The 2016 Pennsylvania Presidential and U.S. Senate Elections COMMONWEALTH, Volume 19, Issue 2 (2017). © 2017 The Pennsylvania Political Science Association. ISSN 2469-7672 (online). http://dx.doi.org/10.15367/com.v19i2.152. All rights reserved. The 2016 Pennsylvania Presidential and U.S. Senate Elections Breaking Pennsylvania’s Electoral Habits BERWOOD YOST Franklin & Marshall College JACKIE REDMAN Franklin & Marshall College SCOTTIE THOMPSON Franklin & Marshall College This article uses pre-election survey data, post-election survey data, and voter registration and election data to interpret the outcomes of the 2016 presidential and U.S. Senate races in Pennsylvania. This analysis shows how changes in voter registration and voter turnout in specific areas of the Commonwealth, driven in large part by less-educated voters, those dissatisfied with the cur- rent direction of the country, and the performance of the incumbent president, explain the 2016 election results. That a Republican won the White House in 2016 was not, according to political science forecasters, a breathtaking surprise. The surprise was that Donald Trump was the Republican nominee and that he won not only the White House but also Pennsylvania, a state that had supported the Democratic presidential candidate in six straight elections. This article uses pre-election survey data, post-election survey data, and voter registration and election data to interpret the outcomes of the 2016 presidential and U.S. Senate races in Pennsylvania. This analysis shows that significant changes in voter registration and voter turnout in specific areas of the Commonwealth, driven in large part by less-educated voters, those dissatisfied with the current direc- tion of the country, and the performance of the incumbent president, explain the 2016 election results. 4 Berwood Yost, Jackie redman, and scottie thompson Data Sources This article includes data from three primary sources. The pre-election poll- ing data on candidate favorability, candidate preferences, and candidate char- acteristics come primarily from Franklin & Marshall College Polls conducted during July, August, September, and October 2016. Interviewing dates and sample sizes for each survey are as follows: July 29–August 1 (n = 661), August 25–29 (n = 736), September 28–October 2 (n = 813), and October 26–October 30 (n = 863).1 The post-election polling data come from re-interviews with respondents from these four pre-election polls.2 The registered voter samples for these surveys were obtained from L2 and all sampled respondents were notified by letter about the survey. Interviews were completed over the phone and online depending on each respondent’s preference. Survey results were weighted (by age, gender, region, education, and party registration) using an iterative weighting algorithm to reflect the known distribution of those char- acteristics as reported by the Pennsylvania Department of State. The data on polling averages leading up to Election Day come from the Huffington Post Pollster website. County-level election and registration data come from the Pennsylvania Department of State. The election results for 2016 were not official at the time the manuscript was being prepared, so final vote totals by county could differ slightly. The Presidential Race Political scientists expected the 2016 race for president to be closely contested because the election fundamentals predicted a tight contest. Still, the Trump candidacy caused some to wonder if such models would be valid because of his nontraditional candidacy—nontraditional both in the way he campaigned (relying on television advertising less and social media more than traditional candidates) and in the issues he emphasized. Some of Trump’s signature issues were at odds with traditional Republican positions, such as his opposition to free trade agreements, his concerns about NATO and foreign intervention, and his desire to undertake massive infrastructure spending. By the end of September, models that relied on the election fundamentals gave Clinton a modest advantage, but it was also clear that economic conditions and voter fatigue with Democrats were liabilities that had the potential to harm her candidacy (Campbell 2016). A great many contemporary newspaper accounts and some scholarship also documented a sizable disaffection among rural, white, non-college-educated voters that had the potential to carry Trump to victory despite his personal liabilities.3 The following section documents the candidate ratings, pre-election polling, and election outcomes for the 2016 presidential race in Pennsylvania. The 2016 Pennsylvania Presidential and U.S. Senate Elections 5 Candidate Ratings Favorability Both presidential candidates began the general election campaign as well- known among voters and with more voters feeling negatively than positively about their candidacies. In July, more than nine in 10 voters were able to offer an opinion about how favorably they viewed both Trump (95%) and Clinton (96%). Donald Trump was perceived far more negatively than positively by Pennsylvania voters with a minus 29 favorability rating. Hillary Clinton was also perceived more negatively than positively, but her rating was only minus two. The ratings of both candidates, although fluctuating a bit during the campaigns with the most notable movement taking place around the time of the party conventions, eventually ended at about the same place they started, as Table 1 illustrates. One of the unusual features of this election was that a large proportion of voters (16%) had an unfavorable opinion of both candi- dates.4 In 2012, only 3% of voters in Pennsylvania had an unfavorable opinion of both presidential candidates.5 Issues and Perceived Capabilities