Understanding Michigan's Electoral Competitiveness: Swing Counties 1976-2018

Donald Trump's victory in the 2016 presidential election surprised many analysts, experts and general observers of U.S. politics. An important part of his victory came from his popular vote wins in what had been part of a “blue wall” of states that consistently had voted for Democratic presidential candidates between 1992 and 2012 (Goldmacher and Karni 2016). Those so-called blue wall states voting for Trump in 2016 included Pennsylvania, Wisconsin, and Michigan. Michigan saw an especially close vote – Trump's popular vote margin of victory was 0.2 percentage points – only 10,704 votes out of 4.8 million cast (State of Michigan 2016).

Early in the election season, many pundits expected Hillary Clinton to win Michigan in 2016. But as the election drew nearer, Michigan was placed by some analysts into the “swing state” category (Mahtesian 2016). The term swing state has different definitions. As we apply that term to Michigan, it means that the state is competitive enough such that the presidential candidate of either major party could win the state's popular vote, and thus potentially affect the outcome of the election. A state could thus “swing” by supporting one party's candidate over the other from one presidential election to the next. Swing states are important to U.S. presidential elections, because most states in recent elections have consistently supported either Democratic or Republican party candidates. But neither the “red states” (i.e., those that traditionally vote for Republican candidates for president) nor the “blue states” (i.e., those that traditionally vote for Democratic candidates for president) have enough Electoral College votes to elect their favorite party's candidate by themselves. This focus on swing states occurs because the voting results in swing states typically determine the winner of the presidential election, and these states tend to see the most candidate visits, campaign advertising, and media attention during the campaign.

Many campaign scholars and observers reasonably focus on state-by-state voting in presidential elections due to the fact that the Electoral College – through individual states' electoral votes – determines the eventual winner. However, an increasing number of researchers also have begun to look at the importance of county-level voting results, especially in these swing states (Morrisey 2016; Schultz 2018; Wilson 2019). Their argument is that the popular vote outcome in swing states is actually determined by a few “swing counties” in those states. For example, in 2016 there were 206 counties across the U.S. (out of over 3,000 total) that saw a plurality of voters choose Barack Obama in both the 2008 and 2012 contests, but then who voted for Donald Trump in 2016; these so-called pivot counties were thought to be key to Trump's Electoral College victory. Of Michigan's total of 83 counties, there were 12 such counties in 2016 – Bay, Calhoun, Eaton, Gogebic, Isabella, Lake, Macomb, Manistee, Monroe, Saginaw, Shiawassee, and Van Buren. In other words, the election results in a few counties may have great weight on a state's results, and therefore the election result as a whole. As such, they recommend that more attention should be given to the voting trends of these counties to understand better the outcome of a swing state's popular vote, and ultimately, the presidential election. In the remainder of this article, we build on the call for additional research by focusing on swing counties in Michigan over a 42-year timeframe: 1976-2018. We believe such a focus will give us a greater understanding of the electoral competitiveness of the state of Michigan.

Our approach to this study is informed by two sets of research interests. First, we are interested in U.S. voter behavior and the factors that influence how individuals vote. There is an extensive body of research that examines voter behavior in the United States. Beginning in the 1940s, research on voting theorized about the antecedents of an individual's voting choice, while also tracking changes in campaigns (e.g., strategies, communications technology, and laws) and elections. Voting research also developed new theories in response to changes in the voting population over time. Second, we summarize previous scholarly and practical interest in county-level voting statistics, especially as it relates to swing voting results. We begin with a brief summary of the research devoted to understanding voter behavior.

Some of the earliest studies of voting were conducted by sociologists, who tended to focus on the longer-term and more stable social and demographic factors they associated with voting choices and turnout (Lazarsfeld 1948). After World War II, nationwide attitudinal surveys were conducted by other scholars, which allowed researchers to track voter attitudes about both long-term factors (party identification) and more election-specific factors that might influence a person's voting choice (Campbell, et al. 1960). The seminal book that grew out of this new approach, The American Voter – and its update over thirty years later, The New American Voter (Miller and Shanks 1996) – identified voter psychological or emotional attachments to a political party, known as party identification. These researchers theorized that party identification was the single most important factor in determining a voter's choice for president. This theory is also referred to as the Psychological Model of Voting Behavior.

According to the original American Voter study that focused on elections in the late 1940s and 1950s, a person's attachment to a political party was developed early in one's life (often as early as eight to ten years old), and for strong partisans especially, continued throughout their voting lives. Therefore, party attachment was fairly stable, and under normal circumstances, those with strong partisan attachments voted for their preferred party's candidates across multiple elections, regardless of other factors. Along with notions of a “majority party” among voters, candidates running on the party label of the majority party always would win, assuming normal circumstances. This dynamic resulted in what the researchers called maintaining elections. But even this early study recognized that not all voters would support the majority party's candidate across all elections. Campaign specific factors, such as issues and voter evaluations of candidates, could cause enough voters to support the minority party's candidate (while retaining their partisan attachment), which would result in what was called a “deviating election.” In rare elections, voters in enough numbers abandoned their partisan attachments to form a new majority party, as occurred during the 1930s, when the Democratic party became the majority party. Partisan attachments were long lasting, according to the American Voter researchers. They also were a more important voting cue than issues or candidates because emotions often outweigh a person's cognitive sense. Because this theory argued that party loyalty or partisanship was an emotional attachment, it was concluded by many that this finding meant voters were not rational – they simply voted for whomever ran on their favorite party's label.

