Description
Scenario: The data file party115cong.csv contains data on all congressional districts1
represented in the U.S. House of Representatives during the 115th U.S. Congress. Each row
represents a district, and the columns are as follows:
state the U.S. state containing the district, or the District of Columbia
district an identifier of Congressional district within state
electedrep name of person elected to the 115th Congress
party the party affiliation of the Representative during the 115th Congress:
D for Democrat, R for Republican
medHouseIncome median household income (dollars)
Use JAGS and R software, and use only the data in party115cong.csv. JAGS code should be
included in the appropriate sections, but all R code and any direct R output listings you choose
to include should be in the Appendix only.
Your report must be neatly typed and can be at most 8 pages, excluding the Appendix. It must
follow this outline:
1. Introduction Provide brief background information about the U.S. House of
Representatives, congressional districts, the 115th U.S. Congress, and U.S. political parties.
(Use footnotes to acknowledge any sources you consult, including web sites.)
2. Data Briefly describe and summarize the variables in party115cong.csv, as appropriate.
Plot a histogram of medHouseIncome.
3. First Model You will fit a Bayesian logistic regression model to explain party based on
the natural logarithm of medHouseIncome:
• The response will be Bernoulli: 1 if Democrat (D), 0 if Republican (R). You will need to
create a variable having this coding.
• The model will be a logistic regression.
• The linear portion of the model will be almost like a simple linear regression: There
will be an ordinary “intercept” term and a “slope” multiplying the (centered and
rescaled) log(medHouseIncome), but, of course, no “error” term.
1For the purposes of this assignment, the federal district of Washington, D.C. is regarded as a congressional
district.
1
• The independent variable is a centered and rescaled version of log(medHouseIncome):
centered to have sample mean of zero, and rescaled to have sample standard deviation
of 0.5 (not 1), as recommended in BDA3. Note that the centering and rescaling should
be after taking the natural logarithm.
• As recommended in BDA3, the prior for the “intercept” should be t1(0, 102
), and the
prior for the “slope” should be t1(0, 2.5
2
) (and these should be independent). Note:
These are expressed in BDA3 notation. Be careful when converting to JAGS code.
This first model will not use state (or any of the other variables).
• For the Bernoulli, use the dbern distribution specifier in JAGS.
• You may wish to consult the JAGS manual to make sure that you are correctly using
the dt distribution specifier (for a t distribution).
Run your analysis (being careful to follow the usual procedures) and report as follows:
(a) List your JAGS model.
(b) Summarize the details of your computation, including number of chains, length of
burn-in, number of iterations used per chain, any thinning (if used), and effective
sample sizes of all parameters. Do not include plots.
(c) Approximate the posterior mean, posterior standard deviation, and 95% central
posterior interval for each parameter.
(d) Approximate the posterior probability that the “slope” exceeds zero. Interpret this
result. (What apparently happens to the probability of electing a Democrat as median
household income increases?)
(e) Approximate the value of (Plummer’s) DIC and the associated effective number of
parameters. Compare the effective number of parameters with the actual number of
parameters.
4. Second Model Now extend your first model by allowing each state to have a separate
additive random effect:
• Create an indexing variable in which the variable state is recoded with the integers 1
to 51. (Refer to the Ebola data example in Week 14.)
• Starting with the first model (as described above), add to the linear portion of the
model a random effect term that varies by state. (For comparison, consider the term
betavirus[virus[i]] in the JAGS model for the Ebola data example in Week 14.)
• Let the prior for these random effects be (conditionally) independent from a normal
distribution with mean zero (since the model already has an intercept) and standard
deviation σstate.
• Let the prior for σstate be approximately flat. (You need to determine how to
implement this. It may require a preliminary run and some adjustment.)
Run your analysis (being careful to follow the usual procedures) and report as follows:
(a) List your JAGS model.
2
(b) Summarize the details of your computation, including number of chains, length of
burn-in, number of iterations used per chain, any thinning (if used), and effective
sample sizes of the top-level parameters. Do not include plots.
(c) Approximate the posterior mean, posterior standard deviation, and 95% central
posterior interval for the “intercept” and for the “slope” coefficient related to median
household income.
(d) Approximate the posterior probability that the “slope” exceeds zero. Interpret this
result. (What apparently happens to the probability of electing a Democrat as median
household income increases, after adjustment for state?)
(e) Which state has the largest (in the positive direction) posterior mean random effect?
Which state has the smallest (in the negative direction) posterior mean random effect?
Interpret in terms of the apparent party preferences of the two states (after adjustment
for median household income).
(f) Approximate the value of (Plummer’s) DIC and the associated effective number of
parameters. Is this second model better than the first?
5. Conclusions Briefly summarize your results in a non-technical manner.
6. Appendix Provide the R code you used to conduct your analysis. Include comments that
label the purpose of each block of code.
NOTES:
• Comma-separated variable (.csv) files can be read into R with read.csv.
• Effective sample sizes of at least 2000 are recommended for accuracy.
• If your computer runs out of memory, consider using thinning (e.g., the thin argument of
coda.samples).
3
Point Allocations
Specifications
2 neatly typed
2 no more than 8 pages (excluding Appendix)
Introduction
4 background given
1 sources acknowledged
Data
2 description/summary of variables
1 histogram
First Model
5 (a)
4 (b)
3 (c)
2 (d)
3 (e)
Second Model
5 (a)
4 (b)
3 (c)
2 (d)
3 (e)
3 (f)
Conclusions
3 brief, clearly stated, appropriate summary of results
Appendix
2 all R code present
2 comments for different blocks of code
Total: 56
4
