Berkeley Law
From the SelectedWorks of Kevin M. Quinn
Fall 2012
Long Live the Exit Poll
D.James Greiner, Harvard Law School Kevin M. Quinn
SelectedWorks1
Available at: http://works.bepress.com/kevin_quinn/38/
Long Live the Exit Poll
D. James Greiner & Kevin M. Quinn
Abstract: We discuss the history of the exit poll as well as its future in an era characterized by increasingly effective and inexpensive alternatives for obtaining information. With respect to the exit poll's future, we identify and assess four purposes it might serve. We conclude that the exit poll's most important function in the future should, and probably will, be to provide information about the administration of the franchise and about the voter's experience in casting a ballot. The nature of this purpose suggests that it may make sense for academic institutions to replace media outlets as the primary implementers of exit polls.
I
D. JAMES GREINER is an Assistant Professor of Law at Harvard Law School.
KEVIN M. QUINN is Professor of Law at the UC Berkeley School of Law.
(*See endnotes for complete contributor biographies.)
s the exit poll intellectually dead? That is, in the foreseeable future, can exit polling serve a purpose other than allowing media operations to "call" elections a few hours earlier than official results become available? This process of calling elections, and the race among media organizations to be the first to do so, may serve a recreational purpose; but whether calling elections contributes much to a thriving democracy is uncertain.
Even if we consider a set of questions crucial to the social sciences and law about the nature of the electorate, it is still not immediately clear that exit polls have much of a future. Suppose we want to learn about the characteristics and motivations of voters. Are we better off with the exit poll-currently around forty-five years old - or with a combination of older (mail, telephone) and younger (Internet) forms of polling, which may now be able to provide a great deal of information more cheaply than exit polls can? The question becomes even sharper when we consider that it may be possible to combine results from the older and younger techniques with information from data aggregators, which compile a vast (and increasing) amount of specific and wide-ranging data on voters and potential voters. In short, we might conclude that the exit poll is unlikely to live much past middle age.
© 2012 by D. James Greiner & Kevin M. Quinn
9
Long Live But another set of questions, perhaps e Exit Poll jess fundamental than those articulated above but nevertheless important, should allow the exit poll to survive into its golden years and perhaps beyond: What is the nature of the voting experience for the sizable majority of U.S. voters who continue to use traditional polling booths? Do voter ID laws prevent persons who reach the voting area from casting valid ballots, and are such laws administered in a race-neutral manner? Is voting equipment sufficiently available, and easy enough to use, to allow ready access to the franchise? What do voters understand when they are told, at least initially, that they cannot cast a valid ballot at a particular polling place ? Is the voting experience affected by the failure to enforce laws that proscribe electioneering from occurring within a certain distance of the voting area ?
In this essay, we suggest that the exit poll's future should lie in the study of this second set of questions - questions central to the field of election administration. This field has long been a subject of interest to political scientists, but it has an even longer history in the legal community. As we discuss below, the law governing how voters vote has been in flux since the early 1600s. Scrolling ahead a couple of centuries, voting in the early United States looked nothing like it does now. And even more recently, rules surrounding voting have constituted an important component of efforts by local, partisan officials to shape the electorate. For at least as long as a substantial portion of the U.S. electorate votes in traditional voting locations, and for as long as local and partisan officials oversee those voting locations, the exit poll should remain a vital tool for compiling information to support regulation of a democracy's most essential function.
^Vt least since the 1634 adoption of the written ballot system in Massachusetts,1
some fraction of voting in the United States has been conducted via the in-per-son presentation of a written indication of the voter's preference.2 The first exit poll in the United States was not, however, conducted until 1967, 343 years after the adoption of the Massachusetts written ballot system. Why did it take so long for exit polling to become an identifiable feature of U.S. elections ? As it turns out, both the law and statistical inference required some development before the exit poll could become either useful or practical.
