Sei sulla pagina 1di 44

# Analysis of Variance (ANOVA)

ANOVA
Analysis of variance (ANOVA) is a method of testing
the equality of three or more population means by
analyzing sample variances.

## One way ANOVA

Two way ANOVA
Multi-factorial ANOVA

Type

Dependent Variable
(Numerical)

Independent variables
(Categorical)

One way

Exam marks

Race
Malay
Chinese
Indian

Two way

Exam marks

Two factors
Gender
Race

Multifactorial

Exam marks

## > Two factors

Gender
Race
Teaching method
..
..

Assumption
The population under study have normal
distributions
The samples are drawn randomly
Each sample is independent of the other
sample
Equal population variances

Experimental design:
For the effective use of ANOVA, experiment has to
be standardized in terms of the randomness and
replications of all variables, including samples,
surrounding and trial protocol.
The experimental design has to be standardized to
reduce the error of all variables mentioned above.

A
D

## Completely Randomized Designed

(CRD):
(e.g.: Six treatments (A, B,..., F) x
Four replications)

F
D
B
B

E
A
F

C
E
F

C
D
F
C
E
B
A
C
E
A
B
D

Block I

D
E
F
B

## Randomized Complete Block Designed

(RCBD):
(e.g.: Six treatments(A, B,..., F) x four
block replications)
Block II

B
C
D
A
F
E

E
C
D

Block III

F
B

F
A

Block IV

C
D
E

STEPS IN ANOVA

1.

## Arrange data. Draft ANOVA table.

ANOVA Standard Table:
DF = degree of freedom
Source of
Variation
*

DF

SS

MS

Fobserved

Ftabulated

5%

Treatment
Error/residual

Total

SS = sum of squares
MS = mean square; MS treatment=SStreatment/DFtreatment..
Fobserved= MStreatment/MSerror

1%

2.

Define and HA
Ho: All means are the same value
HA: There are means which are not the same
3. Determine Source of variation
4. Determine D of F for each source
5. Determine CF
6. Determine SS
7. Determine MS (=SS/D of F)
8. Calculate Fcalculated
9. Obtain Ftabulated from Table-F
10. Compare Ftabulated with Fcalculated
If Fcalculated > Ftabulated reject Ho
If Fcalculated < Ftabulated
accept Ho

## One way ANOVA

To study the effect of one (independent) factor.
Eg: To access whether the mean BMI of patients are
significantly different among races.

Malay
Mean
BMI

Chinese

Indian

Unpaired t test
I

Increase
type I
error!!!!!

Unpaired t test II
Unpaired t test III

ANOVA

## Example 1 (CRD 1 factor)

One trial was conducted to determine the effectiveness of six types of pesticide, RS1-RS6 in four
replications, 1-4, with an experimental design CRD. Does the pesticides increase the grain
yield by preventing the pest deseases?

Treatment

## Grain yield (kg ha-1)

RS1

2537

2069

2104

1797

RS2

3366

2591

2211

2544

RS3

2536

2459

2827

2385

RS4

2387

2453

1556

2116

RS5

1997

1679

1649

1859

RS6

1796

1704

1904

1320

Control

1401

1516

1270

1077

Hypothesis:
Ho: All means are the same value
i.e. 1=2=3=4=5=6=control
HA: There are means which are not the same

Source of
Variation*

DF

Treatment

Error

21

Total

27

SS

MS

Fcalculated

Ftabulated
5%

1%

## * for CRD, 1 factor, SV = T, R and Total

DFtreatment = 7-1 = 6
Dftotal = (total no. data -1) = 28-1=27
Therefore, DFerror = 27- 6 = 21
........next, calculate total and total means for each treatment, grand mean and
grand total. All these are required to calculate SS.....

