![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
ABSTRACT
County-level data are used to estimate the incumbent-party share of the two-party vote in the 2012 and the 2016 U.S. Presidential elections. Using a ‘seemingly unrelated estimation’ procedure the regression results for the two elections show that there were some clear differences in the size of marginal effects for several key covariates. For example, income inequality, the size of the black male and black female populations, the size of the Hispanic male population and percent of the population with a college degree all had significantly larger coefficients in 2016 than in 2012, producing a larger marginal effect in favour of the Democratic candidate’s vote share. On the other hand, counties with increased poverty rates and counties located on the periphery of urban centres had a significantly larger marginal effect favouring the Republican’s vote share in 2016 compared to 2012. Finally, the regression results show that the effects of third-party vote shares, though not statistically different across the two elections, had a positive impact on the Democratic vote share in both elections.
Disclosure statement
No potential conflict of interest was reported by the author.
Dependent variable
Dave Leip’s Atlas of U.S. Elections, available at: https://uselectionatlas.org/.
Independent variables
% third-party vote and lagged vote: same as dependent variable.
Population demographics (including % college) and income gini: U.S. Census Bureau, 2008-2012 and 2012-2016 American Community Survey.
Unemployment data: Bureau of Labour Statistics, Local Area Unemployment Statistics (LAUS), https://www.bls.gov/lau/.
Income, Poverty rates and Rural-Urban Continuum Codes: U.S. Census Bureau, Model-based Small Area Income & Poverty Estimates.
Religious adherence rates: U.S. Religion Census: Religious Congregations and Membership Study, 2010, The Association for Religion Data Archives (ARDA), available at: http://www.thearda.com/.
Employment data for NAICS 21, 31-33: U.S. Census Bureau, County Business Patterns (CBP). In some cases the CBP suppressess actual employment counts for privacy reasons. In such cases, the CBP provides codes that represent count ranges. In these cases, midpoint values for the ranges provided were used.
Notes
1 ‘Grading the 2016 Election Forecasts,’ BuzzFeed News. Accessed on 14 May 2019 at: https://www.buzzfeednews.com/article/jsvine/2016-election-forecast-grades.
2 Stata’s ‘suest’ command is used to carry out the comparisons.
3 There is also a very large body of literature that employs voter surveys, such as the American National Election Studies (ANES) survey, to study electoral outcomes. Many of these studies explore the impact of race, social identity, partisanship and the perception of a candidate’s character (among other things) on voting behaviour. As an example, see D’Elia and Norpoth (Citation2014) which uses the ANES survey to examine the 2012 Presidential election. Another paper by Abramowitz and McCoy (Citation2019) employs the ANES survey to study the 2016 Presidential election.
4 The appendix provides information regarding the sources of the data employed in the empirical analysis.
5 See Hansford and Gomez (Citation2010) for a careful discussion of voter turnout effects on elections.
6 This assumes that the third-party candidate is not a ‘large’ threat to win the election. If the third-party candidate did have a reasonable chance of winning the election then the Democratic and Republican candidates may alter their political positions in an attempt to maximize their vote shares. In such a scenario the ultimate political positions chosen by the three candidates is difficult to identify.
7 In contrast, Alvarez and Nagler (Citation1995) argue that Ross Perot, who garnered 19 percent of the popular vote in 1992, likely did not have a significant impact on either Bill Clinton and President Bush in the election because Perot drew a similar amount of voters away from each candidate.
8 Johnson’s stances on various topics were found at: http://www.ontheissues.org/default.htm, accessed on 17 May 2019.
9 The percent change in income is used to capture how incomes have grown (or shrunk) over the four years leading up to an election. For variables measured in percentage points (the unemployment rate and the poverty rate) the simple change in those rates are employed.
10 Another possibility is that if there were a more liberal, progressive candidate as a third-party choice, this could reduce Democratic share of the two-party vote if more liberal voters were drawn away from the Democratic candidate.
11 Hochschild and Wallace (Citation2011) find mixed support for this proposition when studying voting outcomes in 276 metropolitan statistical areas in the 2000 Presidential election vote share for Green Party candidate, Ralph Nader.
12 ‘Behind Trump’s victory: Divisions by race, gender, education,’ Pew Research Centre, 9 November 2016. Accessed on 22 May 2019 at: https://www.pewresearch.org/fact-tank/2016/11/09/behind-trumps-victory-divisions-by-race-gender-education/.