The Law. Exit polling during much of the colonial period would have been profoundly silly. At that time, and for about one hundred years into our nation's history, voting was public or conducted in such a way that, despite the law as written, a voter's choice could frequently be observed. Some voting was by voice or show of hands.3 Some voting was by means of a written ballot (which, through the first several decades of our history, the voter ordinarily had to supply himself); secrecy in voting was nominally required by a fair number of state constitutions adopted around the time of the Revolutionary War.4 But even with written ballots, the mechanics of the voting process often meant that little was in fact secret. Certainly, the opportunities that open voting provided for bribery and intimidation did not go unexercised. Political parties provided hapless voters with paper ballots of a distinct color, or with a visible symbol or emblem, then posted observers in the voting area. Companies that owned towns generously provided free transportation of voters to polling places, along with free printed ballots, and likewise posted observers. Retaliation against those who voted the "wrong" way could take a variety of forms, from loss of employment, to eviction, to physical violence. Meanwhile, bribery and intimidation were not the only ways in which candidates and their
10
Dcedalus, the Journal of the American Academy of Arts & Sciences
minions influencedballot-casting. In some well-reported instances, the area surrounding the ballot box was a cross between a circus and a boxing match. According to one observer, "Sham battles were frequently engaged in to keep away elderly and timid voters of the opposition."5
Some viewed the lack of secrecy as a good thing. In the early Republic and stretching into the mid-i8oos, at least a few elites argued that public voting was essential to the proper functioning of a democracy. Secret ballots, it was thought, would tempt a man to vote in his own narrow self-interest and would encourage ill-considered choices. More abstractly, voting was believed to be a public act, a declaration for whom and for what a man stood. Virginia politician John Randolph is said to have opined in 1847, "I scarcely believe that we have such a fool in all Virginia as even to mention the vote by ballot, and I do not hesitate to say that the adoption of the ballot would make any nation a nation of scoundrels, if it did not find them so."6
When reform finally came, it moved swiftly on two fronts, both with implications for the exit poll. On the first, from approximately 1888 to the turn of the century, the overwhelming majority of states switched to what was known as the "Australian system" of voting, whereby a government-provided ballot, available only at an official polling area, was given to the voter to mark in secret. That fundamental system is in place in most elections conducted in the United States today, although a recent trend toward "convenience voting" (which takes a variety of forms, from increased use of absentee ballots to early-voting stations at high-volume precincts) continues; and some variation exists even with traditional ballot-casting. On the latter point, for example, state laws differ on whether a voter may voluntarily show another her ballot. Illinois law illustrates
the dominant regime: " [ A]ny person who D. James
knowingly marks his ballot so that it can Gmner & 0 Kevin M.
be observed by another person... shall be Qu[nn
guilty of a Class 4 felony. "7 By requiring the voter to keep her ballot secret, Illinois law attempts to provide a haven to anyone requested to allow voluntary verification that a would-be vote-purchaser has received the benefit of her bargain; the voter can legitimately claim that allowing voluntary verification will expose her to prosecution. West Virginia, in contrast, allows the voter to choose to show her ballot to another.8 In the second reform area, governments asserted control over the location surrounding voting booths. Some went so far as to exclude anyone but a would-be voter or an election official from entering a legally defined halo (say, within 150 feet) around polling booths. Others prohibited certain activities, such as electioneering, from occurring inside a halo but allowed members of the public to occupy spaces close enough to voting booths to observe the administration of the vote. The latter approach eventually dominated when its utility as an antifraud measure became evident.
With respect to the exit poll, the two types of reform operated at cross-purposes. On the one hand, the secret ballot created a need to gain information about voter choices and voter experiences; that need gave rise to the exit poll. On the other hand, laws mandating control and decorum around voting booths provided local administration officials with what they perceived to be a statutory basis for attempting to push exit pollsters outside sometimes large halos around the voting area. Requiring pollsters to stand more than a few feet from the exit to a building in which voting occurs is death to an exit poll because voters often disperse (and sometimes disappear down public transportation entrances) before they can be approached. Not all election administra-
141 (4) Fall 2012
11
Long Live tion officials sought to push exit pollsters e Exit Poll to a prohibitive distance from the polling area, but enough of them did to necessitate a discussion of one other piece of legal history: the development of First Amendment jurisprudence in the court system, particularly in the federal courts.
As is true of the secret ballot and the exit poll, the development of First Amendment jurisprudence occurred far later in our nation's history than many realize. Beginning in the early 1900s, the U.S. Supreme Court began to use the First Amendment as a constitutional basis for scrutinizing federal laws that regulated speech, particularly political speech. During this time, the First Amendment had little to do with political polling or voting but had everything to do with politics. Some of the earliest First Amendment cases were prosecutions against political figures, including the Secretary of the Socialist Party9 and an alleged founder of a wing of the Communist Party.10 In 1925, the Supreme Court broke substantial new ground by holding that the First Amendment provided the judiciary with a textual basis for scrutinizing state (not just federal) laws regulating speech.11
First Amendment law evolved substantially in its first hundred years, and it continues to evolve today. When the first exit polls were conducted in the late 1960s, however, there was a general rule that courts would take a hard look at content-based restrictions on speech, particularly political speech in some kind of public environment, to determine whether such restrictions were truly necessary to meet articulated state goals. Thus, when local officials sought to use halo laws to push exit pollsters to untenable distances from the voting area, media outlets and exit polling firms sued. Courts, particularly federal courts, held almost uniformly that exit polling was a protected form of speech, and that election officials could
not constitutionally prevent exit pollsters from approaching voters within a few feet of the exits to the buildings in which voting took place.12 Legally, then, the ability of exit pollsters to do their work became firmly established.