Treatment

Total

Mean

RS1

2537

2069

2104

1797

8507

2127

RS2

3366

2591

2211

2544

10712

2678

RS3

2536

2459

2827

2385

10207

2552

RS4

2387

2453

1556

2116

8512

2128

RS5

1997

1679

1649

1859

7184

1796

RS6

1796

1704

1904

1320

6724

1681

Control

1401

1516

1270

1077

5264

1316

Grand total
Grand Mean

57110
2040

to calculate SS

## CF = (Grand Total)2/no. observation

= (57110)2/28
= 116,484,004
SStotal=[(2537)2 + (2069)2 + (2104)2 +...+ (1077)2] - CF
= 7,577,421
SStreatment =1/4 (85072+107122+.....+52642) - CF
= 5,587,175
SSerror = SStotal - SStreatment
= 1,990,246

## ANOVA Standard Table

Source
of DF
Variation

SS

Treatment

5,587,175

931,196

Error

21

1,990,246

94,773

Total

27

7,577,421

MS

Fcalculated
9.82

Ftabulated
5%

1%

2.57

3.81

Ftotal = MStreatment/MSerror
Ftable = F (dftreatment, dferror) at a significance level of
= F (x , y) at a significance level of from
Fischers Table
Ho is rejected. There are differences among the means........

## Example 2 (CRD 1 factor)

In one experiment using CRD, three types of fertilizer (B1,
B2, and B3) were tested on its effectiveness on peanuts
yield, in five replications (R1, R2,.....R5). Based on the
results below, determine whether the fertilizers give
different yield.
Fertilizer
type

Yield (g m-2)
R1

R2

R3

R4

R5

B1

86

79

81

70

84

B2

90

76

88

82

89

B3

82

68

73

71

81

Ho: B1=B2=B3
H1: There are mean which are not the same

Fertilizer type

Yield (g m-2)

Total

Mean

R1

R2

R3

R4

R5

B1

86

79

81

70

84

400

80

B2

90

76

88

82

89

425

85

B3

82

68

73

71

81

375

75

Total

1200

CF = (1200)2/15= 96,000
SStotal = [(86)2 + (79)2 +........+(71)2 +( 812) ]- CF= 698
SStreatment = (400)2 + (425)2 + (375)2) CF = 250
5
SSerror = SStotal - SStreatment = 698-250=448
Source of variation

DF

SS

MS

Fcalculated

F(2,12)

5%

1%

treatment
error
total

Accept Ho. All means are the same, the fertilizers do not give any effect on
the yield of peanut.

## In certain one factorial experiment, an

experimental design, RCBD, is used when
we use block.
Block is used in certain cases as follows:
1. The experimental site is non-homogenous (non
uniform). For example, a field experiment involved
planting, there might be uneven amount of soil water
available for the plant, or the site is having granulared
soil or shaded (from light) etc..
2. The experiment was conducted by different people from
day to day with different
accuracy in measurement. In this case, each person can
be considered as one block.

## Example 3 (RCBD 1 factor)

One experiment was conducted on paddy type IR8 to study the the effect of six
germination densities (25, 50,.....,150 kg seed per ha) in four blocks (I IV).
Determine whether germination densities influences the paddy yield based on
the following results.
Yield of paddy variety IR8 at six germination densities
Treatment
(kg seed/ha)

25
50
75
100
125
150

5,113
5,346
5,272
5,164
4,804
5,254

Block I

Ho: 1=2=3=4=5=6
HA: 1 2 3 4 5 6
Ho: b1=b2=b3=b4
HA: b1 b2 b3 b4

yield

(kg/ha)

Block II

Block III

Block IV

5,398
5,952
5,713
4,831
4,848
4,542

5,307
4,719
5,483
4,986
4,432
4,919

4,678
4,264
4,749
4,410
4,748
4,098

Treatment (kg
seed/ha)

(kg/ha)

Treatment

Block I

Block II

Block III

Block IV

total

Mean

25
50
75
100
125
150

5,113
5,346
5,272
5,164
4,804
5,254

5,398
5,952
5,713
4,831
4,848
4,542

5,307
4,719
5,483
4,986
4,432
4,919

4,678
4,264
4,749
4,410
4,748
4,098

20,496
20,281
21,217
19,391
18,832
18,813

5,124
5,070
5,304
4,848
4,708
4,703

T Block

30,953

31,284

29,846

26,947

G Total

119,030

G Mean
M Block

4,960
5158.8

5214.0

4974.3

4491.2

CF = (119030)2/24 = 590,339,204
SStotal = [(5113)2+(5398)2+.........+(4098)2] CF
= 4,801,068
SSblock = (30953)2 + (31284)2 + (29846)2 +(26947)2 - CF
6
= 1,944,361
SStreatment = (20496)2 + (20281)2 +.......+(18813)2 - CF
4
= 1,198,331
SSerror = SStotal SSblock SStreatment
= 4,801,068 1,944,361 1,198,331
= 1,658,376

