Problems 9
Bayesian methods
Data files are at [Link]/imsw/msc/data.
1. In an Asian country, 75% of the population are Chinese and 25% are Indian in origin.
A crime has been committed to which there were no witnesses. The police are
concentrating their investigations among the Chinese community, on the basis that the
offender is 3 times more likely to be Chinese than Indian.
The offender left a blood stain at the crime scene which is found to be a rare genotype.
The proportion of Indians with this genotype is 4.8%, while among the Chinese section of
the population, the proportion is only 0.31%.
Given these data, should the police continue to focus on the Chinese community?
2. Human males have one X and one Y chromosome, while females have two X chromosomes,
each chromosome being inherited from one parent. The disease hæmophilia exhibits
X-chromosome linked recessive inheritance: a male who inherits the disease-causing gene is
affected, whereas a female carrying the gene on only one chromosome is unaffected. The
disease is generally fatal for women who inherit two such genes.
Consider a woman whose brother is hæmophiliac but whose father is unaffected. Her
mother must be a carrier with one ‘good’ and one ‘bad’ gene, and the prior probability that
the woman is a carrier is 0.5 (this is the probability that she receives the defective gene
from her mother).
Given that the woman has two sons, neither of whom is affected by the disease, what is
the posterior probability that the woman is a carrier?
How does this probability change if the woman has a third son, who is also unaffected?
3. A coin has a probability θ of falling heads. The prior probability distribution for θ is
Be(2, 2). The coin is tossed 20 times and falls heads 10 times, tails 10 times.
(a) What is the posterior distribution of θ?
(b) Find a 95% Bayesian confidence interval for θ, (i) using tables of the F distribution,
(ii) using Cambridge Table 29.
(c) Plot the posterior distribution of θ over the range of the confidence interval. (E.g.
using the dbeta function.)
4. If Y is binomial (n,θ), prior for θ is Be(a, b), and neither Y + a nor n − Y + b is small,
posterior distribution of log [θ/(1 − θ)] is approximately normal with
mean = log [(Y + a)/(n − Y + b)] , variance = (Y + a)−1 + (n − Y + b)−1
Use this result to re-do question 3, part (b).
5. Theory predicts that experimental fruit flies are type A, B or C with probabilities
pA = (2 + θ)/4, pB = (1 − θ)/2, pC = θ/4
Among 197 flies, 125 are type A, 38 are type B, and 34 are type C.
(a) Find the ML estimate of θ and its standard error.
(b) Generate a sample of 1000 from the posterior distribution of θ, given a uniform prior
distribution. Display the posterior in the form of a histogram. Are the results
consistent with the ML estimate?
Note that in this case θ is not restricted to be less than 0.25.
6. The gene giving black coat colour (A) is dominant to that giving red (a). Carriers (with
genotype Aa) are black, but carry one copy of the recessive allele.
A black bull calf is the offspring of two carriers. He is mated to a known carrier and the
offspring is black. We require the probability that the bull is a carrier.
Calculate the full conditional distributions required to solve this problem by Gibbs sampling
(the probability distribution of the offspring’s genotype, conditional on the bull’s genotype,
and vice-versa). Find the required probability, (a) directly from the conditional
distributions, (b) by simulation.