The Statistics. An additional step in the exit poll's historical development has been the evolution of polling itself. Survey sampling had its genesis before the 1800s, and political polling that focused on an election was used at least as early as 1824 to allow newspapers to report projections of the results of that year' s presidential elections. x3 But like most forms of statistical inference, polling did not mature until the Great Depression. A breakthrough came in 1934, with the publication of a paper by Russian-Polish mathematician Jerzy Ney-man, one of a handful of true founders of modern probability and statistics.14 Until this paper, statisticians attempting to make inferences about a large population pur-posively chose the units that would be observed (that is, the sample). Sophisticated statisticians allocated observations to categories thought to be salient (for example, large versus small units in industrial samples, men versus women in samples involving people), but the statistician himself controlled that allocation. Ney-man's proposal was to apply the idea of randomization (developed earlier by R. A. Fisher in the context of agricultural experiments) to the selection of a subset of units in a large population for observation-that is, to create a random sample. Neyman showed that good things happened when the statistician gave up at least some control of the selection of the sample to a randomizer. Among these good things was the ability to create ranges within which the true value of some population parameter was reasonably likely to fall. These ranges are now known as error bands, or more technically, confidence intervals.1^ The probability sample ranks next to the
12
Dcedalus, the Journal of the American Academy of Arts & Sciences
randomized experiment as one of the most important creations in the field of statistics. Survey samplers of all kinds now make extensive use of the tool Ney-man gave them.
In the years after Neyman's article, survey sampling grew increasingly sophisticated. When exit polls finally came along, they involved a mixture of randomized and nonrandomized sampling. Statisticians in charge of exit polls ordinarily use randomization to determine which precincts to send pollsters to, but the selection of voters to be approached within each precinct was rarely, if ever, truly random. For instance, in the dominant method of selecting voters, the exit pollsters approach every kth voter, where k is some fixed integer. We have adopted this practice ourselves in our own exit-polling work. Other exit polls sample according to an exiting voter's characteristics (for example, sample every kth African American male).16 Even with these techniques, however, the idea of randomization serves as the backbone for analysis of polling results (particularly confidence intervals). Essentially, those who run exit polls pretend that they randomize even though they do not.
The First Exit Poll. In 1967, political pollster Warren Mitofsky designed and administered the first exit poll in the United States, focusing on Kentucky's gubernatorial contest.17 Oddly, Mitofsky was inspired by the practice of canvassing moviegoers as they left theaters to obtain viewer opinions.18 Mitofsky's client was CBS, which was attempting to find ways to project election results and to break down voter preferences by demographics.19 Apparently, CBS was satisfied, and as other media outlets sought to compete, the exit poll's popularity grew. Mitofsky himself was said to have conducted exit polls for more than three thousand electoral contests, some of them occurring abroad.20
One final aspect of U.S. election admin- D. James
istration deserves mentioning: "A long- ^reine{f
b b   Kevin M.
standing peculiarity of US federalism has Qu[nn
been to reject in principle the Federal definition of voting rules. In practice the system remained profoundly fragmented until the 1960s and was indeed unified at that time only with respect to certain features, specifically those relating to blatant forms of racial discrimination."21 In other words, voting administration has always been controlled by (i) locals who are (ii) political partisans. The United States has never professionalized or centralized election administration. And to state the obvious, partisans have powerful incentives to manipulate voting rules to their advantage.
