MATH3014-6027 Design (and Analysis) of Experiments
SEMESTER 2 EXAMINATION 2022/23
1. [25 marks]
An experiment was carried out to compare the effect of the type of catalyst used in the reactions on concentration of one component of a liquid mixture. Four catalysts were considered, and the results are given in Table 1.
Table 1: Catalyst experiment: percentage concentration from four different catalysts.
Catalyst 1
|
Catalyst 2
|
Catalyst 3
|
Catalyst 4
|
58.2
|
56.3
|
50.1
|
52.9
|
57.2
|
54.5
|
54.2
|
49.9
|
58.4
|
57.0
|
55.4
|
50.0
|
55.8
|
55.3
|
|
51.7
|
54.9
|
|
|
|
(a) [5 marks] Write down a suitable unit-treatment model for the data yij ,
i = 1, . . . , 4; j = 1, . . . ni and state any assumptions your model makes.
(b) [10 marks] Complete the ANOVA Table 4, comparing a unit-treatment model to the null model. Using the table, test the hypothesis of no effect of catalyst at the 5% level.
Table 2: Catalyst experiment: ANOVA table
Source
|
Degrees of freedom
|
Sum of squares
|
Mean square
|
F-ratio
|
Treatment
|
|
|
|
|
Residual
|
|
34.562
|
|
-
|
Total
|
|
|
-
|
-
|
(c) [10 marks] Test all pairwise differences at an exact experiment-wise level of 5%.
You may find the following quantities from R useful.
qf(0.95 , 3 , 12)
## [1] 3.49
qt(1 - 0.025 / 6 , 12) ## [1] 3 . 15
qtukey(0.95 , 4 , 12)
## [1] 4 . 2
2. [25 marks]
An experiment was carried out to compare the effect of the type of catalyst used in the reactions on concentration of one component of a liquid mixture. Four catalysts were considered, and the results are given in Table 3.
Table 3: Catalyst experiment: percentage concentration from four different catalysts.
Catalyst 1
|
Catalyst 2
|
Catalyst 3
|
Catalyst 4
|
58.2
|
56.3
|
50.1
|
52.9
|
57.2
|
54.5
|
54.2
|
49.9
|
58.4
|
57.0
|
55.4
|
50.0
|
55.8
|
55.3
|
|
51.7
|
54.9
|
|
|
|
(a) [10 marks] Complete the ANOVA Table 4, comparing a unit-treatment model to the null model. Using the table, test the hypothesis of no effect of catalyst at the 5% level.
Table 4: Catalyst experiment: ANOVA table
Source
|
Degrees of freedom
|
Sum of squares
|
Mean square
|
F-ratio
|
Treatment
|
|
|
|
|
Residual
|
|
34.562
|
|
-
|
Total
|
|
|
-
|
-
|
(b) [10 marks] Test all pairwise differences at an exact experiment-wise level of 5%.
(c) [5 marks] Test the null hypothesis
H0 : τ1 + τ2 = τ3 + τ4 ,
against the alternative
H0 : τ1 + τ2 τ3 + τ4
at the 5% level, where τi is the effect of the ith catalyst.
You may find the following quantities from R useful.
qf(0.95, 3, 12)
## [1] 3.49
qt(0.975, 12)
## [1] 2.18
qt(1 - 0.025 / 6, 12)
## [1] 3.15
qtukey(0.95, 4, 12)
## [1] 4.2
3. [25 marks]
In an experiment to compare t treatments, it is antcipated that the variance of the response may change with the treatment. To account for this, the model
yij = µ + τi + εij , εij ~ N(0,σi(2)) , i = 1, . . . , t, j = 1, . . . , ni , (1)
is proposed, with all random errors assumed independent and the experiment consisting of runs in total. Compared to the usual linear model for a completely randomised design, the background variance σi(2) is allowed to vary with treatment.
(a) [10 marks] Under model (1), show that the estimator
is unbiased for treatment contrast cT τ and has variance
(b) [15 marks] Assuming t = 3 and σ 1(2) = σ2(2) = 2σ3(2), find an (approximate) optimal
allocation of n = 30 units to the three treatments that leads to minimum average variance for the following two sets of contrasts.
(i) Pairwise differences:
• cT = (1, -1, 0),
• cT = (1, 0, -1),
• cT = (0, 1, -1).
(ii) Linear and quadratic contrasts:
• cT = (-1, 0, 1),
• cT = (-1, 2, -1).
4. [25 marks]
(a) [5 marks] Show it is not possible to construct a balanced incomplete block
design (BIBD) with t = 8 treatments in b = 12 blocks of size k = 4 with each treatment replicated r = 6 times.
(b) [10 marks] Find the smallest BIBD possible for t = 8 treatments in blocks of size k = 4. What values of b and r does your design have?
(c) [10 marks] Now find a larger design with b = 70 blocks. For this design, what are the values for r and λ, the number of times each pair of treatments occur together? What is the efficiency of the design you found in part (b) compared to this larger design for estimating a pairwise treatment difference? Compare this efficiency to the difference in size between the two designs.
5. [25 marks]
(a) [5 marks] Show it is not possible to construct a balanced incomplete block
design (BIBD) with t = 8 treatments in b = 12 blocks of size k = 4 with each treatment replicated r = 6 times.
(b) [10 marks] Find the smallest BIBD possible for t = 8 treatments in blocks of size k = 4. What values of b and r does your design have?
(c) [10 marks] The complement of a BIBD with t treatments, b blocks of size r and treatments occuring together in λ blocks is a design with b blocks, each containing the t - k treatments not included in the corresponding block of the BIBD. Show this complement design is also a BIBD and state the number of times each treatment is replicated and the value of λ . You may assume that b ≥ 2r in the original design.
6. [25 marks]
(a) [10 marks] A factorial experiment with three factors, A, B and C each at two
levels, is to be carried out. Two batches of experimental units are available, each containing four units. Write down the treatment combinations from the block design that confounds the highest order interaction with blocks, and use this example to explain the meaning of confounding.
(b) [5 marks] Give the partition of the degrees of freedom for the analysis of the design in part (a).
(c) [10 marks] It has been decided to add a fourth factor, also at two levels, into the experiment described in part (a), using the same two batches of experimental units. Design an experiment that allows all four main effects to be estimated independently of two-factor interactions and blocks. Give the defining relation of your design, the full aliasing scheme and indicate which factorial effects are confounded with blocks.
7. [25 marks]
A 26 factorial experiment with factors A, B , C, D, E and F was performed in b = 8 blocks of size k = 8. The interactions BCDE, ACDF and ABC have been chosen to be confounded with blocks.
(a) [5 marks] Write down all the interactions which are confounded with blocks.
(b) [5 marks] Write down all the treatments in one block of the design.
(c) Now suppose only one block from the design can be run.
(i) [5 marks] Write down the defining relation of the resulting fractional factorial design.
(ii) [5 marks] Can this fraction be split into two blocks of size four without
confounding any main effects with blocks? If so, which factorial effects could you confound with blocks?
(iii) [5 marks] Can this fraction be split into four blocks of size two without
confounding any main effects with blocks? If so, which factorial effects could you confound with blocks?
Learning objectives:
LO1 Apply theory and methods to a variety of examples.
LO2 Evaluate designs using common optimality criteria and use them to critically compare competing designs.
LO3 Explore the general theory of factorial and block designs and understand this theory sufficiently to find appropriate designs for specific applications.
LO4 Use the R statistical programming language to design and analyse common forms of experiments.
LO5 Encounter the principles of randomisation, replication and stratification, and under- stand how they apply to practical examples.
LO4 is primarily assessed via coursework.