7.

Split plot

A split plot design is a special case of a factorial treatment structure. It is used when some
factors are harder (or more expensive) to vary than others. Basically, a split plot design consists
of two experiments with different experimental units of different “size”, e.g., in agronomic
field trials certain factors require “large” experimental units, whereas other factors can be easily
applied to “smaller” plots of land.

Advantages and Disadvantages of split plot design

Advantages of split plot design
(i) It permits to introduce new treatments into an experiment which is already in progress.
(ii) Treatment factors that need large experimental units can be used efficiently with other
factors which require relatively small experimental units.
(iii) Increased precision over randomized block design is attained on the subplot treatments
and the interaction between subplot and main plot treatments.
(iv) The overall precision of the split plot design relative to the randomized complete block
design may be increased by designing the main plot treatments in a latin square design
or in an incomplete latin square design.
Disadvantages of split plot design
(i) The increase in complexity of the analysis for the split plot design is greater than the
randomized block design if missing data occur in the experiment.
(ii) The precision for the treatments in main plot are less than they are in randomized block
design.
(iii) Different treatment comparisons have different basic error variances which make the
analysis more complex than with the randomized complete block design, especially if
some unusual type of comparison is being made.
Appropriate use of split-plot designs:
1. When the practical limit for plot size is much larger for one factor compared with the
other, e.g., in an experiment to compare irrigation treatments and population densities;
irrigation treatments require large plots and should, therefore, be assigned to the main
plots while population density should be assigned to the subplots.
2. When greater precision is desired in one factor relative to the other e.g., if several
varieties are being compared at different fertilizer levels and the factor of primary
interest is the varieties, then it should be assigned to the subplots and fertilizer levels
assigned to the main plots.

Layout of split plot design

There are two separate randomization processes in a split plot design-one for the main plots
and another for the subplots. In each replication, main plot treatments are first randomly
assigned to the main plots followed by a random assignment of the subplot treatments within
each main plot.
For illustration, a two-factor experiment involving four levels of irrigation (main plot
treatments) and three nitrogen levels (subplot treatments) in three replications is used. Here,
irrigation levels were chosen for the main plots mainly for the convenience in its application,
easiness in controlling the leaching effect and to detect the presence of possible interaction
between irrigation and the fertilizer. The steps in the randomization and layout of a split plot
design are shown, using a as the number of main plot treatments, b as the number of subplot
treatments, and r as the number of replications.
Step 1. Divide the experimental area into r = 3 blocks, each of which is further divided into a
= 4 main plots, as shown in following Figure 1

Main Plots Main Plots Main Plots

1 2 3 4 1 2 3 4 1 2 3 4

Replication I Replication II Replication III

Figure 1. Division of the experimental area into three blocks
(replications) each consisting of four main plots, as the
first step in laying out of a split plot experiment
involving three replications and four main plot
treatments.

Step 2. Following the RCBD randomization procedure with a = 4 treatments and r = 3

replications randomly assign the 4 irrigation treatments to the 4 main plots in each of
the 3 blocks. The result may be as shown in Figure 2.

I3 I1 I0 I2 I1 I0 I2 I3 I2 I0 I3 I1

Replication I Replication II Replication III

Figure 2. Random assignment of four irrigation levels (I0, I1, I2

and I3) to the four main plots in each of the three
replications of Figure 1.

Step 3. Divide each of the ra = 12 main plots into b = 3 subplots and following the RCBD
randomization procedure for b = 3 treatments and ra = 12 replications, randomly assign
the 3 clones to the 3 subplots in each of the 12 main plots. The result may be as shown
in Figure 3.

n3 n1 n0 n2 n1 n0 n2 n3 n2 n0 n3 n1
n2 n1 n2 n1 n1 n3 n3 n1 n3 n2 n1 n2
n1 n3 n3 n2 n3 n1 n2 n2 n2 n1 n2 n3
n3 n2 n1 n3 n2 n2 n1 n3 n1 n3 n3 n1
Replication I Replication II Replication III
Figure 3. A sample layout of a split plot design involving three
nitrogen (n1, n2 and n3) as subplot treatments and four
irrigation levels (I0, I1, I2 and I3) as main plot treatments,
in three replications.

Note that the field layout of a split plot design as illustrated by Figure 3 has the following
important features: (i) The size of the main plot is b times the size of the subplot. In our example
with 3 nitrogen (b = 3) the size of the main plot is 3 times the subplot size (ii) Each main plot
treatment is tested r times whereas each subplot treatment is tested ar times. Thus, the number
of times a subplot treatment is tested will always be larger than that for the main plot and is the
primary reason for more precision for the subplot treatments relative to the main plot
treatments. In our example, each of the 4 levels of irrigation is tested 3 times but each of the 3
nitrogen is tested 12 times.

