Sei sulla pagina 1di 14

Experimental Design and

Analysis of Variance

Experimental Design and Analysis of


Variance

10.1 Basic Concepts of Experimental Design


10.2 One-Way Analysis of Variance
10.3 The Randomized Block Design
10.4 Two-Way Analysis of Variance

1
10.1 Basic Concepts of Experimental
Design
Objective: To compare and estimate the effect of
different treatments on the response variable.
Example 10.1 The Gasoline Mileage Case
Does gasoline mileage vary with gasoline type?
Type A Type B Type C
xA1=34.0 xB1=35.3 xC1=33.3
xA2=35.0 xB2=36.5 xC2=34.0
xA3=34.3 xB3=36.4 xC3=34.7
xA4=35.5 xB4=37.0 xC4=33.0
xA5=35.8 xB5=37.6 xC5=34.9

Response Variable: Gasoline mileage (in mpg)


Treatments: Gasoline types – A, B, C

10.2 One-Way Analysis of Variance


Are there differences in the mean response µ1, µ2,…,µ
µp
associated with the p treatments?

H0: µ1 = µ2 = … = µp
Ha: At least two of the
µ1, µ2,…,µp differ

Or, is the between-


treatment variability
large compared to the
within-treatment
variability?

2
Partitioning the Total Variability in the
Response

Total = Between + Within


Variabilityy Treatment Treatment
Variabilityy Variabilityy

Total Sum = Treatment Sum + Error Sum


of Squares of Squares of Squares

SSTO = SST + SSE


p ni p p ni

∑∑ (xij − x)2 = ∑ ni(xi − x)2 + ∑∑(x ij


− xi )2
i=1 j=1 i=1 i=1 j=1

F Test for Difference Between


Treatment Means
H0: µ1= µ2 =…= µp (no treatment effect)
Ha: At least two of the µ1, µ2, …, µp differ

Test Statistic:
MST SST/(p - 1)
F= =
MSE SSE/(n - p)
Reject H0 if
F > Fα or
p-value < α
Fα is based on p-1 numerator and n-p denominator
degrees of freedom.

3
The One-Way Analysis of Variance
Table
Degrees Sum of Mean F
Source of Freedom Squares Squares Statistic
Treatments p-1 SST MST = SST F = MST
p-1 MSE
Error n-p SSE MSE = SSE
n-p
Total n-1 SSTO

Example 10.5 The Gasoline Mileage Case (Excel Output)


Source of Variation SS df MS F P-value F crit
Between Groups 17.0493 2 8.5247 12.7424 0.0011 3.8853
Within Groups 8.0280 12 0.6690
Total 25.0773 14

Pairwise Comparisons, Individual


Intervals

Individual 100(1 - α)% confidence interval for µi - µh


1 1 tα/2 is based on n-p
(x i − x h ) ± t α/2 MSE  + 
 ni nh  degrees of freedom.

Example 10.6 The Gasoline Mileage Case (A vs B , α = 0.05)

1 1
(34.92 − 36.56) ± 2.179 0.669 +  = −1.64 ± 1.127
5 5
= [-2.7671, - 0.513]
Groups Count Average Variance MSE
Type A 5 34.92 0.587 0.669
Type B 5 36.56 0.723 n-p
Type C 5 33.98 0.697 12

4
Pairwise Comparisons, Simultaneous
Intervals
Tukey simultaneous 100(1 - α)% confidence interval for µi - µh
qα is the upper α percentage point of the
MSE studentized range for p and (n-p) from
(x i − x h ) ± qα Table A.9.
m m denotes common sample size.
Example 10.6 The Gasoline Mileage Case (A vs B, α = 0.05)
0.669
(34.92 − 36.56) ± 3.77 = −1.64 ± 1.379
5
= [−3.019, − 0.261] (A v B)
Groups Count Average Variance MSE [−0.439, 2.319] (A v C)
Type A 5 34.92 0.587 0.669
Type B 5 36.56 0.723 n-p
Type C 5 33.98 0.697 12 [ 1.201, 3.959] (B v C)

Estimation of Individual Treatment


Means
Individual 100(1 - α)% confidence interval for µi

MSE tα/2 is based on n-p degrees of


x i ± tα / 2
ni freedom.

