SPLIT PLOT DESIGN

Two-Factors Experiment

Dr. Yousaf Hayat

Professor
Department of Mathematics, Statistics and Computer Science

01/25/24 Dr. Yousaf Hayat 1

Split-Plot Design
 It is a type of two-factors factorial experiments but only the difference is
that of assignments of treatments/factors to the experimental plot. Also, in
most of the experiments, some factors need larger plots as compared to
others. In addition, in some cases the experimenter is interested to
measure the effect of one factor with more precision as compared to other
factor.

 A design in which one factor levels are assigned to the larger plots (main
plots or whole plots) and the levels of other factor are assigned to the sub-
plots within the main plots, is called split-plot design.
01/25/24 Dr. Yousaf Hayat 2
Split-Plot Design

 Similarly, if an experimenter is interested to measure the effect of one

factor with low precision, their levels are assigned to the main plots, and
the other factor is required to measure with greater precision, their levels
are assigned to the sub-plots within the main plots. In such a case the
design used for conducting the experiment is called split-plot design.
 Split type of design allow the INTRODUCTION of another factor or any
other level of already considered factors if the experiment is already in
progress which is not possible in factorial experiment.

01/25/24 Dr. Yousaf Hayat 3

Example
 An agronomist is interested to determine the effect of planting dates on
the yield of different varieties of wheat. In such case, planting dates are
assigned to the main plots whereas the varieties are assigned to the sub-
plots.

 A researcher is interested to determine the effect of four different doses

of fertilizers on the yield of wheat varieties. In such a case, Fertilizer
doses are assigned to the main plots and varieties are randomly
assigned to the sub-plots, if more interest lies in varieties.

01/25/24 Dr. Yousaf Hayat 4

Experimental Errors ion Split Plot Design
 Since, split-plot design have two types of plots (main plots and sub-plots),
so it involves two experimental errors i.e. one each for main plots and
sub-plots respectively.

 The degrees of freedom for the main plots error is less than the degrees of
freedom for sub-plots errors, so the main plot factor is estimated with less
precision as compared to the subplot and interaction effects

 Note: Precision is an important consideration in deciding which factor to

assign to the main plot

01/25/24 Dr. Yousaf Hayat 5

Randomization and Layout

In case of RCBD, randomization of factors can be performed in two steps:

 The levels of the whole-plot factor are randomly assigned to the main

plots within a block.

 The levels of the subplots are randomly assigned within each main plot

by using a separate randomization procedure for each main plot.

01/25/24 Dr. Yousaf Hayat 6

Layout of Split-plot Design

Block-I Block-II
Main Plots Sub-plots Main Plot Sub-plots
D1 V1 V3 V4 V2 D3 V4 V1 V2 V3
D3 V3 V1 V4 V2 D2 V1 V3 V4 V2
D2 V4 V1 V2 V3 D1 V2 V4 V1 V2
Block-III
Main Plots Sub-plots Three planting Dates (D1, D2, D3):
D1 V1 V2 V4 V3 Main Plots
D2 V3 V2 V4 V1 Four Varieties (V1, V2, V3, V4):
D3 V4 V1 V2 V3 Sub-plots
01/25/24 Dr. Yousaf Hayat 7
Advantages and Limitations of Split Plot Design
Advantages
 It allows the use of some factors that require different sizes of plot for
their application.
 It permits the introduction of new factors or any level of already
considered factors into the experiment if the experiment is already in
progress
Limitations
 Main plot factor is estimated with less precision as compared to the sub-
plot factor as well as interaction effect.
 Statistical analysis is more complex due to different experimental errors
for computing the F-ratios.
01/25/24 Dr. Yousaf Hayat 8
Statistical Model for Split-plot Design
 Statistical Model for Split-plot Design with factors A and B conducted in
RCBD:
 Yijk =  + i + Aj + Error (I)ij + Bk + (AB)jk + Error (II)ijk
  = mean effect
 i = effect of ith block
 Aj = effect of jth level of factor A (Main plots)
 Error (I)ij = Main plot error
 Bk = effect of kth level of factor B (Sub plots)
 (AB)jk = Interaction between factor A and B
 Error (II)ijk = random error of the sub-plots
01/25/24 Dr. Yousaf Hayat 9
Results Interpretations
 If the AB interaction is significant:
 the main effects may have no meaning whether or not they test
significant
 summarize in a two-way table of means for the various AB
combinations and interpret them in terms of their statistical
significance by applying LSD test (or may be any other multiple
comparison test)
 If the AB interaction is not significant:
 test the independent factors for significance
 summarize in a one-way table of means for the significant main
effects
01/25/24 Dr. Yousaf Hayat 10
ANOVA For Split-Plot Design: RCBD

Main plot factor = A having “a” levels

Sub-plot factor = B having “b” levels
Number of blocks = r
SOV df SS MS F-ratio
Block (r-1) SSR MSR Fr = MSR/MSE(I)
Factor A (a-1) SSA MSA Fa = MSA/MSE(I)
Error (I) (r-1)(a-1) SSE(I) MSE(I)
Factor B (b-1) SSB MSB Fb = MSB/MSE(II)
AxB (a-1)(b-1) SSAB MSAB Fab = MSAB/MSE(II)
Error(II) a(r-1)(b-1) SSE(II) MSE(II)
Total rab-1 SSTotal
01/25/24 Dr. Yousaf Hayat 11
Computation of Sum of Squares
 Let us consider that the factor A and B has “a” and “b” levels
respectively, and the experiment is conducted in RCBD with “r” blocks.
Then various sum of squares can be computed as:
CF = (GT)2/rab
SSTotal = 2ijk – CF
SSR = SSBlock = 1/ab Y2i. – CF
SSA = 1/rb Y2j.. – CF
SSE (I) = 1/b Y2ij. – CF – SSR – SSA
SSB = 1/ra Y2..k – CF
SSAB = 1/r Y2.jk – CF – SSA – SSB
SSE(II) = SSTotal – SSR – SSA – SSE(I) – SSB –SSAB
01/25/24 Dr. Yousaf Hayat 12
Numerical Example
 An agronomist was interested to determine the effect of planting dates on
the yield of different varieties of wheat. In such case, planting dates are
assigned to the main plots whereas the varieties are assigned to the sub-
plots. Let the two factors are given as:

 Two factors:

 Planting date (Oct 15, Nov 1, Nov 15) = (D1, D2, D3)

 Variety (V1, V2, V3, V4)

01/25/24 Dr. Yousaf Hayat 13

Data in Usual Layout of Split Plot Design
Dates Variety R1 R2 R3
V1 25 31 28
D1 V2 19 14 16 Grand Total = 674
V3 22 20 17 Grand Mean = 18.72
V4 11 14 14
V1 30 32 28
D2 V2 24 20 24
V3 19 18 16
V4 15 13 19
V1 17 20 19
D3 V2 20 16 20
V3 12 17 15
V4 8 13 8
01/25/24 Dr. Yousaf Hayat 14
Analysis of Variance Table

Source df SS MS F-ratio P-value

Block 2 1.56 0.78
Dates (D) 2 227.06 113.53 32.180 0.003
Error (I) 4 14.11 3.53
Variety (V) 3 757.89 252.63 37.790 0.000
DxV 6 146.28 24.38 3.650 0.015
Error (II) 18 120.33 6.69
Total 35 1267.22

CV (Main plots) = 10.03; CV (Sub-plots) = 13.81; Grand Mean = 18.72

01/25/24 Dr. Yousaf Hayat 15
Table of Means (D x V)

Varieties
Mean
Planting Date V1 V2 V3 V4

D1 28.00 a 16.33 cde 19.67 bc 13.00 ef 19.25

D2 30.00 a 22.67 b 17.67 cd 15.67 cde 21.50

D3 18.67 bcd 18.67 bcd 14.67 de 9.67 f 15.42

Mean 25.56 19.22 17.33 12.78 18.72

01/25/24 Dr. Yousaf Hayat 16
Report of Analysis

 F-test for interaction between planting dates and varieties is significant

(P < 0.05), suggesting that significant differences among the yield of

varieties depends on the planting dates. Similar significant results hold

for planting dates and varieties.

 Variety 1 produced maximum yield at all the planting dates.

 On the average, V4 produced minimum yield at all the planting dates.

01/25/24 Dr. Yousaf Hayat 17

 Blcok Date Variety Yield
2 3 4 13
1 1 1 25

1 1 2 19 3 1 1 28

1 1 3 22 3 1 2 16

1 1 4 11
3 1 3 17
1 2 1 30

1 2 2 24 3 1 4 14

1 2 3 19
3 2 1 28
1 2 4 15

1 3 1 17 3 2 2 24

1 3 2 20
3 2 3 16
1 3 3 12
3 2 4 19
1 3 4 8

2 1 1 31 3 3 1 19
2 1 2 14
3 3 2 20
2 1 3 20

2 1 4 14 3 3 3 15
2 2 1 32
3 3 4 8
2 2 2 20

2 2 3 18

2 2 4 13

2 3 1 20

2 3 2 16

2 3 3 17
01/25/24 Dr. Yousaf Hayat 18
Split-Split-Plot Design

Split-split-plot design contain three factors where:

 The levels of one factor are random assigned to Main plots

 The levels of 2nd factor are assigned to the sub-plots within the main plots

 The levels of 3rd factor are assigned to sub-sub-plots within the sub-plots

Note: Randomization and Analysis procedure is same like Split-plot design

with additional information of 3rd factor in the sub-sub-plots.

01/25/24 Dr. Yousaf Hayat 19

Statistical Model for Split-Split-plot Design with factors A and B conducted in
RCBD:

 Statistical Model for Split-Split-plot Design with factors A and B conducted in

RCBD:
 Yijkl =  + i + Aj + Error (I)ij + Bk + (AB)jk + Error (II)ijk + Cl+ (AC)jl + (BC)kl +
(ABC)jkl + Error(III)ijklk
  = mean effect
 i = effect of ith block
 Aj = effect of jth level of factor A (Main plots)
 Error (I)ij = Main plot error
 Bk = effect of kth level of factor B (Sub plots)
 (AB)jk = Interaction between factor A and B
01/25/24  Error (II)ijk = random error of
Dr. the sub-plots;
Yousaf Hayat and similarly other terms has 20
 variety P-levels Zn-levels RI RII RIII

1 1 1 5 6 6

1 1 2 6 6 7

1 1 3 5 5 7

1 1 4 6 5 7

1 2 1 5 6 7

1 2 2 6 6 7

1 2 3 5 5 6

1 2 4 5 5 6

1 3 1 6 5 8

1 3 2 5 7 8

1 3 3 6 5 7

1 3 4 5 6 7

1 4 1 6 5 5

1 4 2 5 6 6

1 4 3 6 5 5

1 4 4 7 6 6

2 1 1 6 4 5

2 1 2 6 8 8

2 1 3 7 6 6

2 1 4 5 5 8

2 2 1 6 6 7

2 2 2 7 7 6

2 2 3 7 8 8

2 2 4 7 9 7

2 3 1 5 8 7

2 3 2 7 7 6

2 3 3 5 6 5

2 3 4 6 6 6

2 4 1 5 6 5

2 4 2 6 7 6

2 4 3 5 8 5

2 4 4 6 7 6
01/25/24 Dr. Yousaf Hayat 21
 DISCUSSION

01/25/24 Dr. Yousaf Hayat 22