EXPERIMENT – 10

Question: A structural engineer is studying the strength of aluminium alloy purchased from
three vendors. Each vendor submits the alloy in standard size bars of 1.0, 1.5 and 2.0
inches. The processing of different sizes of bars stock from a common ingot involves
different forging techniques and so this factory may be important. Furthermore, the
bar stock is forged from ingots made in different heats. Each vendor submits one test
specimen of each bar stock from 3 heats. The resulting strength data is as follows:

  Vendor 1 Vendor 2 Vendor 3

Heat 1 2 3 1 2 3 1 2 3
Bar Size                  
1 1.23 1.346 1.235 1.301 1.346 1.315 1.247 1.275 1.324
1.5 1.316 1.329 1.25 1.274 1.384 1.346 1.273 1.26 1.392
2 1.287 1.346 1.273 1.274 1.362 1.336 1.301 1.28 1.319

Analyze the data assuming the vendors and bar size are fixed and heats are random
effects.

REPORT PAGE
1
Object: To analyze the given data regarding the study of strength of aluminium alloy
purchased from three vendors assuming the vendors and bar size as fixed effect and
heats as random effects.

Procedure / Formula Used:

1. The model for split plot design is:

yijk = µ + αi + βj + eij + γk + δjk + eijk
(i = 1, 2, … r; j = 1, 2, …, p; k = 1, 2, …, q)

where, µ : general effect

αi: random block effect
βj: fixed j-th whole plot treatment effect
γk: fixed k-th sub plot treatment effect
δjk: fixed interaction effect between j-th whole plot treatment and k-th sub plot
treatment
eij , eijk: random errors
Here, βj, γk and δjk are fixed effects such that:
p q p q

∑ β j = ∑ γ k = ∑ δ jk = ∑ δ jk = 0
j=1 k =1 j=1 k =1

And αi ~ N (0,σ 2α), eij ~ N (0,σ 2e) and eijk ~ N (0,σ 2e' ).
2. The test hypothesis is:
Null Hypothesis:
H01: βj = 0 ∀ j = 1, 2, …, p
H02: γk = 0 ∀ k = 1, 2, …, q
H03: δjk = 0 ∀ (j = 1, 2, …, p; k = 1, 2, …, q)
Against Alternative Hypothesis:
H11: At least two of βj’s are different.
H12: At least two of γk’s are different.
H13: At least two of δjk’s are different.
3. The analysis of split plot design is done in two parts - Whole plot analysis and Sub
plot analysis.
4. For the whole plot analysis, the analysis of RBD with r replicates/blocks and p
whole plot treatments is conducted. Calculate correction factor (C.F), total sum of
square, replication sum of square SSR, sum of square due to whole plot treatment
SSA and whole plot error SSEI as follows:
G2
Correction Factor C.F =
rpq
r p q
2
Total sum of square TSS = ∑ ∑ ∑ y ijk – C.F
i=1 j=1 k=1

2
 r

∑ R2i
Replication sum of square SSR = = – C.F
i=1
pq
p

∑ A 2j
Sum of square due to whole plot treatment SSA = – C.F
j =1
rq
2
Whole plot error SSE1 = ∑ ( RA) – C.F – SSR – SSA
q
5. For the analysis of sub plot within the whole plots table of whole plot treatments
and sub plot treatments is drawn. Calculate the sum of square due to sub plot
treatment SSB, sum of square due to interaction SSAB and sub plot error SSEII as
follows:
q

Sum of square due to sub plot treatment SSB =

∑ B 2k – C.F
k=1
rp
2
( AB)
Sum of square due to interaction SSAB = ∑ – C.F – SSA – SSB
r
Sub plot error SSE2 = TSS – SSR – SSA – SSB – SSE1 – SSE2
6. Calculate the mean sum of square due to replication MSR, whole plot treatment
MSA, sub plot treatment MSB, interaction MSAB, whole plot error MSE1 and
sub plot error MSE2 as :

Mean sum of square =

∑ of square
Degree of freedom
7. Compute the Calculated and Tabulated F values.
MSA MSB MSAB
F A= , F B= , F AB=
MSE 1 MSE 2 MSE 2
8. Draw the ANOVA Table.
9. Accept or reject the test hypothesis according to the F values and conclude the
result.

Result:

The ANOVA Table is drawn in Table-10.4. The calculated F value for whole plot treatment
is 4.0757 and the tabulated F value at 5% level of significance is 6.94. Since the calculated
value is less than the tabulated value the null hypothesis is accepted, and it can be concluded
that whole plot treatments are alike.

The calculated F value for sub plot treatment is 1.3750 and the tabulated F value at 5% level
of significance is 3.89. Since the calculated value is less than tabulated value the null
hypothesis is accepted, and it can be concluded that sub plot treatments are alike.

3
The calculated F value for interaction is 0.0653 and the tabulated F value at 5% level of
significance is 3.26. Since the calculated value is less than the tabulated value the null
hypothesis is accepted, and it can be concluded that interaction effects are alike.

CALCULATION

Here, heat is treated as replication, bar size is treated as whole plot treatment and vendors as
sub plot treatment.
Table-10.1

    Vendor
Heat Bar V1 V2 V3
Siz
e
1 B1 1.23 1.301 1.247
B2 1.316 1.274 1.273
B3 1.287 1.274 1.301
2 B1 1.346 1.346 1.275
B2 1.329 1.384 1.26
B3 1.346 1.362 1.28
3 B1 1.235 1.315 1.324
B2 1.25 1.346 1.392
B3 1.273 1.336 1.319

Table-10.2 (Replication X whole plot)

  Bar Size    
Heat B1 B2 B3 Total( R) R2
1 3.778 3.863 3.862 11.503 132.319
2 3.967 3.973 3.988 11.928 142.2772
3 3.874 3.988 3.928 11.79 139.0041
Total(A) 11.619 11.824 11.778    
A2 135.0012 139.807 138.7213    

Table-10.3(main plot X sub plot)

  Bar Size    
Vendor B1 B2 B3 Total(B) B2

4
 1 3.811 3.895 3.906 11.612 134.8385
2 3.962 4.004 3.972 11.938 142.5158
3 3.846 3.925 3.9 11.671 136.2122

r=3, p=3, q=3, N=27

G= 35.221
G2 = 1240.518841
C.F.= 45.94514226
Raw SS= 45.996023
Total SS= 0.050880741
∑ R 2= 413.600293
∑ A 2= 413.529421
∑ B2 = 413.566629
∑ (RA)2= 137.878263
∑ ( AB)2= 137.865167

SSR = 0.010445852
SSA = 0.002571185
SSE 1 = 0.001261704
SSB = 0.006705407
SSAB = 0.000636815
SSE 2 = 0.029259778
Table-10.4(ANOVA Table)

degree of Sum of Mean sum of

Source of variation
freedom Squares squares F-ratio
Blocks (R) 2 0.010446 0.005223  
Main plot treatment(A) 2 0.002571 0.001286 FA = 4.075735
Error(E1) 4 0.001262 0.000315  
Sub plot treatment(B) 2 0.006705 0.003353 FB = 1.375009
Interaction (AB) 4 0.000637 0.000159 FAB=0.065293
Error(E2) 12 0.02926 0.002438  
Total 26      

Tab FA = 6.94
Tab FB = 3.89
Tab FAB = 3.26