In some respects, using the notion of partisanship to explain voting behavior has tended to raise the question among many voter behavior researchers since the 1960s: are voters informed and rational? Put another way, do voters simply vote according to party, or do they have other influences that come into play when making their voting choices? One early counter-argument to the American Voter study was offered by V.O. Key, Jr. (1966), who believed that “voters are not fools,” and that voters might look at previous party or candidate performance (called retrospective voting) to influence their voting choice. Moreover, using survey data, Key categorized voters as “standpatters” (those voting for the same political party in two consecutive presidential elections) and “switchers” (those who voted for one party in one election and then for the other party in the next election). He also recognized that voters could be casual (“in and outers”) or could be voting for the first time (“new voters”). Switchers (whom we might now call “swing voters”) constituted between 13 and 20 percent of the voting population at the time of his investigation. This approach is generally considered to be the Rational Voter model, although there is a Rational Choice model as well, which sees voters (and non-voters) as rational entities who make voting choices based on their best economic or cost-benefit interest (Lewis-Beck and Stegmaier 2007).

Beginning in the 1960s, the increasing importance of television (and media in general) in campaign communications led to a new era of “candidate-centered campaigns” and politics, in which campaign communication focused more on candidate image, personality, and messaging, while party attachments as a voting cue became less relevant (Wattenberg, 1991; Menefee-Libey, 2000). By the 1990s, considerable scholarly effort was made in trying to understand and measure split-ticket voters – those who voted for a candidate at the top of the ticket such as president, but voted for the candidate of the other major party for U.S. Senator or U.S. House in the same election. Several possible reasons could explain this phenomenon, including weaker partisanship among voters (and party dealignment), as well as incumbency effects (name recognition, superior campaign war chests, ability to mobilize supporters) that provided counter-influences for the congressional offices. But some scholars also suggested that split-ticket voting could be an intentional strategy by voters to balance power in order to produce a divided government (Fiorina 2002).

However, most scholars and pundits have argued that a substantially increased partisan polarization now exists, which began to be felt more clearly by the mid-to-late 1990s. Increased polarization has meant that far fewer voters over the past 20 years have been ticket-splitters (Skelley 2019) or swing voters – those who vote for one party's candidate in one election than the other party's candidate in the subsequent election (Abramowitz 2010). Swing voters are those “who could go either way; a voter who is not so solidly committed to one candidate or the other….” (Mayer 2008, 2). Moreover, some scholars argue that swing voters (also called floating voters) represent a relatively small percentage of the voting population, especially in the current era of polarization (Smidt 2017).

One of the stumbling blocks in understanding individual swing voter behavior is that there is not much, if any, longitudinal data on individual-level voting behavior that tracks how and why an individual votes (or even chooses not to vote), especially over a long period of time. Therefore, scholars often rely on aggregate data – statewide elections results, or perhaps look at voting results by congressional district. If aggregate voting statistics must be used, we would argue that county-level results are even more useful to scholars, since by using county-level voting statistics, we can shed light on an even smaller subsection of the voting population. While some large-population counties have more than one congressional district in them or parts of more than one, most U.S. counties have populations that are smaller than the size of a congressional district (which is about 710,000 people). This fact allows for a greater exploration of voter behavior using smaller units of analysis. In Michigan (as of 2020), 74 of the state's 83 counties have populations under 200,000 people, and only four counties have populations that are at least the size of a congressional district – Wayne, Oakland, Macomb, and Kent. Therefore, using county-level voting statistics may be one of the best available ways to analyze swing voting across multiple elections.

Drawing conclusions about individual voting behavior from county-level results can lead to false assumptions about how individuals think and act, which is known as the ecological fallacy, and it plagues virtually all researchers who use aggregate data to infer individual behavior in their studies. There have been some attempts to address this problem in political science (King 1977) and to specific aspects of voting behavior such as split ticket voting (Burden and Kimball 1998), but it remains a persistent challenge for scholars. Yet, we believe that county-level analysis is superior to state-level or congressional-level analysis, largely due to the smaller number of people within each unit of analysis. Despite some limitations associated with ecological inference, we would argue that analyzing county voting statistics offers the best available opportunity for understanding swing voting behavior over our 40-year period of study.

While analysis of swing counties shows some promise to scholars, there has been very little research attention given to county-level swings in U.S. elections. A notable exception to this void includes Morrisey (2016) who identified seven key swing counties in swing states with a total of about 2 million voters that he predicted could determine the outcome of the 2016 election. In addition, Schultz (2018) has argued that we can better understand presidential election outcomes by examining election results from swing counties located in swing states. He defines four types of swing counties, which he notes are counties “where candidates compete, which are bellwethers, have flipped, and which are competitive” (383). He then identifies 23 swing counties in 14 swing states that served as bellwethers for their states in 2016. Similarly, Wilson (2019) provides a prediction that identifies the “10 counties that will decide the 2020 election,” which includes some of the pivot counties from 2016, some that are Republican counties where party support is declining, and some that are long-time Democratic areas that voted for Trump in 2016.

Drawing inspiration from these few but useful studies of swing counties, we investigate presidential and gubernatorial election outcomes by county in Michigan between 1976 and 2018. This large and rich data set allows us to identify which counties have had election results that swung from one presidential election to the next, from a presidential election to the following gubernatorial election two years later, and from gubernatorial to gubernatorial elections every four years held during mid-term election years. We believe that when all of the counties' results are examined together across multiple elections, trends and patterns can emerge and help explain the dynamics that contribute to a swing result at the county level. These patterns can be part of a longer trend that occurs over multiple elections, such as a state's or county's changing demographics. They also can be short-term, often election-specific factors, such as candidate quality and personal characteristics, election-specific issues, candidate campaign messaging, campaign spending and advertising, and events that might occur during a campaign that could affect how individuals vote. Moreover, counties could have swung from one election to the next due to a combination of reasons. These reasons can include actual individual swing voting, but also changes in voter turnout, and trends such as the changing political geography of counties or demographic changes occurring within counties over time. When an election outcome is close – as it was in Michigan in 2016 – it is quite likely that even one, or any combination of two or more factors, could contribute to a county-level election outcome, regardless of whether the county swung or not.