Statistical Model and ANOVA structure

The model for split plot design is
𝑌𝑖𝑗𝑘 = 𝜇 + 𝑅𝑘 + 𝛼𝑖 + 𝜀𝑖𝑘 + 𝛽𝑗 + (𝛼𝛽)𝑖𝑗 + 𝜀𝑖𝑗𝑘

for 𝑘 = 1, 2, . . . , 𝑟; 𝑖 = 1, 2, . . . , 𝑎 ; 𝑗 = 1, 2, . . . , 𝑏
Here 𝑌𝑖𝑗𝑘 is the ith observation for the jth treatment at kth replication, 𝜇 is the grand
mean, the ith treatment effects 𝛼𝑖 are subject to the restriction
Let there be treatments planned in RBD. A main plot treatments replicated r times each.

Table 7.1 Structure of ANOVA for split plot design

Source of Variation Degree of Sum of Mean Sum of F
Freedom Square Square
Whole plot Analysis
Replications 𝑟−1 SSR RMS
Main plot treatment (A) 𝑎−1 SSA MSA
2
Error (a) (𝑟 − 1) (𝑎 − 1) SSE(a) MSE(a) =𝑆𝑒(𝑎)
Sub plot Analysis
Sub plot treatment (B) 𝑏−1 SSB MSB
Interaction (A×B) (𝑎 − 1)(𝑏 − 1) SSAB MSAB
Error (b) 𝑎(𝑟 − 1)(𝑏 − 1) SSE(b) 2
MSE(b) =𝑆𝑒(𝑏)
Total 𝑟𝑎𝑏 − 1 SSTO

Since the main plot (Factor A) is randomized in the whole experimental area and each subplot
is randomized in each main plot, the error used to test the main plot should be different from
the error to test the subplot (and interaction). Therefore, there should be two (2) error terms in
split– plot.

Analytical procedure
The analysis of variance of a split plot design is divided into the main plot analysis and the
subplot analysis. The computations are shown with the data from a two-factor experiment in
eucalyptus involving two silvicultural treatments (pit size) and four fertilizer treatments. The
data on height of plants after one year of planting are shown in Table 7.2.

Table 7.2 Data on height (cm) of Eucalyptus tereticornis plants from a

field trial under split plot design
Height (cm)
Fertilizer Average
Replication I Replication II Replication III
P0 - Pit size (30 cm x 30 cm x 30 cm)
f0 25.38 61.35 37.00 41.25
f1 46.56 66.73 28.00 47.10
f2 66.22 35.70 35.70 45.87
 f3 30.68 58.96 21.58 37.07
P1 - Pit size (40 cm x 40 cm x 40 cm)
f0 19.26 55.80 57.60 44.22
f1 19.96 33.96 31.70 28.54
f2 22.22 58.40 51.98 44.20
f3 16.82 45.60 26.55 29.66

Let A denote the main-plot factor (pit size) and B, the subplot factor (fertilizer treatments).
Carry out the analysis of variance as follows:
Whole plot Analysis
Step 1. Construct two tables of totals as:
(i) The replication x factor A two-way table of totals, with the replication totals, Factor
A totals and grand total : For our example, the replication x pit size table of totals
((RA)ki), with the replication totals (Rk), pit size totals (Ai) and the grand total (G)
computed, is shown in Table 7.3

Table 7.3 The replication x pit size table of height totals computed from
data in Table 7.2
Pit size Rep I Rep II Rep III Total (Ai)
P0 168.84 222.74 122.28 53.86
P1 78.26 193.76 167.83 439.85
Rep. Total (Rk) 247.10 416.50 290.10
Grand Total (G) 953.70

(ii) The factor A x factor B two-way table of totals, with factor B totals : For our
example, the pit size x fertilizer treatment table of totals (AB), with the fertilizer
treatment totals (Bj) computed, is shown in Table 7.4.