Registered voters, when asked about the reasons they supported their candi- date, most often said their candidate was the better option; Clinton voters also believed she was more qualified, while Trump’s voters preferred him because he was not a typical politician. Specifically, in response to the question “What is the main reason you plan to vote for [the candidate]?” Clinton’s supporters believed she was the better option (35%), was the more qualified (30%), was more aligned with their political views (17%), or had some desirable personal table 1. Favorability ratings of presidential candidates, pennsylvania 2016 candidate month Favorable Unfavorable net Favorable rating aware of candidate Donald Trump July 0.33 0.62 –0.29 0.95 Aug. 0.37 0.58 –0.21 0.95 Sept. 0.32 0.60 –0.28 0.92 Oct. 0.35 0.62 –0.27 0.97 Hillary Clinton July 0.47 0.49 –0.02 0.96 Aug. 0.38 0.54 –0.16 0.92 Sept. 0.47 0.50 –0.03 0.97 Oct. 0.46 0.52 –0.06 0.98 Source: July, August, September, and October 2016 Franklin & Marshall College Polls. 6 Berwood Yost, Jackie redman, and scottie thompson characteristic (11%). Trump’s supporters most often preferred him because they thought he was the better option (30%), because he was not a politician (17%), because his political views better aligned with theirs (16%), and because they believed he could fix the country’s problems (13%). Registered voters provided consistent responses throughout the campaign about their reasons for supporting each candidate. Registered voters were also relatively consistent in which candidate they believed was better described by a series of statements about their perceived strengths. Secretary Clinton had a clear advantage in having sufficient expe- rience to be president and in her ability to handle foreign policy, as Figure 1 shows. Clinton and Trump were rated similarly, however, in the ability to encourage change and to fix the economy. Figure 1. Percentage of Respondents Who Believe Trump or Clinton Is Best Described by Each Statement, by Month of Survey. Poll question: “Regardless of how you plan to vote, which of these Presidential candidates do you think is best described by each of the following statements?” (Franklin & Marshall College Polls.) The 2016 Pennsylvania Presidential and U.S. Senate Elections 7 Pre-election Polling Poll Averages The average of polls conducted in Pennsylvania suggested that Hillary Clinton led Donald Trump throughout the entire fall campaign. Her monthly average lead was four points in July, eight points in August, five points in September, seven points in October, and three points in November.6 Only rarely did indi- vidual polls show either candidate with support from a majority of voters; the averages in the final week showed Clinton’s expected vote share at 47% and Trump’s expected vote share at 44%. The monthly polling averages showed a sizable number of undecided and other voters that, given the tightening of the race over the final weeks, suggested movement toward Trump. The undecided and third-party candidates totaled nearly 20 points in July, 19 in August and September, 14 in October, and 11 in November. The rate of undecided and third party voters was much higher in 2016 than in 2012. Polls released in the final month of the 2012 campaign found only 5% of voters were undecided or planning to vote for a third-party candidate.7 Figure 2 shows the results of Figure 2. Pennsylvania 2016 Presidential Polling Results, July–November 2016. (Compiled and created by the authors using data downloaded from Huffpost Pollster, December 6, 2016.) 8 Berwood Yost, Jackie redman, and scottie thompson pre-election polls conducted during the fall campaign. Clinton’s share of the vote appeared to be increasing during the course of the fall campaign at the same time the margin between the two candidates converged in November. Support within Demographic Subgroups Much of the media narrative during the 2016 election had to do with sizable differences in the votes of white and nonwhite voters and the votes of college- educated and non-college-educated voters. These differences were appar- ent in Pennsylvania. Besides his expected support among conservatives and Republicans, Donald Trump’s strongest support throughout the campaign came from white, non-college-graduates. In addition, there were geographic differences likely reflecting some cultural or social features amplified by this demographic support within specific areas of the state. Notably, pre-election polls showed Trump performing strongly in western Pennsylvania outside of Pittsburgh. Figure 3 presents the demographic support for the candidates during the fall campaign. Figure 3. Clinton Advantage or Disadvantage in Percentage Points among Demographic Subgroups, by Month of Survey. (Franklin & Marshall College Polls.) The 2016 Pennsylvania Presidential and U.S. Senate Elections 9 Election Results Donald Trump won a narrow and to many a surprising victory over Hillary Clinton in Pennsylvania, 48.6% to 47.9%. Trump’s triumph came from having a significant advantage among late-deciding voters. Nearly all (97%) of the respondents who planned to vote for Trump in Franklin & Marshall’s pre- election polls and who made their final decision in the last week of the cam- paign did vote for him, while only three in four (74%) who planned to vote for Clinton and made their final decision in the last week of the campaign voted for her.8 The other sizable advantage for Trump came from voters who were undecided in our pre-election polls: Trump had a sizable advantage whether these undecided voters decided in the final week or earlier.9 Table 2 shows the composition of presidential voters pre- and post-election in Pennsylvania. Most voters had consistent preferences pre- and post-election. Trump