To determine what future the exit poll should have, we need to assess the purposes it can serve, how well it can serve them, what alternatives exist, and what the relative costs and benefits of exit polls are vis-á-vis alternative techniques. The exit poll's most familiar purpose is to provide the backbone of a system that can be used to predict winners of elections earlier than the official results are available. As we noted above, it is not immediately clear that enabling media outlets to call elections a few hours earlier than official results are available contributes to the democratic process or to any other process of serious social concern. Even for those inclined to believe that calling elections has democracy-promoting value, evaluating how well the exit poll serves this purpose is nonetheless complicated. In our view, the record of exit poll-based22 projections of winners and losers is both helpful and frustrating. The record is helpful because reasonably well-executed and well-analyzed exit polls, particularly when used with other available information, generally do allow pollsters to predict the right winners. That is, those who
141 (4) Fall 2012
13
Long Live run and use exit polls are right more often e Exit Poll (mucn more often, in fact23) than they are wrong. The record is frustrating because exit polls are most prone to fail when we most want them to work. Specifically, exit polls run into the greatest difficulties when predicting close elections. Here, various hard-to-adjust-for biases, weaknesses in field operations or analysis techniques, and the uncertainty inherent in any kind of sampling scheme can overwhelm slim margins of victory. These shortcomings have led to some spectacular failures of exit poll-based predictions and erroneous election calls by media organizations. A notable example is the 2000 presidential election, when the contest came down to Florida, which was - or rather, should have been24 - too close to call.
A second purpose that exit polls might serve is to provide information about the electorate, specifically its characteristics, thinking, and motivations. Such information is valuable. True, democracies can function without it, and additional information is not inevitably democracy-promoting. Nevertheless, information about the electorate can further short- and long-term purposes. With respect to the short term, to the extent that we want politicians to do what the electorate wants them to do, and to the extent that we want the electorate to be able to punish politicians when they fail to do so, it is probably better that politicians know what the electorate wants - or at least that they know more than they would from the raw results of elections in which, most often, no more than two candidates seek each office.25 Thinking long term, greater information can support academic study of the electorate's characteristics and desires; one hopes that such study would lead to deeper understanding and, concomitantly, a better-functioning democracy.
How well have exit polls furthered this second purpose? Again, the record is
mixed. Turning, for example, to a subject of our own research, exit polls have conclusively established that voting in United States presidential elections is racially correlated, meaning that African American voters have preferences noticeably and predictably different from those of Caucasian voters, and that both have noticeably different preferences from those of Hispanic/Latino voters (with the latter's preferences harder to predict).26 This fact seems so obvious at present that one might ask why we need exit polls to keep proving it; but it is obvious in large part because exit polls have established it so conclusively for so long. Exit polls have also contributed to the information we have about offices below the presidency, where again, racially correlated voting appears to be stubbornly persistent in some jurisdictions. And depending on one's world-view, this stubborn persistence of racially correlated voting may have consequences for subjects as varied as candidate strategy and constitutional law.27 Nevertheless, exit polls are limited in the complexity of the subjects they can probe; as we discuss below, one source of such limitations is the fact that exit poll questionnaires must be short and individual questions simple. In terms of information production, then, exit polls may allow finer cuts than the raw results of two-party elections, but they hardly allow scalpel-like precision.
A third purpose the exit poll might serve is as a check against official shenanigans. Here, the theory is that if the official results do not match the exit poll results, then the official results might be the result of tampering. At least in the United States, and thinking systematically (as opposed to focusing on an occasional freak occurrence), we view the exit poll's ability to serve this purpose as almost a nonstarter. First, exit polls are visible to election administrators, so the presence of an exit poll might deter the behavior it is at-
14
Dcedalus, the Journal of the American Academy of Arts & Sciences
tempting to detect.28 Second, the margin of error involved in exit polls, and the hard-to-adjust-for biases that plague any complex field operation, mean that fraud would have to be large for an exit poll to detect it. Yet "the entire art of electoral fraud, as ample evidence from history shows, is to manipulate the outcome only to the extent required."29 Third, on what basis would one conclude that a discrepancy between official and polling results indicates problems with the former as opposed to the latter? Given the difficulty that exit polls have had in predicting the results of some high-profile elections in which fraud was never seriously alleged, one would need substantial additional evidence external to the exit poll to suggest that the official count, not the exit poll, is suspect, in which case it is not clear how much value an exit poll adds.3°
A fourth purpose that exit polls might serve is to allow study of the voting experience. In the 2008 presidential election, at least 70 percent of the civilian electorates1 voted via the Australian ballot system (with some technological bells and whistles added) - that is, by visiting in person an officially run polling location and casting, in secret, a written or electronic ballot. As noted above, elections in the United States are administered by local partisans, who have powerful incentives to manipulate laws and practices governing election administration. Registration, purging of voting lists, ballot design, waiting times (which may increase relative to a jurisdiction's failure to respond to changing demographics by redrawing precinct lines),32 the presence or absence of interpreters and multilingual ballot materials, operation of voting machines, and voter identification requirements: these are just a few of the areas of law that might be altered to make voting more - or less - difficult, either across the board or for identifiable groups. Moreover, poll workers
are often volunteers or poorly paid tem- D. James porary employees who are given one- to ^J™^ two-hour training sessions - this despite Quinn the complex overlay of law that governs the voting process. Indeed, relevant law comes from the U.S. Constitution, federal statutes, federal regulations, state constitutions, state statutes, and state regulations, all as interpreted in state and federal court cases. Meanwhile, even more pedestrian concerns, such as the layout of the room in which voting occurs, may determine ease of access. Exit polls can provide valuable information about such aspects of the voting experience. And as we explain in the next section, they may be uniquely situated to do so.
