Scribd
Upload
49 views

Sampling Distributions

The Central Limit Theorem (CLT) states that the sum of independent random variables will be approximately normally distributed for large sample sizes, regardless of the underlying distributions. Specifically, if X1, X2, ..., Xn are independent and identically distributed random variables with mean μ and standard deviation σ, then the distribution of their mean X= (X1 + X2 + ... + Xn)/n will be approximately normal with mean μ and standard deviation σ/√n for n≥30. The CLT allows probabilities of sample means and sums to be estimated using the normal distribution for large samples, enabling statistical inference about populations.

Uploaded by

Yuleta Puspa Melati
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
49 views

Sampling Distributions

The Central Limit Theorem (CLT) states that the sum of independent random variables will be approximately normally distributed for large sample sizes, regardless of the underlying distributions. Specifically, if X1, X2, ..., Xn are independent and identically distributed random variables with mean μ and standard deviation σ, then the distribution of their mean X= (X1 + X2 + ... + Xn)/n will be approximately normal with mean μ and standard deviation σ/√n for n≥30. The CLT allows probabilities of sample means and sums to be estimated using the normal distribution for large samples, enabling statistical inference about populations.

Uploaded by

Yuleta Puspa Melati
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 18

The Central Limit Theorem

Summer 2003
The Central Limit Theorem (CLT)
If random variable Sn is defined as the sum of n
independent and identically distributed (i.i.d.)
random variables, X1, X2, …, Xn; with mean, µ,
and std. deviation, σ.
Then, for large enough n (typically n≥30), Sn is
approximately Normally distributed with
parameters: µSn = nµ and σSn = n σ

This result holds regardless of the shape of the X


distribution (i.e. the Xs don’t have to be normally
distributed!)
15.063 Summer 2003 2
Examples
Exponential n=2 n=5 n = 30
Population

Uniform n=2 n=5 n = 30


Population

15.063 Summer 2003 3


U Shaped n=2 n=5 n = 30
Population

Normal n=2 n=5 n = 30


Population

15.063 Summer 2003 4


An Example
Each person take a coin and flip it twice (a pair)
The distribution of two heads vs. other is binomial
Now flip your coin 3 new pairs, report #(two heads)
New variable H3 = sum of n=3 binomial (p=.25)
Now flip each coin 10 new pairs, report #(two heads)

.75

.42
.28

0 1 0 1 2 3 0 1 2 3 4 5 6 7 8 9 10

15.063 Summer 2003 5


For any binomial r. v. X (n,p)
X can be seen as the sum of n i.i.d. (independent,
identically distributed) 0-1 random variables Y, each with
probability of success p (i.e., P(Y=1)=p).
X = Y1 + Y2 + … Yn
In general we can approximate r.v. X binomial (n,p) using
r.v. Y Normal: µ = np ; σ = np(1-p)

0.35
0.30
p=.8, n=10 0.35
0.30
p=.8, n=25
0.25 0.25
0.20 0.20
0.15 0.15
0.10 0.10
0.05 0.05
0.00 0.00
0

5
10

15

20

25

10

15

20

25
15.063 Summer 2003 6
Using the Normal Approximation to
The Binomial…
If r.v. X is Binomial (n, p) with parameters:
E(X) = np; VAR(X) = np(1-p);
We can use Normal r.v. Y with mean np and variance
np(1-p) to calculate probabilities for r.v. X (i.e., the
binomial)
The approximation is good if n is large enough for the
given p, i.e, must pass the following test:

Must have : np ≥ 5 and n(1 - p) ≥ 5


15.063 Summer 2003 7
Small Numbers Adjustment
To calculate binomial probabilities using the normal
approximation we need to consider the “0.5 adjustment”:
1. Write the binomial probability statement using “≥” and “≤”:
e.g. P(3<X<9)= P(4 ≤ X ≤ 8)
2. Draw a picture of the normal probability Y you want to
calculate and enlarge the area making a 0.5
adjustment(s) to the edge(s). This is because each
discrete probability is represented by a range in the
normal probability, e.g., P(X=4) is ~ P(3.5<Y<4.5)
3. Calculate the size of the area (Normal probability)
(The book ignores this adjustment. The example on page
139 should have ~ P(Y≥9.5))
This adjustment is less important as n becomes larger.

