Scribd
Upload
48 views

Module04 RandomizedCompleteBlockDesign

This document discusses randomized complete block design (RCBD) and Latin square design. RCBD is used to study the effect of a treatment factor while controlling for a blocking (nuisance) factor. It involves randomly assigning treatments to experimental units (blocks) that control for the nuisance factor. Latin square design controls for two nuisance factors by arranging treatments in a square such that each treatment appears once in each row and column. The document provides examples and explains how to perform ANOVA to analyze the effects of treatments for these designs.

Uploaded by

asdf
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
48 views

Module04 RandomizedCompleteBlockDesign

This document discusses randomized complete block design (RCBD) and Latin square design. RCBD is used to study the effect of a treatment factor while controlling for a blocking (nuisance) factor. It involves randomly assigning treatments to experimental units (blocks) that control for the nuisance factor. Latin square design controls for two nuisance factors by arranging treatments in a square such that each treatment appears once in each row and column. The document provides examples and explains how to perform ANOVA to analyze the effects of treatments for these designs.

Uploaded by

asdf
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 28

Randomized Complete

Block Design and Latin


Square Design
Prof. Sayak Roychowdhury
Motivation
• Blocking is a technique to understand the extent of
influence of nuisance factors
• A nuisance factor is a factor that may have an effect on
the response variable, not of primary importance to the
experimenter. E.g Variability due to raw material,
operator, shift, etc.

Unknown and
Nuisance factors

Randomize
Uncontrollable

Known and uncontrollable ANCOVA

Known and controllable Blocking


Randomized Complete Block Design
(RCBD)
• RCBD has 3 components
• Block: Experimental runs are conducted in blocks for different
levels of nuisance factors e.g. 2 blocks for Operator 1 and
Operator 2 or 3 blocks for Shift A, Shift B and Shift C
• Complete: Each block should contain all the different treatment
levels. E.g. There are 5 treatment levels of a factor, and 2
operators are involved. Each operator should conduct
experiments at all 5 treatment levels.
• Randomized: Within each block, order of experimental runs
should be randomized, to minimize the effect of other
factors(such as environmental, fatigue) not considered.
Example 1
• Think about an engineering application for which
i. There is a response variable (Y) e.g Yield%, Viscosity, Purity(%)
etc.
ii. There is a factor that you are interested (X), which can be
varied at 3 levels, a1, a2, a3 e.g Concentration of a particular
component, temperature, pressure
iii. A blocking (nuisance) factor (B) with 2 levels: b1, b2, e.g.
experimenter, machines
• The purpose is to study the variability due to the levels of factor X
only, and to isolate the error component present due to the
nuisance factor
• Assume there are 30 experimental runs to be conducted; 10 for
each levels
• There are 2 ways to design this a. completely randomized design ;
b. randomized block design
Completely Randomized Design

Level a1, 10 runs

Randomized
Design (Assign b1 Level a2, 10 runs
and b2 randomly)

Level a3, 10 runs


Randomized Block Design
Level a1, 5
runs

Blocking Level a2, 5


level b1 runs

Level a3, 5
runs
Randomized
Block Design
Level a1, 5
runs

Blocking Level a2, 5


level b2 runs

Level a3, 5
runs
Example 2
• We want to study whether the effect of different tips on
measurement of hardness
• 1 factor : tip type, number of levels: 4
• You may take 4 replications for each levels (total 4 X 4 = 16
runs) and apply ANOVA to determine the effect due to levels
• Nuisance factor : test coupons, as they maybe produced
from different batches of raw materials
• An example of RCBD in this case:
Test Coupon
1 2 3 4
Tip3 Tip3 Tip2 Tip1
Tip1 Tip4 Tip1 Tip4
Tip4 Tip2 Tip3 Tip2
Tip2 Tip1 Tip4 Tip3
Data Collection for RCBD

Each of the a treatment levels is repeated in each of the b


blocks, in random order
Each data point is 𝑦𝑖𝑗 , where 𝑖 = 1, . . , 𝑎; 𝑗 = 1, . . 𝑏
Model for RCBD
• Model for RCBD
𝑦𝑖𝑗 = 𝜇 + 𝜏𝑖 + 𝛽𝑗 + 𝜖𝑖𝑗
∀ 𝑖 = 1 , … , 𝑎; 𝑗 = 1, … , 𝑏
𝜇: Grand mean
𝜏𝑖 : Effect due to 𝑖 𝑡ℎ treatment
𝛽𝑗 : Effect of 𝑗𝑡ℎ block
𝜖𝑖𝑗 : Error ~𝑁(0, 𝜎 2 )
We consider treatment and block levels as deviation from
mean; hence σ𝑎𝑖=1 𝜏𝑖 = 0 ; σ𝑏𝑗=1 𝛽𝑗 = 0
Hypothesis Statement
• The same model may be stated as
𝑦𝑖𝑗 = 𝜇𝑖𝑗 + 𝜖𝑖𝑗 ∀ 𝑖 = 1 , … , 𝑎; 𝑗 = 1, … ,
𝜇𝑖𝑗 = 𝜇 + 𝜏𝑖 + 𝛽𝑗
• We are interested in testing the equality of treatment
means 𝜇𝑖 = 𝜇 + 𝜏𝑖
𝐻0 : 𝜇1 = 𝜇2 = ⋯ = 𝜇𝑎 (null hypothesis)
𝐻1 : at least one 𝜇𝑖 ≠ 𝜇𝑗
• Alternatively we want to test if all the treatment effects
are zero
𝐻0 : 𝜏1 = 𝜏2 = ⋯ = 𝜏𝑎 = 0
𝐻1 : at least one 𝜏𝑖 ≠ 0
ANOVA Table for RCBD

𝑆𝑆𝑇 = 𝑆𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝑆𝑆𝐵𝑙𝑜𝑐𝑘𝑠 + 𝑆𝑆𝐸


ANOVA for RCBD
• Because there are N observations, SST has N - 1 degrees of
freedom.

