Who Is Really Voting?

Why you shouldn’t trust exit polls—and why the 2020 electorate is much older than you think.

It’s a familiar lament: the Democratic electorates in the nation’s first presidential contests are quite different from those elsewhere in America. Far less known and much less discussed: how wildly un-representative those voting so far in Iowa, New Hampshire, and Nevada,  are when compared to those states’ typical Democratic electorates.

The caucus system itself is the most visible culprit. Just 30 percent of Iowa’s 600,000 registered Democrats participated in the February 3 caucus. Meanwhile, Democratic turnout in Nevada was closer to 15 percent. Former Senator Harry Reid is right: both caucuses need to go.

But the largely ignored, more insidious problem involves caucus “entrance” and election “exit” polls—or more precisely, the political journalists, pundits, and analysts who can’t separate their reliable data from that which is highly suspect, if not downright wrong.

The most widely cited of these surveys are conducted by the New Jersey-based Edison Research, which has conducted hundreds of such polls since 2004. Accept them at face value, and here’s what you’d be forced to conclude, based on official state voting files:

  • In Iowa, 18-44 year-old Democrats supposedly comprised 45 percent of all caucus goers, which means their turnout rate was significantly higher (34 percent) than either their 45-64-year-old parents (21 percent) or 65+ grandparents (25 percent). Edison’s sample of Iowa caucus goers shows a median age of 48. Iowa voter files put the median age of registered Democrats at about 56.
  • In Nevada, nearly two of three Democrats 65 and older, who’d voted in the November 2018 midterms, apparently stayed home, even with five days of early voting along with the February 22 caucus. These older voters’ apparent turnout rate was just 28 percent — compared to 70 percent in the 2018 midterms, and 43 percent in the low-profile June 2018 Nevada state primary.
  • In New Hampshire’s primary election, 18-44 year-olds supposedly composed 35 percent of the electorate, accounting for 105,000 of the 300,000 votes cast. The exit poll “found” that about 82,000 of them also voted in the 2016 presidential primary, though state voter files show only about 50,000.

Polling is an increasingly tough business. Cell phones, caller ID, and uncooperative voters—especially older ones—typically result in the over-sampling of younger voters.  But while pollsters now routinely weight their samples to compensate, too many unquestioning journalists simply run with  results that misrepresent the actual electorate in major ways.

This isn’t just an arcane discussion of statistical sampling methodology. A blizzard of upcoming presidential primaries—and more exit polls—is fast approaching, and history holds some harsh lessons for how relatively small miscalculations in this regard can have potentially enormous consequences.

Indeed, one remarkable New York Times story by Nate Cohn, published in June 2016, should be required reading. Cohn’s own skepticism about the accuracy of exit poll samples led him to delve into reams of U.S. Census data and voter files that weren’t published or accessible until long after the 2012 election (and its exit polls) had passed into history’s rear view mirror. His story was arguably the most prescient of the entire 2016 election cycle. Its headline told the entire story (almost): “There are More White Voters Than People Think. That’s Good News for Trump.”

Another revelation Cohn’s in-depth reporting made abundantly clear: “Voters Are (a Lot) Older, too.

Unlike their pre-election cousins—which identify and interview “likely voters” to predict future election results—exit pollsters interview actual voters. Exit poll data are especially valuable in determining how various candidates fared among key demographic groups, including those based on age, gender, education, race, ethnicity, and income. Edison’s 2016 exit poll, for example, revealed how a once Democratic-leaning voting bloc – non-college educated, white voters – dramatically lurched rightward in 2016, giving Trump a 66-29 percent margin.

The large sample sizes usually drawn for these polls—more than 1,600 voters in Edison’s Iowa survey, and more than 2,500 in both New Hampshire and Nevada—also ensure high reliability in determining key voter groups’ preferences for particular candidates. There’s little reason to question Buttigieg’s 24 percent support among Iowa’s college-educated caucus goers, or Sanders’s 51 percent support among Nevada’s Latino voters. But it’s an entirely different question as to whether the first group constituted 46 percent of Iowa’s caucus electorate, or that the latter comprised 18 percent of Nevada’s.

It’s difficult to credibly question, much less independently validate, the accuracy of exit polls’ samples for most demographic factors, at least in anything resembling real time. But the benefit of more time and more accurate methodology reveals their fallibility.

About six months after each general election, the U.S. Census Bureau publishes its biennial “Voting and Registration” report, based on a survey of voting habits among 100,000 Americans. The party affiliation and candidate preferences aren’t part of the mix, but key demographics are, including race/ethnicity, education, and income.

It was largely from this data—supplemented by official state voting files compiled by Catalist, a national list vendor—that Cohn derived his story’s most telling statistic. In 2012, non-college educated white voters, 45 and older, constituted closer to 29 percent of the voting electorate, not the 23 percent as suggested by that year’s exit poll. That wasn’t a trivial difference. It meant 10 million more of these voters probably cast ballots in 2012–and likely would again in 2016.