Example 10.6 The Gasoline Mileage Case (Type B , α = 0.05)

0.669
36.56 ± 2.179 = 36.56 ± 0.797 = [35.763, 37.357]
5

Groups Count Average Variance MSE


Type A 5 34.92 0.587 0.669
Type B 5 36.56 0.723 n-p
Type C 5 33.98 0.697 12

5
10.3 The Randomized Block Design
A randomized block design compares p treatments (for example,
production methods) on each of b blocks (or experimental units; for
example, machine operators.) A generalization of the paired difference
design, this design controls for variability in experimental units by
comparing each treatment on the same (not independent) experimental
units.

Blocks
1 2 3… b
1
2 xij = response from
. treatment i and
.
.
block j
p

Example: Randomized Block Design

Example 10.7
The Defective Cardboard Box Case

Block
Treatment Machine Operator
Prod Meth 1 2 3 Trt Mean
1 9 10 12 10.3333
2 8 11 12 10.3333
3 3 5 7 5.0000
4 4 5 5 4.6667
Blk Mean 6.00 7.75 9.00 7.5833

6
The ANOVA Table, Randomized
Blocks

Degrees Sum of Mean F


Source of Freedom Squares Squares Statistic
Treatments p-1 SST MST = SST F(trt) = MST
p-1 MSE
Blocks b-1 SSB MSB = SSB F(blk) = MSB
b-1 MSE
Error (p-1)(b-1) SSE MSE = SSE
(p-1)(b-1)
Total pb-1 SSTO

F Test for Treatment Effects

H0: No difference between treatment effects


Ha: At least two treatment effects differ

Test Statistic:
MST SST/(p - 1)
F= =
MSE SSE/[(p - 1)(b - 1)]
Reject H0 if
F > Fα or
p-value < α
Fα is based on p-1 numerator and (p-1)(b-1) denominator
degrees of freedom.

7
F Test for Block Effects

H0: No difference between block effects


Ha: At least two block effects differ

Test Statistic:
MSB SSB/(b-1)
F= =
MSE SSE/[(p-1)(b-1)]
Reject H0 if
F > Fα or
p-value < α
Fα is based on b-1 numerator and (p-1)(b-1) denominator
degrees of freedom.

Example: Randomized Block


ANOVA
Example 10.7 The Defective Cardboard Box Case Minitab Output
Analysis of Variance for Defects
Source DF SS MS F P
Method 3 90.917 30.306 47.43 0.000
Operator 2 18.167 9.083 14.22 0.005
Error 6 3.833 0.639
Total 11 112.917

Block
Treatment Machine Operator
Prod Meth 1 2 3 Trt Mean
s = MSE
1 9 10 12 10.3333 = 0.639 = 0.7994
2 8 11 12 10.3333
3 3 5 7 5.0000
4 4 5 5 4.6667
Blk Mean 6.00 7.75 9.00 7.5833 Data Summary

8
Estimation of Treatment Differences
Under Randomized Blocks, Individual
Intervals
Individual 100(1 - α)% confidence interval for µi•• - µh••

2 tα/2 is based on (p-1)(b-1)


(x i• − x h • ) ± t α/2 s
b degrees of freedom.

Example 10.8 The Defective Cardboard Box Case (4 vs 1)


2
(4.6667 − 10.3333) ± 2.447(0.7994) = −5.6666 ± 1.5971
3
Treatment
Block
Machine Operator
= [−7.2637, − 4.0695]
Prod Meth 1 2 3 Trt Mean
1 9 10 12 10.3333
2 8 11 12 10.3333 t.025 with (3-1)(4-1) = 6
3 3 5 7 5.0000
4 4 5 5 4.6667 degrees of freedom.
Blk Mean 6.00 7.75 9.00 7.5833

Estimation of Treatment Differences


Under Randomized Blocks,
Simultaneous Intervals
Tukey simultaneous 100(1 - α)% confidence interval for µi•• - µh••
s qα is the upper α percentage point of the
(x i• − x h • ) ± qα studentized range for p and (p-1)(b-1)
b from Table A.9.

Example 10.8 The Defective Cardboard Box Case (4 vs 1)


0.7994
(4.6667 − 10.3333) ± 4.90 = −5.6666 ± 2.2621
3
Block
Treatment Machine Operator
= [-7.9278, - 3.4054]
Prod Meth 1 2 3 Trt Mean
1 9 10 12 10.3333
2 8 11 12 10.3333
3 3 5 7 5.0000
q.05 for 4 and 6.
4 4 5 5 4.6667
Blk Mean 6.00 7.75 9.00 7.5833

9
10.4 Two-Way Analysis of Variance
A two factor factorial design compares the mean response for a levels
of factor 1 (for example, display height) and each of b levels of factor 2
( for example, display width.) A treatment is a combination of a level of
factor 1 and a level of factor 2.