had two major advantages: more voters who supported Clinton pre-election moved away from her than moved away from him, and more voters who preferred neither candidate prior to the election voted for Trump than for Clinton. Post-election Survey Results: Regression Analysis A logistic regression was performed to determine the effects of demographic characteristics (gender, age, education, race, marital status, employment, union membership, and veteran status), political characteristics (partisan affiliation and political ideology), and attitudinal characteristics (Obama job performance, expected future financial condition, and direction of the country) on the likelihood that respondents voted for Donald Trump for president. The logistic regression model was statistically significant, χ2(23) = 1456.02, p < .0001. The model explained 82.3% (Nagelkerke R2) of the vari- ance in presidential vote choice and correctly classified 92% of cases. Model sensitivity was 89.1% and model specificity was 92.6%. Table 3 provides the logistic regression coefficients and confidence intervals for the variables in the model. table 2. change in Voter preferences pre- and post-election as proportion of all Voters, pennsylvania 2016 pre-election preference post-election preference clinton trump neither Clinton 43.5% 0.6% 0.2% Trump — 41.5% 0.2% Neither 4.2% 6.7% 3.3% Source: July, August, September, and October 2016 Franklin & Marshall College Polls. 10 Berwood Yost, Jackie redman, and scottie thompson Republicans were more likely to vote for Trump than were Democrats, and conservatives were more likely to vote for Trump than were liberals. Mod- erates were also less likely than conservatives to vote for Trump. Educational attainment also predicted vote choice: those with a high school degree or less and those who attended some college were more likely to vote for Trump than were college graduates. Those who believed the United States is on the wrong track were more likely to vote for Trump, and those who believed Obama was doing a poor job as president were more likely to vote for Trump. table 3. Logistic regression model for trump preference, pennsylvania 2016 dependent Variable: probability Voting for trump Party independent 0.134 (–0.566, 0.834) Party Republican 1.616*** (1.116, 2.116) Male 0.300 (–0.162, 0.761) Over 55 –0.239 (–0.830, 0.351) Under 35 –0.624 (–1.490, 0.243) White 0.479 (–0.507, 1.466) High school or less 1.238*** (0.654, 1.822) Some college 0.877*** (0.363, 1.391) Liberal –1.758*** (–2.460, –1.057) Moderate –0.760*** (–1.218, –0.301) Urban county –0.193 (–0.702, 0.317) Not married 0.299 (–0.291, 0.890) Single, never married 0.336 (–0.361, 1.032) Not retired or employed –0.375 (–0.968, 0.218) Retired 0.300 (–0.262, 0.863) United States on wrong track 1.120*** (0.515, 1.726) Obama poor job 3.597*** (2.951, 4.244) Obama excellent job –1.520** (–3.009, –0.031) Obama fair job 1.804*** (1.220, 2.388) Better finances next year 0.402 (–0.150, 0.955) Worse finances next year 0.074 (–0.568, 0.716) Labor union Household –0.169 (–0.715, 0.377) Veteran –0.278 (–0.877, 0.321) Constant –3.797*** (–5.203, –2.392) Observations 1,556 Log Likelihood –314.399 Akaike Inf. Crit. 676.798 Note: *p < 0.1; **p < 0.05; ***p < 0.01 Source: July, August, September, and October 2016 Franklin & Marshall College Polls. The 2016 Pennsylvania Presidential and U.S. Senate Elections 11 Comparison with 2012 by County There were strong and consequential regional differences in the presiden- tial election results in Pennsylvania. Hillary Clinton outperformed Barack Obama’s vote totals in nonrural counties by 18,789 votes, but underperformed his vote totals by 82,606 votes in the state’s rural counties. These stark dif- ferences in voter preference are evident regionally. Clinton outperformed Obama by 52,258 votes in the Southeast and by 14,930 votes in Allegheny County, but underperformed his vote in all other parts of the state: Central (34,419), Northwest (34,072), Northeast (29,351), Southwest (28,382), and Phil- adelphia (4,781). Only 11 of the state’s 67 counties yielded more votes for the Democratic presidential candidate in 2016 than 2012. Figure 4 presents the county-level changes in support for the Democratic presidential candidate in Pennsylvania between 2012 and 2016. The counties in Figure 4 are color- coded by geographic region.10 Figure 4. Change in County Presidential Votes, Pennsylvania, 2012 and 2016. Counties are color-coded by geographic region. (Compiled and created by the authors using data downloaded from the Pennsylvania Department of State website, January 24, 2017.) 12 Berwood Yost, Jackie redman, and scottie thompson The U.S. Senate Race Incumbent Senator Pat Toomey was considered one of the most vulnerable Senate Republican incumbents in the 2016 election cycle. Although Toomey had significant crossover appeal, there was a belief that a strong Clinton vic- tory in the state would be too much for him to overcome. During the cam- paign, Toomey emphasized his ability to work with Democrats and cited his efforts to close loopholes in gun background checks as proof. Toomey’s challenger, Katie McGinty, touted her working-class roots and ran as a tra- ditional Democrat while wholeheartedly embracing Hillary Clinton’s candi- dacy. Toomey, on the other hand, seemed deeply concerned that his party’s candidate might harm his chances of re-election. Toomey never endorsed Trump and he would not say whether he intended to vote for Trump until he admitted voting for the candidate one hour prior to the polls closing on election night (Stolberg 2016). Because of Toomey’s perceived vulnerabilities, money from both sides poured into Pennsylvania and the Pennsylvania Sen- ate race became the most expensive in the 2016 cycle; spending in