Evan Katz

*This post is a modified version of my POLS 8150 research paper, The Impact of Third Party Candidates on the 2016 Presidential Election.

Democrat Hillary Clinton lost the 2016 presidential election despite winning the popular vote by nearly three million votes. Razor-thin margins in Michigan, Wisconsin, and Pennsylvania, traditionally blue states that flipped to the Republican column, decided the election in Republican Donald Trump’s favor, netting him 75 electoral votes and contributing to his 304-227 Electoral College victory. In each of those three states, prominent third party candidates Gary Johnson (Libertarian) and Jill Stein (Green) garnered more votes than Trump’s margin of victory. Some pundits have speculated that the presence of Johnson and Stein on those states’ ballots swung the election to Trump because the candidates siphoned votes away from Clinton. Others have dismissed this claim, pointing instead to Trump’s economic message that resonated with Rust Belt voters as well as Clinton’s decision not to allocate resources to defending those states.

Disagreement over the effect of third party candidates on the outcome of a presidential election is not unique to 2016. Scholars have spent years arguing over the effect of Ross Perot and Ralph Nader on their respective elections. I hypothesize that the presence of Johnson and Stein on the ballot negatively affected Clinton, but not to the extent that it cost her the election. First, I will look at characteristics of American presidential elections as well as previous elections where third party or independent candidates played a major role. I will then review different models used by scholars to test the impact of third party candidates. Finally, I will conduct a simulation based on polling data to determine if Johnson and Stein indeed cost Clinton the election.


The American political system is relatively unique among other western democracies in that two major parties completely control the political landscape. At the start of the 115th Congress, Democrats and Republicans controlled every seat in the House of Representatives and all but two seats in the Senate (Angus King of Maine and Bernie Sanders of Vermont, the lone Independent senators, both caucus with the Democrats). Moreover, every president dating back to Franklin Pierce has been affiliated with either the Democratic or Republican Party.

Two-party dominance at the presidential level can be explained by a couple of key factors. First, Duverger’s law holds that plurality-rule electoral systems, like the first-past-the-post system used to decide presidential elections in the United States, tend to produce two dominant parties. In such systems, in order to prevent the least desirable candidate from winning, voters may engage in tactical voting in an election by voting for their second or third choice in the event that their most preferred candidate cannot feasibly win. Second, the Electoral College helps perpetuate the two-party system because the winner-take-all nature of most state contests diminishes the opportunity for independent or third party candidates to win electoral votes. Minor candidates must have strong appeal in specific geographic regions to be able to actually win states; otherwise, even candidates with broad support nationwide can end up with zero electoral votes (e.g. Ross Perot in 1992). In spite of these factors, independent and third party candidates have occasionally found success and influenced presidential election outcomes.

1912: Best Third Party Showing in American History

The 1912 presidential election featured the best performance by a third party candidate in American history. Former President Theodore Roosevelt disapproved of the performance of his successor, incumbent Republican William Taft, and entered the race, initially challenging him in the Republican primaries. When he failed to secure the Republican nomination, he created the Progressive Party and ran on its ticket. Roosevelt’s presence in the race not only siphoned votes away from Taft, aiding Democrat Woodrow Wilson to victory in the process, but also discouraged Taft to the point that he hardly campaigned leading up to the election. Roosevelt captured 88 electoral votes and over 27 percent of the popular vote. Socialist Eugene Debs also competed for the presidency and managed to win over six percent of the popular vote, but zero electoral votes.

