The Measure — and Mismeasurement — of Universal Vote by Mail

In three U.S. states—Oregon, Washington, and Colorado—the traditional polling place is now extinct. Every registered voter instead receives his or her ballot in the mail about two weeks before election day. Voters can return their marked ballots with a 49-cent postage stamp, or personally deliver them to special ballot drop sites. Fraud and other electoral mischief have proven to be total nonissues—but the democracy-enhancing results are dramatic. Had America’s other forty-seven states averaged the same active registered voter turnout rate, 110 million—not 83 million—ballots would have been cast in November 2014. And based on original research we’re doing at Portland State’s Center for Public Service, using complete voter history databases, “universal vote by mail” (UVBM) systems seem to help eighteen- to thirty-four-year-old and minority voters cast ballots at roughly double the rates found in more traditional systems.

Since Oregon adopted UVBM in 1998, dozens of academic studies have ostensibly focused on how it and various other election reforms might boost voter turnout. So why does a reform that seems to have worked so well in Oregon and elsewhere — especially in midterm and party primary elections — get such an indifferent if not chilly reception among a relatively small coterie of prominent and oft-quoted election researchers?

The core question these academic studies address seems straightforward enough. Controlling for all other relevant factors, how much difference does Election Reform X seem to make? Every election is different—by year, between states, and within states. Was Reform X, first used in 2010, really responsible for that 8 percent higher voter turnout compared to the 2006 midterms? Or were other factors responsible? In fact, might turnout have otherwise been 12 percent higher, except for X?

Every electorate is different, too. Election researchers have long known that four demographic factors significantly correlate with the propensity of a given state’s population to register and to vote: age, personal income, educational attainment, and race/ethnicity. Latino and Asian citizens, for example, have voting rates—especially in non-presidential elections like midterms and primaries—that are roughly half the rates of both black and white citizens, who now vote at roughly equal rates in such elections.

Researchers use various assumptions and statistical models to try to account for these and other differences between elections and electorates. But there’s no one broadly accepted framework for making these adjustments; assumptions and models vary widely among academics. And the resulting studies—rife with various correlations, weighted least squares logit models, and regression analyses—are often opaque to nonspecialists.

Far and away the best summary I’ve found of the major academic research about voter-turnout-focused reforms was published this past April by Tova Andrea Wang. Wang, a longtime election analyst who’s worked with Demos and other progressive organizations, doesn’t endorse UVBM. In a thorough and fair-minded discussion of the most oft-cited studies, she characterizes the conclusions of the academic research as “extremely mixed” with regard to Oregon’s fifteen years of experience.

But in trying to separate out the true “signals” in all this research from the statistical noise and irrelevancies, Wang’s discussion is notable for several reasons.

First, she recognizes that labels matter. While Wang calls Oregon’s system simply “Vote-by-Mail,” she notes that much of the research is really based on various types of “absentee ballot voting” (“voluntary,” “permanent,” and “no excuse”) and thus is largely irrelevant in assessing Oregon’s system. Wang also does an excellent job of identifying and succinctly summarizing the findings—and limitations—of the major studies most often referred to in discussing the subject. For example, during the mid-2000s Dr. Adam Berinsky of the Caltech-MIT Voting Technology Project published several papers whose two intertwined conclusions are often echoed by others: that while increased use of mailed-out ballots might slightly increase voter turnout, it’s likely to “reinforce the demographic compositional bias of the electorate and may even heighten that bias.”

Translation: even if turnout does increase, it will just mean more older, affluent, and/or mostly white voters casting ballots. This a major reason, Wang notes, that Project Vote, a leading election reform advocacy group, has cautioned that “voting by mail … does not always serve underrepresented or vulnerable populations as well as traditional polls.”

Other academics, however, strongly disagree. A 2013 study by a team led by the Yale political scientist Dr. Alan Gerber focused on Washington State and found a significant overall turnout effect, with the largest increases occurring among “rarely participating registrants.”