Below we build on previous work and offer a set of swing county influences based on factors such as individual voter behavior and the impact of larger contexts and trends within a county. Recall our earlier discussion about the role of partisan attachment as a stable force in influencing voting behavior. For voters with strong partisan attachment, they are loyal partisans who rarely – if ever – vote for a candidate of the major party to which they are opposed. However, there are those who have weaker partisan attachment, who do not always vote straight party ticket, and are more likely to be influenced by candidate personality and quality, incumbency, issues, campaign spending and advertising, and other campaign effects. The following points summarize the factors associated with a county's election results that may contribute to a county's swing from one party's candidate to another across two or more elections. The factors that contribute to swing county voting results include:

• Individual voting behavior, such as split ticket voters and swing voters. Ticket splitters (in enough numbers) vote for one party's candidate at the top of a ballot, but for the other major party's candidate for other offices in the same election. Ticket splitting was common through the mid-1990s, but has decreased in the late 20^th and early 21^st century. Swing voters (in enough numbers), who likely have a weaker partisan loyalty, are considered as likely to vote for one party's candidate as much as the other. Sometimes independent voters and undecided voters are included in this category.

• Voter turnout differences. Such differences can occur due to increased turnout by one party's supporters over the other – which can be due to enthusiasm for a particular candidate. Often there is an associated lack of enthusiasm for the other party's candidate. In sufficient numbers, this trend can cause a swing at the county level. Turnout differences also could be due to superior voter mobilization by one candidate and party over the other. There also is a long-standing pattern of “surge and decline” voting between presidential and midterm elections, in which voter turnout is consistently lower in midterm elections, which affects election outcomes and often penalizes the party of the sitting president.

• Campaign effects, which are election-specific. These are factors related to a given campaign and election cycle. These effects influence voters enough so they ignore their partisan attachment in favor of a preferred candidate regardless of party label – in enough numbers to change the election outcome from the previous election. Weaker partisans are more likely to be influenced by these effects. Campaign effects can include candidate personality and quality, incumbency, candidate messaging, issue positions, campaign spending and advertising, and candidate debate performance.

• Changing demographics, such as increasing diversity or population changes. Growing counties and those increasingly diverse have been trending blue in most areas of the United States. Counties declining in population and which are becoming older on average, are trending redder, as younger people have moved away from home to larger cities for college, for more cultural amenities, and for jobs. While these trends tend to be unidirectional (going blue or getting redder), there can be several elections in which the county's partisan balance is fairly even, leaving the county competitive across one or more elections.

The Michigan counties that switched their partisan support from 2012 to 2016 were part of a nationwide phenomenon that was considered important to understanding the 2016 presidential election outcome. These key swings raised questions about the behavior of county voters in previous elections. We use county-level election results data for presidential elections since 1976 and for the Michigan gubernatorial elections since 1978 (Leip 2019) in order to address several questions that arose after the 2016 presidential election. This data set allows us to measure the number of county swings from presidential to presidential election, from gubernatorial to gubernatorial election, from presidential to gubernatorial, and calculate an overall total number of swings since 1976. We have several research interests as they relate to swing county voting results. Based on previous research and historical voting statistics, we expect that the most county swings occur in the presidential elections to gubernatorial elections. We draw this conclusion because we know that the president's party almost always loses congressional seats during the midterm elections held two years after the presidential election, regardless of the president's party (Campbell 1960). We also are interested simply in how frequently counties swing or flip. To measure county flips, we investigate the total number of county flips since 1976. Therefore, we look at which counties have had the most total flips since 1976. We also are interested in tracking the trend in the number of county swings over the past 42 years. In looking at the combination of elections available to us (presidential to presidential, gubernatorial to gubernatorial, and presidential to gubernatorial), in which pairs of elections or election scenarios are counties more likely to swing? Finally, as part of this analysis, we also identify key counties in Michigan, which we argue are those with larger populations who are frequent flippers and can thus more likely affect an election outcome. Finally, we seek to explore the implications of our study's findings for future campaign strategies and future election outcomes.

Keeping in mind the various possible influences on county-level aggregate swings in election results from one election to the next, we discuss and analyze the 40-year history of county election swings in the state of Michigan. Using such a large data set, we hope to identify any patterns that extend beyond a single election or two. We begin with presidential-to-presidential election county-level swings.

In order to understand the recent history of Michigan's swing counties in presidential elections, we asked a basic question: How many times between 1976 and 2016 have a majority of Michigan's statewide voters voted for one major party's candidate in one presidential election, then voted for the other major party's candidate in the next presidential election? Given the somewhat surprising results in 2016, these elections are a good place to begin. We know Michigan overall has not swung very much since 1976 – only twice (1992 and 2016). Aside from those two elections, the state consistently supported Republican presidential candidates in the 1970s and 1980s (Ford, Reagan twice, and George H.W. Bush), or was part of the blue wall that supported Democratic candidates from 1992 through 2012.