E
xit polls are only one kind of poll. To assess what the exit poll's future will or should be, we need an overview of its strengths and weaknesses vis-á-vis the alternatives. We limit our discussion to polling techniques, although we concede that polls are only one way to collect information that might further one or more of the four purposes listed in the preceding section.
One strength of the exit poll, and the argument most often made by its proponents, is that comparatively speaking, pollsters conducting an exit poll are more likely to request information from a person who has actually voted or attempted to vote. This advantage can be overstated. Refusal rates in exit polls are high, particularly in the current era; in a well-executed exit poll, about half of persons approached will refuse to participate. Moreover, a less appreciated problem is that a great deal of voting occurs in schools, churches, apartment buildings, elderly residences, malls, and other high-traffic multiuse buildings. Exit pollsters, who ordinarily must stand outside a building's exit, can have trouble distinguishing between a voter exiting the building and a non-voter who came to
141 (4) Fall 2012
15
Long Live the building on other business.33 Still, at e Exit Poll present| exit polls target the electorate (or the would-be electorate) much more effectively than polls using the mail, the telephone, or the Internet. These other forms of polling must target a set of people who say they will vote or did vote.
But the exit poll has a second strength that proponents articulate less frequently : namely, the fact that respondents are approached within minutes of having experienced the voting process and having cast their ballots. Their memories of whom they voted for are still fresh. Perhaps more important, their memories of the voting experience are still fresh. With respect to voter choices, freshness matters particularly for state and local contests, such as elections for state legislative representative, mayor, city council, or school board. To illustrate: as we write this essay, neither of us can remember for whom we voted in the most recent set of state representative and municipal contests. Moreover, catching voters shortly after they have cast ballots can help prevent certain kinds of biases that may be related to - but more complicated than - mere recall problems. For example, there is evidence that if voters are polled (via, say, the telephone or the mail) after the winner of the contest has been announced, they overreport having voted for the winner or the incumbent.34
Memory issues are even more critical with respect to voter experiences. To cite an example from our own work, an exit poll we conducted documented how poll workers in one jurisdiction requested voter IDs from minority voters at a higher rate than they did white voters, with the disparity difficult to explain on grounds other than racial bias. This was true even though there was supposedly no element of discretion in whether to request IDs from voters. That is, under applicable state law, poll workers were to request an ID if
a would-be voter had one of two particular symbols next to her name on the registrant list, but not to request an ID if no such symbol appeared.35 Our point is that we would have difficulty trusting a finding of this nature had it come from a polling technique other than an exit poll. We suspect that for many a voter, being asked to show an ID is a low-salience event, one she is not likely to recall accurately a few hours after leaving the voting area. And yet, currently, it is hard to find an issue in voting administration that occupies more attention in the press than voter ID laws,36 attention we think is well deserved.
So the exit poll has advantages - big ones. It also has big disadvantages. Each exit poll requires a complicated, expensive, and delicate field operation that includes the temporary hiring and training of hundreds of personnel. Because of the length of time between elections, there is no feasible way to keep pollsters permanently on staff. The expense involved in running, say, a national exit poll puts pressure on poll architects to cut corners in the field operation. For example, pollster training for the 2004 presidential election exit poll -a poll performed on behalf of major media operations - ran into difficulties.37 Training had been carried out via the telephone and was surprisingly short in duration.38 Our experience, consistent with that of others, is that in-person training of pollsters is important.39 Similarly, cost concerns have driven professional exit polls to place only one pollster at each precinct; our experience (which, again, dovetails with that of others40) is that multiple pollsters - preferably at least three at a time in two shifts, or six per precinct - are necessary. In busy precincts, more may be needed. Greater numbers allow pollsters simultaneously to keep an accurate interval count, to administer questionnaires, and to take care of basic requirements such as the need for food or rest.