Source
Variation

DF

SS

Block
Treatment
Error

3
5
15

1,944,361
1,198,331
1,658,376

Total

23

4,801,068

MS
648,120
239,666
110,558

Fcalculated
5.86**
2.17n.s.

Ftable
5%

1%

3.29
2.90

5.42
4.56

## Block Ftable= (3,15)

Treatment Ftable =(5,15)
Block= we accept Ha at both sig. Level 5% & 1%
Treatment= we accept Ho, no sig dif.......

Exercise
Four researchers (P I P IV) were asked to measure the photosynthesis rate in
an extensive experiment which involved five light treatments (T1- T5).
Results was reported as follows:
Light
treatment

PI

P II

P III

P IV

T1

3.8

2.9

1.1

3.6

T2

6.7

6.8

3.2

7.0

T3

9.9

10.1

6.1

10.2

T4

12.5

11.8

7.4

13.0

T5

13.1

14.0

8.2

14.1

## Using ANOVA and an appropriate method (at 1%), can the

differences among the treatments (if different) be accepted
without being sceptical on the researchersaccuracies?

PI

P II

P III

P IV

T1

3.8

2.9

1.1

3.6

T2

6.7

6.8

3.2

T3

9.9

10.1

6.1

10.2

T4

12.5

11.8

7.4

13

T5

13.1

14

8.2

14.1

46

45.6

26

47.9

CF

SS total

SSblock

SStreatment

SV

DF

SS

MS

F t (1%)

Block
treatment
error
Total

Conclusion:
Among light treatments: there are differences between mean (accept HA)
Among the researchers (Block): there are differences between mean (accept HA)

## LSD, Least Significant Difference

(Perbezaan Bererti Terkecil)
The conclusion in ANOVA is general:
e.g. If Ho is accepted, no further test is required as all
means are the same.
e.g. If HA is accepted, the difference in means (which
pairs?) cannot be determined through F test in
ANOVA.

Further step that is LSD test is carried out to determine which pairs are
significantly different. If the treatments were conducted with the same no.
replications (=r), the formula to calculate LSD at the significance level of
is as follows:
LSD = t (2MSe/r)

## = ttabulated for 2 tail at significance level of and

degree of freedom of error (dferror), from ANOVA
MSe = error mean square, from ANOVA
If no. replication are not the same (= r1 and r2 for two paired means), the
formula above will be changed to:
t

## Example 1 (CRD 1 factor)

One trial was conducted to determine the effectiveness of six types of pesticide, RS1-RS6 in four
replications, 1-4, with an experimental design CRD. Does the pesticides increase the grain
yield by preventing the pest deseases?

Treatment

## Grain yield (kg ha-1)

RS1

2537

2069

2104

1797

RS2

3366

2591

2211

2544

RS3

2536

2459

2827

2385

RS4

2387

2453

1556

2116

RS5

1997

1679

1649

1859

RS6

1796

1704

1904

1320

Control

1401

1516

1270

1077

Hypothesis:
Ho: All means are the same value
i.e. 1=2=3=4=5=6=control
HA: There are means which are not the same

Interpretation of Result
If the difference between means > LSD
The different is significant at a significance level of
If the difference between means < LSD
The different is not significant at a significance level of
Example 1: A trial on pesticides (CRD)
Source of
variation

DF

SS

MS

Fobserved

Ftabulated
5%

Treatment

5,587,175

931,196

Error

21

1,990,246

94,773

Total

27

7,577,421

9.82

1%

2.57 3.81

Conclusion:
Treatment

Mean

## Different with control

Significant

RS2

2678

1362

**

RS3

2552

1236

**

RS4

2128

812

**

RS1

2127

811

**

RS5

1796

480

RS6

1681

365

ns

Control

1316

** significant at 1%
* significant at 5%
ns, the difference is not significant

## Example 2: An experiment on the effect of the light

treatment on photosynthesis rate
SV

DF

SS

MS

Ftabulated 1%

Block

63

21

36.2

5.95

Treatment

242

60.5

104.3

5.41

Error

12

0.58

Total

19

312

## Determine the difference in mean pairs between the treatments at

significance level of 1%.

