PSTAT W 120A: Normal Distribution August 30, 2012

Normal Density
2
• Another special density that comes up a lot is e−t
• It looks like a Bell curve
• Gauss showed that ∞
Z
2 √
e−t /2
dt = 2π
−∞

so it is often called the Gaussian distribution.

• Z is a random variable such that
1
φ(z) = √ exp[−z 2 /2]
2π
Then Z is a standard normal random variable.

• The integral can’t be worked out exactly but it is tabled in the back of the text.
1. Symmetric around 0, P{Z < t} = P{Z > −t}.
2. About 95% of the probability is between -2 and 2 (1.96 to be more accurate).

Expectation and Variance

• We can do an explicit calculation of the expectation of the normal random variable.
d −x2 /2 2
e = −xe−x /2
dx

• Therefore Z ∞ ∞
2 2
xe−x /2
dx = −e−x /2 
 =1
0 0

• Likewise, Z 0 0
2 2
xe−x /2
dx = −e−x /2 
 = −1
−∞ −∞

• Thus,
Z ∞
1 2
E(Z) = z √ e−z /2 dz
−∞ 2π
Z 0 Z ∞ 
1 −z 2 /2 −z 2 /2
=√ ze dz + ze dz
2π −∞ 0
1
= √ [1 − 1] = 0
2π

• We could have deduced this from the symmetry of the distribution. Φ(x) = 1 − Φ(−x).

1
 • To calculate the variance requires integration by parts.
Z ∞
1 2
2
Var(Z) = E(Z ) = z 2 √ e−z /2 dz
−∞ 2π
u=z du = dz
2 2
dv = ze−z v = −e−z /2
/2

1 2
∞ Z
1 2
=⇒ E(Z 2 ) = −z √ e−z /2  + √ e−z /2 dz

2π −∞ 2π
=1

• So the variance is 1.

Standard Normal Calculations

• P(Z > −0.5) = P(Z < 0.5) = 0.6915 by the symmetry of the distribution.
• P(Z < −3) = P(Z > 3) by symmetry of the distribution. P(Z > 3) = 1 − 0.9987 = 0.0013.
• P(Z < 0.03) is right from the table as 0.5120.
• P(Z > 6.3) < 0.0001 because the endpoint is greater than 3.5. A computer calculation approximates
it by 1.5 × 10−10 .

General normal random variables

New Random variables can be constructed by linear functions of Z.

X = σZ + µ

(µ = mean, and σ = standard deviation).

Then
E(X) = σ E(Z) + µ = µ
Var(X) = σ 2 Var(Z) = 1
To standardize X, we just invert this formula
X −µ
Z=
σ
Any set of X’s corresponds to a set of Z’s
 
6−µ
P{X > 6} = P{σZ + µ > 6} = P Z >
σ
OR  
1−µ 3−µ
P{1 < X < 3} = P <Z<
σ σ

As Variance Gets Small

In the limit as σ 2 ↓ 0, there is no variance nor variability. The random variable X becomes in fact a fixed
value equal to the mean µ, because X = µ + 0 · Z = µ (recall the transformation from Z to X).
In the limit there is a unit probability mass at x = µ. The value of the density function at x = µ becomes
higher and higher in fact goes off to an infinite value as σ → 0 so that the density does not exist in the limit.

2
Example
• An IQ test is such that its mean is 100 and it standard deviation is 30. What is the probability of
scoring greater than 200?
•  
X −µ 200 − 100
P{X > 200} = P > = P{Z > 10/3} = 0.00043
σ 30

• The probability of being between 75 and 90 is

   
75 − 100 X −µ 90 − 100 5 1
P{75 < X < 90} = P < < =P − <Z<−
30 σ 30 6 3

• This is equal to
P{Z < −1/3} − P{Z < −5/6} = 0.3694 − 0.2023 = 0.1671.

Normal probability calculations

1. Write down the event as an inequality.

If the random variable is binomial, use the appropriate continuity correction

X−µ
2. Standardize: σ , on both sides of the inequality.

3. Look up the standard normal probabilities in the table (using symmetry if necessary.)

Limit Theorems
There are two main important limit theorems in statistics. If we have n independent random variables,
n
!
1X
E Xi = E(X)
n i=1

n
!
1X Var(X)
Var Xi =
n i=1 n

1. Law of large numbers, the mean converges to the expected value.

n
1X
Xi → E(X)
n i=1

2. Central Limit Theorem, the distribution of the mean is normal

n
1 X Xi − E(X)
√ p →Z
n i=1 Var(X)