Table 7.4 The pit size x fertilizer treatment table of height totals
computed from data in Table 7.2
Pit size Fertilizer treatment
f0 f1 f2 f3
P0 123.73 141.29 137.62 111.22
P1 132.66 85.62 132.60 88.97
Total (Bj) 256.39 226.91 270.22 200.19

Step 2. Compute the correction factor and sums of squares for the main plot analysis as follows.
Let 𝑦𝑖𝑗𝑘 refer to the response observed ith main plot, jth subplot in the rth replication.
𝐺2 953.702
Correction factor (CF) = 𝑟𝑎𝑏 = (3)(2)(4) = 37897.92

𝑆𝑆𝑇𝑂 = ∑𝑎𝑖=1 ∑𝑏𝑗=1 ∑𝑟𝑘=1 𝑦𝑖𝑗𝑘

2
− 𝐶𝐹
= [25.382 + 46.562 + ⋯ + 26.552 ] − 37897.92
= 6133.10
∑𝑟𝑘=1 𝑅𝑘2 247.102 +416.502 +290.102
𝑆𝑆𝑅 = − 𝐶𝐹 = − 37897.92 = 1938.51
𝑎𝑏 (2)(4)
 ∑𝑎 2
𝑖=1 𝐴𝑖 513.862 +439.852
𝑆𝑆𝐴 = − 𝐶𝐹 = − 37897.92 = 228.25
𝑟𝑏 (3)(4)

∑𝑎 𝑟
𝑖=1 ∑𝑘=1((𝑅𝐴)𝑖𝑘 )
2
𝑆𝑆𝐸𝑎 = − 𝐶𝐹 − 𝑆𝑆𝑅 − 𝑆𝑆𝐴
𝑏
168.842 +⋯+167.832
= − 37897.92 − 1938.51 − 228.25
4

= 1161.70
Sub plot Analysis
Step 3. Compute the sums of squares for the subplot analysis as
∑𝑏 2
𝑗=1 𝐵𝑗 256.392 +226.912 +270.222 +200.192
𝑆𝑆𝐵 = − 𝐶𝐹 = − 37897.92 = 488.03
𝑟𝑎 (3)(2)
2
∑𝑎 𝑏
𝑖=1 ∑𝑗=1((𝑀𝑆)𝑖𝑗 )
𝑆𝑆𝐴𝐵 = − 𝐶𝐹 − 𝑆𝑆𝐴 − 𝑆𝑆𝐵
𝑟
123.732 + … +88.972
= − 37897.92 − 488.03 − 1161.70 = 388.31
(3)

𝑆𝑆𝐸𝑏 = 𝑆𝑆𝑇𝑂 − 𝑆𝑆𝑅 − 𝑆𝑆𝐴 − 𝑆𝑆𝐸𝑎 – 𝑆𝑆𝐵 − 𝑆𝑆𝐴𝐵

= 6133.10 − 1938.51 − 228.25 − 1161.70 − 488.03 − 388.31 = 1928.30
Step 5. For each source of variation, compute the mean square by dividing the SS by its
corresponding df. The F value for each effect that needs to be tested is to be computed
by dividing each mean square by the corresponding error term as shown in Table 7.1.

Table 7.5 Structure of ANOVA for split plot design

Source of Variation Degree of Sum of Mean Sum of F Test
Freedom Square Square
Whole plot Analysis
Replications 2 1938.51 969.25
Main plot treatment (A) 1 228.25 228.25 0.39 NS
Error (a) 2 1161.70 580.85
Sub plot Analysis
Sub plot treatment (B) 3 488.03 162.68 1.01 NS
Interaction (A×B) 3 388.31 129.44 0.81 NS
Error (b) 12 1928.30 160.69
Total 23 6133.10

Step 6 Compare each computed F value with the tabular F value at the desired level of
significance.
Step 7. Compute the two coefficients of variation, one corresponding to the main plot analysis
and another corresponding to the subplot analysis.
2
√𝑆𝑒(𝑎)
√580.85
𝐶𝑉(𝑎) = 𝐺𝑟𝑎𝑛𝑑 𝑀𝑒𝑎𝑛 × 100 = × 100 = 60.65%
39.7375

2
√𝑆𝑒(𝑏)
√160.69
𝐶𝑉(𝑏) = 𝐺𝑟𝑎𝑛𝑑 𝑀𝑒𝑎𝑛 × 100 = × 100 = 31.90%
39.7375

The value of cv(a) indicates the degree of precision attached to the main plot factor. The value
of cv(b) indicates the precision of the subplot factor and its interaction with the main-plot factor.
The value of cv(b) is expected to be smaller than that of cv(a) because, as indicated earlier, the
factor assigned to the main plot is expected to be measured with less precision than that
assigned to the subplot. In our example, the value of cv(b) is smaller than that of cv(a) but both
of them were high enough to mask any possible treatment differences turning the all the factor
effects in the ANOVA non-significant.