T1

## (t 12, 0.005 = 3.055)

mean

Significant

vs.

T2

3.08

**

T1 vs.

T3

6.23

**

T2

T3

3.15

**

T5

1.17

n.s.

vs.

.......
........

........
T4

vs.

Concl:
....
LSD for block = 1.65
......
on the table above, only the mean pair of T4 and T5 does not show a
significant difference at significant level of 1%. Other mean pairs show
significant differences.

## USE LSD WHEN NECESSARY!

1. Use LSD when F test in ANOVA shows significant
difference between the mean.
2. DONT USE LSD to differentiate between means in
which its treatments are more than FIVE.
3. If the experiment/trial is conducted with CONTROL
treatment, then the difference between means can be
done by comparing the CONTROL with other
treatments with no limit. However, if there is no
CONTROL treatment, find the difference for EACH
pair as much as possible.

## Duncans Multiple Range Test (DMRT)

(Ujian Berbilang Duncan)
Steps in DMRT:
1.
Arrange all means in descending order.
2.
Calculate Standard Deviation for Mean Treatment, Sx
Sx = (MSe/r)

3.

Terdekat), Rp.
Rp = rp Sx

## (rp = from New Multiple Range Test Table)

4.
5.

Arrange Rp in sequence next to its mean. (Rp must follow the mean
sequence i.e. descending order)
Calculate D = (Mean Rp) . Draw a conclusion (see the following
examples)

## Example 1: An experiment on pesticides

grain yield (kg ha-1)

treatment

RS1

2537

2069

2104

1797

RS2

3366

2591

2211

2544

RS3

2536

2459

2827

2385

RS4

2387

2453

1556

2116

RS5

1997

1679

1649

1859

RS6

1796

1704

1904

1320

Control

1401

1516

1270

1077

Source
variation

of

DF

SS

MS

treatment

5,587,175

931,196

error

21

1,990,246

94,773

total

27

7,577,421

Fcalculated
9.82

Ftabulated
5%

1%

2.57

3.81

DMRT method:
Sx = (MSe/r) = (94773/4) = 153.93
Rp = rp Sx
rp from table New Multiple Range Test for p= 2-7 (df =21) at significance
level of 5%as follows:

rp

Rp = rp Sx

2.94

453

3.09

476

3.18

490

3.25

500

3.30

508

3.33

513

treatment

Mean

Rp

D
Mean - Rp

RS2

2678 a

513

2165

RS3

2552 ab

508

2044

RS4

2128 bc

500

1628

RS1

2127 bc

490

1637

RS5

1796

476

1320

RS6

1681

cd

453

1228

Control

1316

For RS2 treatment, assume all means >2165 as same with this mean treatment. Thus,
mean treatments RS2 and RS3 are the same. These means are given same symbol
(i.e. a).
For RS3 treatment, assume all means >2044 as same with this mean treatment. Thus,
mean treatments RS3, RS4 and RS1 are the same. These means are given same
symbol (i.e. b).
For RS4 treatment, assume all means >1628 as same with this mean treatment. Thus,
mean treatments RS4, RS1, RS5 and RS6 are the same. These means are given
same symbol (i.e. c).
For RS1 treatment, assume all means >1637 as same with this mean treatment. Thus,
mean treatments RS1, RS5 and RS6 are the same.
Next, for RS5 treatment, assume all means >1320 as same with this mean treatment.
Thus, mean treatments RS5 and RS6 are the same.
Lastly, for RS6 treatment, assume all means >1228 as same with this mean treatment.
Thus, mean treatments RS6 and control are the same. These means are given same
symbol (i.e. d).