the race was nearly $125 million.11 The following section documents the candidate ratings, pre-election polling, and election outcomes for the 2016 U.S. Senate race in Pennsylvania. Candidate Ratings Favorability Neither U.S. Senate candidate began the general election with high name recognition and incumbent Senator Pat Toomey had clear disadvantages in his personal popularity and job performance ratings as the fall campaign began. In July, fewer than two in three voters were able to offer an opinion about how favorably they viewed Toomey (63%) and only one in two (50%) were able to offer an opinion about McGinty. By the end of the campaign more than three in four (77%) voters were able to offer an opinion about both candidates. Senator Toomey was perceived far more negatively than his challenger throughout the campaign; he had a net favorability rating of minus 17 by the end of October compared to McGinty’s net favorability of plus one.12 Mir- roring the course of the presidential campaign, the ratings of both Senate candidates eventually ended at about the same place they started, as Table 4 illustrates. More of a concern for Toomey than his lack of personal popularity was that only one in three (29%) voters believed he was doing an “excellent” or “good” job as U.S. senator by late October. The 2016 Pennsylvania Presidential and U.S. Senate Elections 13 Pre-election Polling Poll Averages The polling averages in the U.S. Senate race in Pennsylvania showed a rela- tively tight race throughout the fall until there was some late movement toward McGinty as Election Day approached. The polling averages suggested the race was tied in July, but that McGinty’s lead increased to about three points in August, slipped to one point in September and October, and then rose to three points in November.13 Only rarely did individual polls show either candidate with support from a majority of voters; the averages in the final week showed McGinty’s expected vote share at 47% and Toomey’s expected vote share at 44%. The estimated vote share in the Senate race mirrored the presidential vote shares precisely, suggesting there would be a strong relationship between the presidential and Senate votes. Another key feature of this race was the large proportion of undecided and third-party voters reported throughout the cam- paign. Undecided and third-party voters totaled 21 points in August, 24 points in September, 15 points in October, and 13 points in November. Figure 5 shows the results of pre-election polls conducted during the fall campaign. As with Clinton, McGinty’s share of the vote appeared to be increasing during the course of the fall campaign although, unlike the presidential race, the margin between the two candidates appeared to be expanding in November. Support within Demographic Subgroups Pat Toomey’s performance in pre-election polling among demographic sub- groups was similar to Trump’s. He did well as expected among conservatives table 4. Favorability ratings of U.s. senate candidates, pennsylvania 2016 candidate month Favorable Unfavorable net Favorable rating aware of candidate Katie McGinty July 0.25 0.25 0.00 0.50 Aug. 0.28 0.29 –0.01 0.57 Sept. 0.33 0.33 0.00 0.66 Oct. 0.39 0.38 0.01 0.77 Pat Toomey July 0.23 0.40 –0.17 0.63 Aug. 0.29 0.37 –0.08 0.66 Sept. 0.30 0.39 –0.09 0.69 Oct. 0.30 0.47 –0.17 0.77 Source: July, August, September, and October 2016 Franklin & Marshall College Polls. 14 Berwood Yost, Jackie redman, and scottie thompson and Republicans and his pattern of support geographically was also similar. Figure 6 presents the demographic support for the candidates during the fall campaign. Election Results Pat Toomey won a narrow victory over Katie McGinty in Pennsylvania, 48.8% to 47.3%. Table 5 shows the composition of U.S. Senate voters pre- and post- election in Pennsylvania. Most voters had consistent preferences pre- and post-election. More voters who preferred neither candidate prior to the elec- tion voted for Toomey than McGinty. Post-election Survey Results: Regression Analysis A logistic regression was performed to determine the effects of demographic characteristics (gender, age, education, race, marital status, employment, union membership, and veteran status), political characteristics (partisan Figure 5. Pennsylvania 2016 U.S. Senate Polling Results, July–November 2016. (Compiled and created by the authors using data downloaded from Huffpost Pollster, December 6, 2016.) The 2016 Pennsylvania Presidential and U.S. Senate Elections 15 Figure 6. McGinty Advantage or Disadvantage in Percentage Points among Demographic Subgroups, by Month of Survey. (Franklin & Marshall College Polls.) table 5. change in Voter preferences pre- and post-election as proportion of all Voters, pennsylvania 2016 pre-election preference post-election preference mcGinty toomey neither McGinty 38.4% 1.4% 0.2% Toomey 1.1% 36.3% 0.8% Neither 8.5% 10.3% 3.0% Source: July, August, September, and October 2016 Franklin & Marshall College Polls. affiliation and political ideology), and attitudinal characteristics (Obama job performance, expected future financial condition, and direction of the coun- try) on the likelihood that respondents voted for Pat Toomey for U.S. Senate. The logistic regression model was statistically significant, χ2(23) = 1458.47, p < .0001. The model explained 81.7% (Nagelkerke R2) of the variance in U.S. 16 Berwood Yost, Jackie redman, and scottie thompson Senate vote choice and correctly classified 91% of cases. Model sensitivity was 89.3% and model specificity was 93%. Table 6 provides the logistic regression coefficients and confidence intervals for the variables in the model. Republicans and independents were more likely to vote for Toomey than Democrats. Conservatives were more likely to