Page 1 Lecture 5-6 MEEN 260 Dr. R.

Tafreshi

Lecture 5&6: Statistical Data Analysis:

Central Limit Theorem
Introduction to t-Distribution (t-Statistic)

Lecture Outline:
Probability Distributions
Confidence Intervals
o For the expected value of measurements (as we saw in the last lecture)
o For the Mean
Sampling Approximations
o Distribution of means
o Central Limit Theorem
Examples
Introduction to t-Distribution (t-Statistic)

Question of the day:

What is the objective of the use of statistical analysis in measurement?

Announcements / Reminders / Discussions:

HW#1&2&3 (&4 Not covered yet): was sent to you

Reading Assignments:
Chapter 3 of textbook by Beckwith
(PP46 54)
Study the below Useful Online Material:

Some Useful Online Material:

1. Video: Lecture on Sampling distribution of the sample mean (10 min):
http://www.khanacademy.org/math/probability/statistics-
inferential/sampling_distribution/v/sampling-distribution-of-the-sample-mean

2. Video: Lecture on Central limit theorem (11 min):

http://www.khanacademy.org/math/probability/statistics-
inferential/sampling_distribution/v/central-limit-theorem

3. Extra:
Online Statistics Education: An Interactive Multimedia Course of Study:
http://onlinestatbook.com
Page 2 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Two Different Problems:

1. Confidence Interval (for the expected value of measurements):

Assume knowing and
Note: we use and (of population) to predict x

Review: As discussed in the last Lecture:

Consider the Gaussian distribution shown below. If we magically know and of the
population we could calculate a % confidence interval on the expected value of our
measurements:

P( x ) p( x)dx 0.68

1.96

P( 1.96 x 1.96 ) p( x)dx 0.95

1.96
3

P( 3 x 3 ) p( x)dx 0.99
3

The probability of a
measurement being within:
1 = 68.3%

2 = 95.5%

3 = 99.7%
6 = 99.9999998%

1.96 = 95.0%

In general,
zc/2

P( zc /2 x zc /2 ) p( x)dx
zc/2

c: Confidence interval

Note: The textbook (and many statistics textbooks) use the notation zc /2
Example: zc /2 = z0.475 c: Confidence interval
This refers to the value of z when:
zc/2

p( z )dz c / 2 0.95 / 2 0.475

0
Page 3 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Real-life situation: What if we want to use x to predict ?

2. Confidence Interval (for the Mean)

zc /2 x zc /2

Subtract from each side: zc /2 x zc /2

Subtract x from each side: x zc /2 x zc /2
Change signs: x zc /2 x zc /2 or
x zc /2 x zc /2

Replace with the sample mean: x zc /2 x x zc /2 x

If we know x and x we can estimate the uncertainty (confidence interval) for .

Any practical issue?!
Page 4 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Distribution of Averages (Central Limit Theorem)

A fundamental result in statistics: the distribution of sample means always follows a

normal distribution (Central Limit Theorem)

Example: Consider multiple sets of measurements:

o Could be based on any probability distribution (normal, exponential, etc.)
o Compute the averages of points 1 through 10, 11 through 20, 21 through 30,

Calculate the sample mean for:

1 10
o Data points 1-10 x1 xi
10 i 1
1 20
o Data points 11-20 x2 xi
10 i 11
1 30
o Data points 21-30 x3 xi
10 i 21
o

If we plot the distribution of x1 , x2 , x3 , the will always be normally distributed

and they will have a std. dev. of:
x
x
n

Use of the Sample Deviation

Recall, if we know x and x we can calculate a confidence interval for
o But x depends on x !
o If we have a large number of samples (measurements) then we can assume
2 1 n
Sx x S x ( xi x )2
n 1i1

What is a large sample?

o Roughly > 30 samples
Sx Sx
Putting it all together x zc /2 x zc /2
n n
Sx Sx
For example at 95% confidence x 1.96 x 1.96
n n