Factor 2
1 2 3… b
1
2 xijk = response for the
. kth experimental unit (k=1,…,m)
. assigned to the
. ith level of Factor 1 and the
a jth level of Factor 2

Example: Two-Way Analysis of


Variance
Example 10.9 The Shelf Display Case

Width
Height Reg Wide Mean
Bottom 58.2 55.7
53.7 52.5
55.8 58.9
Mean 55.9 55.7 55.8
Middle 73.0 76.2
78.1 78.4
75.4 82.1
Mean 75.5 78.9 77.2
Top 52.4 54.0
49.7 52.1
50.9 49.9
Mean 51.0 52.0 51.5
Mean 60.8 62.2 61.5

10
Example: Graphical Analysis of Bakery
Demand

Possible Treatment Effects in


Two-Way ANOVA

11
Two-Way ANOVA Table

Degrees Sum of Mean F


Source of Freedom Squares Squares Statistic
Factor 1 a-1 SS(1) MS(1) = SS(1) F(1) = MS(1)
a-1 MSE
Factor 1 b-1 SS(2) MS(2) = SS(2) F(2) = MS(2)
b-1 MSE
Interaction (a-1)(b-1) SS(int) MS(int) = SS(int) F(int) = MS(int)
(a-1)(b-1) MSE
Error ab(m-1) SSE MSE = SSE
ab(m-1)
Total abm-1 SSTO

Example: Two-Way ANOVA

Example 10.9 The Shelf Display Case Minitab Output

Analysis of Variance for Demand


Source DF SS MS F P
Height 2 2273.88 1136.94 185.62 0.000
Width 1 8.82 8.82 1.44 0.253
Interaction 2 10.08 5.04 0.82 0.462
Error 12 73.50 6.12
Total 17 2366.28

Width Data Summary


Height Reg Wide Mean
B 55.9 55.7 55.8
M 75.5 78.9 77.2
T 51.0 52.0 51.5
Mean 60.8 62.2 61.5

12
F Tests for Treatment Effects

H0: No difference between treatment effects


Ha: At least two treatment effects differ

Test Statistics:
Main Effects
MS(1) SS(1)/(a - 1) Fα is based on a-1 and ab(m-1)
F(1) = =
MSE SSE/[(ab(m - 1)] degrees of freedom.

MS(2) SS(2)/(b - 1) Fα is based on b-1 and ab(m-1)


F(2) = =
MSE SSE/[(ab(m - 1)] degrees of freedom.
Interaction
MS(int) SS(int)/[(a - 1)(b - 1)] Fα is based on (a-1)(b-1) and
F(int) = =
MSE SSE/[(ab(m - 1)] ab(m-1) degrees of freedom.
Reject H0 if
F > Fα or p-value < α

Estimation of Treatment Differences


Under Two-Way ANOVA, Factor 1

Individual 100(1 - α)% confidence interval for µi•• - µi’••


 2  tα/2 is based on ab(m-1)
(x i• − x i'• ) ± t α/2 MSE  
 bm  degrees of freedom.

Tukey simultaneous 100(1 - α)% confidence interval for µi•• - µi’••

 1  qα is the upper α percentage point of the


(x i• − x i'• ) ± qα MSE   studentized range for a and ab(m-1)
 bm  from Table A.9.

Example 10.10 The Shelf Display Case (M vs B)


 1 
(77.2 − 55.8) ± 3.773 6.12  = 21.4 ± 3.8105
 2(3) 
= [17.5895, 25.2105]

13
Estimation of Treatment Differences
Under Two-Way ANOVA, Factor 2

Individual 100(1 - α)% confidence interval for µ•j - µ•j’


 2  tα/2 is based on ab(m-1)
(x • j − x • j ' ) ± t α/2 MSE  
 am  degrees of freedom.

Tukey simultaneous 100(1 - α)% confidence interval for µ•j - µ•j’


qα is the upper α percentage point
 1 
(x • j − x • j ' ) ± qα MSE   of the studentized range for b and
 am  ab(m-1) from Table A.9.

Summary
•For more than two means comparison we use
ANOVA technique
•One way ANOVA is used when there is one
group variable
•Two way ANOVA is used when there are two
group variables

14

Potrebbero piacerti anche