How Exceptional Was Turnout in 2004?

by Scott L. Althaus

Two years ago in this issue of Political Communication Report, Tom Patterson presented a detailed analysis of the historical declines in American voter turnout. This article revisits Patterson’s analysis to ask whether the 2004 presidential election marked a turning point in this long-term decline. Judging from the television coverage of long lines at polling places around the country, November’s election might seem destined to break historical records for levels of voter involvement. After all, more people voted for president than in any previous election. But did these voters constitute a substantially larger share of the eligible population than in previous years?

In the analysis below, I suggest that while 2004 turnout levels were indeed somewhat higher than those in 2000—particularly in battleground states and in states with gay marriage referenda, as will be detailed below—they did not appear to break any records. An important reason that popular accounts were often so confused about the level of turnout is that there are now several different ways of measuring turnout, compared to one generally accepted method that had been used prior to 2001. To assist political communication researchers in making sense of recent turnout trends, the main part of this article therefore examines the new methods of estimating voter involvement.

Did We Turn Up the Turnout?

On November 2nd, slightly more than 121 million votes were cast for president among a national population of 204 million eligible voters. This 59.6% turnout rate [1] sounds impressive at first glance to those familiar with past levels of turnout, but in context it may seem less so. In 2000, the turnout rate among eligible voters was indeed lower—at 54.2%—but neither of these recent contests surpassed the 60.6% turnout in the three-way 1992 race.

Astute observers of the political process will quickly object that the turnout rates were 51.2% in 2000 and 55.0% in 1992, implying that the 2004 election had substantially higher turnout than any contest since 1968, when unrest at home and war abroad drove nearly 61% of eligible voters to cast a ballot, a figure never surpassed since.

Only if we compare apples to oranges.

The main reason that a 59.6% turnout sounds impressive is that these earlier measures of voter turnout divided the number of ballots cast by a different denominator: the number of voting-age persons rather than the number of eligible voters.

Turnout Ain’t What It Used to Be

There is so much confusion today about measuring turnout that it is worthwhile to take a closer look at the alternative methods, each of which yields a different apparent level of turnout.\

Turnout estimates can use any of three denominators:

• The voting-age population (VAP), which includes non-citizens and felons ineligible to vote, and excludes expatriate citizens who could legally vote from overseas: VAP estimates provide the lowest turnout levels because they underestimate actual turnout.
  
• The vote-eligible population (VEP), a measure developed by McDonald and Popkin (2001), which is voting-age population minus disenfranchised felons minus noncitizens plus eligible overseas citizens: VEP estimates provide (correctly) higher measures of turnout than VAP.
  
• The number of registered voters: REGISTERED estimates should tend to produce the highest apparent levels of turnout because they improperly exclude eligible voters not registered.

My home state of Illinois provides substantive illustration of these different ways of calculating turnout. Illinois cast a total of 5,271,049 ballots on November 2nd. VAP gives an underestimated turnout of 55.4%, VEP gives the correct turnout of 60.3%, and REGISTERED gives an overestimated turnout of 69.9%. Turnout can seem high or low depending on which denominator is being used, so the important thing is to always compare apples with apples.

Back in 2000 there was only one accepted denominator for calculating turnout—voting-age population—because due to variations in the way states handle registration procedures, there is no national estimate of the number of registered voters. But research by political scientists Michael McDonald and Samuel Popkin published after the 2000 election (McDonald and Popkin 2001) showed that using the number of voting-age persons as the turnout denominator shorted actual levels of voting because many adults are felons ineligible to vote, and many more are non-citizens. Using voting-age persons therefore made turnout levels appear lower than they actually were. To correct this problem, turnout estimates made in the last three years have begun using vote-eligible population rather than voting-age persons to calculate turnout levels.

The consensus now in the research community is that VEP is the correct formulation. But since all of the “old” turnout numbers are based on VAP, this discrepancy alone gives a misleading impression that turnout has suddenly risen, when mainly this apparent rise comes from the new denominator that is now the norm.

An additional source of confusion is that there are now two alternate estimates of the vote-eligible population. This means that in 2004 there were three alternative ways of calculating voter turnout, which produce three slightly different turnout trends between 1968 and 2004 (see Table 1).

The first is the “old” measure, which divides votes cast for president by the voting-age population. VAP estimates produce what are to most of us the familiar turnout numbers for past elections, but they are now widely seen as problematic measures that underestimate actual turnout levels.