1948 and 1968: Segregationists Win States in the South

Both the 1948 and 1968 presidential elections saw third party candidacies emerge from the fractured Democratic Party. In 1948, incumbent Democrat Harry Truman attempted to win another term in his own right after succeeding Franklin D. Roosevelt following his death in 1945. However, ideological disputes within the party, which had controlled the presidency since 1932, produced two other competitors. Southern Democrats disaffected with the national party’s embrace of civil rights moved to create a third party—the States’ Rights Democratic Party—as a means of continuing de jure racial segregation, nominating then-South Carolina Governor Strom Thurmond to head the ticket. Thurmond won 2.4 percent of the popular vote and carried four states totaling 39 electoral votes. At the other end of the party, Henry A. Wallace established a new reincarnation of the Progressive Party supporting stronger regulation and opposing the anti-Communist hysteria that he believed undermined civil liberties. Wallace also won just over two percent of the popular vote, but unlike Thurmond he garnered zero electoral votes.

Despite the fragmentation within his party and losing states in the once-Solid South, Truman managed to pull off one of the most shocking upsets in American history, beating Republican favorite Thomas E. Dewey. The final result was so close that had California, Illinois, and Ohio—states that Truman carried by less than one point each—flipped to the Republican column, Dewey would have been elected president, and if Dewey had carried any two of those states, Thurmond would have denied either major party candidate an Electoral College majority, sending the election to the House of Representatives.

The 1968 election unfolded in a similar manner to 1948. The Democratic Party fractured along factional lines, producing a gruesome nomination process that saw the assassination of once-favorite Robert F. Kennedy on the campaign trail, utter chaos at the Democratic National Convention in Chicago, and the end of the New Deal coalition that had dominated American politics since Franklin D. Roosevelt’s election in 1932. President Lyndon B. Johnson was embroiled in an extremely unpopular war in Vietnam, contributing to his unpopularity within the party and his ultimate decision not to seek reelection. Vietnam dominated much of the nomination process, helping propel the antiwar candidates, Kennedy and Eugene McCarthy, to early successes. However, Johnson’s control over the Democratic machinery helped him to secure the nomination for his vice president—and preferred candidate—Hubert Humphrey.

Democrats’ lock on the South unraveled as the national party pushed issues like the Civil Rights Act, leading to yet another Southern Democrat—Alabama Governor George Wallace—splitting off and running as a third party, pro-segregation candidate. Wallace intended to win enough states and influence enough state contests to deny both major nominees an Electoral College majority and force concessions. Wallace’s presence in the race combined with Republican Richard Nixon’s “Southern Strategy” to win disaffected white conservatives helped permanently realign the South out of the Democrats’ column. Indeed, 90 percent of white conservatives in the region voted for either Wallace or Nixon, a strong rebuke of Humphrey.

Wallace fell short in his goal to deny Nixon and Humphrey electoral majorities; his 46 electoral votes from five states in the South could not prevent Nixon from winning 301 electoral votes nationwide. However, the close nature of the election—Nixon only won the popular vote by 0.7 percent—has raised questions as to whether Wallace’s presence in the race tipped the election in Nixon’s favor. In an eerily similar fashion to 1948, California, Illinois, and Ohio—all won by Nixon by less than three points—played the deciding role in the election. Wallace secured nearly 12 percent of Ohio’s vote, which mostly consisted of union workers that might have otherwise voted for Humphrey. Had Humphrey won both Ohio and Illinois, or California alone, Wallace would have forced the election to the House of Representatives. Had Humphrey won all three, he would have been elected the 37th president.

1992: Ross Perot

The outcome of the 1992 election initially seemed like a foregone conclusion as incumbent President George H.W. Bush boasted an 89 percent approval rating after the successful conclusion of the First Gulf War. This fact discouraged a number of high profile Democrats from challenging the president. However, a sputtering economy and going back on his promise not to raise taxes lowered Bush’s odds of reelection and helped lead outsider and Arkansas Governor Bill Clinton to a 370-168 Electoral College victory.

Complicating the picture was billionaire businessman Ross Perot, an independent candidate from Texas whose populist, anti-establishment message translated into an upswell of grassroots support. Perot campaigned against the North American Free Trade Agreement, played up the federal deficit as a major issue, and advocated for direct democracy via electronic town halls. At one point, Perot led the three-way race, but his ill-advised decision to drop out of contention in July doomed his chances of upending the two-party system. Nevertheless, he reentered the race in October and managed to win almost 19 percent of the popular vote on Election Day.

