Description
1. Please read: On the Effects of COVID-19 Safer-At-Home Policies on Social Distancing, Car Crashes and Pollution, by Brodeur, Cook and Wright. http://ftp.iza.org/dp13255.pdf
a. Consider equation 1. Explain why ???????????????????????? is a DD estimator. What are the two differences being considered? (2 marks)
b. Why is it important for identification that states/counties introduced the safer-at-home policies at different times? Could the same empirical strategy have worked if the country went into lockdown all at once? Why/why not? (2 marks)
c. The estimated effect of safer-at-home policies on pollution was found to be heterogenous across regions. List the sources of heterogeneity considered, and summarize the results. (3 marks)
d. The authors also look at safer-at-home policies on trips pattern (equation 2 and Figures 5-). We can see that the confidence intervals get larger the farther from 0 we are. Can you hypothesize why estimates are less precise further away from 0? (2 marks)
e. In many cases (ex Fig 5) there is evidence of changed behavior leading up to the date the policy takes effect. What might explain this behavior? **How might the fact that people respond early to a policy change affect the DD estimates from equation 1? (Hint: see** below) (4 marks)
2. Please read: COVID-19, Lockdowns and Well-Being: Evidence from Google Trends, by Brodeur, Fleche, Clark and Powdthavee. http://ftp.iza.org/dp13255.pdf
a. This paper also uses a DD strategy. Explain the two levels of difference considered. (2 marks)
b. Why is it not enough to compare data pre and post lockdown? (2 marks)
c. Is the common trend/parallel trend assumption justified? Explain. (2 marks)
d. How is mental health measured in this paper? Compare these measures (in terms of both advantages and disadvantages) to other more common measures used in the literature. Ex: mental health related hospitalizations as found in administrative data, or self-rated health from survey data. (4 marks)
e. One perhaps surprising finding is that while many measures of well-being were found to get worse post lockdown some measures including suicide searches, went down. Suggest an explanation, or two, for this result. (Hint: one explanation can be found in the paper from question 1!) Explain how one might go about testing your hypothesis. (3 marks)
3. The data used in these papers includes GOOGLE location tracking data and GOOGLE search information. These are super rich sources of data. Researchers are just starting to fully explore the possibilities they provide. Please suggest a possible extension to the existing papers that would make use of one or both of these data sources. Be clear about what the research question would be, and what empirical approach you would use. If DD, explain the differences to be used. If IV, briefly explain and justify the instrument. If other methodology is appropriate, please be sure to explain how the method will answer your research question. (4 marks)
**Aside: For anyone thinking of using DD to estimate a causal effect of a COVID Policy for their proposal please see: Using Difference-in-Differences to Identify Causal Effects of COVID-19 Policies