One big reason for such a disparate finding is that two seemingly similar—but ultimately very different—questions are being asked. The first, and most common, question is this: “How is voter turnout affected when lots of registered voters choose to apply for and use mailed-out ballots rather than go to traditional polling places to receive (and then vote) their ballots?” The second question is, “How is voter turnout affected when all registered voters receive their ballots two weeks prior to election day and then return them either via return mail or by going to official drop sites?”

It turns out that many of the most oft-cited studies really address the first question. In so doing, they quickly fall victim to what we might label the “Arizona fallacy.”

In November 2014, about 1.5 million Arizonans cast ballots. Nearly 77 percent (1.2 million) of these votes came via absentee ballots, the highest such rate in the U.S. Arizona’s turnout rate of active registered voters was still below the national average—and voters’ median age in this deep-red state, as revealed by our voting-record-based research here at PSU, was sixty-two years. Yet the median age (forty-seven) of Arizona’s citizen voting-age population is actually now slightly below the national average, in large part due to the state’s burgeoning Latino population.

This seems like compelling evidence for the Berinsky thesis. But the Arizona fallacy manifests itself here in two ways. First, there’s the “substitution effect”—specifically, the failure to acknowledge that the vast majority of these absentee-voting seniors would likely have voted at the polls anyway. Offer already-motivated voters a more convenient alternative to schlepping to the polls to receive and cast their ballots, and they’ll usually take it. But it doesn’t mean the “alternative” itself caused a big change in overall voting behavior.

The second, more important, leap of illogic is to simply extrapolate that if even more voted ballots are “mailed out,” it will simply amplify the same underlying turnout patterns. Indeed, Oregon’s Universal Vote by Mail system can best be understood—and perhaps even more precisely labeled—as “universal ballot delivery.” The key question isn’t what Arizona’s electorate might have looked like had 100 percent of its 1.5 million votes in 2014 been cast via mailed-out ballots. Rather, it’s contemplating the voter turnout implications if every one of the state’s other 1.7 million active registered voters who failed to vote—a group that was disproportionately younger, poorer, and far more Latino compared to those who applied for and then cast an absentee ballot—at least started with their ballots already in hand two weeks prior to the election, giving candidates and campaigns plenty of time to try to persuade them to vote.

While most academic researchers tend to carefully word their findings, that doesn’t prevent simplistic headlines and too-sweeping conclusions based on their work. A good example is an award-winning 2013 research project by a team led by the University of Wisconsin professor Barry Burden that came to this provocative conclusion: Early voting reforms such as liberalized absentee ballot laws and early in-person voting (EIPV) seemed to reduce voter turnout, by about 3-4 percent.

There’s actually abundant evidence that some “early voting” reforms might have little to no ultimate effect on turnout, perhaps for the same “substitution effect” reasons. (Minority and young voters clearly use these options more than their older, whiter counterparts—but again, might these same voters have cast ballots regardless?) But as reported by the Pew Research Center on Elections, Burden and his team then went a step further and surmised that this was due to early voting “robbing Election Day of its stimulating effects, reducing social pressure to vote and giv[ing] less reason for campaigns to motivate their supporters and get them to the polls.”

The Burden study only looked at two recent presidential elections, where the power of UVBM will always be least apparent. Most important, dig a bit deeper and one learns that Burden’s team ended up excluding Oregon and Washington because of their “unusual mail-in balloting rules.” (To be precise—and this does matter, as per the Arizona fallacy noted earlier—what’s truly unusual are these states’ “mail-out-ballots-to-every-registered-voter” rules.)

Burden’s study contains another common but problematic assumption: that calculating voter turnout should be based on all “eligible citizens,” including the unregistered, because of the variability in how states maintain their voter registration rolls. For more than twenty years, Dr. Michael McDonald of the U.S. Election Project (now at the University of Florida) has painstakingly calculated “voter eligible population” (VEP) numbers for all fifty states. In a nutshell, take the U.S. Census data for each state’s “voting-age population” eighteen and older, then subtract non-citizens and those who’ve lost their voting rights due to states’ widely varying rules about incarcerated citizens and ex-felons. According to this VEP-based yardstick, 2014’s voter turnout—among both registered and unregistered “potential voters”—was just 36.3 percent, the lowest since 1942.