The other two trends divide votes cast by measures of vote-eligible population, which correct for the errors introduced by VAP alone. The first VEP estimate is the McDonald and Popkin measure. The second VEP estimate is being used by Curtis Gans for the Committee for the Study of the American Electorate report, and was originally developed by Walter Dean Burnham. This way of estimating the vote-eligible population uses a less sophisticated formula than that used by McDonald and Popkin: voting-age population minus non-citizen adults.

These two VEP estimates yield fairly similar results for the last three elections. But the McDonald and Popkin VEP measure produces a 1992 turnout of 60.6%, which is still a full point higher than the 59.6% turnout rate their measure yields for 2004. In contrast, the Gans and Burnham measure suggests a 1992 turnout of just 58.1%, which would be more than two points lower than the 60.3 that their measure yields for 2004.

Thus, the claim that 2004 was a year of record-breaking turnout turns entirely on an arcane methodological debate about the proper way to estimate turnout for 1992. Only the Gans and Burnham formulation suggests that this election’s turnout levels were exceptional. Using either of the other two ways of measuring turnout—VAP or McDonald & Popkin’s VEP—1992 had the higher levels of voter participation.

Which measure is the correct one? McDonald and Popkin’s has been rigorously peer-reviewed, unlike (to my knowledge) the Gans and Burnham measure. Moreover, the McDonald and Popkin formulation makes more intuitive sense than Gans and Burnham’s, which ignores both ineligible felons as well as eligible expatriates. Gans claims that since the number of ineligible felons tends to be about the same as the number of eligible expatriates, both numbers should tend to cancel out and thus can be ignored (Committee 2004: Part II, pp. 1-4). However, this claim is only generally true for the 1994-2004 period, and even in this period it does not always hold. According to the analysis by McDonald and Popkin (2001, Table 1) in every year from 1948 to 1992 the number of ineligible felons was always substantially lower than the number of eligible expatriates. This means that the Gans and Burnham measure, which is based on a flawed premise, incorrectly underestimates actual levels of vote-eligible turnout for the period before 1994.

These factors convince me that the McDonald and Popkin formula provides the superior measure of vote-eligible population. By that measure, 1992 is still the year to beat.

Buoyed in the Battlegrounds

National turnout didn’t set any records this past year, but having advanced this claim, it is nonetheless important to point out that turnout was quite a bit higher than usual in the battleground states.

If we divide the country into battleground and non-battleground states, and then calculate aggregate turnout separately for these two groups [2], we are presented with an interesting contrast. Among the nine battleground states—Florida, Iowa, Minnesota, Nevada, New Hampshire, New Mexico, Ohio, Pennsylvania, and Wisconsin—the total number of votes cast divided by the total population of eligible voters yields an aggregate turnout rate of 66.3%, up 8.4 percentage points from their aggregate 2000 turnout of 57.9%. Among the other 42 states (including the District of Columbia), the aggregate 2004 turnout was 58.9% of eligible voters, up 4.7 points from the 54.2% turnout rate in 2000 for the combined vote in those states. Seen in this way, the relative gain for battleground states was an additional 3.7 percentage points in turnout over non-battleground states.

Mobilized by Gay Marriage

Turnout was also higher in states that voted on referenda banning homosexual marriage. But although this issue has been cited as a major mobilizing factor in the 2004 election, the numbers suggest a somewhat more limited impact.

Among the 11 states with gay marriage referenda—Arkansas, Georgia, Kentucky, Michigan, Mississippi, Montana, North Dakota, Ohio, Oklahoma, Oregon, and Utah—the total number of votes cast divided by the total population of eligible voters yields an aggregate turnout rate of 61.9%, up 7.5 percentage points from their aggregate 2000 turnout of 54.4%. Among the other 40 states (including the District of Columbia), the aggregate 2004 turnout was 60.2% of eligible voters, up 5.1 points from the 55.1% turnout rate in 2000 for the combined vote in those states.[3] Thus, having gay marriage on the state’s agenda appeared to boost relative turnout by 2.4 percentage points. This is a smaller effect than for being targeted as a battleground state, but—especially in the case of Ohio—important nonetheless.

Do We Have a Mandate?