Perot forced all but two of the 51 individual state contests to be decided by pluralities. Seventeen states had margins of victory smaller than five percent, including two—Georgia and North Carolina—with margins smaller than one percent. The general closeness of so many state contests has generated numerous counterfactuals. For example, had Bush flipped 300,000 from Clinton across ten different states, he could have won the electoral vote, 271-267, despite trailing Clinton by more than five million votes. Had fewer than 300,000 votes flipped from Clinton to both Bush and Perot across eleven different states, no candidate would have secured an Electoral College victory.

Perot’s effect on the outcome of the election is hotly debated. The Bush camp has repeatedly claimed that Perot ate significantly into Bush’s support given that he was a former Republican that touted a conservative economic message, therefore swinging the election to Clinton. However, various studies dispute this claim. For example, one study by Voter Research & Surveys used exit polls to construct a counterfactual Perot-less election and concludes that Ohio would have been the lone state to flip from Clinton to Bush absent Perot. The study also finds that more Perot voters would have voted for Bush in a handful of other states Clinton carried—Georgia, New Hampshire, New Jersey, and Tennessee—though not enough to swing the results.

2000: Florida, Nader, and Buchanan

The 2000 presidential election turned out to be the closest in American history. Texas Governor George W. Bush beat Vice President Al Gore in the Electoral College, 271-266 (with one abstention from a D.C. elector), despite losing the popular vote by more than 500,000 votes. The election hinged on Florida, whose 25 electoral votes were enough to push either candidate over the 270-electoral vote threshold. Florida’s minuscule margin of 1,784 votes triggered an automatic recount that shrunk the margin to approximately 900 votes. Gore, in accordance with Florida election law, requested a manual recount in four counties, but the Bush campaign argued that the recount violated the Fourteenth Amendment’s equal protection clause. After a contentious legal battle, the Supreme Court ruled in favor of Bush, effectively declaring him the winner of Florida by 537 votes.

Much attention has been paid to the role of two third party candidates, Ralph Nader of the Green Party and Pat Buchanan of the Reform Party, on the final vote totals in Florida and the election as a whole. Members of the Gore campaign have argued that Nader, a left-wing candidate who received nearly 100,000 votes in Florida, disproportionately siphoned votes away from Gore, handing the state to Bush. While many studies have concluded that Nader spoiled the election for Gore, the relative partisanship of Nader (and Buchanan) voters has been the subject of debate. A 2007 study used ballot images from ten Florida counties and concluded that approximately 40 percent of Nader voters preferred Bush as a second choice, indicating that Nader’s appeal cut across traditional partisan identities. Accounting for the number of Nader voters that would have stayed home on Election Day had Nader not been in the race, a 2006 study revealed that approximately a quarter of Nader voters would have voted for Gore, 11.8 percent would have voted for Bush, and the remaining share would not have voted.

2016: Unpopular Nominees and Protest Votes

The 2016 election occurred against the backdrop of perpetual gridlock, creating disdain for the establishment and “politics as usual” within the electorate. This disdain gave rise to two prominent anti-establishment candidates: on the right, billionaire real estate mogul Donald Trump, and on the left, self-proclaimed democratic socialist Senator Bernie Sanders, an Independent from Vermont. Trump’s bombastic rhetoric was perceived by many to be bigoted against immigrants, women, and Muslims, which contributed to his broad unpopularity. However, the inability of the Republican establishment to rally behind one candidate during the nomination process enabled Trump to defeat each of his 16 competitors and secure the Republican nomination by early May.

Within the Democratic Party, Sanders presented a formidable challenge to Barack Obama’s heir presumptive and former Secretary of State Hillary Clinton. He quickly turned the nomination process into a two-candidate race and pressed Clinton on her ties to Wall Street, neoconservative foreign policy record, and numerous scandal allegations, rallying millions of young voters and independents that were frustrated with the establishment in the process. Those factors, combined with the rumor that the Democratic National Committee rigged the nomination for her, contributed to Clinton’s broad unpopularity. However, Clinton eventually secured the Democratic nomination in early June.