How about county swings in Michigan during this period? Some counties swung more than twice as much than the state did overall. Michigan was not necessarily the secure blue state that many believed it to be prior to 2016. The fact that the state of Michigan overall supported Democratic presidential candidates in the six elections from 1992 through 2012 arguably hides two facts. First, several of those presidential elections were quite competitive in Michigan. Except for the relative landslides in 1996 and 2008 (both of which also were large nation-wide victories for the Democrats), the other four elections between 1992 and 2012 saw a popular vote difference of less than ten percent in Michigan, and in two of those elections, the vote difference was less than five percent (in 2000 and 2004). Michigan remained a relatively competitive state during the 1990s and into the 2000s despite being considered part of the Democratic blue wall (Dulio and Klemanski 2018). In fact, some observers have considered Michigan to be a “traditional swing state,” which generally means the election outcome is expected to be close and that the state receives increased attention from candidates and the media (Silver 2016).

Second, as we alluded to above, the consistent support of Democratic presidential candidates statewide masked a surprising number of county flips during this time. After Donald Trump's election and his narrow popular vote win in Michigan, many media outlets gave considerable attention to the 12 counties in Michigan that flipped in 2016 – and rightly so. However, of the 12 counties that flipped to Trump in 2016, half have had a consistent history of flipping since 1976. Six of the pivot counties – Monroe, Macomb, Manistee, Shiawassee, Calhoun, and Van Buren – have flipped between 40 percent or 50 percent of the time from one presidential election to the next in the period between 1976 and 2016.

There have been ten possible flips in Presidential elections between 1976 and 2016 (i.e., in the period we investigated, there were 11 presidential elections with the first possible occurring in the 1980 election. Table 1 indicates all of the 25 Michigan counties that flipped at least 40 percent of the time during this period.

As Table 1 shows, presidential-to-presidential election swings have occurred on a somewhat regular basis, with 25 counties (out of 83) flipping at least 40 percent of the time. These four or five changes compare favorably to five nationwide flips and would seem to suggest that these Michigan counties were bellwethers – voting for the eventual winning candidates – since 1976 (1980, 1992, 2000, 2008, and 2016).

However, only two flips have occurred statewide in Michigan during that time (1992, 2016). The greater number of county-level flips has occurred because most of the counties listed in the table above have relatively small populations. Several of these counties are in the sparsely populated Upper Peninsula (Alger, Delta, Schoolcraft, and Ontonagon). Others are located in the northern Lower Peninsula, which also is mostly rural with relatively small populations (Benzie, Alpena, Presque Isle, and Iosco). As such, these swings have had little effect on the state's overall total vote, and Michigan remained part of the blue wall of Democratic support from 1992 through 2012 in presidential elections. We now turn to possible swings from one gubernatorial election to the next.

As we saw above, Michigan's county-level voting results from presidential cycle to presidential cycle showed a bit more variation than the blue wall from 1992 to 2012 would have suggested. But most of the swings occurred in counties that had relatively small populations, so the overall statewide vote remained consistently blue. But this first investigation into county swings suggests that Michigan has some potential as a swing state, due to the state's electoral competitiveness over time. In addition to the presidential elections held every four years, Michigan governors are elected every four years in what are considered midterm election years. Midterm elections are those that take place at the midpoint of a president's term. In the gubernatorial elections between 1978 and 2018, there have been a fairly large number of county flips from one gubernatorial election to the next. Even counties that are known to be strongholds for one party or the other have seen at least one flip during this period of time. Table 2 shows the number and percentage of county flips in gubernatorial elections since 1978.

Compared to presidential-to-presidential elections, Michigan counties have tended to flip much more in gubernatorial election cycles. Out of ten possible flips between 1978 and 2018, 37 counties (45 percent) have seen a flipped result from one cycle to the next at least 40 percent of the time during this period from one gubernatorial election to the next. One county – Manistee – located in northern Lower Peninsula southwest of Traverse City – has flipped six out of ten possible times in gubernatorial elections. Thirteen counties have flipped five times between 1978 and 2018. These counties range from some of the largest in population (e.g., Oakland, Macomb) to very small (e.g., Alger, Baraga), and are located widely throughout the state. Another twenty-three counties have flipped 40 percent of the time (i.e., four out of ten elections) during this period. The location of these flipped counties ranges throughout the state as well, from the Upper Peninsula counties of Chippewa and Marquette, to northern Lower Peninsula counties such as Alpena and Benzie, down to the far southwest (Van Buren) and southeast (Monroe) areas of the state. In terms of statewide gubernatorial results, Michigan has flipped a total of five – out of ten – times since 1978 (1982, 1990, 2002, 2010, 2018). Gubernatorial flips have been more common during this period than presidential flips. More county flips, including some by larger population counties, have led to more statewide flips as a result.

While county-level voting patterns from presidential election to presidential election and gubernatorial election to gubernatorial election are important, both of these types of elections take place four years apart and do not consider what happens from election cycle to election cycle every two years. In other words, important voter behavior reflected by county-level patterns may be hidden by looking only at elections for the same office. When campaign managers and consultants advise candidates, they typically compare only similar elections (i.e., presidential to presidential or gubernatorial to gubernatorial) because of the different dynamics and contests that are often present (e.g., higher or lower levels of voter turnout; different degrees of media coverage and voter interest, different levels of campaign resources available to campaigns). But for a fuller picture of what Michigan looks like politically, we offer an assessment of the 2-year cycle of presidential to gubernatorial election outcomes.

The history of voting behavior in the U.S. has shown a consistent pattern between presidential elections and midterm elections. The president's party almost always loses congressional seats in the midterm election that follows the president's election two years before. The so-called presidential midterm slump has been consistent with only a few exceptions for many years and election cycles. There remains some debate about the causes of the midterm slump. One theory argues that the slump is due to the surge and decline of total votes – and in the composition of the electorate that votes in presidential elections compared to those who vote in midterm elections (Cover 1985; Campbell 1987). Another approach argues that the president's party is punished by voters unhappy with the president's performance (Tufte 1975) and that negative voting against the president is a major feature of midterm congressional elections (Kernell 1977).

