Materials Engineering 14

Analysis of Variance

Recall:
Variable Screening

What are the two types of experiments?

Optimization Experiments

Variable Screening

Optimization

Complete Randomized Design (CRD)

vs
Randomized Block Design (RBD)

Some Terms
Response Variable
dependent variable

Factors
qualitative or quantitative

Factor Levels
values of the factor utilized

Treatments
factor-level combinations

Experimental Unit
sample for which a data can be obtained

The Completely Randomized Design (CRD)

a design for which random samples of experimental units are independently selected for each treatment objective: compare the treatment means
H0: 1=2=...=a H1: at least two of the a treatment means differ

Example
A manufacturer of paper used for making grocery bags is interested in improving the tensile strength of the product. Product engineering thinks that the tensile strength is a function of the hardwood concentration in the pulp and that the range of hardwood concentrations of practical interest is between 5% and 20%.

Example
A team of engineers responsible for the study decides to investigate 4 levels of hardwood concentration: 5%, 10%, 15% and 20%. They decided to make up 6 test specimens at each concentration level, using a pilot plant. All 24 specimens are tested on a laboratory tensile tester, in random order.

Typical Data for a Single-Factor Experiment

Treatment

Observations

Totals

Averages

1 2
a

y11 y21
ya1

y12 y22
ya2

... ...
...

... ...
...

... ...
...

y1n y2n
yan

y1. y2.
ya. y..

y1. y2.
y2. y..

Table of Results
Hardwood Conc. (%) 5 10 15 20

Observations 1
7 12 14 19

2
8 17 18 25

3
15 13 19 22

4
11 18 17 23

5
9 19 16 18

6
10 15 18 20

Totals Averages

Step 1: Formulating Hypotheses

H0: 1=2=3=4
Different hardwood concentrations do not affect the mean tensile strength of the paper

H1: at least two means are not equal

Different hardwood concentrations affect the mean tensile strength of the paper

Step 2: Deciding on the confidence interval

The usual values of are 0.25, 0.10, 0.05, and 0.01
For the example, we will be using an of 0.01 (99% confidence)

Use F-tables
f(0.01,v1=n-1,v2=N-a)

Computing for Sum of Squares

Sum of Squares Total (SST)

measures the total variability in the data

y SSTotal y N i 1 j 1
2 ij

2 ..

Computing for Total Sum of Squares

Hardwood Conc. (%) 5 10 15 20 Observations 1 7 12 14 19 2 8 17 18 25 3 15 13 19 22 4 11 18 17 23 5 9 19 16 18 6 10 15 18 20 Totals 60 94 102 127 383 Averages 10.00 15.67 17.00 21.17 15.96

(383) SST (7) (8) ... (20) 24 512.96

2 2 2

Sum of Squares for Treatments (SSTreatments)

measures the variations between treatment means

y y SSTreatments N i 1 n

2 i

2 ..

Computing for Treatment Sum of Squares

Hardwood Conc. (%) 5 10 15 20 Observations 1 7 12 14 19 2 8 17 18 25 3 15 13 19 22 4 11 18 17 23 5 9 19 16 18 6 10 15 18 20

Totals Averages
60 94 102 127 383 10.00 15.67 17.00 21.17 15.96

(60) (94) (102) (127) (383) SSTreatments 6 24 382.79

2 2 2 2

Sum of Squares for Error (SSE)

measures the sampling variability within the treatments accounts for the sampling error

SS E SST SSTreatments

Computing for Error Sum of Squares

Hardwood Conc. (%) 5 10 15 20 Observations 1 7 12 14 19 2 8 17 18 25 3 15 13 19 22 4 11 18 17 23 5 9 19 16 18 6 10 15 18 20

Totals Averages
60 94 102 127 383 10.00 15.67 17.00 21.17 15.96

SS E SST (512.96) SSTreatments(382.79) 130.17

Step 4: Summarizing of Results

Perform Analysis of Variance (ANOVA)
Source of Variation Treatments Sum of Squares SSTreatments Degrees of Freedom a-1 Mean Squares MST Computed f MST/MSE

Error Total

SSE SST

N-a (a-1)+(N-a)

