The folder congress has election outcomes and incumbency for U.S. congressional election races in the 1900s.
(a) Take data from a particular year, t, and estimate the effect of incumbency by fitting a regression of vi,t, the Democratic share of the two-party vote in district i, on vi,t−2
(the outcome in the previous election, two years earlier), Iit
(the incumbency status in district i in election t, coded as 1 for Democratic incumbents, 0 for open seats, −1 for Republican incumbents), and Pit (the incumbent party, coded as 1 if the sitting congressmember is a Democrat and −1 if he or she is a Republican). In your analysis, include only the districts where the congressional election was contested in both years, and do not pick a year ending in “2.” (District lines in the United States are redrawn every ten years, and district election outcomes vit and vi,t−2
are not comparable across redistrictings, for example, from 1970 to 1972.)
(b) Plot the fitted model and the data, and discuss the political interpretation of the estimated coefficients.
(c) What assumptions are needed for this regression to give a valid estimate of the causal effect of incumbency? In answering this question, define clearly what is meant by incumbency as a “treatment variable.”
See Erikson (1971), Gelman and King (1990), Cox and Katz (1996), Levitt and Wolfram (1997), Ansolabehere, Snyder, and Stewart (2000), Ansolabehere and Snyder (2002), and Gelman and Huang (2006) for further work and references on this topic.