|
Variation |
Degree of Freedom |
Sum of Squares |
Mean Square |
F |
|
Explained by treatments (between columns) |
c-1 |
SST = # rows x
[(column avg-Grand mean)2 + (column avg Grand mean)2} |
MST=SST/c-1 |
MST/MSE |
|
Error or unexplained (within columns) |
(r-1)c |
SSE = (treatment
- column avg)2 + (treatment column avg)2 Or (SS-SST) |
MSE = SSE/(r-1)c |
|
|
Total |
Rc-1 |
SS = (treatment Grand mean)2
+ (treatment Grand mean)2 Or (SST + SSE) |
|
|
|
Variation |
Degree of Freedom |
Sum of Squares |
Mean Square |
F |
|
Explained by treatments (between columns) |
c-1 |
SST =# rows x[ (column avg-Grand mean)2 + (column avg
Grand mean)2 ] |
MST=SST/c-1 |
MST/MSE |
|
Explained by blocks (between rows) |
r-1 |
SSB =# columns
x[ (row avg - Grand mean)2
+ (row avg Grand mean)2] |
MSB = SSB / r 1 |
MSB/MSE |
|
Error or unexplained (within columns) |
(r-1)(c-1) |
SSE = SS-SST-SSB |
MSE = SSE/(r-1)(c-1) |
|
|
Total |
Rc-1 |
SS = (treatment Grand mean)2
+ (treatment Grand mean)2 |
|
|
F table (c-1 , error of
unexplained within columns) 5% significance level
Column Across
Top (c-1)
error of unexplained within
columns 5%answer
Example:
Hypothesis: Average examination score for each package comes
from populations with the same mean. It
will therefore indicate whether the packages make a difference to eventual
examination results.
|
|
A |
B |
C |
Average |
|
School 1 |
64 |
61 |
55 |
60 |
|
School 2 |
69 |
63 |
63 |
65 |
|
School 3 |
72 |
68 |
70 |
70 |
|
School 4 |
58 |
65 |
60 |
61 |
|
School 5 |
64 |
60 |
59 |
61 |
|
School 6 |
63 |
61 |
59 |
61 |
|
Average |
65 |
63 |
61 |
Grand mean = 63 |
SST = #
rows x[(column avg-Grand mean)2 + (column avg Grand mean)2]
SST = 6 x [ (65-63)2
+63-63)2 + (61-63)2]
SST = 6 x [ (2)2
+ (0)2 + (-2)2]
SST = 6 x [4+4]
SST = 6 x 8
SST = 48
SSB = # columns x[(
(row avg - Grand mean)2 + (row avg Grand mean)2]
SSB = 3 x [ (60-63)2
+(65-63)2 + (70-63)2 + (61-63)2 + (61-63)2
+ (61-63)2]
SSB = 3 x [ (-3)2
+ (2)2 +(7)2 + (-2) 2 + (-2)22 +
(-2)2]
SSB = 3 x [9+4+49+4+4+4]
SSB = 3 x 74
SSB = 222
SS = (treatment
Grand mean)2 + (treatment Grand mean)2
|
|
A |
|
|
B |
|
|
C |
|
|
School 1 |
(64-63)2= |
1 |
|
(61-63)2= |
4 |
|
(55-63)2= |
64 |
|
School 2 |
(69-63)2= |
36 |
|
(63-63)2= |
0 |
|
(63-63)2= |
0 |
|
School 3 |
(72-63)2= |
81 |
|
(68-63)2= |
25 |
|
(70-63)2= |
49 |
|
School 4 |
(58-63)2= |
25 |
|
(65-63)2= |
4 |
|
(60-63)2= |
9 |
|
School 5 |
(64-63)2= |
1 |
|
(60-63)2= |
9 |
|
(59-63)2= |
16 |
|
School 6 |
(63-63)2= |
0 |
|
(61-63)2= |
4 |
|
(59-63)2= |
16 |
|
|
|
144 |
|
|
46 |
|
|
154 |
SS = 144 + 46 + 154
SS = 344
Or for 1 way Anova
SS = SST + SSE
SS = 48 + 296
SS = 344
SSE = (treatment -
column avg)2 + (treatment column avg)2
|
|
A |
|
|
B |
|
|
C |
|
|
School 1 |
(64-65)2= |
1 |
|
(61-63)2= |
4 |
|
(55-61)2= |
36 |
|
School 2 |
(69-65)2= |
16 |
|
(63-63)2= |
0 |
|
(63-61)2= |
4 |
|
School 3 |
(72-65)2= |
49 |
|
(68-63)2= |
25 |
|
(70-61)2= |
81 |
|