MSE

SSTreatments MSTreatments a 1 SS Error MS Error N a

Step 4: Summarizing of Results

Source of Variation
Hardwood concentration

Sum of Squares
382.79

df
3

Mean Squares
127.60

Computed f
19.6

Error
Total

130.17
512.96

20
23

6.51

SSTreatments MSTreatments a 1 SS Error MS Error N a

Step 5: Decision
H0 is true if the ratio f=MST/MSE is a value of the random variable F having the F-distribution with n-1 and (N-a) degrees of freedom
The null hypothesis is rejected at the value of significance when

f > f[a-1, N-a)]

Example 2
In many integrated circuit manufacturing steps, wafers are completely coated with a layer of material such as silicon dioxide or a metal. The unwanted material is then selectively removed by etching, whose rate is dependent on the radiofrequency (RF) power setting. An engineer is interested in investigating the relationship between the RF power setting and the etch rate. Four test levels of RF power: 160, 180, 200, and 220 W (with 5 wafers at each level of RF) were prepared.

Example 2
RF Power (W) Observed Etch Rate (/min) Totals 1 2 3 4 5 Averages

160
180 200 220

575
565 600 725

542
593 651 700

530
590 610 715

539
579 637 685

570
610 629 710

2,756
2,937 3,127 3,535 12,355

551.2
587.4 625.4 707.0 617.75

Use =0.05

Example 2
ANOVA for the experiment
Source of Variation RF Power Error Total Sum of Squares df Mean Square F0 Ftable

Example 2
ANOVA for the experiment
Source of Variation RF Power Error Total Sum of Squares 66,870.55 5,339.20 72,209.75 df 3 16 19 Mean Square 22,290.18 333.70 F0 66.80 Ftable 3.24

The Analysis of Variance

Description
more number of levels of a factor is tested examples:
the effect of curing temperature on the compressive strength of concrete blocks the effect of dosage size of a certain drug on its curative properties

Statistically-based experiments will lead to

better and improved processes
development of new processes

Control Variables are used to describe the said processes. Examples of which are time, temperature, feed rate, amount of material, concentration, etc.

The Pros

improved process yield reduced variability and closer conformance to specifications reduced design and development time reduced cost

In Engineering Design
evaluation and comparison of configurations and materials selection of design parameters that will make the product work well under varying field conditions
determination of key product design parameters that affect product performance

Steps involved in an experiment

Conjecture

Experiment
Analysis Conclusion

Exercise
An engineer is interested in how the mean absorption of moisture in concrete varies among 5 different concrete aggregates. The samples are exposed to moisture for 48 hours. It is decided that 6 samples are to be tested for each aggregate, requiring a total of 30 samples to be tested.

Summary of Results
Absorption of Moisture in Concrete Aggregates
Aggregate
Run 1

1
551

2
595

3
639

4
417

5
563

Run 2
Run 3 Run 4 Run 5 Run 6

457
450 731 499 632

580
508 583 633 517

615
511 573 648 677

449
517 438 415 555

631
522 613 656 679

RBD

Randomized Block Design

blocks, or matched sets of experimental units, are formed each block consists of p experimental units; each block (b) should contain units which are as similar as possible
an experimental unit from each block is assigned to each treatment

General ANOVA Table for RBD

Source of Variation Treatment SS SST df a-1 MS MST f MST/MSE

Block Error
Total

SSB SSE
SS(Total)

b-1 N-a-b+1
N-1

MSB MSE

Formulas
SSTreatments b( xTi x ) 2
i 1 a

SS Blocks p ( xBi x )
i 1

SSTotal ( xi x )
i 1

SS Error SSTotal SSTreatments SS Blocks

Example:
Suppose the USGA wants to compare the mean distances associated with four brands of golf balls when struck by a driver, and will employ human golfers. Assuming that 10 balls of each brand are to be utilized in the experiment
Suppose that an RBD is used, utilizing a random sample of 10 golfers with each golfer using a driver to hit four balls, one of each brand, in random sequence
Use one way ANOVA to see if there is a difference among the mean distance for the brands of golf balls

Table of Results
Golfer (block) Brand A Brand B Brand C Brand D Totals 832.9 991.2 895.1 947.7 821.8 1012.8 859.3 974.7 909.4 1018