Because both Clinton and Trump were the two of the least favorable nominees in recent history, a number of voters sought out alternatives. One such alternative was former New Mexico Governor Gary Johnson, the Libertarian nominee. Johnson, a former Republican running on the ticket with former Republican Massachusetts Governor Bill Weld, offered some disaffected establishment Republicans a way out of voting for Trump without having to vote for Clinton, and his liberal social positions made him a more attractive candidate to anti-Clinton moderate Democrats. Another alternative, Green Party nominee Jill Stein, carried the torch for Sanders’ movement after Sanders conceded to and endorsed Clinton. A sizable portion of the “Bernie or Bust” felt betrayed by Sanders’ move and flocked to Stein in protest.

Despite polls suggesting otherwise, Trump emerged victorious on Election Day, winning the electoral vote, 304-227 (7 electors voted for other candidates in protest), even though he lost the popular vote by nearly three million votes. His electoral victory stems in large part from his ability to secure narrow wins in Michigan, Wisconsin, and Pennsylvania, three states that had voted for the Democratic candidate in each election dating back to 1992, which pushed him over the 270-electoral vote threshold. Both Johnson and Stein alone covered Trump’s margin of victory in those three states, leading some to blame Clinton’s loss on the presence of third party candidates on the ballot.

Literature Review

Traditionally, political scientists have used spatial models to analyze elections. In such models, candidates are placed along a unidimensional continuum and voters are predicted to vote for the candidate that lies closer on the continuum to their ideal point. Usually these models determine the ideological position of both voters and candidates by using input data from questionnaires that ask voters to place themselves and the candidates along the liberal/conservative spectrum. While spatial models perform well for two-candidate races, the picture becomes much more complicated when other candidates enter the mix. Spatial models impose the property of independence of irrelevant alternatives (IIA) on voters, which holds that the addition of other candidates into a race should not affect the probability of a voter choosing one party over the other. As such, numerous political scientists have dedicated work to developing models specifically to analyze third party voters, including their demographic makeup, why they chose to vote for a third party or independent candidate, and who their second choice candidate would be.

In examining the impact of Ross Perot on the 1992 election from a voter choice perspective, Alvarez and Nagler (1995) conduct a multinomial probit analysis testing prevailing explanations, including an economic explanation that predicts voters will base their decision on how the candidates’ parties have run the economy in the past, an ideological explanation that predicts voters will decide based on which candidate is ideologically closest to their ideal point, and an angry voter explanation that posits that angry voters turn out when there is an anti-status quo candidate in the race. With respect to 1992, the economic explanation holds that voters spurned Bush in response to the recession, the ideological explanation holds that Clinton was more moderate than previous Democrats and managed to adopt policy positions that appealed to broader swaths of the electorate, and the angry voter hypothesis holds that Ross Perot helped activate voters because of his anti-status quo message. Their model concludes that the economy best explains the 1992 election, not the angry voter or ideological explanations.

Magee (2003) runs regressions based on two sources—voter data collected from the American National Election Studies (NES) on the 2000 election and CBS/New York Times polls from July 2000—to determine who Nader voters would have preferred in a hypothetical two-way race. Both sources asked voters their second choice preferences. He finds that, in line with other studies, approximately a third of Nader voters would have voted for Gore, while between 14 and 17 percent would have voted for Bush. He runs counterfactual election predictions based on demographic characteristics and finds that the predictions fall largely in line with polling data. The issue with Magee’s study is that it covers a small sample size (just 33 Nader voters are surveyed in the NES survey while 85 are surveyed in the CBS poll) and does not break down the vote based on state of residence.