16
Dcedalus, the Journal of the American Academy of Arts & Sciences
Even well-funded and well-executed exit polls have severe limits; we discuss two here. First, exit poll questionnaires must be short, typically no more than one page; questionnaires can certainly be longer, but if they are, the response rate usually declines. The questions themselves typically must be short and simple enough to be read and answered accurately by an impatient voter standing outside a voting area (perhaps in the rain), delaying the next activity in her day. Complex questions are not an option. Second, exit polls have no effective way to incentivize persons approached to respond to the poll. In contrast, the business models of online polling services include substantial incentives.
So what are the alternatives to an exit poll, and what are their strengths and weaknesses? For the most part, the strengths of the exit poll are the weaknesses of the alternatives, and vice versa. Consider surveys conducted via mail. Mail surveys typically struggle with low response rates. On the issue of whether questionnaires reach actual voters, statisticians who take the time to pull voter lists from official records can be assured that their questionnaires arrive at the right destination. But this process takes time, so voter memories will be at least somewhat stale. Addresses will be out of date. Results will be delayed, although again, delay on its own (apart from what it does to the integrity of the responses) may not be a serious concern. Mail surveys have some advantages. Pollsters can sometimes risk longer and/or more complicated questions, on the supposition that respondents might have longer than two minutes to respond (an admittedly risky supposition). Mail polls are comparatively cheap and easy to administer. And mail polls remove the aspect of human pollsters-to-respondent interaction, which can obviate the concern that answers might change depending on interviewer/ pollster characteristics.41
The analysis for telephone polls is sim- D. James ilar. Telephone polls can be administered ^J™1^ before, during, or after elections. If ad- Quinn ministered after an election, official voter lists can be obtained and matched to commercially available telephone lists in an attempt to reach actual voters. Such matching inevitably misses a sizable fraction of actual voters, and the pattern of such misses might be worrisome; one might be concerned, for example, that matching is more likely to miss transient voters, and that transience might be correlated with a result of interest. Even with the right telephone numbers, reaching voters by phone and persuading them to respond to the poll is an uncertain business. Again, what causes potential respondents to fail to answer the telephone or refuse to take the poll after a connection is made is always uncertain. In fact, uncertainties associated with matching voter lists are so great that many telephone polls abandon a strategy based on official voting lists in favor of random digit dialing. Putting aside matching and response rates, telephone polls have other drawbacks. Questions must be read aloud, which puts a limit on their complexity. One might suppose that interviewer characteristics would not matter in telephone polls; alas, it is not certain that this is the case.42 In terms of advantages, telephone surveys can be quickly and cheaply administered.
A final option is the Internet. Some firms now maintain stables of Internet users who are willing to respond to polls in return for some form of compensation, such as free high-speed Internet service.^ A firm can attempt to match the demographics of a subset of its stable to those of some target population (such as voters or potential voters). The process can be tricky, particularly for a characteristic like age or income, because Internet usage varies widely among different age or income groups. For instance, how readily
141 (4) Fall 2012
17
Long Live would one believe that the political pref-the Exit Poll erences 0f sixty-five-year-old Internet users are representative of those of all sixty-five year olds, given that, at present, less than half of persons sixty-five or older regularly use the Internet?44 Nevertheless, Internet polls can be fielded quickly and comparatively cheaply. And questions can potentially be much more complex than in other formats. Specifically, the poll can be programmed so that the questions a respondent sees depend on prior answers in a way that engenders minimal confusion. As before, the weakness of the Internet poll is the exit poll's strength: at present, the Internet poll's ability to reach actual voters, or persons who attempted to vote, is uncertain.
We close our discussion of strengths and weakness with one recent development: the rise of information aggregation firms. These firms either already do or have the potential to make available vast (some might say disturbing) amounts of information on individuals or sets of individuals, such as registered voters.45 Depending on how these information aggregators evolve, and the extent to which the government regulates them, these firms have the capacity to revolutionize polling (and indeed may have done so already). Imagine, for example, a world in which official voter lists are electronically matched to files from information aggregation firms, and that these files are then electronically matched to an Internet polling firm's databases. Such a system could potentially allow polling of actual voters within, say, a day or so of their having voted. And the vast array of demographic and other information available from the information aggregation firm could allow statisticians to compare the characteristics of the sample respondents to those of the voting populace. Such a system would have holes, no doubt. At some point, for example, one might worry about the effect on survey
responses of the near-total loss of anonymity on the part of respondents. Exit polls go to great lengths not just to preserve anonymity but also to demonstrate that anonymity to respondents.46 But again, as we now know, exit polls have holes, too.