vote for Toomey than were liberals or moderates. Those who believe the United States is on the wrong track were more likely to vote for Toomey, and those who believed Obama was table 6. Logistic regression model for toomey Vote share, pennsylvania 2016 dependent Variable: probability Voting for toomey Party independent 0.658** (0.046, 1.269) Party Republican 1.799*** (1.343, 2.255) Male –0.077 (–0.516, 0.362) Over 55 –0.385 (–0.968, 0.198) Under 35 –0.185 (–1.033, 0.663) White 0.492 (–0.410, 1.394) High school or less –0.212 (–0.755, 0.332) Some college 0.091 (–0.410, 0.593) Liberal –3.069*** (–3.751, –2.387) Moderate –1.228*** (–1.696, –0.760) Urban county 0.109 (–0.398, 0.616) Not married –0.171 (–0.739, 0.397) Single, never married –0.405 (–1.078, 0.268) Not retired or employed –0.930*** (–1.507, –0.354) Retired –0.207 (–0.762, 0.347) United States on wrong track 1.076*** (0.546, 1.606) Obama poor job 2.314*** (1.720, 2.908) Obama excellent job –1.655*** (–2.743, –0.568) Obama fair job 1.495*** (0.958, 2.032) Better finances next year 0.347 (–0.174, 0.868) Worse finances next year 0.121 (–0.544, 0.787) Labor union household –0.531** (–1.052, –0.010) Veteran –0.414 (–1.001, 0.173) Constant –1.310** (–2.580, –0.040) Observations 1,556 Log Likelihood –333.978 Akaike Inf. Crit. 715.956 Note: *p < 0.1; **p < 0.05; ***p < 0.01 Source: July, August, September, and October 2016 Franklin & Marshall College Polls. The 2016 Pennsylvania Presidential and U.S. Senate Elections 17 doing a poor job as president were more likely to vote for Toomey. Labor union households were less likely than nonunion households to vote for Toomey. Comparison with 2012 by County As in the presidential race, the change in support for the Democratic can- didate at the county level was substantial. McGinty received fewer votes than Senator Bob Casey received in 2012 in all but five Pennsylvania coun- ties. McGinty received significantly fewer votes than Casey in Lackawanna and Luzerne counties in the northeastern part of the state and from western counties, including Beaver, Cambria, Erie, Washington, and Westmoreland. Regionally, McGinty underperformed Bob Casey by 5,009 votes in Allegheny County, by 43,107 votes in central Pennsylvania, by 62,708 votes in western Pennsylvania, and by 12,077 votes in Philadelphia and the Southeast. Figure 7 displays the differences in the Democratic U.S. Senate voting by county between 2012 and 2016. Figure 7. Change in County U.S. Senate Votes, Pennsylvania, 2012 and 2016. Counties are color-coded by geographic region. (Compiled and created by the authors using data downloaded from the Pennsylvania Department of State website, January 24, 2017.) 18 Berwood Yost, Jackie redman, and scottie thompson Split-Ticket Voting Increased partisanship has produced less split-ticket voting in recent elec- tions (Beck et al.1992), but Donald Trump’s nontraditional positions on key issues raised the possibility that split-ticket voting might be more common in 2016. Post-election survey data show that among those who cast a vote for both president and Senate, one in 10 (9%) split their tickets.14 A split- ticket voter is defined as a voter who voted for Trump for president but not Toomey for senator, or as a voter who voted for Toomey for Senate but not Trump for president. Figure 8 displays the geography of split-ticket voting in Pennsylvania for the 2016 election. Voters in southeastern Pennsylva- nia cast more votes for Senator Toomey than President Trump, while many more voters in western and northeastern Pennsylvania voted for President Trump.15 Figure 8. Difference in County Presidential and U.S. Senate Votes, Pennsylvania, 2016. Counties are color-coded by geographic region. (Compiled and created by the authors using data downloaded from the Pennsylvania Department of State website, January 24, 2017.) The 2016 Pennsylvania Presidential and U.S. Senate Elections 19 Table 7 presents the regional variation evident in the vote differences between the Republican presidential and Senate candidates in 2012 and 2016 and demonstrates a likely increase in Republican ticket splitting. In 2012, Mitt Romney’s vote totals exceeded Senate candidate Tom Smith’s in every region of the state, with a difference between the two candidates of about 146,000 votes, which was about 5% of the votes cast for Romney. In 2016, Toomey actually outperformed Trump in Allegheny County, Philadelphia, and in the Southeast. The absolute difference in votes cast for the two candidates was about 182,000 votes or about 6% of the presidential vote total. In raw votes, Romney outperformed Trump in Allegheny County and the Southeast, while Toomey outperformed Smith in all parts of the state. These differences hint at the slightly different coalitions assembled by the two candidates; these dif- ferences are discussed in the next section. table 7. regional Vote totals for republican presidential candidates in pennsylvania, 2012 and 2016 region total Vote for president 2016 registered Voters 2016 trump Votes toomey Votes turnout trump minus toomey Votes Allegheny 650,114 920,346 259,480 261,316 70.6% (1,836) Central 1,600,415 2,267,666 979,503 957,106 70.6% 22,397 Northeast 723,607 1,066,205 378,653 352,135 67.9% 26,518 Northwest 515,001 743,183 324,435 307,329 69.3% 17,106 Philadelphia 707,631 1,082,240 108,748 116,714 65.4% (7,966) Southeast 1,345,460 1,788,298 553,906 625,499 75.2% (71,593) Southwest 573,218 791,562 366,039 331,672 72.4% 34,367 region total Vote for president 2012 registered Voters 2012 romney Votes smith Votes turnout romney minus smith Votes Allegheny 634,608 916,470 262,039 236,546 69.2% 25,493 Central 1,491,431 2,167,498 860,579 820,589 68.8% 39,990 Northeast 665,181 1,038,882 305,170 282,261 64.0% 22,909 Northwest 496,141 722,563 275,170 266,370 68.7% 8,800 Philadelphia 690,724 1,055,319 96,467 84,461 65.5% 12,006 Southeast 1,274,517 1,712,173 566,653 534,492 74.4% 32,161 Southwest 543,385 801,859 314,356 284,413 67.8% 29,943 Source: Pennsylvania Department of State, available at http://www.dos.pa.gov/Pages/default.aspx. Calculations by the authors. 