VEP is a highly useful tool, especially for analyzing the entire continuum of states’ election system components. But using it to rank state voter turnout can arguably illustrate each state’s demographics more than the way state elections are conducted. Consider Oregon’s and Colorado’s 2014 VEP-based turnout rates in 2014—53.5 percent and 54.5 percent. That’s still impressive, ranking these two states fifth and third respectively. But the selective manner in which VEP is used to overlook (or even disparage) UVBM is perhaps best reflected in those who champion other ideas as far better alternatives.

The liberal think tank Demos, for example, often cites how “election-day/same-day” (EDR/SDR) voter registration laws — which allow eligible citizens to register and then vote at the polls all the way through election day — result in 7 to 10 percent higher turnout (albeit in presidential years) compared to VEP-based turnout in non-EDR/SDR states. Wang goes so far as to label these laws the only “consensus” reform researchers seem to generally agree on. (It’s a fine idea—and in 2015 Oregon went even further by becoming the first U.S. state to automatically register eligible citizens based on drivers’ DMV records.) But point out that in 2014, the nation’s three UVBM states’ voter turnout—based on the identical VEP denominator—was 16 percent higher than non-UVBM states, and the response is often something along the lines of, “Yes, but that’s irrelevant because Oregon (and presumably now Washington and Colorado) have always been high-turnout states.” As if Minnesota, Maine, and Wisconsin are any different?

But perhaps the biggest shortcoming of most of these studies involves the core data itself, especially when it comes to the key question of how UVBM affects the vote-casting behavior of key portions of the electorate. For example, no study of Oregon’s fifteen-year experience has been based on statewide voter files and individual voter history that would help determine what kind of voters do—or don’t—end up voting when all registered voters receive their ballots. Such individual-centric research is typically difficult and expensive; for demographic information beyond age and location it requires surveying a large enough sample of actual voters to allow statistically valid conclusions about how voting patterns vary by key demographic factors such as income, education, and race/ethnicity that aren’t part of public voting records.

The far easier path is to rely on sources such as the U.S. Census Bureau’s biennial election report based on about 100,000 respondents. But even with that large a sample, the margin of error in doing any state-to-state demographic group comparisons can be huge, even without the obvious errors due to self-reporting. (Rely on the 2014 census data for something as basic as “registered voter turnout” and the answer—92 million out of 142 million—is laughably different from the real number of 83 million out of 190 million.) As Wang notes, these challenges pose major limitations to virtually all turnout-related research, something our age-cohort study based on states’ complete voter files is working to address. Our research so far has been revealing: for example, 45.6 percent of Oregon’s eighteen- to thirty-four-year-old registered voters cast ballots in the 2014 midterm elections, the highest rate among twenty states in our study and one that’s more than double those of key battleground states like North Carolina (22 percent), Pennsylvania (20 percent), and Ohio (16 percent).

When academics do focus on elections where all registered voters receive mailed-out ballots, Oregon is still frequently ignored. One oft-cited 2011 study by a team led by California State University’s Elizabeth Bergman gave this reason for ignoring Oregon’s experience: it’s a state “without the challenges posed by population demographics, high density, or language diversity.” (As noted elsewhere, Oregon ranks at or below the national average in income, age, and educational attainment—and now has a higher Latino population, at 12 percent, than all but thirteen states).

Bergman and her colleagues studied voter turnout in a handful of California counties in 2006-08 where some small precincts were allowed to close their polls and simply mail out ballots to all their voters, while other precincts’ voters continued to have polling-place access. While the methodology is virtually impenetrable to non-statisticians, the study’s bottom line, as Bergman described it in a December 2015 Washington Post op-ed, is that “when you can only vote by mail, voter turnout actually drops by 13 percent.”

