AG 401 Advanced Design and Analysis of Experiments
Unit 5 Split Plot Arrangements
Lecture by: Sanjay Anand
Introduction
• The split plot structure is used when one of the factors
needs more experimental material than the second
factor.
• These treatment structures are frequently used for
factorial experiments.
• In many two-factorial experiments, one of the
treatment factors to be applied may not be easily
replicated.
• In such experiments it is necessary or advisable to use
different types of plots, a feature that which leads to
the split-plot design.
• The underlying principle or feature is that there are
different sized plots dedicated to the various factors.
• In essence, a split-plot design is one in which the plots
of one experiment are used as the blocks of another.
• As a result one factor comprising the main treatments
is applied to the larger sized plots, distinguished as the
main plots.
• The other factor comprising of the sub-treatments is
applied to the sub-plot, which are divisions of the
main plots.
Hence the split-plot is applicable when the effects of two
factors are organised in the following manner:
- Experimental material is divided into several main units, to
which the levels of the first factor are randomly assigned.
- Further, each of the main units is again divided into sub-
units to which the levels of the second factor are randomly
assigned.
- To obtain repetitions, the whole setup is replicated on
several blocks with independent randomisation.
• The split-plot arrangement plan can include combinations of
completely randomised designs, randomised complete block
design, or Latin square designs, which can be applied either on
plots or subplots.
• Each replicate block is divided into main plots (one for each level
of the first treatment factor, A) and each of the main plots is then
further subdivided into an equal number of similar units (subplots)
to accommodate all the levels of the second factor (treatment B),
in some orthogonal manner. The allocation requires a separate
randomisation for each main plot.
As a result of the randomisation process, there are three
strata of error variation:
• Variation between the block within the whole
• Variation between main plots within the blocks and
otherwise within the whole
• Variation between sub-plots (split-plots) within the
main plots.
Interpretation of results
• In general, the interpretation process is very similar to that
of a factorial without the splitting. However, there are some
unique features.
- First, one would test the A-main effect with the Error(a).
- Second, the standard errors for comparing interaction
means are different when we are comparing 2 A’s within a
B than 2 B’s within an A. One has to be very careful about
this.
- The other point is that when there are more than 2-levels of
the factors, the interpretation would depend upon whether
each factor is qualitative or quantitative just as it does
with any factorial experiment.
• It implies that the treatment of much interest in
placed on the smaller experimental unit (the split-
plot) and that of less interest on the larger unit (main
plot).
• The interaction is also measures with a higher
precision.
• The fact that there are two experimental units implies
that there are two experimental errors, hereby referred
to as error (a) and error (b).
• That is, the main plot error (MP residual ) is used for
comparison between main plot effects; however, both the
sub-plot factor effects and the interaction effects between
the factors are tested against the sub-plot or split-plot
error (SP residual).
• In three factor experiments, it is possible to extend the
process through the division of the sub plots into sub-sub-
plots if that seems appropriate. Such an experiment has
an additional stratum.
Uses of Split-Plot Treatment Structure
1. Variable natural plot sizes for different factors in a
factorial.
2. In situations where a second factor is brought into an
experiment to increase its scope.
3. In situations where it is known that larger differences
will occur for some factors than others.
4. Where greater precision is desired for some factors
than others. It enables emphasis to be put where it is
needed.
• Another reason for using split-plots is convenience in the
field. Some cultural treatments (e.g. differential soil
management, sprays and systems of planting) cannot
reasonably be applied to small areas whereas others (e.g.
varieties and fertilisers) can be applied to individual plots.
• When one of the treatment factors needs more
replication or experimental units (material) than
another or when it is hard to change the level of one
of the factors, these design become important.
• The primary disadvantage of these designs is the loss
in precision in the whole plot treatment comparison
and the statistical complexity.