20 Berwood Yost, Jackie redman, and scottie thompson Regression Analysis: Split-Ticket Voting A logistic regression was performed to determine the effects of demographic characteristics (gender, age, education, race, marital status, employment, union membership, and veteran status), political characteristics (partisan affiliation and political ideology), and attitudinal characteristics (Obama job performance, expected future financial condition, and direction of the country) on the likelihood that respondents split their tickets between the Republican presidential and Senate candidates. The logistic regression model was statistically significant, χ2(23) = 139.2, p < .0001. The model explained 18.5% (Nagelkerke R2) of the variance in split-ticket voting and correctly clas- sified 90.6% of cases. Model specificity was 90.6%. Table 8 provides the logistic regression coefficients for the model. Independent voters were more likely than Republicans or Democrats to split their tickets. Moderates were more likely and liberals less likely than conservatives to split their tickets. Those who rated Obama’s job performance as excellent were less likely to split their tickets, and those who rated Obama’s job performance as fair were more likely to do so. Voter Turnout High voter turnout was a determinative factor in the presidential and Sen- ate races. More than seven in 10 (70.6%) registered voters cast a ballot in the presidential election in 2016, which was higher as a proportion of registered voters than either 2008 or 2012. Table 7 earlier showed the changes in regional voter turnout between 2012 and 2016. That table showed increased turnout in every region of the state except Philadelphia and that turnout in the Northeast (3.8%) and Southwest (4.7%) showed the largest comparative change. Table 9 shows the regional changes in voter registration from 2012 to 2016. Since 2012, Republican registration increased markedly in western and central Pennsylvania. The patterns of voter turnout were remarkable and are apparent in Figure 9. Figure 9 plots the turnout among registered voters in relation to Republican performance relative to Republican registration in each county. The upper- right-hand quadrant of Figure 9 (turnout above the state average of 71% on the x-axis and Republican performance that exceeded Republican registra- tion, a ratio of 1.0 on the y-axis) shows that the Republican victory was made possible primarily by a surge in Republican performance in western Penn- sylvania. The central Pennsylvania counties, the traditional Republican “T,” underperformed relative to registration. Given the relatively high turnout in The 2016 Pennsylvania Presidential and U.S. Senate Elections 21 southeastern Pennsylvania, these overperforming counties coupled with low turnout in Philadelphia and Centre counties, made the difference for both Trump and Toomey. The changes in turnout and in the Democratic share of the presidential vote between 2012 and 2016 display the county-level dynamics underlying the election. Table 10 shows the changes in turnout and Democratic vote share for table 8. Logistic regression model for ticket splitting, pennsylvania 2016 dependent Variable: probability of splitting ticket for republican candidates Party independent 0.839*** (0.258, 1.419) Party Republican 0.290 (–0.215, 0.795) Male –0.253 (–0.653, 0.146) Over 55 –0.207 (–0.717, 0.303) Under 35 0.265 (–0.443, 0.973) White 0.644 (–0.438, 1.725) High school or less –0.417 (–0.942, 0.107) Some college –0.306 (–0.761, 0.150) Liberal –1.097*** (–1.838, –0.355) Moderate 0.394* (–0.058, 0.846) Urban county 0.101 (–0.359, 0.561) Not married 0.017 (–0.518, 0.551) Single, never married 0.046 (–0.535, 0.628) Not retired or employed –0.061 (–0.574, 0.453) Retired 0.101 (–0.394, 0.597) United States on wrong track 0.064 (–0.470, 0.598) Obama poor job –0.450 (–1.090, 0.190) Obama excellent job –1.946*** (–3.006, –0.887) Obama fair job 0.683** (0.118, 1.249) Better finances next year –0.018 (–0.486, 0.449) Worse finances next year 0.131 (–0.447, 0.708) Labor union household –0.371 (–0.892, 0.149) Veteran 0.425 (–0.095, 0.946) Constant –2.642*** (–4.011, –1.273) Observations 1,556 Log Likelihood –414.800 Akaike Inf. Crit. 877.600 Note: *p < 0.1; **p < 0.05; ***p < 0.01 Source: July, August, September, and October 2016 Franklin & Marshall College Polls. table 9. changes in Voter registration between 2012 and 2016 by region, pennsylvania region dem 2016 rep 2016 dem 2012 rep 2012 dem change rep change % dem change % rep change Allegheny 538,103 258,946 553,099 249,086 (14,996) 9,860 –2.7% 4.0% Central 808,739 1,128,228 803,384 1,063,054 5,355 65,174 0.7% 6.1% Northeast 520,439 384,266 531,068 357,499 (10,629) 26,767 –2.0% 7.5% Northwest 308,086 341,818 325,078 314,900 (16,992) 26,918 –5.2% 8.5% Philadelphia 839,165 123,307 824,130 129,720 15,035 (6,413) 1.8% –4.9% Southeast 788,803 726,612 740,803 717,744 48,000 8,868 6.5% 1.2% Southwest 385,344 319,258 436,961 282,865 (51,617) 36,393 –11.8% 12.9% totals 4,188,679 3,282,435 4,214,523 3,114,868 (25,844) 167,567 –0.6% 5.4% Source: Pennsylvania Department of State, available at http://www.dos.pa.gov/Pages/default.aspx. Calculations by the authors. Figure 9. Registered Voter Turnout and Republican Presidential Performance by County, Pennsylvania, 2016. Counties are color-coded by geographic region. (Compiled and created by the authors using data downloaded from the Pennsylvania Department