But look closer and several key details render suspect the study’s applicability to a true UVBM system. First, that “13 percent decrease” isn’t the difference between, say, a 50 percent voter turnout rate and a 37 percent rate. Rather, it’s a 13 percent difference in something called the “odds ratio” of voting—a much different (and smaller) gap. (An even older California county-based study, by Dr. Thad Kousser of the University of California, San Diego, estimated the nominal turnout rate decline at “up to 3 percent”). Second, all permanent absentee voters and those who returned their mailed-out ballots in person were excluded, to be consistent with the study’s stated purpose: to determine “What would happen if Voting by Mail became compulsory.” (My emphasis; there’s utterly no requirement in UVBM states to return ballots via the mail.) Perhaps most important, and as Wang notes, “even the researchers conceded that the result might be due to voters largely being unaware and campaigns failing to mobilize their efforts accordingly.”

Indeed, Bergman’s 2015 op-ed piece focused on California’s San Mateo County’s experience using UVBM for its November 2015 local elections. San Mateo election officials did something blindingly obvious: they made a meaningful effort to reach out and inform voters of the new system ahead of time. Bergman wrote that her earlier work suggested the change should have “reduced … Latino turnout by 27 percent”—but instead, it rose dramatically, “yielding dramatic spikes in turnout among young people and minorities,” according to a report in the San Jose Mercury News.

Bergman’s pre- and post-election surveys also put the lie to those who insist that traditional polling places—no matter how lonely and depopulated they become—must be retained because of their overwhelming popularity among tradition-loving voters. Before the election, more than 50 percent of “polling place only” voters disapproved of switching ballot delivery methods. Afterward nearly two-thirds of them (and, of course, an even higher portion of absentee ballot voters) gave the new system a thumbs up—what readers of the late Dr. Seuss might dub the Green Eggs and Ham effect.

One of the few Oregon-centric studies of the last decade, published in 2011 by Drs Paul Gronke and Peter Miller of Reed College’s Early Voting Information Center, concluded that Oregon’s new system hadn’t increased Oregon registered voter turnout in presidential and midterm elections, and that initially reported boosts of 10% or more were simply a passing “novelty effect.”

But again, a closer look is revealing. The study’s conclusion was based on its failure to replicate an earlier 2001 academic study that found a 10 percent nominal boost in Oregon voter turnout, for several special elections in the mid-1990s. The pair’s study wasn’t based on individual voter data, but instead largely relied on a “visual inspection of turnout trends over time.” More specifically, the study’s key chart shows that between 1960 and 1998, Oregon’s average registered voter turnout—81 percent in presidential races, 71 percent in midterms—wasn’t much different than the average from 2000 to 2010.

But during this half-century time frame, profound demographic and electoral changes swept across Oregon. Per capita personal income went from well above the national average to just 90 percent of it, and the state’s Latino population more than quadrupled. Visually inspect a more contemporary time frame—say, the decade prior to UVBM’s enactment and the decade since—and a far different picture emerges. Presidential registered voter turnout fell from about 81 percent to 71 percent during the 1988-96 period. The decline then reversed dramatically, beginning with Oregon’s first UVBM presidential election in 2000 (80 percent), and turnout has further climbed since then to average 84 percent over four presidential contests. (This has happened despite Oregon’s losing its national “battleground” status after 2000.) Midterm registered voter turnout prior to UVBM took a far steeper nosedive, from 77 percent in 1990 to just 59 percent in 1998. Again, it rebounded by 10 percent beginning in 2002 (69 percent) and has averaged a steady 71 percent since then. This happened despite the drop-off in ballot measure initiatives and a notable lack of highly competitive U.S. Senate and gubernatorial races—and during a period in which midterm election turnout dramatically declined most everywhere else.

Wang wisely recognizes that it’s impossible to referee between highly technical academic studies and their competing methodologies. And researchers like Bergman and Gronke deserve credit for trying to focus on the right question, rather than evaluate UVBM in context of it simply being “absentee ballots on steroids,” as some have dismissively labeled it.

