Basic Designs

Dr Abdul Nasir Khan

School of Mathematics and Statistics
MIT WPU, Pune
 Complete randomized design
The CRD is the simplest design. Suppose there are v treatments to be
compared.
• All experimental units are considered the same and no division or grouping
among them exist.
• In CRD, the v treatments are allocated randomly to the whole set of
experimental units, without making any effort to group the experimental
units in any way for more homogeneity.
• Design is entirely flexible in the sense that any number of treatments or
replications may be used.
• All the variability among the experimental units goes into experimented
error.
• CRD is used when the experimental material is homogeneous.
• CRD is well suited for the small number of treatments and for the
homogeneous experimental material.
Layout of CRD
• Following steps are needed to design a CRD
Latin Square Design
 cont…..

• In Latin square design (LSD), the experimental material is

divided into rows and columns, each having the same number of
experimental units which is equal to the number of treatments.
• The treatments are allocated to the rows and the columns such
that each treatment occurs once and only once in each row and
in each column.
• In LSD, the experimental units are grouped according to two
factors. Hence two effects (like as two block effects) are removed
from the experimental error.
 Example
• Suppose different brands of petrol are to be compared with
respect to the mileage per liter achieved in motor cars.
• Important factors responsible for the variation in mileage are
‐ the difference between individual cars.
‐ the difference in the driving habits of drivers.
• We have three factors – cars, drivers and petrol brands. Suppose
we have,
• ‐ 4 types of cars denoted as 1, 2, 3, 4.
• ‐ 4 drivers that are represented by a, b, c, d.
• ‐ 4 brands of petrol are indicated as A, B, C, D.
 Example (Medical Research)
• Research Objective: Imagine you are conducting a clinical trial to
investigate the effects of three different medications (Medication A,
Medication B, and Medication C) on blood pressure in hypertensive
patients. However, you want to control for potential confounding factors,
such as patient age, that might influence the results.
 Randomized Complete Block Design
• If a large number of treatments are to be compared, then a large
number of experimental units are required.
• This will increase the variation among the responses and CRD may not
be appropriate to use.
• In such a case when the experimental material is not homogeneous
and there are v treatments to be compared, then it may be possible
to group the experimental material into blocks of sizes v units.
• Blocks are constructed such that the experimental units within a block are
relatively homogeneous and resemble to each other more closely than the
units in the different blocks.
• If there are b such blocks, we say that the blocks are at b levels.
• Similarly, if there are v treatments, we say that the treatments are
at v levels.
• The responses from the b levels of blocks and v levels of
treatments can be arranged in a two‐way layout. The observed
data set is arranged as follows:
 Randomized Block Design: Layout

• A two‐way layout is called a randomized block design (RBD) or a

randomized complete block design (RCB) if, within each block, the
v treatments are randomly assigned to v experimental units such
that each of the v! ways of assigning the treatments to the units
has the same probability of being adopted in the experiment and
the assignment in different blocks are statistically independent.
• The RBD utilizes the principles of design ‐ randomization,
replication and local control ‐ in the following way:
• 1. Randomization:
• Number the v treatments 1,2,…,v.
• Randomly allocate the v treatments to v experimental units in
each block.
• 2. Replication
• Since each treatment is appearing in each block, so every
treatment will appear in all the blocks. So each treatment can
be considered as if replicated the number of times as the
number of blocks.
• Thus in RBD, the number of blocks and the number of
replications are same.
• 3. Local control
• Local control is adopted in RBD in the following way:
‐ First form the homogeneous blocks of the experimental
units.
‐ Then allocate each treatment randomly in each block.
• The error variance now will be smaller because of
homogeneous blocks and some variance will be parted
away from the error variance due to the difference
among the blocks.
 Example
• Suppose there are 7 treatment denoted as 𝑇1 , 𝑇2 , … , 𝑇7
corresponding to 7 levels of a factor to be included in 4 blocks.
• So one possible layout of the assignment of 7 treatments to 4
different blocks in a RBD is as follows:
 Advantages of the RCBD
• 1. Generally more precise than the CRD.
• 2. No restriction on the number of treatments or replicates.
• 3. Some treatments may be replicated more times than
others.
• 4. Missing plots are easily estimated.
• 5. Whole treatments or entire replicates may be deleted from
the analysis.
• 6. If experimental error is heterogeneous, valid comparisons
can still be made
 Disadvantages of the RCBD

• 1. RCBD may be give misleading results if blocks are not

homogeneous.
• 2. RCBD is not suitable for large number of treatments
because in that case the block size will increase and it may not
be possible to keep large blocks homogenous.
• 3. If the data on more than two plots is missing, the statistical
analysis become quite tedious and complicated.
• NOTE: The most important item to consider when choosing a
design is the uniformity of the experimental units.
 Statistical Analysis
• in this case, the linear model is a two‐way model as

• Where 𝜇 is an overall mean

• 𝜏𝑗 is the effect of the jth treatment
• 𝛽𝑖 is the effect due to ith block
• where 𝜀𝑖𝑗 are identically and independently distributed random
errors following a normal distribution with mean 0 and variance 𝜎 2 .
Such that
𝑦𝑖.=σ𝑣𝑗=1 𝑦𝑖𝑗 = 𝐵𝑖
𝑦.𝑗=σ𝑏
𝑖=1 𝑦𝑖𝑗 =𝑇𝑗
• If 𝐻𝑜𝑏 is accepted, then it indicates that the blocking is not
necessary for future experimentation.
• If 𝐻𝑜𝑡 is rejected then it indicates that the treatments are
different.
• Then the multiple comparison tests are used to divide the
entire set of treatments into different subgroup such that the
treatments in the same subgroup have the same treatment
effect and those in the different subgroups have different
treatment effects.
ANOVA
Example
• i). Under the null hypothesis: 𝜏𝑖 − 𝜏𝑗 = 0

Where t follows t-distribution with (r-1)(t-1) d.f.

2
• Find the Cretical difference (C.D.)=𝑡𝛼/2,𝑑𝑓(𝑒𝑟𝑟𝑜𝑟) 2𝑆𝐸 /𝑟
Where t is the table value at 5% level of significance with error df. R
2
is the number a treatment is replicated. 𝑆𝐸 is the MSSE (Given in
ANOVA).
• calculate the difference of each treatment means. If any
pairwise mean difference i.e. (𝑇ത𝑖 -𝑇ത𝑗 ) is greater than critical
difference. This pair is significantly differ.
• ii). Efficiency of RBD relative to CRD
 Example
• Let k be the number of treatments and 𝑦ത1 and
𝑦ത2 be the mean yields of the 2 blocks. Show
that sum of squares between blocks
𝑘
• SSB= (𝑦ത1 − 𝑦ത2 )2 .
2
 Missing Value
• The estimated missing value is obtained by using formula
𝑟𝑦.𝑗 ′ +𝑡𝑦.𝑖 ′ −𝑦.. ′
𝑥=
(𝑟−1)(𝑡−1)

Where

• Reasons for missing data include:

1. Animal dies
2. Break a test tube.
3. Animals eat grain in the plot.
4. Spill grain sample.
 Statistical Analysis
• ANOVA is performed in usual way after substituting the
estimated values of the missing observations. For each
observation 1 d.f. is subtracted from total and consequently from
error d.f.

• The adjusted treatment S.S. is obtained by subtracting the

adjustment factor given as.

′ ′ ′
(𝑦.𝑗 +𝑡𝑦𝑖. −𝑦.. )2
•
𝑡(𝑡−1)(𝑟−1)2
Example