While most of the surge and decline studies have focused on congressional elections, one notable study looks at the interplay between presidential and midterm gubernatorial elections (Erickson et al. 2015). The Erikson study shows that voters punish the president's party in midterm gubernatorial elections (as a party balancing strategy) and that a presidential candidate from the same party as a sitting governor loses several percentage points of the vote share in the presidential election following a gubernatorial election.

Whatever the causes of county flipping, Michigan has experienced a consistent pattern of election outcomes when comparing presidential elections results with the state's gubernatorial elections results. Since 1976, only one pair of consecutive elections saw Michigan vote for a presidential candidate (George H.W. Bush in 1988) and a gubernatorial candidate of the same party (John Engler in 1990). In every other set of elections, Michigan voted for the gubernatorial candidate who was of the opposite party as the president. This pattern is worth exploring and by using county-level data, we hope to understand further the factors that contribute to swing voting and swing county election results.

Based on the historical patterns of midterm slumps and the scholarly research that has sought to explain those patterns, we expect to see substantially more swings in presidential to gubernatorial elections than in elections involving the same office. Indeed, almost half of Michigan counties (38 counties, or 46 percent of all counties) had at least five swings (or at least 45 percent of the 11 swing possibilities) from presidential to gubernatorial elections between 1976 and 2018. Table 3 shows the number and percentage of counties that had swings from a presidential election to the following gubernatorial election.

As Table 3 shows, some of these counties had many swings – a total of 38 Michigan counties (42 percent) had at least five flips out of a total of 11 possible elections. At the high end, eight counties had eight or nine swings out of eleven pairs of elections, or roughly three-quarters or more of the elections. Nineteen counties had six or seven swings, flipping in a little over half to almost two-thirds of the cycles. Eleven counties had five swings, which approached half of all the election pairs from the period under study. As with the other swings investigated, the counties with a large number of swings were located throughout the state. In the case of presidential to gubernatorial elections, however, they also included some of the more populous counties in the state, such as Macomb, Ingham, Muskegon, Monroe, Kalamazoo, and Saginaw, which likely contributed heavily to the flipped outcomes. County swings from gubernatorial to presidential elections showed a similar pattern – a fair number of counties flipped (two counties flipped 70 percent of the time, and 18 counties flipped in at least 50 percent of possible election pairs). Moreover, the counties that frequently flipped from gubernatorial to presidential elections also included some of the more populous counties in the state – Oakland, Macomb, Kalamazoo, Ingham, Saginaw, Monroe, and Calhoun.

The fact that almost half of Michigan counties have been frequent flippers between presidential and gubernatorial elections is notable. One obvious example of a presidential-to-gubernatorial flip occurred in the 1984 presidential election and the 1986 gubernatorial election. In 1984, Ronald Reagan won a national landslide victory over Walter Mondale. The State of Michigan gave Reagan a 59 – 40 percent victory (Gibbons 2019). President Reagan had a somewhat controversial first term, but by the time of his re-election, he had the support of many different groups of voters. Among them were “Reagan Democrats” – those voters who considered themselves Democrats but who voted twice for Ronald Reagan. Indeed, Macomb county blue-collar workers were emblematic of the Reagan Democrats identified during the 1980s (Greenberg 1996). In 1984, President Reagan won the popular vote in all in but four Michigan counties. Gogebic, Keweenaw, and Iron counties in the Upper Peninsula, along with Wayne, were the few counties whose majorities voted blue in 1984. However, just two years later, Michigan voters re-elected Democratic Governor James Blanchard, who took 68.1 percent of the vote across the state in 1986. Blanchard's Republican opponent, William Lucas, won a plurality of votes in only one county (Ottawa) that year. The contrasts between these two elections are remarkable in such a short period of time. There are many likely factors contributing to the presidential-to-gubernatorial swing in Michigan in the 1980s. However, in both elections, it was likely that battle-tested incumbents were viewed as more capable of dealing with difficult challenges compared to their less-experienced challengers. With such substantial differences in partisan support in just two short years, it's likely that some combination of factors that contribute to county swings had a hand in causing this swing in 1986.

A more recent cycle with a large number of county flips in the plurality of voters county-wide occurred between the 2008 presidential and 2010 gubernatorial elections. The 2008 presidential campaign was an open seat race that featured Democratic nominee Barack Obama and Republican nominee John McCain. In Michigan, the state's popular vote advantage for Obama was 16 percentage points (57-41). Obama won a plurality of votes in 46 Michigan counties, including the top seven largest counties by population. His opponent John McCain, won a plurality of votes in 37 counties. Figure 1 illustrates the county-level support for Barack Obama in the 2008 election, including a fairly widespread geographic support from county voters across the state.

A short two years later, the 2010 Michigan gubernatorial election was held. This election was “open seat” (with no incumbent running), which pitted Republican businessman Rick Snyder against Democratic Lansing Mayor Virgil “Virg” Bernero. There were a number of national issues at play during this election (e.g., the Wall Street “bailout,” Obamacare, and a struggling economy) as well as related state-level issues in Michigan (especially the state's economy, which was faring worse than the national economy on measures such as the unemployment rate). Snyder handily beat Bernero, with a 58 to 40 percent victory in the statewide vote and capturing the plurality of votes in all but four counties. This substantial turnaround from Democratic support in 2008 to Republican support 2010 occurred in other U.S. states (e.g., Democrats lost a total of 63 seats in the U.S. House that year). But the large number of county swings in only two years shows the competitive nature of Michigan electoral politics. Figure 2 illustrates the county-level support for Rick Snyder in 2010. The reasons for the many county swings from 2008 to 2010 include the influence of issues, some swing or split-ticket voting due to candidate messaging and quality, and surge and decline turnout differences (2008 voter mobilization by Democrats and the Obama campaign, with 2010 voter mobilization by Tea Party activists and Republicans).