1 2
3 4 5 6 7 8 9 10 Totals

202.4 242.0
220.4 230.0 191.6 247.7 214.8 245.4 224.0 252.2

203.2 248.7
227.3 243.1 211.4 253.0 214.8 243.6 231.5 255.2

223.7 259.8
240.0 247.7 218.7 268.1 233.9 257.8 238.2 265.4

203.6 240.7
207.4 226.9 200.1 244.0 195.8 227.9 215.7 245.2

2270.5

2331.8

2453.3

2207.3

Computing for the Sum of Squares Total

Golfer (block) Brand A Brand B Brand C Brand D Totals 832.9 991.2 895.1 947.7 821.8 1012.8

1
2 3 4 5 6 7 8 9 10 Totals

202.4
242.0 220.4 230.0 191.6 247.7 214.8 245.4 224.0 252.2 2270.5

203.2
248.7 227.3 243.1 211.4 253.0 214.8 243.6 231.5 255.2 2331.8

223.7
259.8 240.0 247.7 218.7 268.1 233.9 257.8 238.2 265.4 2453.3

203.6
240.7 207.4 226.9 200.1 244.0 195.8 227.9 215.7 245.2 2207.3

859.3
974.7 909.4 1018

2 ( 9262 ) SSTotal (202.4) 2 (203.2) 2 ... (245.2) 2 40 15919.2

Computing for the Sum of Squares Treatments

Golfer (block) 1 2 3 4 5 6 7 8 9 Brand A 202.4 242.0 220.4 230.0 191.6 247.7 214.8 245.4 224.0 Brand B 203.2 248.7 227.3 243.1 211.4 253.0 214.8 243.6 231.5 Brand C 223.7 259.8 240.0 247.7 218.7 268.1 233.9 257.8 238.2 Brand D 203.6 240.7 207.4 226.9 200.1 244.0 195.8 227.9 215.7 Totals 832.9 991.2 895.1 947.7 821.8 1012.8 859.3 974.7 909.4 1018

10
Totals

252.2
2270.5

255.2
2331.8

265.4
2453.3

245.2
2207.3

(2270.5) 2 ...(2207.3) 2 (9262) 2 SSTreatments 10 40 3298.7

Computing for the Sum of Squares Blocks

Golfer (block) 1 2 3 4 5 6 7 8 9 10 Totals Brand A 202.4 242.0 220.4 230.0 191.6 247.7 214.8 245.4 224.0 252.2 2270.5 Brand B 203.2 248.7 227.3 243.1 211.4 253.0 214.8 243.6 231.5 255.2 2331.8 Brand C 223.7 259.8 240.0 247.7 218.7 268.1 233.9 257.8 238.2 265.4 2453.3 Brand D 203.6 240.7 207.4 226.9 200.1 244.0 195.8 227.9 215.7 245.2 2207.3 Totals 832.9 991.2 895.1

947.7
821.8 1012.8 859.3 974.7

909.4
1018

(832.9) 2 (991.2) 2 ... (1018) 2 (9262) 2 SS Blocks 4 40 12073.9

Computing for the Sum of Squares Error

Golfer (block) 1 2 3 4 5 6 7 8 9 10 Totals Brand A 202.4 242.0 220.4 230.0 191.6 247.7 214.8 245.4 224.0 252.2 2270.5 Brand B 203.2 248.7 227.3 243.1 211.4 253.0 214.8 243.6 231.5 255.2 2331.8 Brand C 223.7 259.8 240.0 247.7 218.7 268.1 233.9 257.8 238.2 265.4 2453.3 Brand D 203.6 240.7 207.4 226.9 200.1 244.0 195.8 227.9 215.7 245.2 2207.3 Totals 832.9 991.2

895.1
947.7 821.8 1012.8 859.3

974.7
909.4 1018

SS Error SST (15919.2) SSTreatments(3298.7) SS Blocks (12073.9) 546.6

ANOVA Table
Source of Variation df SS MS f

Treatment Block
Error Total

3 9
27 39

3298.7 12073.9
546.6 15919.2

1099.6 1341.5
20.2

54.31 66.26

3298.7 MSTreatments 1099.6 3 12073.9 MS Blocks 1341.5 9 546.6 MS Error 20.2 27

1099.6 FTreatments 54.31 20.2 1345.1 FBlocks 66.26 20.2

Announcement

Midterms is on February 7 (Thurs), 6-9pm at the P&G Room Melchor Hall