Conclusion:

Treatment

Mean

RS2

2678 a

RS3

2552 ab

RS4

2128 bc

RS1

2127 bc

RS5

1796 c

RS6

1681 cd

Kawalan

1316 d

## An CRD experimental designed was conducted to examine the

effectiveness of plant hormonal treatments (H1, H2 dan H3) on
pineapple yield in four replications (R1-R4). The following results were
obtained: Hormonal
Pineapple yield (seed ha-1 week-1)
treatment

R1

R2

R3

R4

H1

87

79

83

92

H2

80

73

75

71

H3

94

87

90

92

SV

DF

SS

MS

Fcalculated

Ftable (0.01)

treatment

529

265

14.6

7.56

error

164

18.2

total

11

693

Using ANOVA, test whether the hormonal treatment gives different yield
at a significance level of 1%:

Conclusion: Fcalculated > Ftabulated. There are differences among the means, but
we dont know which treatments.
To solve this problem, used LSD or DMRT.
LSD method
LSD = t (2MSe/r)
LSD 0.01 = t 0.005 x (2 x 18.2)/4 = 3.25 x 9.1 = 9.8
Conclusion:
..........
mean
Mean
treatment

Significant

H1 & H2

10.5

**

H1 & H3

5.5

n.s.

H2 & H3

16

**

DMRT method
Sx = (MSe/r) = (18.2)/4 = 2.13
*

Rp = rp
Sx

rp

4.60

9.8

4.86

10.4

## rp at a significance level of 1%, df =9

Treatment

Mean

Rp

D
Mean - Rp

H3

90.8 a

10.4

80.4

H2

85.3 a

9.8

75.5

H1

74.7 b

Conclusion...........................

