240C:
Problem Set 4
Data
We will be using the Correlates of State Policy dataset (Version 2.1, available on the course bCourse files page) to examine the relationship between Democratic control of state government and state poverty rates. The variable ranney4 control says whether a state’s government is under unified Democratic control(1), unified Republican control(0), or divided control(0.5). The variable povrate is the state poverty rate (estimated percent of individuals living in poverty). Additional variables to consider are region (indicating whether the state is located in the South(1), West(2), Midwest(3), or Northeast(4)) and statemin (the state minimum wage in dollars).
Problem 1
A) Use ranney4 control to create three dummy variables: dem control (states controlled by Democrats), rep control (states controlled by Republicans), and divided control (states controlled by neither) .
Next, lead the variable povrate by four years (because we might expect it to take some time for party control to impact economic outcomes). You can do this using the ’slide’ function from the DataCombine package. Call your new variable lead4_povrate .
B) Regress lead4 povrate on dem control, controlling for region and statemin. Provide a regression table.
C) What does the coefficient for dem control mean, and how should you interpret it?
D) Do you think that your regression from part B is able to isolate the causal effect of Democratic control on the state poverty rate? Why or why not?
E) Your regression by default reports standard errors of type HC0, which require homoscedasticity. What does the assumption of homoscedasticity (also spelled homoskedasticity) mean in substantive terms? In other words, what must be true for these standard errors to be “correct”?
F) Now report heteroscedasticity-corrected standard errors of type HC1. Note that where k is the number of independent variables in your regression. You may use the provided formula or the ’lm_robust’ command in R.
Problem 2
A) Next, we want to use a difference-in-differences design to estimate the effect of Democratic control on the state poverty rate. Run a two-way fixed effects model to test the hypothesis of whether Democratic control affects the poverty rate.
Since you are doing a diff-in-diff in this problem, you must cluster your standard errors by state.
B) Write the equation of your model from part A.
C) What does the coefficient for dem control mean in plain language? Explain the role of unit and time fixed effects in the analysis.
D) Do you think that your model from part A is able to isolate the causal effect of Democratic control on the state poverty rate? Why or why not?
E) It is possible to adjust for covariates in your diff-in-diff. Recall the control variables we used in Problem 1, region and statemin. Explain whether it makes sense to include each of these variables in your model. (Hint: Think about multicollinearity and posttreatment/collider bias.)