There’s only one key exit poll demographic that can be cross-checked against Census data and ground-truthed against official voting files for all 50 states: the age of voters. (Rightfully, voters aren’t asked about their race, income, religion or income-level.) For this piece, I’ve used data available directly from states or from L2, another major national list vendor.

Only a handful of states, Iowa among them, publishes age-cohort based registration and turnout data, something every state could and should to. Fortunately, there’s a wealth of accessible Census and voting file data that reveals America’s actual electorate to be far older than many people realize.

Here’s a few things we can learn simply from various Census Bureau tables:

  • Most of America’s potential voters were born before 1973 .The median age of the nation’s “citizens of voting age population”—CVAP in Census parlance—today is 48. Half are younger, and half older, including those not even registered to vote.
  • The American electorate is growing older, not getting younger. Census projections show the median age of eligible voters steadily rising, reaching 52 by age 2060.
  • Actual voters are older still, even in high-turnout presidential contests. The median age of 2016 voters was 51. That year’s Edison exit poll sample, incidentally, wrongly pegged voters’ median age at 47—as if the 100 million non-voters that year were older than their ballot-casting counterparts.
  • Actual voters are even older in midterms. Median voter age in the 2018 midterms—which set a record for midterm turnout, at 50 percent of eligible voters—was 53.

The Census Bureau doesn’t collect data for primary elections — arguably the “terra incognito” of election research, despite their enormous importance. Political insiders are keenly aware that for the 7,000 most prominent of the nation’s partisan offices (U.S. Senate, Congress, Governor, and state legislator) the vast majority of races are essentially over once the dominant party nominee is selected in these notoriously low-turnout contests.

But one of the most time-tested axioms in American politics is that the lower the turnout, the older the typical voter. Actual primary election data from all three states readily confirms this, while busting another common myth: that Democratic primary voters tend to be far younger than their Republican counterparts.

In Iowa’s June 2018 regular state primary, virtually the same number of Iowa Democrats cast ballots—180,000—as participated in the 2020 caucus. Age-related data on the Iowa secretary of state’s website for that election suggests a median age for Democratic voters of 62—nearly a generation older than the 48-year median for caucus goers. Meanwhile, Nevada’s June 2018 non-presidential primary election attracted substantially more Democratic voters (133,000) than the roughly 100,000 who participated in the February 22 caucus. Democratic ballot-casters in the 2018 primary election had a median age of 65—compared to 53 for supposed 2020 caucus-goers.

To be sure, presidential primary elections generally attract higher turnout among younger voters, but usually among older ones, too.

In 2016, Hillary Clinton and Bernie Sanders battled for the nomination well into June. Among the 31 U.S states that register voters by party, 12 held “stand-alone” presidential contests that year, while 12 “piggy backed” the presidential contest onto their regular state primary elections. (The other 7 states held caucuses).

Based on data from L2,  just 39 percent of current registered Democratic voters, with a median age of 60, cast ballots in these 24 contests. Democrats under 45 had a turnout rate of 18%, and accounted for just 18% of those primary election ballots. Democratic voters 65 and older had triple that turnout rate, and cast 38 percent of all ballots –a profoundly different (and much older) electorate than the one portrayed in 2020’s first three exit polls

While it won’t be the only important factor in 2020, voter age today seems more closely correlated with partisan leaning than at any time in modern political history. In 2018, the national Edison poll found 18-44-year-old voters—estimated then at a 35 percent share, but closer to 31 percent in state records—preferring Democratic House candidates by a net 25-point margin. Meanwhile, voters 45 and older showed just a one-point preference for Republicans—but cast almost 70% of all votes that year.

This November, it will be the exact mix of actual voters, especially in key states, that will matter most. That further underscores the importance of demanding that the nation’s major news organizations and their exit poll partners do a far better job in helping Americans understand who’s really voting now, not to mention, who most likely will be this fall.

Indeed, as we all brace for a 2020 election unlike any other in American history, a paraphrase of Donald Rumsfeld’s famous axiom on the eve of the Iraq war is worth reflecting on: “You go to war with the electorate you have, not the electorate you might wish you have.”

Support Nonprofit Journalism

If you enjoyed this article, consider making a donation to help us produce more like it. The Washington Monthly was founded in 1969 to tell the stories of how government really works—and how to make it work better. Fifty years later, the need for incisive analysis and new, progressive policy ideas is clearer than ever. As a nonprofit, we rely on support from readers like you.

Yes, I’ll make a donation

Phil Keisling

Phil Keisling, a Washington Monthly contributing editor, served as Oregon secretary of state (1991–99) and is currently the director of the Center for Public Service at Portland State University.