Rather, the major point here is that far too much of the research cited in discussing UVBM suffers from one or more key shortcomings. As noted earlier, much of it lazily tries to extrapolate absentee ballot-based data, inappropriately uses VEP-based yardsticks, and/or focuses on presidential contests only—ignoring midterm and primary elections, where the approach is best suited to show dramatic results. And almost all studies pre-date Washington and Colorado’s decision to follow Oregon’s adoption of the system.

There’s also one other factor worth noting: an undoubtedly sincere (but irrelevant) commitment to the “civic ritual” of the polling place.

Consider the outsized role played by MIT’s Dr. Charles Stewart III and others associated with the Caltech-MIT Voting Technology Project. Its members’ extensive (and often quite useful) work pops up in many places, such as helping make the case for online voter registration laws (now in thirty states, including many red ones). Stewart and his colleagues’ work was also heavily relied upon by the Presidential Election Commission, appointed by President Barack Obama in 2013 to recommend ways to shorten voting lines that plagued millions of voters in the presidential election. (As noted in our main story, in every other election, a far bigger threat to democracy’s health is the notable absence of voting lines.)

But as the group’s very name implies, its focus is on building even better polling places, and stocking them with even more sophisticated (and expensive) voting machines. The group’s first report in 2001 cited the five “dangers” of increased reliance on mail-in ballots: coercion; fraud and security; accuracy; speed; and “the loss of the public ceremony of voting.” The authors warned of the many threats to the “real and perceived” integrity of any state election system that chose to rely on more mailed-out ballots, much less any that might follow Oregon’s lead in “making mail-in absentee ballots mandatory.” Instead, the report strongly recommended that states “[r]estrict or abolish on-demand absentee voting in favor of in-person early voting.”

Stewart’s 2004 report recommending ways to improve election administration and voter turnout in Massachusetts didn’t even mention Oregon’s then four years of experience with UVBM. Stewart’s advice was terse: “Massachusetts would do well to steer clear of this alternative.” (In 2014, just 58 percent of Massachusetts’s active registered voters cast ballots, compared to Oregon’s 71 percent. Even using the VEP denominator, Oregonians outperformed Massachusetts’s far more affluent and better educated citizens by 9 percent).

In 2012, the Caltech-MIT group published “Voting: What’s Changed, What Hasn’t.” The group cited the same five dangers, almost verbatim. “A decade later, our concerns have only grown,” the authors said, calling on lawmakers to “[d]iscourage the continued rise of no-excuse absentee balloting and resist pressures to expand all-mail elections.”

This same worldview—that there’s no viable alternative to the traditional polling place remaining the center of the electoral universe—reveals itself in many other venues, too. You can certainly find it in the Presidential Election Commission’s final report, where UVBM was given the shortest of shrifts, despite its quite elegant solution to long presidential election voting lines (“Just Abolish them”). Instead, the commission’s final report endorsed more “Early, in Person Voting” at polling places, though it sternly warned jurisdictions against the temptation to “overcompensate by significantly reducing the number of polling places, staff, and other resources available for Election Day.”

This bias against mail-based ballot delivery is also evident in the “Election Performance Index” (EPI) project recently launched by the Pew Foundation. This is an ambitious and valuable effort to benchmark key aspects of all fifty states’ election systems across a wide range of components, from voter registration and ballot delivery to ballot counting and tallying. But among the EPI’s seventeen ranking criteria – which Stewart played a key role in selecting — what’s the only one that relates to voter turnout? Ballots cast as a function of the voting eligible population (VEP). The ability of each state’s election systems to facilitate the ability of getting its already registered voters to actually cast ballots was determined to be irrelevant and/or too hard to measure.

Contrast that with the same weight given to each of four other criteria, all involving “mailed out ballots.” Oregon’s 83 percent registered voter turnout in 2012 meant that 17 percent of its “mailed out” ballots went unreturned; that earned the state a dismal thirty-ninth ranking. (There was no comparable penalty for states where 30 or even 40 percent of their registered voters failed to show up at their designated polling places.)

