Week 13: Data Joining

Author

Joe Ornstein

It is rare that all the variables you need for your research will be found in a single dataset. More often, you will need to merge information from multiple datasets before you can begin your analyses. By the end of this week, you will be able to:

Reading

  • R4DS Chapter 19

Problem Set

Gelman et al. (2007) find an interesting paradox in which rich people are more likely to vote for Republican presidential candidates, but rich US states are more likely to vote for Democrats. In this problem set, we will see if this paradox still holds true in the year 2020.

  1. Load the raw CES 2020 dataset and tidy the family income, 2020 presidential vote choice, and state of residence variables as necessary.
  2. Were rich people more likely to vote Republican in 2020? Compute the percent of respondents at each family income level that reported voting for Joe Biden, and plot the relationship.
  3. Use the tidycensus package to get a dataset with 5-year estimates of median income from the 2020 American Community Survey.
  4. Join the individual-level survey data with the median income variable by state.
  5. Were rich states more likely to vote Democratic in 2020? Create a state-level dataset with each state’s median income and the percent of CES respondents in that state that reported voting for Joe Biden. Plot the relationship and fit a linear model, reporting the estimated slope.
  6. Bonus. Another finding from the Gelman et al. (2007) paper is that the relationship between individual-level income and vote choice is strongest in poor states and weakest in rich states. Let’s see if that result still holds up in 2020. Create a variable that splits the dataset into three groups: poor states (poorest third of the US states), middle-income states (middle third), and rich states (richest third of the US states). Then repeat the individual-level analysis in problem 2, faceting by this new income-group variable.

Additional Resources

  • For a primer on fuzzy record linkage (merging datasets when the key variables do not contain exact matches), might I self-promote a bit and suggest this paper and R package?

References

Gelman, Andrew, Boris Shor, Joseph Bafumi, and David Park. 2007. “Rich State, Poor State, Red State, Blue State: What’s the Matter with Connecticut?” Quarterly Journal of Political Science 2 (June 2006): 345–67. https://doi.org/10.1561/100.00006026.