So where does all this leave us ? We surmise that the only important purpose that exit polls will be able to serve in the foreseeable future - and the one they should serve - is the fourth one on our list: namely, to provide information about the voting experience. As we suggest above, the early calling of elections serves only to provide entertainment value. Given the financial pressure placed on traditional newspaper and television organizations in recent years, we wonder how much longer these media outlets will choose to finance exit polling for this purpose. The process of cutting costs by cutting corners, already under way in the form of measures such as a single pollster per location and reduced pollster training, may degrade the information obtained beyond the point of usefulness. With respect to information about the electorate's characteristics and its views, we speculate that other polling sources, particularly Internet polling augmented by information from aggregation firms, will evolve to the point that this method of polling provides a cheaper and easier way to produce information of equivalent quality. And given the uncertainties and challenges associated with the exit poll, we cannot reasonably expect it to serve as a fraud-detection device.
The exit poll is, however, uniquely well suited to provide information about the voting experience. In our view, the details of election administration matter. As our short history illustrates, the method in which voting occurs has been the frequent subject of political battles since colonization. And in the past half-century or so,
18
Dcedalus, the Journal of the American Academy of Arts & Sciences
the election administration battle has been ugly. Poll taxes, literacy tests, and other "details" of the voting experience led to passage of the Voting Rights Act of 196547; the constitutionality of reauthorizing a portion of this statute promises to be among the nastier fights waged in the U.S. Supreme Court over the next several years.48 Moreover, assuming the Supreme Court takes its own decision in Bush v. Gore^ seriously (a debatable assumption), miniscule differences in voting procedures may have constitutional significance. And we have already mentioned the ferocity of the debate on voter ID laws. These developments demonstrate the need to seek information about the voting experience-information that exit polls are well positioned to provide.
The question then arises: who should conduct the exit polls? If the primary role of exit polling becomes to document, evaluate, and (perhaps) combat partisan efforts to manipulate the electorate and the voting experience, what institution can best pursue these goals? The numerous
failings (in the business sense) of local D.James and regional newspapers, together with ^J™^ consolidation in the exit polling industry, Quinn suggest that the press - the traditional watchdog over governance - is probably unable or unwilling to finance the gathering of needed information. Our view is that academia should step in. As we mentioned above, some of the highest-quality exit polling in the nation is currently done in Utah, and it is overseen by academics who use the experience as a "teachable moment" for college and graduate students. Our own experience with student-based exit polling was similarly positive; it revealed startling and potentially unpopular (with election administrative officials) data on voter identification requests. In short, while the practicality behind exit polling may make it at first seem like an unusual task for academics, we believe that exit polling provides an opportunity for academia to perform one of its noblest functions: to speak unpopular truth to power.
ENDNOTES
* Contributor Biographies: D. JAMES GREINER is an Assistant Professor of Law at Harvard Law School. His research focuses on statistics and litigation, and his current projects involve re-districting, election administration, adjudicative system design, and the evaluation of delivery of legal services. His work has been published in the Review of Economics and Statistics and the Harvard Law Review, among other journals.
KEVIN M. QUINN is Professor of Law at the UC Berkeley School of Law. His research, which focuses on judicial decision-making and statistical methodology, has been published in such journals as the Columbia Law Review, the American Journal of Political Science, and the Stanford Law Review. He currently serves as an Associate Editor for the Journal of the American Statistical Association.
1 Eldon Evans, A History of the Australian Ballot System in the United States (Chicago: University of Chicago Press, 1917), 1.
2 Nick Moon, Opinion Polls: History, Theory, and Practice (Manchester, U.K.: Manchester University Press, 1999), 6.
3 Paper ballots began to replace a show of hands in most of the original colonies in the mid-to late 1600s; Evans, A History of the Australian Ballot System in the United States, 1-6.
4 Joseph P. Harris, Election Administration in the United States (Washington, D.C.: Brookings Institution, 1934), 16 -17.