Comparison of treatments
In a split plot design, there are four different types of pair comparisons. Each requires its own
set of LSD values. These comparisons are:
Type-(1). Comparisons between two main plot treatment means averaged over all subplot
2
2𝑆𝑒(𝑎)
treatments. 𝑆𝐸𝑑(𝐴̅𝑖 −𝐴̅𝑗) = √ 𝑟𝑏

Type-(2). Comparison between two subplot treatment means averaged over all main plot
2
2𝑆𝑒(𝑏)
treatments. 𝑆𝐸𝑑(𝐵̅𝑖 −𝐵̅𝑗) = √ 𝑟𝑎

Type-(3). Comparison between two subplot treatment means at the same main plot
2
2𝑆𝑒(𝑏)
treatment. 𝑆𝐸𝑑(𝐴𝑖 𝐵𝑗−𝐴𝑖 𝐵𝑘) = √ 𝑟

Type-(4). Comparison between main plot means at the (i) same level of sub plot measured
as (𝐴𝑖 𝐵𝑗 − 𝐴𝑘 𝐵𝑗 ) or (ii) different level of sub plot (any two treatment means)
2 +𝑆 2 ]
2[(𝑏−1)𝑆𝑒(𝑏)
measured as (𝐴𝑖 𝐵𝑗 − 𝐴𝑘 𝐵𝑙 ). 𝑆𝐸𝑑 = √ 𝑒(𝑎)
𝑟𝑏
2 2
Note : 𝑆𝑒(𝑎) = MSEa, 𝑆𝑒(𝑏) = MSEb, r = number of replications, a = number of main plot
treatments, and b = number of subplot treatments.
These standard errors are used to compute critical differences just as in factorial
experiments. To observe table value for t α at α level of significance for main plot or sub plot
the degrees of freedom for standard errors involving only Error(a) is (r-1)(a-1), while those
involving only Error(b) is [a(r-1)(b-1)]. Table value for the computation of critical difference
for 4th type of comparison where the standard error which contains both Error(a) and Error(b)
has no exact value for the degree of freedom associated with it. To obtain an approximate value
t following formula has to be applied as
2
(𝑏−1)𝑆𝑒(𝑏) 2
𝑡𝑏 +𝑆𝑒(𝑎) 𝑡𝑎
𝑡= 2 +𝑆 2
(𝑏−1)𝑆𝑒(𝑏) 𝑒(𝑎)

where 𝑡𝑎 and 𝑡𝑏 are tabulated t values, at the chosen level of significance, corresponding
to the degrees of freedom in Error(a) and Error(b), respectively.
Although the analysis of variance (Table 7.5) shows all the three effects (the two main effects
and the interaction effect) as non-significant, for the purpose of illustration, consider the case
where there is a significant interaction between pit size and fertilizer indicating the variation in
fertilizer effect with the changing pit size. In such a case, comparison between the means of pit
size levels pooled over all fertilizer levels or that between fertilizer levels averaged over all
levels of pit size will not be valid. The more appropriate comparisons will be those between
fertilizer means under the same pit size levels or between pit size means at same fertilizer level.
Thus the steps involved in the computation of the LSD for comparing two subplot means at the
same main plot treatment are:
Step 1. Compute the standard error of the difference between means following the formula for
Type-(3) comparison.
2
2𝑆𝑒(𝑏) 2(160.69)
𝑆𝐸𝑑(𝐴𝑖 𝐵𝑗−𝐴𝑖 𝐵𝑘) = √ = √ = 10.35
𝑟 3

Step 2. Following the formula 𝐿𝑆𝐷𝛼 = (𝑡𝛼,𝑒𝑟𝑟𝑜𝑟 𝑑.𝑓. )(𝑆𝐸𝑑) compute the LSD value at 5%
level of significance using the tabular t value with 12 degrees of freedom of Error(b).
𝐿𝑆𝐷0.05 = (2.18)(10.35) = 22.56
Step 3. Construct the pit size x fertilizer two-way table of mean differences in height as shown
in Table 7.6. Compare the observed differences in the mean height among the fertilizer
levels at each pit size with the LSD value computed at Step 2 and identify significant
differences if any.

Table 7.6 Difference between mean heights of eucalyptus plants with

four fertilizer levels at different pit sizes based on the data
in Table 7.2.
f0 f1 f2 f3
Difference in mean height (cm) at P0
f0 0.00 -5.85 -4.63 4.17
f1 0.00 1.22 10.02
f2 0.00 8.80
f3 0.00
Difference in mean height (cm) at P1
f0 0.00 15.68 0.02 14.56
f1 0.00 -15.66 -1.12
f2 0.00 14.54
f3 0.00
In the above table -5.85 = 41.25 - 47.10, -4.63 = 41.25 – 45.87, 1.22 = 47.10 – 45.87 etc.