While the levels of turnout in 2004 weren’t historically exceptional, many commentators have suggested that President Bush might claim a mandate on the basis of receiving more votes than any other president in American history, and on the basis of being the first president since his father to win more than 50% of the popular vote. Although the first claim is technically correct, it is also true that John Kerry received more votes than any previous president in American history. Since the first claim becomes an embarrassingly weak leg to stand on, attention has shifted to the second.

It is clear by any measure that President Bush had a clean win in 2004, leading his opponent by 3.2 million votes. But no one to my knowledge has noticed something peculiar about this 3.2 million vote advantage: more than half of those votes come from President Bush’s home state of Texas, where Bush won with a landslide 1.7 million votes over his opponent from Massachusetts.

If we remove the home state advantage, we are left with a Bush lead whittled down to 1.5 million votes, or 1.2% of ballots cast for president. To put this in context, it is instructive to recall that President Clinton was widely perceived to have failed to receive a mandate in 1996 because he won with slightly less than 50% of the vote. However, Clinton’s lead in the popular vote was 8.2 million ballots over Bob Dole, and even if we take away his meager home state advantage (a 150,000-vote margin in Arkansas), he was left with an 8 million vote lead over his opponent, constituting 9.3% of the popular vote cast in 1996. If President Clinton couldn’t claim a mandate out of a lead like that, it is difficult to imagine how President Bush could claim one with a margin of victory less than one-seventh the size of President Clinton’s re-election margin.

Historians may continue to debate the mandate issue—and, it should be pointed out, nearly every president tries to claim one—but no one should conclude that President Bush was re-elected on an unprecedented tide of new enthusiasm for matters political. To the contrary, we as a people were only slightly more motivated to vote in 2004 than we were when Bush was running against Gore. In 2004, as in the previous election, the larger story is the nearly half of eligible voters who stayed at home despite an uncertain situation in Iraq, growing concerns about terrorism on American soil, the prospect of new faces on the Supreme Court, the pressing need for Social Security reform, and a host of important choices that will shape the future of this great country in ways big and small. For this group of stay-at-homes, it was just another Tuesday.

References

Committee for the Study of the American Electorate. 2004. President Bush, Mobilization Drives Propel Turnout to Post-1968 High: Kerry, Democratic Weakness Shown. Washington D. C.: Committee for the Study of the American Electorate. Available here.

McDonald, Michael P., and Samuel L. Popkin. 2001. The myth of the vanishing voter. American Political Science Review 95 (4):963-74.

Scott L. Althaus is Associate Professor of Speech Communication and Political Science at the University of Illinois, Urbana-Champaign.

1. Unless otherwise noted, all of the turnout measures in this article are calculated by dividing the number of votes cast for president by the number of eligible voters. Votes cast for president were obtained by locating the official, certified presidential vote for each state from the web sites maintained by secretaries of state. Official vote totals were collected from the Internet on December 8, 2004. The estimates of vote-eligible population used throughout this article were developed by Michael McDonald and Samuel Popkin (2001). National estimates of vote-eligible population for the 1968–2000 period come from Table 1 of their APSR article, while state-level estimates and the national estimate for 2004 come from the U.S. Elections Project web page maintained by Michael McDonald at George Mason University. The author thanks McDonald and the U.S. Elections Project for making these data available to the scholarly community. [return]
2. In other words, this way of comparing turnout lumps all votes cast in all the battleground states together and divides by the total number of eligible voters in all of those states combined. I refer to this measure as aggregate turnout to differentiate from an alternative method: comparing the average level of turnout in battleground states with the average level of turnout in non-battleground states. Using this approach, the average vote-eligible turnout in battleground states was 66.5%, while the average turnout in non-battleground states was 60.4%. But note that both of these averages are higher than the national turnout rate of 59.6%. The reason for this discrepancy is that average turnout over-represents small states relative to the size of their populations, and small states tend to have higher levels of turnout. It thus gives a misleadingly high estimate of overall turnout, and for that reason I report the population-sensitive measure of aggregate turnout instead. [return]
3. As detailed in note 2, these estimates are also using aggregate turnout. The comparable 2004 figure for average turnout in states with gay marriage referenda was 60.9% of eligible voters, compared to an average of 61.5% for the states without such referenda. This is opposite the pattern in aggregate turnout: using average turnout, voter involvement appears lower in states with gay marriage referenda. As discussed above, these average levels present a misleading picture because they fail to account for between-state differences in population. Most of the states with gay marriage initiatives were small states that did little to influence aggregate vote totals. [return]