• There are a treatments and b blocks, so SSTreatments and SSBlocks


have a - 1 and b -1 degrees of freedom, respectively.

• The error sum of squares is just a sum of squares between


cells minus the sum of squares for treatments and blocks.

• There are ab cells with ab - 1 degrees of freedom between


them, so SSE has ab - 1 - (a - 1) - (b - 1) = (a - 1)(b - 1) degrees
of freedom.
Calculations

𝑆𝑆𝑇
𝑆𝑆𝐵𝑙𝑜𝑐𝑘𝑠

𝑆𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠
𝑆𝑆𝐸
Calculations
• For ease of calculations the following simplifications
maybe used:
Results and Conclusion
• If 𝐹0 > 𝐹 1−𝛼 , 𝑎−1 , 𝑎−1 𝑏−1 then conclude that null
hypothesis is rejected and the treatments have significant
effects
𝑀𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠
where 𝐹0 =
𝑀𝑆𝐸
𝐹 1−𝛼 , 𝑎−1 , 𝑎−1 𝑏−1 is found from the F-distribution table,
usually 𝛼 = 0.05
• Effect of blocking factor may also be estimated from the ratio
𝑀𝑆𝐵𝑙𝑜𝑐𝑘𝑠
(but not usually compared with F-distribution, as
𝑀𝑆𝐸
blocking puts a restriction on randomization)
• Usually if the ratio is much greater than 1, it suggests the
blocking factor has a significant contribution in the variability
of the response, and suggests that blocking was a good idea
Explanation
• Considering the error term 𝜖𝑖𝑗 ~𝑁(0, 𝜎 2 ), the expected
values of mean squares can be derived as

𝑀𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠
• Hence if 𝐹0 = ≫ 1 it suggests significance of
𝑀𝑆𝐸
treatment effects. Similar conclusion can be made for the
blocking factor as well
Estimation of Parameters
• The estimates of the parameters are given by:
𝜇ො = 𝑦.
ത.
𝜏ෝ𝑖 = 𝑦ത𝑖∙ − 𝑦.
ത . . 𝑖 = 1, . . , 𝑎
𝛽መ𝑗 = 𝑦ത∙𝑗 − 𝑦.
ത . . 𝑗 = 1, . . , 𝑏
𝑦ො𝑖𝑗 = 𝑦ത𝑖∙ + 𝑦ത∙𝑗 − 𝑦.ത..
Latin Square Design
• There are several other types of designs that utilize the blocking principle.

• For example, suppose that an experimenter is studying the effects of five


different formulations of a rocket propellant used in aircrew escape
systems on the observed burning rate.

• Each formulation is mixed from a batch of raw material that is only large
enough for five formulations to be tested.

• Furthermore, the formulations are prepared by several operators, and


there may be substantial differences in the skills and experience of the
operators.

• Thus, it would seem that there are two nuisance factors to be “averaged
out” in the design: batches of raw material and operators.
Latin Square Design

• Notice the square arrangement, 5 levels each of raw material,


operators and treatments
• Latin Square Design is used to eliminate 2 nuisance factors
• Rows and columns impose 2 restrictions on randomization
Latin Square Design
• Some examples of latin squares are
Latin Square Design
• Following the similar formulation as RCBD, for 𝑝 levels
each of row factor, column factor and the treatment, the
statistical model is given by:

𝑦𝑖𝑗𝑘 denotes 𝑖 𝑡ℎ row, 𝑘 𝑡ℎ column and 𝑗𝑡ℎ treatment


• This is an effects model, the model is completely
additive, and there is no interaction between rows,
columns and treatments
ANOVA of Latin Square Design
• We are interested in determining the significance of
treatment levels
𝐻0 : 𝜏1 = 𝜏2 = ⋯ = 𝜏𝑎 = 0
𝐻1 : at least one 𝜏𝑖 ≠ 0
• The analysis of variance consists of partitioning the total
sum of squares of the N = p2 observations into
components for rows, columns, treatments, and error, for
example,
𝑆𝑆𝑇 = 𝑆𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝑆𝑆𝑅𝑜𝑤 + 𝑆𝑆𝐶𝑜𝑙𝑢𝑚𝑛 + 𝑆𝑆𝐸
ANOVA for Latin Square Design
ANOVA for Latin Square Design with
Replication
• A disadvantage of small latin squares is the lack of error degree of freedom.
• To resolve this, replication is used. Suppose 𝑛 replicates are used for each
combination.
• Case 1: design with same operator and raw material in each replicate
ANOVA for Latin Square Design with
Replication
• Case 2: Use same operator and different raw materials in each replicate or vice
versa
ANOVA for Latin Square Design with
Replication
• Case 3: Use different operator and different raw materials in each replicate
Graeco Latin Square Design
• Graeco-Latin Square Designs are used to control 3 sources
of variability, i.e. to block in 3 directions

The statistical model is given by: (row 𝑖, col 𝑙, Latin letter 𝑗,


Greek letter 𝑘)
ANOVA of Graeco-Latin Square

You might also like