These results appear to be similar to the surge and decline pattern that we have previously discussed. Candidates of the president's party do not do well in the elections for governor that immediately follow that president's election. It is entirely possible that the effects felt at the national level trickle down to the gubernatorial level in Michigan where the voters are in a mood to punish the party of the president. Indeed, the pattern is too clear to ignore. Another possible explanation is that Michigan voters might want to see a partisan balance of government executives as they make their voting choices. There are likely a number of other factors in play here (including partisan control of the Michigan Legislature), but those are beyond the scope of this article. The differences in election results between presidential and gubernatorial elections are powerful examples of how competitive Michigan has been in the past, and how quickly voters in Michigan can swing from a candidate of one party to a candidate of the other party.

There are other possible explanations. The quality of candidates and the effectiveness of their campaign messages could potentially influence voters. These factors are possible contributors to the county swings between 2008 and 2010, especially in counties that frequently flip from one election to the next. Frequent flippers are more likely to be responding to election specific factors such as candidate quality, campaign messaging, and campaign advertising. Moreover, voters who frequently swing between elections are more likely to be weaker partisans. Without a strong partisan loyalty, these voters will tend to vote for the candidate, not the party, when making their voting decisions.

Surprisingly, there has been only one county in Michigan that has never flipped in gubernatorial or presidential elections during our period of study – a plurality of voters in Ottawa, located in the western Lower Peninsula, and whose largest city is Holland, has chosen the Republican candidate for governor and president in every election since 1976. All other counties have experienced at least two flips during this time period. The evidence presented thus far is clear: a fair number of Michigan counties show swings in presidential election cycles and in gubernatorial election cycles, but especially in presidential-to-gubernatorial election cycles. Given the patterns in statewide outcomes from election cycle to election cycle, we also wanted to explore the overall pattern of swings across all elections that were held during our period of study. Examining all elections during the period under investigation can give us even more insight into how Michigan votes and may uncover other identifiable trends in county flips from the 1970s on.

There have been 22 presidential and gubernatorial elections between 1976 and 2018. As such, there have been a total of 21 possible flips during this time. We begin counting possible flips after 1976, which is the first election in our investigation period. We then consider each subsequent election as another possibility that a county could flip. In looking at all possible flips during this period, 29 Michigan counties (35 percent) experienced flipped county-wide outcomes in about half or more of the elections held. Baraga County, in the Upper Peninsula, has seen flipped outcomes in 16 out of 21 elections – a remarkable three-quarters of all elections investigated (76 percent). Monroe County, located in the far southeastern part of the state, has flipped in 15 out of 21 elections (71 percent). Table 4 shows the total number of flips by counties that have swung in about half or more of the 21 possible elections since 1976.

The counties in Table 4 can be regarded as frequent flippers. We have previously speculated about why these counties flip more frequently than others. There could be demographic explanations – some of these counties have been very competitive politically and could have roughly equal numbers of Republican and Democratic voters living there, so only small increases in turnout supporting one candidate or the other can make a difference in who wins the county. There seems to be a consistent pattern of electing the gubernatorial candidate of the party opposite to the sitting president – across virtually all elections. This trend could be part of the surge and decline phenomenon that we have discussed earlier. Enough voters in these counties might be weaker partisans who do not simply vote straight party ticket in every election. As such, voters in these counties also might be more responsive to campaign effects – for example, candidate characteristics and quality, candidate messages, voter enthusiasm and turnout differences – more than in other counties. In other words, more voters in these counties, more so than voters in other counties, may make their voting choices based on their evaluation of the candidates and the issues, or because of other election-specific factors.

As mentioned above, the Michigan counties that deserve special attention in this analysis are those that both have relatively large populations and that swing frequently. These counties can make a big difference in statewide election results for both president and governor. In addition, some of these counties can be considered key counties in our analysis (some scholars and practitioners use the term “bellwether” to indicate these counties can predict the statewide outcome). David Schultz includes bellwether counties as examples of swing counties in his study (2018). For purposes of this study, one important type of key county is simply a large population county. By its sheer size, such a county can influence a statewide election outcome – by either a larger or smaller turnout in a given election. The five largest population counties are Wayne, Oakland, Macomb, Kent, and Genesee. These five counties comprise almost half of the state's total population. As such, some observers argued that a relatively low turnout by Wayne (especially in Detroit) and Genesee (Flint) County voters helped Donald Trump achieve his narrow popular vote victory in Michigan in 2016 (Livengood et al. 2016). This finding has obvious implications for candidates and campaigns running for election in Michigan. In that regard, we would argue that campaigns thinking strategically would focus their campaign efforts on counties such as Macomb, Monroe, Muskegon, Calhoun, and Saginaw.

