EDtest 22

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

UNIVERSITY OF CAPE TOWN

DEPARTMENT OF STATISTICAL SCIENCES


STA2005S : CLASS TEST 2 : EXPERIMENTAL DESIGN
30 September 2022

TIME: 90 mins AVAILABLE MARKS: 51


PAGES: 4 MAXIMUM MARKS: 50

1. Experimental and comparative studies suggest that the striped coats of zebras can prevent
biting fly attacks. Biting flies are serious pests of livestock that cause economic losses in
animal production.
A group of researchers is interested in investigating whether painting stripes on cows could
be a better way to protect cows than pesticides.
They are planning to use three treatments: painting white stripes (see picture), painting
black stripes and a control with no stripes (all cows are black).
They can only use 3 cows in October 2022, and 3 different cows in October 2023 (October
being the month when the flies are most active).
Once a treatment is applied, the cow is observed and measured several times over the next
three days by taking 3 pictures per day and counting the number of flies.

(a) Design an experiment. State clearly what the experimental unit is, any blocking
and treatment factors, how the response will be calculated, how many replicates per
treatments will be available make a sketch of your design describe clearly how you
will randomize. Carefully explain how you would design an appropriate experiment,
by considering the following points, not necessarily in the same order.
i.Name all treatment and blocking factors, and their levels.
ii.What design are you using?
iii.What is the experimental unit?
iv. How will the response be measured?
v. Are there any other factors you need to keep constant? Briefly explain how you
will do this.
vi. How should randomization be done in your experiment?

1
vii. Make a sketch of your design’s layout. (11)
(b) Why is the control treatment important in this experiment? (1)
[12]

2. Briefly describe how one should randomize for a 3 × 4 factorial experiment conducted
as a randomized block design. You can assume that there are 2 blocks, each with 24
experimental units. [3]

3. Give two approaches for avoiding confounding between a treatment factor and intrinsic
differences between experimental units. Exactly how do these approaches avoid confound-
ing? [3]

4. Name two features of designed experiments that allow one to make causal inference.
Explain HOW each of these features contributes to this advantage (causal inference).
[3]

5. This question is about an experiment investigating the effect of consuming chocolate on


cardiovascular (heart) health. The experiment consisted of using three different types of
chocolates: 20 g of dark chocolate, 20 g of dark chocolate with 50 ml of full-cream milk,
and 20 g of milk chocolate. A total of 36 subjects were used, 21 women and 15 men, with
an average age range of 32.2 ± 1 years, an average weight of 65.8 ± 3.1 kg, and body-mass
index of 21.9 ± 0.4 kg m−2 . Each treatment was randomly assigned to 12 experimental
units.
A subject consumed one of the chocolate-factor levels and 1 hour later the total antioxidant
levels in their blood was measured (antioxidants are usually good).

120
antioxidant in blood

115

110

105

100

95

DC DC.MK MC

chocolate treatment

2
(a) P
Construct an ANOVA table P based
P on the following information:
2 2
P
i j (Ȳi. − Ȳ.. ) = 1952.6, i j (Yij − Ȳ.. ) = 2296.9

Some quantiles of the relevant F-distribution:

Fν(0.95)
1 ,ν2
= 3.285, Fν(0.99)
1 ,ν2
= 5.312, Fν(0.999)
1 ,ν2
= 8.579, Fν(0.9999)
1 ,ν2
= 12.334

(3)
(b) Write down the model underlying the above ANOVA table. Define all terms and
include all necessary constraints. (3)
(c) Write down the line of the design matrix corresponding to an observation from treat-
ment 3 (MC) without incorporating the constraint. (1)
(d) State the null and alternative hypotheses that are tested with this ANOVA table and
give a conclusion based on the above ANOVA table. (2)
(e) The means for the three treatments were:

chocolate
DC DC.MK MC
116.06 100.70 100.18

Construct a 95% confidence interval for the difference between dark chocolate and
milk chocolate. Interpret.
The relevant quantile of the t-distribution is 2.035, but you should still state its
distribution, with degrees of freedom. (4)
(f) Predict how the above results would change if we had only used 2 replicates per treat-
ment. Specifically, predict how MSE, treatment sum of squares, error and treatment
degrees of freedom, the F-statistic, p-value and conclusion would change. (5)
(g) Below is the output from a randomization test (using 10000 randomly generated
randomizations, and each time calculating the statistic M SM
treatment
SE ). Obtain a p-
value from this randomization distribution for the test conducted here. Compare
this result to that obtained from the ANOVA table. Is this what you would have
expected? Give a reason for your answer. (3)

0.8

0.6
Density

0.4

0.2

0.0

0 2 4 6 8 10

F−values from randomizations

3
(h) Estimate the standard error for the mean response with the dark chocolate treatment.
What does this tell you about the statistic in relation to the true parameter value
and its expected value? (2)
(i) What is the distribution of an observation from the dark chocolate distribution? (1)
(j) Construct a 95% prediction interval for a future observation from the dark chocolate
treatment. (2)
(k) Write down the likelihood function for the model in (b). (1)
(l) Use the following information / R output to construct a likelihood ratio test to test
for differences in blood antioxidant levels between the chocolate treatments. Give an
expression for the distribution of the likelihood ratio statistic and explain how you
would calculate a p-value from this. (3)

> a1 <- aov(antiox ~ chocolate)


> a0 <- aov(antiox ~ 1)

> logLik(a1)
’log Lik.’ -91.72599 (df=?)
> logLik(a0)
’log Lik.’ -125.8865 (df=?)
[30]

You might also like