Error

Protection

df

level

10

11

12

13

14

P=

number

of

means

for

range

being

tested

10

12

14

16

18

20

0.05

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

18.0

0.01

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

90.0

0.05

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

6.09

0.01

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

14.0

0.05

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

4.50

0.01

8.26

8.5

8.6

8.7

8.8

8.9

8.9

9.0

9.0

9.0

9.1

9.2

9.3

9.3

0.05

3.93

4.01

4.02

4.02

4.02

4.02

4.02

4.02

4.02

4.02

4.02

4.02

4.02

4.02

0.01

6.51

6.8

6.9

7.0

7.1

7.1

7.2

7.2

7.3

7.3

7.4

7.4

7.5

7.5

0.05

3.64

3.74

3.79

3.83

3.83

3.83

3.83

3.83

3.83

3.83

3.83

3.83

3.83

3.83

0.01

5.70

5.96

6.11

6.18

6.26

6.33

6.40

6.44

6.5

6.6

6.6

6.7

6.7

6.8

0.05

3.46

3.58

3.64

3.68

3.68

3.68

3.68

3.68

3.68

3.68

3.68

3.68

3.68

3.68

0.01

5.24

5.51

5.65

5.73

5.81

5.88

5.96

6.0

6.0

6.1

6.2

6.2

6.3

6.3

0.05

3.35

3.47

3.54

3.58

3.60

3.61

3.61

3.61

3.61

3.61

3.61

3.61

3.61

3.61

0.01

4.95

5.22

5.37

5.45

5.53

5.61

5.69

5.73

5.8

5.8

5.9

5.9

6.0

6.0

0.05

3.26

3.39

3.47

3.52

3.55

3.56

3.56

3.56

3.56

3.56

3.56

3.56

3.56

3.56

0.01

4.74

5.00

5.14

5.23

5.23

5.40

5.47

5.51

5.5

5.6

5.7

5.7

5.8

5.8

0.05

3.20

3.34

3.41

3.47

3.50

3.52

3.52

3.52

3.52

3.52

3.52

3.52

3.52

3.52

0.01

4.60

4.86

4.99

5.08

5.17

5.25

5.32

5.36

5.4

5.5

5.5

5.6

5.7

5.7

0.05

3.15

3.30

3.37

3.43

3.46

3.47

3.47

3.47

3.47

3.47

3.47

3.47

3.47

3.48

0.01

4.48

4.73

4.88

4.96

5.06

5.13

5.20

5.24

5.28

5.36

5.42

5.48

5.54

5.55

0.05

3.11

3.27

3.35

3.39

3.43

3.44

3.45

3.46

3.46

3.46

3.46

3.46

3.47

3.48

0.01

4.39

4.63

4.77

4.86

4.94

5.01

5.06

5.12

5.15

5.24

5.28

5.34

5.38

5.39

0.05

3.08

3.23

3.33

3.36

3.40

3.42

3.44

3.44

3.46

3.46

3.46

3.46

3.47

3.48

0.01

4.32

4.55

4.68

4.76

4.81

4.92

4.96

5.02

5.07

5.13

5.17

5.22

5.24

5.26

0.05

3.06

3.21

3.30

3.35

3.38

3.41

3.42

3.44

3.45

3.45

3.46

3.46

3.47

3.47

0.01

4.26

4.48

4.62

4.69

4.74

4.84

4.88

4.94

4.98

5.04

5.08

5.13

5.14

5.15

0.05

3.03

3.18

3.27

3.33

3.37

3.39

3.41

3.42

3.44

3.45

3.46

3.46

3.47

3.47

0.01

4.21

4.42

4.55

4.63

4.70

4.78

4.83

4.87

4.91

4.96

5.00

5.04

5.06

5.07

15

16

17

18

19

20

22

24

30

60

0.05

3.01

3.16

3.25

3.31

3.36

3.38

3.40

3.42

3.43

3.44

3.45

3.46

3.47

3.47

0.01

4.17

4.37

4.50

4.58

4.64

4.72

4.77

4.81

4.84

4.90

4.94

4.97

4.99

5.00

0.05

3.00

315

3.23

3.30

3.34

3.37

3.39

3.41

3.43

3.44

3.45

3.46

3.47

3.47

0.01

4.13

4.34

4.45

4.54

4.60

4.67

4.72

4.76

4.79

4.84

4.88

4.91

4.93

4.94

0.05

2.98

3.13

3.22

3.28

3.33

3.36

3.38

3.40

3.42

3.44

3.45

3.46

3.47

3.47

0.01

4.10

4.30

4.41

4.50

4.56

4.63

4.68

4.72

4.75

4.80

4.83

4.86

4.88

4.89

0.05

2.97

3.12

3.21

3.27

3.32

3.35

3.37

3.39

3.41

3.43

3.45

3.46

3.47

3.47

0.01

4.07

4.27

4.38

4.46

4.53

4.59

4.64

468

4.71

4.76

4.79

4.82

4.84

4.85

0.05

2.96

3.11

3.19

3.26

3.31

3.35

3.37

3.39

3.41

3.43

3.44

3.46

3.47

3.47

0.01

4.05

4.24

4.35

4.43

4.50

4.56

4.61

4.64

4.67

4.72

4.76

4.79

4.81

4.82

0.05

2.95

3.10

3.18

3.25

3.30

3.34

3.36

3.38

3.40

3.43

3.44

3.46

3.46

3.47

0.01

4.02

4.22

4.33

4.40

4.47

4.53

4.58

4.61

4.65

4.69

4.73

4.76

4.78

4.79

0.05

2.93

3.08

3.17

3.24

3.29

3.32

3.35

3.37

3.39

342

3.44

3.45

3.46

3.47

0.01

3.99

4.17

4.28

4.36

4.42

4.48

4.53

4.57

4.60

4.65

4.68

4.71

4.74

4.75

0.05

2.92

3.07

3.15

3.22

3.28

3.31

3.34

3.37

3.38

3.41

3.44

3.45

3.46

3.47

0.01

3.96

4.14

4.24

4.33

4.39

4.44

4.49

4.53

4.57

4.62

4.64

4.67

4.70

4.72

0.05

2.89

3.04

3.12

3.20

3.25

3.29

3.32

3.35

3.37

3.40

3.43

3.44

3.46

3.47

0.01

3.89

4.06

4.16

4.22

4.32

4.36

4.41

4.45

4.48

4.54

4.58

4.61

4.63

4.65

0.05

2.83

2.98

3.08

3.14

3.20

3.24

3.28

3.31

3.33

3.37

3.40

3.43

3.45

3.47

0.01

3.76

3.92

4.03

4.12

4.17

4.23

4.27

4.31

434

4.39

4.44

4.47

4.50

4.53