In 2012, Washington, Oregon, and Colorado ranked forty-eighth, forty-ninth, and fiftieth, respectively, in their “rejected mail ballot” rates. (Oregon’s rate was 0.73 percent.) That’s hardly a surprise: when a lot more voters at least cast ballots, less frequent voters will invariably be more prone to making basic errors like forgetting to sign their return envelopes. (Low-voter turnout Georgia earned a #1 rank here with a reject rate of just .02 percent among its 1.4 percent of voters that received and cast absentee ballots.) At least when it came to “voting waiting time” Oregon was able to make up lost ground in the rankings, right? No such luck. Top honors there went to Vermont; Oregon got a “Not Applicable.” Perhaps no other foundation has funded and supported such a wide range of high-quality election research in the last decade as Pew. So the double standards that reveal themselves in the details of the EPI’s scoring system are especially disappointing.

In writing this article, I was heartened to run across a handful of (mostly) younger academic researchers who seem willing to ask a different set of questions, with some fresh eyes.

A 2014 book—The Measure of American Elections, coedited by Stewart and Burden—contains several articles of interest. Dr. Lisa Schur of Rutgers University found that no other system came close to helping disabled voters actually cast their ballots: in “all vote by mail” systems like Oregon’s there was a “13.8 percent higher turnout among people with disabilities.” Buried in the discussion is perhaps an even more revealing statistic: among the “non-disabled” voters used as a control group, Schur still found an 8 percent increase.

Dr. Christopher Mann of Skidmore—who worked in Colorado politics for several years prior to getting his PhD—authored another chapter entitled “Mail Ballots in the U.S.” While Mann’s work didn’t focus on overall turnout rates, it has several other useful findings, especially relevant to rebutting the concerns of those who point to higher error rates and administrative mistakes when states expand the availability of traditional absentee ballots. Such problems tend to “peak” in states, Mann told me, where absentee ballots account for roughly between 5 and 20 percent of all ballots cast—enough to make a difference in the outcome of some races, but not a large enough “critical mass” to force election officials to focus on clear and consistent rules to educate voters and process ballots. But break through that threshold—when far higher portions of ballots are mailed out—and these problems diminish significantly.

Perhaps the most comprehensive study of Oregon’s experience that I found is one that’s seldom discussed or critiqued by others. Even Wang fails to mention it in her otherwise excellent summary. Its author is Georgia State professor Sean Richey, whose PhD thesis focused on Oregon’s system. A summary of his research was published in 2008 in Social Science Quarterly—a respected journal, though it’s hardly in the mainstream for most election-related research.

Richey looked at Oregon voting patterns in both presidential and midterm elections, from 1980 to 2006. In stark contrast to other studies, he recognized that UVBM does nothing—and can do nothing—to get unregistered citizens to register and vote. Accordingly, Richey rejected the approach used by most previous research—much of it reliant on voting-age population data, not even VEP-based information—and instead used registered voter turnout as his yardstick.

Richey also took into account Oregon’s and other states’ demographic characteristics, with a particular eye to how Oregon’s demographics had changed dramatically over this time period. His conclusion: “a significant positive effect from voting by mail of around 10 percentage points of registered voters in both mid-term and presidential elections.”

Richey’s study isn’t necessarily flawless, and, as he himself notes, it only goes through 2006. But when I asked Richey why he thought so much of the academic literature seemed to reach far different conclusions, he had an interesting response. “There’s a great premium among many academics to being a contrarian,” he told me. “If something is supposed to do something—like increase voter turnout—you often get a lot more attention if you can somehow show that it actually doesn’t.”

Richey is enthusiastic about universal vote by mail. “If anything, the effect might be even greater today—and in addition to Oregon, we can now study Washington State and Colorado, too.”

Wang also hopes the addition of two more jurisdictions to the UVBM ranks will spur a new wave of research. Oregon, she notes in her report, is often dismissed (even if inaccurately) as a unique situation and a state “not particularly representative or reflective of what might occur in other places given its perceived homogeneity.”

Fifteen years after largely being alone, Oregon’s long experience with universal vote by mail might finally get some long-overdue attention from those bringing far better research techniques to the task—and hopefully a willingness to challenge a different type of conventional wisdom.

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.