15.063 Summer 2003 8


Example: An electrical component is guaranteed by its
suppliers to have 2% defective components. To check a
shipment, you test a random sample of 500 components. If
the supplier’s claim is true, what is the probability of finding 15
or more defective components?
X = number of defective components found during the test.
X is Binomial(500, 0.02).
We want P(X≥15)= P(X=15) + P(X=16) …. + P(X=500)
Can we use r.v. Y Normal with:
mean=500(0.02) = 10 and sd = sqrt{500*.02(1-.02) = 3.13 ?
Yes! np = 500 * 0.02 = 10 and n (1-p) = 500 * 0.98=490 (> 5)
Using the “.5 adjustment” we see that P(X≥15) ~ P(Y≥14.5)
Easiest way is to calculate P(Y ≥ 14.5) = 1-P(Y<14.5)
= 1-F(z) =(14.5-10)/3.13=1.44)= 1-F(1.44) = 1- 0.9251
= 0.0749 15.063 Summer 2003 9
The Central Limit Theorem (for the mean)
If random variable X is defined as the average of n
independent and identically distributed random
variables, X1, X2, …, Xn; with mean, µ, and Sd, σ.
Then, for large enough n (typically n≥30), Xn is
approximately Normally distributed with
parameters: µx = µ and σx = σ/ n

Again, this result holds regardless of the shape of


the X distribution (i.e. the Xs don’t have to be
normally distributed!).

15.063 Summer 2003 10


The CLT for the mean and statistical
sampling: (chapter 4)
Point estimate: Interval Estimate:
σ

X=
∑ X X ±Z
n
n or
σ σ
X −Z ≤µ ≤ X +Z
n n

Idea: If we take a large enough random sample (i.e. n>=30) for r.v. X
(i.e., the population of interest), then we can estimate the mean, µ , for
r.v. X even if we do not know the distribution of X. Note: use the sample
SD, s, if the population sd, σ, is not known: ∑ (X − X )
2

=
2
S n −1
More on s vs. σ later
S=
2
S
The value of z is determined by the confidence level assigned to the
interval (see next slide)

15.063 Summer 2003 11


Values of Z for selected confidence levels:
Confidence
Z Value
Level

90% (α=0.1) 1.645


.025 .025
95%
.4750 .4750 95% (α=0.05) 1.96
µ X 98% (α=0.02) 2.33
Z
-1.96 0 1.96 99% (α=0.01) 2.575

We would for example say that we are 95% confident the true
mean for x falls in the interval:

σ σ
X −1.96 ≤ µ ≤ X +1.96
n n

15.063 Summer 2003 12


Confidence Limits Are a Way of
Knowing What We Know
Estimates and forecasts are difficult to
evaluate for quality or degree of confidence

What will the Dow Jones be


six months from now?

15.063 Summer 2003 13


Estimation and Confidence Limits
How many employees (in total) did IBM have
worldwide on Dec. 31, 2002?
After making your best estimate, give a low
estimate and high estimate so you are 95%
sure that the true answer falls within these
limits

Low___________ High____________

315,889
15.063 Summer 2003 14
Overconfidence
Respondant Topic Target Result

Harvard Trivia facts 2% 46%


MBAs

Kellogg Starting
MBAs salary 49% 85%

Chemical Industry &


employees co. facts 10% 50%

Computer Business: 5% 80%


managers co. facts: 5% 58%

15.063 Summer 2003 15


More Overconfidence
“A severe depression like that of 1920-1921 is
outside the range of probability”
Harvard Econ. Soc’ty W’kly Letter, Nov. 16, 1929
“With over 50 foreign cars already on sale here, the
Japanese auto industry isn’t likely to carve out a big
slice of the U.S. market for itself”
Business Week, August 2, 1968
“There is no reason anyone would want a computer
in their home”
Ken Olson, DEC founder, 1977

15.063 Summer 2003 16


Overcoming Overconfidence

Commercial Loan Officer, Midwest Bank: “Are


we overconfident about the competition?”
First, convince the boss.
Tactic: overconfidence test
Second, avoid the mistakes
Tactic: competitor alert file
Result: within 3 weeks, saved $160K account

15.063 Summer 2003 17


Summary and Look Ahead
The Central Limit Theorem allows us to use the Normal
distribution, which we know a lot about, to approximate
almost anything, as long as some requirements are met
(e.g., n>=30)
Confidence limits are a way of estimating our degree of
knowledge
People typically think they know more than they do (we
don’t like uncertainty)
Next class we use the same tools to look at statistical
sampling
Homework #2 is due!!

15.063 Summer 2003 18

You might also like