Assignment 3: Due before midnight on
Wednesday, April 13 on D2L
March 23, 2022
Note that the due date for this assignment is pushed back until midnight on Wednesday
following the final meeting of 702 (tutorial #6) in Week 12. After that, I’ll be posting
the grades and feedback for the assignment 3s that have been submitted, and late marks
will begin accruing for anybody who has not yet submitted. No exceptions, i.e. no
extensions will be given behind April 12 for Assignment 3 since the feedback will start
going out. The extra time beyond the syllabus due date should allow you to process
the final topics in 702 at your own pace, but of course you can submit the finished
assignment at any time before April 13. People who submit early will of course get
their feedback first.
number year state
1 1982 Alabama
2 1983 Iowa
3 1983 Kansas
4 1984 Calinfornia
5 1984 Texas
6 1985 Ohio
7 1986 Maine
8 1987 Missouri
9 1987 Georgia
0 1988 Arizona
Load the ROADs data set into R. We will use this data set for the entire assignment.
1. (25 points) Load the ROADs data set into R. Choose the two years from the
table corresponding to the 6th and 7th digits of your Ryerson student number. (If
they’re the same year, use the 8th digit in place of the 7th.) Run the regression
of traffic fatalities per 10000 residents against beer tax for the two years as two
separate cross sectional OLS regressions, then run the “changes” regression of
∆ fatalities against ∆ beer tax between the first and second year by OLS with
an intercept. What happens to the coefficient, βˆ1, on beer tax? Explain why βˆ1
changes between the two specifications (cross sectional vs “changes”) and how
this compares to the Worked Example from the Chapter 10 notes.
1
2. (25 points) Now show that your estimate of βˆ1 from the “changes” regression is
identical to the βˆ1 from a fixed effect regression using the same two years of data
and in which you allow for both a time and a state fixed effect. Why are these
βˆ1s the same?
3. (25 points) Now run the fixed effects regression of fatalities per 10,000 residence
on the beer tax and the state unemployment rate using the 336 state-year obser-
vations in the ROADS data set, allowing for both state and year fixed effects in
your model. (There are several ways to do this covered in tutorial #5 and you can
pick whichever way you find most intuitive.) Summarize the results in R with
the correct standard errors. Interpret the results (i.e. the economic implications
of the estimated coefficients) of this multiple (k = 2) regression. What is the
state- and time-specific intercept for the year-state combination corresponding
to the 6th and 7th digits of your student number? (Note: you may need to google
the fips codes.)
4. (25 points) Suppose now you have only a cross section of the 48 states from the
year corresponding to the 6th digit of your student number. You want to regress
spirit consumption per capita on the beer tax and the state unemployment rate,
but you’re worried about reverse causality: states with higher alcohol consump-
tion might have higher beer taxes through the political process, e.g. if the state
government sees beer sales as a good way to raise revenue OR states whose pop-
ulations enjoy drinking elect politicians who keep alcohol taxes low. Find the
most plausible instrument for beer tax you can from the ROADS data set and
run the corresponding IV regression. Explain why you believe your instrument
is potentially valid using your intuition and whatever diagnostic tests you find
helpful.
2