All of these counties are among the top quintile of the largest population counties in Michigan. Four of these larger population counties – Macomb, Calhoun, Saginaw, and Monroe – were among the 12 pivot counties that helped Donald Trump win the popular vote in Michigan in 2016. In looking ahead, we believe that Muskegon County also should be considered in this analysis. While it so far has consistently supported Democratic presidential candidates since 1992, Hillary Clinton won the county by only 1,200 votes in 2016 (Wilson 2019). For smaller counties that frequently flip (Manistee, Baraga, Clare, Roscommon), their support for one party's candidate over another across multiple elections will not have the same impact in a statewide election as much large counties will. So, while more numerous, the counties in the northern part of the Lower Peninsula and all of the Upper Peninsula are relatively small by comparison. As such, their contribution to the statewide total vote and impact on the overall result in Michigan is quite modest. Whether these counties swing or not therefore will not necessarily have a major influence on overall statewide vote totals.

The swing counties that contribute many more votes to a statewide total tend to be located in the southern portion of the Lower Peninsula. Macomb is Michigan's 3^rd largest county (almost 900,000 population as of 2018), Saginaw is the 11^th largest (191,000), Muskegon is 12^th largest (173,000) Monroe is the 16^th largest (150,000), and Calhoun is the 17^th largest (135,000). Of this group, Macomb County stands out as historically important; the county vote totals frequently flip and it casts many votes. Historically, it also has served as a bellwether county, as Macomb has voted for the ultimate winner in most elections since 1976. For presidential elections, Macomb has voted for the statewide winner nine out of 11 times (82 percent) between 1976 and 2016. For gubernatorial elections, Macomb voters have supported the ultimate winner an impressive ten out of 11 times (91 percent) since 1978.

Figure 3 illustrates the relative sizes of counties in terms of their contributions to the total state vote in the 2016 election. Note the relatively larger sizes of counties in the metro Detroit area compared to the smaller counties in northern Lower and in the Upper Peninsula.

County population size certainly does matter in these elections. For example, Wayne and Oakland are the state's two largest counties, with a combined 2018 population of about 3 million people. This total is about 30 percent of the entire state's population. By contrast, the state's smallest county by population is Keweenaw, the state's northernmost county. This county, which juts out into Lake Superior, had a 2018 population of only 2,113 (U.S. Census Bureau 2018). Keweenaw's county-wide vote total in the 2018 election was only 1,348. For comparison, this total number of voters in Keweenaw might compare to the average voting precinct size in a large city. For example, most of the 45 precincts in the City of Sterling Heights in Macomb County had between 1,000 and 1,500 voters cast a ballot in the 2018 general election (State of Michigan 2018b).

Indeed, the combined votes for the smallest ten counties in the 2018 election totaled only 38,007 votes. Of these smallest ten counties, eight are located in the Upper Peninsula. In addition to Keweenaw, the small Upper Peninsula counties included Ontonagon, Luce, Schoolcraft, Baraga, Alger, Montmorency, and Mackinac. The two Lower Peninsula counties in the bottom ten were Alcona and Oscoda (Mack 2018). Table 5 shows the vote totals in 2018 for the top ten counties by population, as well as the percent each county contributed to the state's overall vote.

As Table 5 illustrates, the ten most populous counties in Michigan delivered almost two-thirds of the total statewide vote (64 percent) in 2018. If these counties swing in a given election, their impact will be much greater than most of the counties in Michigan. Of these ten most populous counties, Macomb had the most flips, with 13 out of 21, followed by Ingham and Kalamazoo counties, with 11 each. Previous studies have indicated that the more recent era of partisan polarization has likely reduced the number of swing counties and increased the number of counties solidly in support of one party over the other. Approximately 60 percent of all U.S. counties were won in a landslide – 20 percentage points or more by one of the major party candidates in 2016 (Aisch et al. 2016). Below, we examine party competition by each decade within the scope of our study (1980s, 1990s, 2000s, and 2010s).

While some Michigan counties have experienced election results that have flipped from favoring one party's candidate to the other, a remaining question is whether the frequency of these flips has changed over time. After all, for many years, observers have noted a strong association between urban areas and support for Democratic candidates and rural areas and support for Republicans. More recently, Americans may also be “sorting” themselves into geographic locations where they choose to live with others who share their values and political beliefs (Bishop 2009). We have previously noted how increased partisan polarization has reduced the number of swing voters and competitive counties as well. With these factors taken into account, it is likely that the number of competitive counties, and therefore flipped county-wide results, will have decreased over time, and the number of non-competitive counties will have increased.

To find any trends over time, we examined the number of times a county flipped from one party's candidate to another (in both presidential and gubernatorial elections) and analyzed flips by the decades of the 1980s, 1990s, 2000s, and 2010s. The total possible number of elections in each decade is five, so a party's presidential or gubernatorial candidate could have won a plurality county-wide votes in all five elections, in four out of five, or three out of five (with the loser winning none of the five elections, or one out of five, or two out of five). We would label those counties that see a 3-2 split in election outcomes “competitive counties,” those where a party's candidates win four of five contests as “somewhat competitive,” and counties where candidates of one party win all five contests as “one-party dominant counties.” Given the increased sorting of voters described by Bishop and others, we hypothesized that the number of counties that had flipped would decrease over time, with a corresponding increase in one-party dominant counties. Table 6 shows the competitiveness of parties in counties by the decades we investigated.

As Table 6 illustrates, over time there is a substantial decline in the number of counties that were “competitive” during a decade and a sharp increase in counties that were “one-party dominant” in a decade. In the 1980s, only four counties voted consistently for one party's candidates in all of the five elections of that decade. Almost one-third (31 percent) of the counties were split three elections to two and could be considered competitive. By the 1990s, over one-quarter of Michigan's counties (27 percent) were one-party dominant – winning all five elections. But over half of the state's 83 counties were still competitive in the 1980s (splitting the five elections 3-2). By the decade of the 2000s, almost half of Michigan's counties were one-party dominant (46 percent). The number of competitive counties dropped to 24 (29 percent), but this level of competitiveness still was similar to the 1980s, which had 26 competitive counties. However, by the 2010s, almost three-quarters (73 percent) of all Michigan counties were one-party dominant. While other counties might flip from time to time, only eight counties (10 percent) could be considered competitive during this decade. Table 6 shows a decline in competitive counties (which also means they are less likely to flip), but illustrates an even more consistent and clear increase in the number of one-party dominant counties with each passing decade.

