤ = Φ ( 2.5 ) =
Pr Z ≤ = 0.9938
2 2 2
Suppose X ~ Norm(15, 32), find Pr(X > 14).
X − 15 14 − 15
Pr ( X > 14 ) =
1 − Pr ( X ≤ 14 ) = 1 − Pr ≤
3 3
1
= 1 − Pr Z ≤ −
3
1
= 1− Φ −
3
= 1 − 0.3707 = 0.6293
Finding Probabilities for Normals - II
Suppose X ~ Norm(-10, 102), find Pr (-10 ≤ X ≤ 10).
Pr ( −10 ≤ X ≤=
10 ) Pr ( X ≤ 10 ) − Pr ( X ≤ −10 )
10 − ( −10 ) −10 − ( −10 )
= Φ −Φ
10 10
20 0
= Φ −Φ
10 10
= Φ ( 2) − Φ ( 0)
.5 0.4772
= 0.9772 −=
Z-values – I
Instead of obtaining probabilities for a standard normal, one may be
interested in finding the number which yields a given probability.
Let Z be a standard normal rv. The value, zα, is the number such that
Pr(Z ≥ zα) = α (i.e. α is the area under the normal pdf and to the right
of zα).
1
2π
The area under the pdf is α.
0 zα
The number zα is referred to the upper α cutoff value of the standard
normal.
Z-values – II
Below are some commonly encountered upper z-cutoff values.
Pr(Z > 2.575) = 0.005, so z0.005 = 2.575
Pr(Z > 2.33) = 0.01, so z0.01 = 2.33
Pr(Z > 1.645) = 0.05, so z0.05 = 1.645
Pr(Z > 1.28) = 0.1, so z0.1 = 1.28
Pr(Z > 0.84) = 0.2, so z0.2 = 0.84
Pr(Z > 0) = 0.5, so z0.5 = 0
The usefulness of this concept will become evident once we get to
confidence intervals and hypothesis testing.
Example – Getting
Z cutoff Values in
Excel