of State website, January 24, 2017.) The 2016 Pennsylvania Presidential and U.S. Senate Elections 23 five counties, three with above-average turnout and above-average Republican performance and two with below-average turnout and below-average Republi- can performance. Turnout was markedly higher in the three western Pennsyl- vania counties included in the table, while the share of votes won by Hillary Clinton declined substantially in these counties compared to Barack Obama. In the two counties expected to favor Clinton, turnout actually declined in both and Clinton’s share of the vote compared to Obama’s decreased in one of them. This dynamic perfectly captures what happened in Pennsylvania in 2016: counties with more working-class voters turned out in greater numbers and gave less support to Democratic candidates than in previous elections, while areas that should have been supportive of Democrats had lower turnout and offered little change in support. Discussion This analysis shows that significant changes in voter registration and voter turnout in specific areas of the Commonwealth, driven in large part by less- educated voters, those dissatisfied with the current direction of the country, and the performance of the incumbent president, best explain the 2016 elec- tion results in Pennsylvania. Trump overcame his pre-election polling deficit for two major reasons: more voters who supported Clinton pre-election moved away from her than moved away from him, and more voters who preferred neither candidate prior to the election voted for Trump rather than for Clinton. The high number of undecided and third-party voters throughout the campaign should have been a warning that such movement was possible, particularly in light of common electoral models that highlighted Clinton’s many potential vulnerabilities. table 10. turnout and democratic Vote share for selected pennsylvania counties, 2012 and 2016 county region clinton % turnout 2016 obama % turnout 2012 Beaver Southwest 38.9% 73.8% 45.3% 71.3% Cambria Southwest 29.9% 73.0% 40.1% 69.9% Centre Central 48.7% 62.8% 48.5% 65.3% Philadelphia Philadelphia 82.5% 65.4% 85.2% 65.5% Westmoreland Southwest 32.8% 74.1% 37.4% 71.9% Source: Pennsylvania Department of State, available at http://www.dos.pa.gov/Pages/default.aspx. Calculations by the authors. 24 Berwood Yost, Jackie redman, and scottie thompson The foundational importance of Trump’s working-class supporters mani- fested itself by shifting traditional voting patterns in the state. Educational attainment predicted vote choice: those with a high school degree or less and those who attended some college were more likely to vote for Trump than were college graduates. The desire for change was also clear in the election results: those who believed the United States is on the wrong track were more likely to vote for Trump, and those who believed Obama was doing a poor job as president were more likely to vote for Trump. Taken together, these factors led to strong and consequential regional differences in the presidential elec- tion results in Pennsylvania compared to prior elections. Only 11 of the state’s 67 counties yielded more votes for the Democratic presidential candidate in 2016 than 2012. As in the presidential race, the change in support for the Democratic Senate candidate between 2012 and 2016 at the county level was substantial. McGinty received fewer votes than Senator Bob Casey received in 2012 in all but five Pennsylvania counties. Toomey voters were similar to Trump vot- ers in their desire for change: those who believed the United States is on the wrong track were more likely to vote for Toomey as were those who believed Obama was doing a poor job as president. But Toomey had a slightly different coalition than Trump. There were no educational differences among Toomey supporters, he attracted fewer inde- pendents than Trump, and union members were also less likely to support him. His support was more conservative and traditionally Republican than was Trump’s, although there can be little doubt that the surge in Republican turnout driven by Trump’s candidacy made the difference for Toomey. High voter turnout was a determinative factor in the presidential and Sen- ate races. More than seven in 10 (70.6%) registered voters cast a ballot in the presidential election in 2016, which was higher as a proportion of registered voters than either 2008 or 2012. Since 2012, Republican registration increased markedly in western and central Pennsylvania. The Republican victory was made possible primarily by that surge in Republican performance in western Pennsylvania. This dynamic perfectly captures what happened in Pennsylvania in 2016: counties with more working-class voters turned out in greater numbers and gave less support to Democratic candidates than in previous elections, while areas that should have been supportive of Democrats had lower turnout and offered little change in their support for Democratic candidates. Trump’s victory in Pennsylvania scrambles what has been the state’s tra- ditional electoral habits and raises significant and important political ques- tions about future state elections. The main questions include the durability The 2016 Pennsylvania Presidential and U.S. Senate Elections 25 of Trump’s electoral coalition, his ability to transfer his voters’ energy to other Republican candidates in future elections, and the calculus that other elected and aspiring Republicans make in embracing his nontraditional positions. The answers to these questions will make clear the battle lines for future state races, beginning with the 2018 gubernatorial race. NOTES 1. Most of the interviewing for the October survey (625 of the 863 interviews) was conducted prior to the FBI’s statement about Secretary Clinton’s emails on October 28. 