Description
1. From the class survey, y = 12 out of n = 70 sampled students had pets. R Example 8.1
(ex8.1.R, posted under Lecture Materials) illustrates how to approximate the posterior
mean of the population proportion π of people like us who have pets. It assumes a binomial
model and Jeffreys prior. Using the same binomial model and Jeffreys prior, you will
approximate the posterior variance of π.
(a) [2 pts] Write out all of the mathematical formulas for the integrals you will compute
using R.
(b) [2 pts] Perform the integrations using function integrate() in R. (You may either use
the exact value for the posterior mean of π, or approximate it using integrate().)
(c) [2 pts] Now compute the posterior variance analytically (with the help of conjugacy
and Table A.2 in Cowles), and compare this answer to your approximation.
2. Official combined city/highway energy consumption ratings (MPGe, miles per gallon
gasoline equivalent) are given below for 2019 model all-electric vehicles in two categories:1
Small Cars
Hyundai Ioniq Electric 136
Volkswagen e-Golf 119
Honda Clarity EV 114
BMW i3 113
BMW i3s 113
Nissan Leaf 112
Fiat 500e 112
smart EQ fortwo (coupe) 108
smart EQ fortwo (convertible) 102
Sport Utility Vehicles (SUVs)
Hyundai Kona Electric 120
Tesla Model X 75D 93
Tesla Model X 100D 87
Tesla Model X P100D 85
Jaguar I-Pace 76
Regard MPGe as independent between vehicles and normally-distributed within category,
with both mean and variance possibly differing by category. Use “independent” “standard”
(product-Jeffreys) priors, as illustrated in R Example 8.3 (ex8.3.R, posted under Lecture
Materials). Use at least 100000 simulation samples for all of your approximations.
(a) [1 pt] Compute the sample means and sample standard deviations for the two
categories.
(b) [2 pts] Compute an approximate 95% equal-tailed credible interval for the difference
between the mean for small cars and the mean for SUVs. Do the means appear to differ?
(c) [1 pt] Approximate the posterior probability that the mean for small cars does not
exceed the mean for SUVs.
1Data from https://fueleconomy.gov
1
(d) [3 pts] Compute the (frequentist) Welch two sample t-test one-sided p-value for testing
the null hypothesis that the mean for small cars does not exceed the mean for SUVs.
(Use R function t.test(…, …, alternative=…, var.equal=FALSE), making
sure to select the correct alternative.) Also compute the usual (frequentist) two sample
t-test one-sided p-value that assumes equal variances (t.test(…, …,
alternative=…, var.equal=TRUE)). Compare with the Bayesian probability of the
previous part.
(e) [2 pts] Compute an approximate 95% equal-tailed credible interval for the ratio of the
variance for small cars to the variance for SUVs. Do the variances appear to differ?