: the standard deviation of the population

x

x : the standard deviation of the sample mean values

S x : the standard deviation of sample

Page 5 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Understanding through Example: Distribution of Averages (Central Limit Theorem)

Example: Consider the 200 measurements of the same quantity below (Hints for HW #2).

a. Compute the averages of the first 10, the first 20, and the first 30, and finally all
200 data points.
Page 6 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Example…
Facts
Theoretically, it takes infinite number of measurements to establish the true
value of a quantity.
In practice, sufficiently large (say a 100 or so measurements) do converge
towards the true value.
Question
How do we get a handle on the true value of the measured quantity using a
finite number of measurements?

2. Confidence Interval (for the Mean)

Consider multiple sets of measurements:
b. Compute the averages and standard deviations of the 1-10, 11-20, 21-30, ..., 191-
200.

xi

xi

We mark the average values on a scale, say from 29.8 to 31.1:

Facts
Multiple sets of measurements of the same quantity produce average values, x1 ,
x2 , x3 , that themselves form a Gaussian distribution as shown above
(Central Limit Theorem).
The nominal standard deviation of the cumulative distribution (distribution of
S
x1 , x2 , ) is x where n is the number of measurements in a given set; S x is
n
the standard deviation for one such set.
Page 7 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Example:
In a capacitor manufacturing plant 200 capacitors are randomly sampled from a
particular batch, with sample mean of 0.1 µF and sample deviation of 0.007 µF.
Management requires that the capacitor tolerance be within 0.1% (0.001 µF).
Does this batch meet the specification with a 95% confidence interval?
How about a 99% confidence interval?

Given , , n = 200

From table

Confidence Interval

Is it ok?
Page 8 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

T-Distribution (t-Statistic)
S S
x zc /2 x x zc /2 x if n is sufficiently high: 30 or more
n n

For smaller n we need to use the formula:

S S
x t x x t x where t comes from T-Table
n n
Page 9 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

Example: Consider 200 data points from a simulated pressure measurement experiment
(The same as HW #2).

a. Using the first 100 data points, determine the average pressure and standard
deviation and estimate the true value of pressure with a confidence of 95%.
b. Repeat Part a using the first 50 data points. What is your conclusion?
c. Repeat Part a using the first 30 data points. What is your conclusion?
d. Repeat Part a using the first 20 data points. What is your conclusion?
e. Repeat Part a using the first 10 data points. What is your conclusion?
f. Repeat Part a using the first 5 data points. What is your conclusion?

Solution:
a. Using the first 100 data points, determine the average pressure and standard
deviation and estimate the true value of pressure with a confidence of 95%.
We first find the average and standard deviations to be 29.96 and 0.2704
respectively. Since the number of data points is sufficiently large we simply use the
formula
Sx Sx

b. Repeat Part a using the first 50 data points. What is your conclusion?
For the first 50 points, we have 29.98 and 0.2575 as the mean and standard
deviation:
Sx Sx

c. Repeat Part a using the first 30 data points. What is your conclusion?
Doing the same with 30 points we have 29.90 and 0.2494 as mean and standard
deviation. Note that D.O.F., defined as N 1 = 30 1 = 29. Therefore we have

Sx Sx
Page 10 Lecture 5-6 MEEN 260 Dr. R. Tafreshi

d. Repeat Part a using the first 20 data points. What is your conclusion?
With 20 points we have mean and standard deviation given by 29.94 and 0.2739
respectively. Using the t-table with D.O.F. of 19, we have

Sx Sx

e. Repeat Part a using the first 10 data points. What is your conclusion?
With 10 points we have mean and standard deviation given by 29.92 and 0.2938
respectively. Using the t-table with a D.O.F. of 9, we have

f. Repeat Part a using the first 5 data points. What is your conclusion?
With 5 points we have mean and standard deviation are given by 29.89 and 0.3492
respectively. Using the t-table we have

* Some material based on original slides by Drs. Reza Langari and Bryan Rasmussen, TAMU MEEN.