13 ‘America Is divided by education,’ Adam Harris in The Atlantic, 7 November 2018. Accessed on 22 May 2019 at: https://www.theatlantic.com/education/archive/2018/11/education-gap-explains-american-politics/575113/.
14 See: OnTheIssues.org (http://www.ontheissues.org/Donald_Trump.htm), accessed on 23 May 2019.
15 E&E News, accessed on 23 May 2019 at: https://www.eenews.net/stories/1060036954/.
16 It would be desirable to use a less aggregated measure for these variables, but employment data at the county level are only available at the two-digit NAICS level.
17 See: https://uselectionatlas.org/.
19 In 48 of the 50 states, the term county is used for subdivisions of states. In Louisiana the county-equivalent subdivisions are called parishes. In Alaska they are called boroughs. Washington D.C. is not included.
20 County population weights from 2016 are used in both regressions as the same weights are required for the comparison of the two election’s coefficients (reported in the third column) using Stata’s ‘suest’ command. The correlation between county population values from 2012 and 2016 exceeds 0.99.
21 Furthermore, both regressions produce predicted values that fall into the [0, 100] interval.
22 See: OnTheIssues, ‘Donald Trump on Tax Reform,’ accessed on 18 June 2019 at: http://www.ontheissues.org/Tax_Reform.htm#Headlines.
23 Recall that these variables are measured as the number of adherents per 1000 people, thus the magnitude of the estimated coefficients is small. As such we can consider the effects on the Democratic vote share from a one standard deviation change in the religion measures.
24 See, for example, the Pew Research Centre’s article titled ‘How the faithful voted: A preliminary 2016 analysis,’ at: https://www.pewresearch.org/fact-tank/2016/11/09/how-the-faithful-voted-a-preliminary-2016-analysis/.
25 See, for example the 7 December 2015 New York Times article, ‘Donald Trump Calls for Barring Muslims From Entering U.S.’ accessed on 31 May 2019 at: https://www.nytimes.com/politics/first-draft/2015/12/07/donald-trump-calls-for-banning-muslims-from-entering-u-s/.
26 See, for example, the article in Politico titled, ‘Uttered in 2008, still haunting Obama,’ at: https://www.politico.com/story/2012/04/uttered-in-2008-still-haunting-obama-in-2012-074892.
27 The regression contains a full set of covariates as shown in and uses the same methodology. The estimated coefficients on the covariates not shown in Table 4 are little changed from those shown in . Complete results are available from the author.
28 A one-standard deviation increase in Stein’s total vote share (also a fairly large change) would lead to a two-party vote share increase for the Democratic candidate of about a 0.5 percentage points in 2012 and about 0.8 percentage points in 2016.
29 For example, see CNN’s article: ‘How Gary Johnson and Jill Stein helped elect Donald Trump,’ found at: https://www.cnn.com/2016/11/10/politics/gary-johnson-jill-stein-spoiler/index.html.
30 One possibility is that Stein’s presence in the in the election affected the composition of the voter turnout that ultimately improved the chances of the Democratic candidates.
31 There is some evidence that Johnson did help Clinton in the 2016 election. See the Washington Post article by Sasha Volokh titled, ‘Gary Johnson helped Hillary. Not by enough, but he did,’ at: https://www.washingtonpost.com/news/volokh-conspiracy/wp/2016/11/11/gary-johnson-helped-hillary-not-by-enough-but-he-did/?utm_term=.4617ea4b56bc.
32 John Quincy Adams in 1824, Rutherford B. Hayes in 1876, and Benjamin Harrison in 1888.
33 See, for example, the Wall Street Journal article, ‘Poll Finds Lack of Enthusiasm for Clinton and Trump,’ at: https://www.wsj.com/articles/poll-finds-lack-of-enthusiasm-for-clinton-and-trump-1464037289.
34 See the Pew Research Centre article, ‘Behind Trump’s victory: Divisions by race, gender, education,’ at: https://www.pewresearch.org/fact-tank/2016/11/09/behind-trumps-victory-divisions-by-race-gender-education/.
35 Reny, Collingwood, and Valenzuela (Citation2019) work with survey data from the Cooperative Congressional Election Studies. Their analysis focuses on White voters categorized as both ‘working class’ and ‘non-working class’ and who switched their votes from one party to the other between the 2012 and 2016 presidential elections.