Statistical model defining a split-plot treatment structure in
randomised complete block design - RCBD
Yhij = µ + ρh + αi + еhi + βj + (αβ)ij + ƒhij
Where: µ = is the overall mean
ρh = is the hth block effect
αi = is the ith main plot treatment effect
еhi= is the main plot error variation
βj = is the jth split-plot treatment effect
(αβ)ij = interaction effect of αi and βj
ƒhij = is the split-plot error variation term
Experimental layout and randomisation
Main Plot (Fallow Treatments)
1m
C2 C1 C1 C2 C1
BLOCK I
Split plots
T5 T3 T4 T1 T2
C1 C2 C2 C1 C2 (Cultivars)
C1 C2 C2 C1 C2
BLOCK II T3 T2 T1 T4 T5
C2 C1 C1 C2 C1
C2 C1 C2 C1 C2
BLOCK III
T4 T5 T2 T3 T1
C1 C2 C1 C2 C1
C1 C2 C1 C2 C1
BLOCK IV
T1 T4 T3 T2 T5
C2 C1 C2 C1 C2
ANOVA Skeleton for split plot treatment structure in RCBD
The total variation in a split-plot treatment structure laid in RCBD
experiment is divided into variation among blocks, variation among
main plot treatment factor and variation within main plot treatment
factor (variations among main plots treated alike or experimental
error a); variation among split-plot treatment factors, the interaction
effects between the main plot factor and the split-plot factor and the
variation within the split plot treatment factor (variations among
split-plots treated alike or experimental error b). Generally all
unknown sources of variation between main plots go into the error a
and all unknown sources of variations between sub-plots go into
error b. Known sources of variation here are differences between
blocks and differences between treatments. The outline of the
analysis of variance (ANOVA) table is as follows:
Source of Variation Degrees of Freedom
Block 3
Fallow 4
Error [a] 12
Main plot totals 19
Cultivar 1
Fallow x Cultivar 4
Error [b] 15
Total 39
Experimental layout and randomisation
H1 H2 H4 H8 H6 H7 H5 H3 C1
Main plots
BLOCK I
H3 H5 H7 H4 H2 H1 H8 H6 C2 (Taro Cultivars)
H8 H1 H2 H5 H7 H3 H6 H4 C2
BLOCK II
H6 H7 H8 H4 H5 H1 H3 H2 C1
H4 H2 H1 H8 H3 H6 H7 H5 C2
BLOCK III
H5 H3 H6 H2 H8 H7 H4 H1 C1
H3 H2 H8 H6 H4 H1 H5 H7 C1
BLOCK IV
H1 H7 H2 H5 H3 H6 H4 H8 C2
H7 H6 H3 H8 H4 H2 H1 H5 C2
BLOCK V
H3 H5 H1 H6 H7 H4 H8 H2 C1
Split plots – Eight randomised biomass harvest
Source of Variation Degrees of Freedom
Block 4
Cultivar 1
Error [a] 4
Main plot totals 9
Harvest 7
Cultivar x Harvest 7
Error [b] 56
Total 79
Example Split-plot treatment structure in Randomised
Complete Block Design
Four strains of perennial ryegrass were grown as swards at each of
the two fertiliser levels. The 4 strains were S23, New Zealand, Kent
and X. The fertiliser levels were denoted by H for heavy and A for
average. The experiment was laid out as four blocks of four whole
plots for the varieties, each split to accommodate the two levels of
fertiliser application. The midsummer dry matter yields, in units of
10 lb/acre, were as follows:
Block
Strain Fertiliser
1 2 3 4
H 299 318 284 279
S23
A 247 202 171 183
New H 315 247 289 307
Zealand A 257 175 188 174
H 403 439 355 324
X
A 222 170 192 176
H 382 353 383 310
Kent
A 233 216 200 143
1. Identify the treatment structure and the design structure used
for this research.
2. State the factor(s)/level(s) and the response variable for this
study.
3. Formulate a reasonable statistical model, defining all symbols
used within the context of this study.
4. State an appropriate null hypothesis for this trial and
manually carry out an analysis of variance to test your null
hypothesis, showing all your workings and your completed
ANOVA table.
5. Test the null hypothesis at α = 0.05, that there are no
differences in biomass yield attributable to differences in the 4
strains of ryegrass.
6. Test the null hypothesis at α = 0.05, that there are no
differences in biomass yield attributable to differences in the
two fertiliser levels.
7. Test the null hypothesis at α = 0.05, that there are no
interaction effects between the four strains of ryegrass and the
two fertiliser levels on biomass yields.
8. Depending upon your conclusions in parts d, e and f, above,
calculate and use the appropriate LSD value(s) to compare the
significant means.
9. Based on your conclusions above, outline your
recommendation for farmers to use.
Example of split-plot in animal science research
An experiment was conducted to investigate the effects of four
different treatments of pasture (P1, P2, P3 and P4) and two mineral
supplements (M1 and M2) on milk yield. The total number of cows
needed for the experiment was 24. However, there were eight cows
from each of the three different breeds available for the trial. Hence,
blocking was carried out breed wise. The experiment was designed
as a split-plot, with pasture treatments (Factor A) assigned to the
main plots and mineral supplements (Factor B) assigned to split
plots. The following milk yields were measured:
Breed
Breed 1 Breed 2 Breed 3
Pasture
Mineral Supplements
M1 M2 M1 M2 M1 M2
P1 25 27 31 30 31 24
P2 26 28 32 37 30 31
P3 24 26 33 32 33 32
P4 29 30 34 37 38 36
1. Identify the treatment structure and the design structure used
for this research.
2. State the factor(s)/level(s) and the response variable for this
study.
3. Formulate a reasonable statistical model, defining all symbols
used within the context of this study.
4. State an appropriate null hypothesis for this trial and
manually carry out an analysis of variance to test your null
hypothesis, showing all your workings and your completed
ANOVA table.
5. Test the null hypothesis at α = 0.05, that there are no
differences in milk yield attributable to the 4 pasture species.
6. Test the null hypothesis at α = 0.05, that there are no
differences in milk yield attributable to differences in the two
mineral supplements.
7. Test the null hypothesis at α = 0.05, that there are no
interaction effects between the four pasture species and the two
mineral supplements on milk yields.
8. Depending upon your conclusions in parts d, e and f, above,
calculate and use the appropriate LSD value(s) to compare the
significant means.
9. Based on your conclusions above, outline your
recommendation for dairy farmers to use.