2. The Center for Opinion Research at Franklin & Marshall College completed post- election interviews with 2,287 of the 3,077 individuals (74%) who had participated in our July, August, September, and October 2016 pre-election polls. The post-election interviews were completed over the telephone (n = 1,202) or using a self-administered online (n = 974) or paper and pencil (n = 111) format. Post-election interviews were completed from November 16, 2016 to January 13, 2017. The response rates for the post-election survey by month of pre-election interview were as follows: July (n = 487/661) = 74%, August (n = 541/736) = 73%, September (n = 602/813) = 74%, and October (n = 657/863) = 76%. The post-election survey did not verify whether respondents actually voted, so it is possible that some of those interviewed did not vote. 3. An excellent scholarly description of these voters is Cramer 2016. Examples of contemporary news accounts of these voters included Seib and O’Connor, “Republicans Rode Waves of Populism until They Crashed the Party.” Wall Street Journal, October 26, 2016; Bob Davis and Gary Fields, “The Great Unraveling,” Wall Street Journal, September 15, 2016; and Keith O’Brien, “Uprising in the Rust Belt,” Politico Magazine, June 24, 2016. 4. This estimate is confirmed by the 2016 Pennsylvania exit poll estimate: Edison Research exit polls showed 17% of Pennsylvania voters had an unfavorable opinion of both candidates. The exit polling showed that these voters broke for Trump 56% to 31%. 5 These ratings are from the October 2012 Franklin & Marshall College Poll. The poll showed that Mitt Romney’s net favorability rating was minus four and Barack Obama’s net favorability was plus four. 6. Huffpost Pollster 2016a. 7. Authors calculation from data compiled: Yost and Borick 2013. 8. Although the survey asked voters in our post-election interviews when they decided they would vote for their candidate, it did not specifically ask about the effect on their choice of the FBI director’s October 28 statement about reopening the investigation into Secretary Clinton’s email server. 9. During the campaign, many speculated that there was a “hidden” Trump vote. It is possible that those who made their decision in October or before while claiming to be undecided were, in fact, hiding their support for Trump. 10. The regional breakdowns used in this article are those used by the Franklin & Marshall College Poll. For the counties included in each region, see https://www.fandm .edu/fandmpoll/survey-releases. 11. Data were provided by opensecrets.org, accessed January 29, 2017. 12. In the 2012 Senate race, Republican Tom Smith’s net favorability was even, although only about half (54%) of voters recognized him, and Bob Casey’s net favorabil- ity was plus five, according to the October 2012 Franklin & Marshall College Poll (https:// www.fandm.edu/fandmpoll/survey-releases). 26 Berwood Yost, Jackie redman, and scottie thompson 13. Huffpost Pollster 2016b. 14. The proportion of voters who cast a vote for both parties in this election cycle would likely be much higher had I included all the different races on the ballot. For instance, Democratic candidates won all of the state’s row offices in November, with more than 50% of ballots cast. 15. Although there were sizable differences in votes cast, there was a significant cor- relation in vote share. The correlation in county-level vote share for Toomey and Trump is r(65) = 0.95, p < .001. REFERENCES Beck, Paul Allen, Lawrence Baum, Aage R. Clausen, and Charles E. Smith, Jr. 1992. “Pat- terns and Sources of Ticket Splitting in Subpresidential Voting.” American Political Science Review 86 (December): 916–928. Campbell, James E. 2016. “Seeing the Forest for the Trees: Presidential Election Fore- casts and the Fundamentals.” Sabato’s Crystal Ball 14 (September 22). Available at http://www.centerforpolitics.org/crystalball/articles/seeing-the-forest-for-the-trees -presidential-election-forecasts-and-the-fundamentals/. Accessed September 23, 2016. Center for Opinion Research. 2012. “October 2012 Franklin & Marshall College Poll— Statewide.” Available at https://www.fandm.edu/uploads/files/427891958633141665 -keyoct12-1.pdf. Accessed January 27, 2017. Cramer, Katherine. 2016. The Politics of Resentment: Rural Consciousness in Wisconsin and the Rise of Scott Walker. Chicago: University of Chicago Press. Huffpost Pollster. 2016a. “2016 Pennsylvania President: Trump vs. Clinton.” Available at http://elections.huffingtonpost.com/pollster/2016-pennsylvania-president-trump-vs -clinton. Accessed December 6, 2016. ———. 2016b. “2016 Pennsylvania Senate: Toomey vs. McGinty.” Available at http://elec tions.huffingtonpost.com/pollster/2016-pennsylvania-senate-toomey-vs-mcginty. Accessed December 6, 2016. O’Brien, Keith. 2016. “Uprising in the Rust Belt.” Politico Magazine, June 24. Stolberg, Sheryl Gay. 2016. “Patrick Toomey Wins Re-election in Pennsylvania Senate Race.” New York Times, November 9. Yost, Berwood, and Christopher Borick. 2013. “Polling Accuracy in Pennsylvania: The 2012 Presidential Election.” Presented at the Annual Meeting of the Pennsylvania Political Science Association, Harrisburg. Berwood Yost is the director of the Center for Opinion Research and the Floyd Institute for Public Policy at Franklin & Marshall College. He is also the head methodologist for the Franklin & Marshall College Poll. Jackie redman is a project manager at the Center for Opinion Research at Franklin & Marshall College. scottie thompson is a project and data specialist at the Center for Opinion Research at Franklin & Marshall College. Structure Bookmarks Data Sources The Presidential Race The U.S. Senate Race Split-Ticket Voting Voter Turnout Discussion