Simmons and Simmons (2006) also use (NES) data from the 2000 election to examine the different demographic and political characteristics of Nader voters, Nader-supporting Gore voters, and non-voters. They found that Nader voters tended to be more economically privileged and politically progressive, and also used second-choice data from the studies to determine that, while roughly a quarter of Nader voters preferred Gore, nearly 12 percent preferred Bush, indicating that support for Nader tended to be less partisan and more the product of other factors. Using those numbers, Simmons and Simmons conclude that Nader spoiled the election for Gore, although not to the degree that some pundits, and Democratic strategists, claim.

Herron and Lewis (2007) use three million ballot images from ten Florida counties to construct partisan profiles of Nader and Buchanan voters—similar to NOMINATE scores—as a means of predicting which candidate they would have voted for in a hypothetical two-way race. Those profiles were then input into a single-dimensional spatial model. Ballot images allowed them to view voting decisions by Nader and Buchanan voters in down-ballot races that generally pitted Democrats against Republicans. Using this model they found that 60 percent of Nader voters would have voted for Gore, while 40 percent would have voted for Bush, enough to swing the election to Gore. Despite lacking demographic data, Herron and Lewis chose this method over the NES method used by Simmons and Simmons due to the unreliability of such data for candidates with little support, like Nader and Buchanan.

The 2016 Election

Various pundits and news outlets have offered opinions on whether or not Johnson and Stein tipped the election in Trump’s favor. Johnson, a former Republican, intuitively would be expected to attract more voters from the right, especially considering Trump’s relative unpopularity among establishment voters. But some sources, including respected statistics and polling analysis website FiveThirtyEight, argued that Johnson may have taken more support from Clinton, given that her lead over Trump shrank when he was included in polls, on average. With Stein, a left-wing progressive running on the Green Party ticket, many expect her presence in the race disproportionately hurt Clinton because she gave disaffected Sanders voters a way out of voting for Clinton without having to vote for Trump.

Voter data

Given that the 2016 presidential election concluded less than a year ago, most of the available voter data is incomplete. The NES has thus far only released a pilot study on the 2016 election studying the characteristics of approximately 1200 primary voters. No variables are included for Johnson or Stein, making it impossible to construct voter profiles based on demographic and political characteristics. Edison Research conducted exit polls on behalf of the National Election Pool (NEP), the successor to the Voter News Service (VNS) used by other studies on the 1992 and 2000 presidential elections, and released certain data to analysts, but raw polling data has not been made available to the public.

Two sources of polling data, a Pew Research Center study released in mid August and an NBC News/SurveyMonkey weekly election tracking poll from September, asked registered voters their candidate preference in a hypothetical two-way race, among other questions. The Pew study surveyed approximately 2,000 adults, including nearly 1,600 registered voters, while the NBC poll sampled over 16,000 registered voters. Though limited in that neither source breaks down voters by state, both can be used to help examine the impact of Johnson and Stein on the election.

The Pew study placed Johnson’s overall support at 10 percent, while 4 percent of those surveyed said they would vote for Stein. Both candidates disproportionately drew support from younger voters (19 percent and 9 percent of all voters aged 18-24 said they supported Johnson and Stein, respectively). Indeed, approximately one third of Johnson supporters in this study are aged 29 or younger. Stein notably drew 9 percent of the Hispanic vote as well. The study asked Johnson voters their preference in a two-way race between Clinton and Trump and found that his support draws evenly from both sides: 43 percent preferred Clinton while 42 percent preferred Trump. There is no indication within the study that Stein voters were asked the same question.

The NBC poll asked voters a variety of questions, including which issue they felt was most important (among eight choices), their thoughts on Obama’s job as president, their partisan identification and lean, and who they would vote for in a two-way race. The poll found that 39 percent of Johnson voters preferred Clinton while 35 preferred Trump, and that 57 percent of Stein voters preferred Clinton to 18 percent that preferred Trump. Given that the NBC poll is more recent, samples a larger portion of the population, and includes statistics on both candidates, this simulation will use the poll to run basic calculations on the vote breakdowns in the four closest states that Trump managed to win (Michigan, Pennsylvania, Wisconsin, and Florida).