School 4 |
(58-65)2= |
49 |
|
(65-63)2= |
4 |
|
(60-61)2= |
1 |
|
School 5 |
(64-65)2= |
1 |
|
(60-63)2= |
9 |
|
(59-61)2= |
4 |
|
School 6 |
(63-65)2= |
4 |
|
(61-63)2= |
4 |
|
(59-61)2= |
4 |
|
|
|
120 |
|
|
46 |
|
|
130 |
SSE = 120 + 46 + 130 = 296
Or for 1 way Anova
SSE = SS SST
For 2 way Anova
SSE = SS - SST SSB
Hypothesis: Average examination
score for each package comes from populations with the same mean. It will therefore indicate whether the
packages make a difference to eventual examination results.
F table (2,15) 5%
significance level
Column Across Top 2
15 3.59
|
Variation |
Degree of Freedom |
Sum of Squares |
Mean Square |
F |
|
Explained by treatments (between columns) |
c-1 3-1=2 |
SST = =# rows x[ (column avg-Grand mean)2 + (column avg
Grand mean)2 ]SST = 48 |
MST=SST/c-1 MST = 48/2= 24 |
MST/MSE 24/19.73 = 1.2164 |
|
Error or unexplained (within columns) |
(r-1)c (6-1)3=15 |
SSE = (treatment
- column avg)2 + (treatment column avg)2 Or (SS-SST) SSE = 296 |
MSE = SSE/(r-1)c MSE = 296/15=19.73 |
|
|
Total |
rc-1 (6 x 3)-1=17 |
SS = (treatment Grand mean)2
+ (treatment Grand mena)2 Or (SST + SSE) SS = 344 |
|
|
Value 1.22 is much
smaller. The hypothesis is
accepted. There is no significant
difference in the examination results.
The packages do not appear to make a difference to examination results.
Above the fact that treatment group came form populations
with the same mean was accepted.
However, the experiment was carried out in several schools. It could well be a large part of the
variation between the groups was accounted for by difference between schools.
A two way anova table will
neutralize and remove from calculation example difference between schools
Hypothesis: Average
examination score for each package comes from populations with different mean
so we will remove the school parameter from the equation the same mean. It will therefore indicate whether the
packages make a difference to eventual examination results.
F table (2,10) 5%
significance level
Column Across Top 2
10 4.10
|
Variation |
Degree of Freedom |
Sum of Squares |
Mean Square |
F |
|
Explained by treatments (between columns) |
c -1 3 -1=2 |
SST = # rows x[ (column avg-Grand mean)2 + (column avg
Grand mean)2 ]SST = 48 |
MST=SST/c-1 MST = 48/2= 24 |
MST/MSE 24/7.4 = 3.2432 |
|
Explained by blocks (between rows) |
r- 1 6 1= 5 |
SSB =# columns
x[ (row avg - Grand mean)2
+ (row avg Grand mean)2] SSB = 222 |
MSB = SSB / r 1 MSB = 222 /5 = 44.4 |
MSB / MSE 44.4 / 7.4 = 6.0 |
|
Error or unexplained (within columns) |
(r-1)(c-1) (6-1)(3-1)=10 |
SSE = SS-SST-SSB SSE = 74 |
MSE = SSE/(r-1)(c-1) MSE = 74/10= 7.4 |
|
|
Total |
rc-1 (6 x 3)-1=17 |
SS = (treatment Grand mean)2
+ (treatment Grand mean)2 SS = 344 |
|
|
Value 3.24 is less than 4.10
. The hypothesis is accepted. There is no significant difference in the
examination results. The packages do
not appear to make a difference to examination results