This trend of decreasing competitiveness of Michigan counties in an era of increased partisan polarization does suggest that campaigns should consider focusing more on increasing voter turnout by their supporters rather than trying to persuade weak partisans or undecided voters. The era of swing voters – or of ticket splitting voters – may be largely over. One increasingly accepted view by political scientists is that turnout is the new swing voter or that “turnout explains everything” (Freedlander 2020).

Our analysis of the voting results in Michigan counties has revealed useful results for understanding Michigan's electoral competitiveness, as seen through the dynamics of presidential and gubernatorial elections in Michigan. First, we found that while Michigan's counties have experienced a substantial number of swings between elections, the number of counties considered to be competitive has declined since the 1980s. Even more obviously, the number of non-competitive counties – which we call “one-party dominant” counties – has consistently increased over the past 40 years.

By the decade of the 2010s, only eight counties (10 percent) are considered to be competitive. This small number can be compared to the decade of the 1980s, when almost one-third of Michigan's counties could be thought of as competitive. At the same time, only four (5 percent) of Michigan's counties in the 1980s were one-party dominant. By the decade of the 2010s, almost three-quarters of the state's counties (61, or 73 percent) were one-party dominant. This trend is similar to the geographic sorting and partisan polarization that has been occurring elsewhere in the United States.

Next, we found identifiable differences in the types of elections in which Michigan counties were most likely to have flipped results. The fewest county flips occurred from one presidential election to the next. While about 25 counties have flipped 40 to 50 percent of the time over the past 40 years, most of them were too small to affect the statewide outcome. Indeed, Donald Trump won a surprise popular election in Michigan in 2016 because regular flipper Macomb County was joined by several larger population counties (Bay, Saginaw, Calhoun, and Monroe) that previously had voted for Barack Obama in 2008 and 2012.

We found that the next most common flips occurred in gubernatorial to gubernatorial elections. Almost half of Michigan counties (37, or 45 percent) flipped in at least 40 percent of the 11 gubernatorial elections between 1978 and 2018. This changing dynamic did create more statewide flips as well, as there were five (out of ten) flipped gubernatorial results since 1978. Michigan voters also have a pattern of electing governors of the opposite party to the sitting president. In over 40 years, the only time a presidential-to-gubernatorial party flip did not happen was in 1990, when Michigan voters elected Republican John Engler, while Republican George H.W. Bush was the sitting president.

We also found that the most frequent county flips occurred between the presidential and the gubernatorial elections that followed them two years later. Almost half of Michigan's 83 counties (38, or 46 percent) flipped in at least five out of the 11 presidential-to-gubernatorial elections beginning in 1976. Eighteen counties (21 percent) flipped at least half the time from a gubernatorial to the following presidential election, and several of those counties had large populations. It is also worth remembering that there is an apparent surge and decline in voter turnout occurring between presidential and gubernatorial elections, which previous scholars have seen in other states. Indeed, Michigan voters have elected gubernatorial candidates of the sitting president's opposing party in every set of elections since 1976, except for the 1988 presidential and 1990 gubernatorial elections.

Given the decreased overall number of swing counties in Michigan, the importance of the counties that do continue to swing make them key factors in determining the state's election outcomes, especially in close races. We identify two types of counties that are crucial in influencing a statewide vote total, which we call “key counties.” One type is comprised of the largest population counties. By virtue of their size, changes in their voter turnout alone can affect a statewide outcome. The top five counties in Michigan have about half of the state's voting population, and the top ten counties have about two-thirds of the state's voting population (see Table 5). Counties don't vote, people do – and the influence of the state's population centers cannot be understated. Many – but not all – of the state's largest counties can be considered one-party dominant. Among the top five largest counties, Macomb County has frequently flipped, and Kent County is becoming bluer in recent elections. While large counties such as Wayne and Oakland appear to be blue now reliably, the election outcomes in shifting counties such as Kent and Macomb will continue to be of interest to campaign professionals and scholars.

The second type of key county is the county that swings regularly or even frequently. While fewer in number than in previous decades, these counties can still influence an election outcome because the state has remained fairly competitive overall. As such, we would argue that any medium-sized to large population county that is still relatively competitive will be a key factor in determining the outcome of a future presidential or gubernatorial election in Michigan. Aside from the top five largest counties, other counties that have been key factors in previous elections include Monroe, Muskegon, and Saginaw. These counties include Bay and Genesee, which also could play an important role in upcoming elections.

Of course, underlying the county flips are important election dynamics such as voter turnout and the partisan loyalties of voters. In both of these cases, voters may respond to issues and candidates in a variety of ways, including exhibiting greater enthusiasm for candidates (and thus turning out to vote) or selecting one candidate over the other based on election-specific factors rather than simply voting according to partisan loyalty. Over the decades, voters in Michigan counties have appeared to set aside party loyalties in order to vote for the candidates that they deem to be the most capable to lead the nation or state. Michigan's status as a battleground state appears to be secure, as influenced especially by two types of key counties – those that have large populations and whose turnout differences can have an impact on election outcomes, and those counties that have a history of swing voting results.