Counterfactual Simulation

Table 1: Trump’s Four Closest States by Popular Vote Margin

Four Closest States
Each of these states voted for Obama in the previous two elections. Johnson beat Trump’s margin of victory in all four states, while Stein beat the margin of victory in the closest three.

Table 1 shows the results from the actual 2016 election in the four key states. Clinton would have needed approximately 80,000 more votes in the three closest states to win the election. To get a rough estimate of the effect of Johnson and Stein’s presence on the ballot, I will simulate a counterfactual election using data from the NBC poll. This simulation will take the results of the four closest states that Trump won—Michigan, Pennsylvania, Wisconsin, and Florida—and calculate new vote totals with Johnson and Stein voters reallocated to their preference in a hypothetical two-way race. For example, Trump’s new vote total would be the sum of his actual vote, 35 percent of that state’s Johnson voters, and 18 percent of that state’s Stein voters. The remaining Johnson and Stein voters that did not answer the survey question will be treated as non-voters.

This simulation is less than ideal for multiple reasons. First, the data from the NBC poll uses a national pool of respondents and does not break down voters by state. Any unique trends relating to the candidates at the state level will therefore have to be ignored. Second, the poll does not account for “other” voters that covered the margin of victory in Michigan, Pennsylvania, and Wisconsin. One cannot assume the preferences of these voters in a two-way race if their candidates were removed from the ballot. By keeping them on the ballot, it is entirely possible that Johnson and Stein voters would have chosen any of the other available candidates on the ballot instead of Trump or Clinton, but ranked preferences for all candidates are unavailable. Third, this poll was taken in September, which doesn’t account for later trends and key events like now former FBI Director James Comey’s November letter to Congress informing them that he was reopening an investigation into Clinton’s private email server.

Table 2: Trump’s Four Closest States With Johnson and Stein Votes Reallocated

Screen Shot 2017-04-24 at 8.04.33 AM
Based on the NBC poll numbers, had Johnson and Stein not been in the race, only Michigan would have flipped to Clinton. However, Trump’s margin of victory shrinks in all four states, indicating that Johnson and Stein on balance hurt Clinton more than Trump.

The simulation yields some interesting results. As expected, Trump’s margin of victory shrinks when reallocating Johnson and Stein voters to their preference in a two-way race. Clinton performs more than half of a point better in Michigan and Wisconsin, and about 0.4 points better in Pennsylvania and Florida. However, based on these results, the argument that both third party candidates’ presence on the ballot is incorrect. Only Michigan switched from Trump to Clinton, which would have decreased Trump’s electoral vote total from 306 to 290, not accounting for faithless electors. Clinton would have needed approximately 25,000 more votes distributed across Pennsylvania and Wisconsin to change the outcome of the election. More data is needed on the preferences of voters who chose other minor candidates like Evan McMullin, Darrell Castle, and others.


Unlike 2000, when Nader cost Gore the presidency by a mere 537 votes, the 2016 election outcome was likely unaffected by the presence of Gary Johnson and Jill Stein on the ballot. Although the combined effect of both third party candidates likely hurt Hillary Clinton, neither pulled a sufficient number of votes to tip the election to Donald Trump. Other explanations for Trump’s victory, like the economy, ideology, the angry voter hypothesis, and campaign tactics, to name a few, have more explanatory power.

With the proper data, future studies can more accurately determine the impact that Johnson and Stein had on the election, if any at all. Alvarez and Nagler’s multivariate analysis from a voter choice perspective can test competing explanations and assess which has the most significance. Because the election came down to the results in a few states, Herron and Lewis’ ballot image study can be used to construct partisan profiles of third party voters to determine if they were more likely to prefer one candidate to another. NES data on demographics and political characteristics can also be used to predict whom third party voters would have preferred in a hypothetical two-way matchup as well.