141 (4) Fall 2012
19
Long Live   5 Burson v. Freeman, 504 U.S. 191, 202 (1992). the Exit Poll &
John Crowley, "The Secret Ballot in the American Age of Reform," in Cultures of Voting: The Hidden History of the Secret Ballot, ed. Romain Bertrand, Jean-Louis Briquet, and Peter Pels (London: Hurst, 2007), 43, 52.
7 10 I.L.C.S. sec. 5/29-9.
8 Const. W. Va. art. IV, sec. 2.
9 Schenckv. United States, 249 U.S. 47 (1919).
10 Whitney v. California, 274 U.S. 357 (1927).
11 Gitlow v. New York, 268 U.S. 652 (1925).
12 See, for example, NBC, Inc. v. Cleland, 697 F. Supp. 1204 (N.D. Ga. 1988).
13 Moon, Opinion Polls, 6.
14 Jerzy Neyman, "On the Two Different Aspects of the Representative Method: The Method of Stratified Sampling and the Method of Purposive Selection," Journal of the Royal Statistical Society 97 (1934): 558. Opinion pollsters appear to have anticipated some of the benefits of random selection of sampling units that Neyman demonstrated in his paper; see Moon, Opinion Polls, 11.
15 Morris Hansen, "Some History and Reminiscences on Survey Sampling," Statistical Science 2 (1987): 180.
16 It might be more accurate to say that polls sample according to pollsters' perceptions of voter characteristics. In our experience, pollsters' perceptions of voter characteristics do not always match voters' self-identification.
17 Mark R. Levy, "The Methodology and Performance of Election Day Polls," Public Opinion Quarterly 47 (1983): 54.
18 J. Michael Brick and Clyde Tucker, "Mitofsky-Waksberg: Learning from the Past," Public Opinion Quarterly 71 (5) (2007): 703-716, http://poq.oxfordjournals.0rg/content/71/5/703.full.
x9 Levy, "The Methodology and Performance of Election Day Polls," 54.
20 Pew Research Center, "Mitofsky on Mexico's Standoff," July 6, 2006.
21 Crowley, "The Secret Ballot in the American Age of Reform," in Cultures of Voting, ed. Bertrand et al., 43, 46.
22 At least in the current era, media outlets rarely rely solely on exit polls to call elections. To the contrary, the predictions they rely on result from a constantly changing mix of official returns, exit poll results, telephone and mail polling (targeted at absentee voters in particular), and the application of complex weighting schemes designed to attempt to correct for over- and underrepresentation of particular categories of voters.
23 See Moon, Opinion Polls, for a review of the record on this score.
24 On the basis of projections that relied in great part on exit polling results, major media outlets called Florida for Gore even before all precincts in the state had closed. Matt A. Barreto, Fernando Guerra, Mara Marks, Stephen A. Nuno, and Nathan D. Woods, "Controversies in Exit Polling: Implementing a Racially Stratified Homogenous Precinct Approach," PS: Political Science&Politics 39 (2006): 477, http ://faculty.washington.edu/mbarreto/papers/exitpoll.pdf. The U.S. Supreme Court ultimately declared Bush the winner in Florida; see Bush v. Gore, 531 U.S. 98 (2000).
25 Of course, some of what polls reveal is contradictory or nonsensical: various polls (none of them exit polls, and each possessing different methodological weaknesses) find that U.S. citizens want the foreign aid share of the federal budget to be simultaneously cut and increased in size by a factor of ten. The problem is that we think that foreign aid makes up about 25 percent of the federal budget and that it should be reduced to about 10 percent. The true share is
20
Dcedalus, the Journal of the American Academy of Arts & Sciences
less than 1 percent. See CNN Opinion Research Poll, April 2011, http://i2.cdn.turner.c0m/ D.James cnn/20ii/images/03/3i/rel4m.pdf; andWorldPublicOpinion.org, "American Public Opinion Greiner & on Foreign Aid," November 2010, http://www.worldpublicopinion.org/pipa/pdf/novio/
Kevin M.
F or eign Aid_No vi o _quair e. p df. Quinn
2^See, for example, CNN Election Center, http://www.cnn.com/ELECTION/2008/results/polls/ #val=USPoopi.
27 With respect to constitutional law, one might think of the upcoming battle in the Supreme Court over the constitutionality of Section 5 of the Voting Rights Act. See Stephen Ansola-behere, Nathaniel Persily, and Charles Stewart, "Race, Region, and Vote Choice in the 2008 Election: Implications for the Future of the Voting Rights Act, "Harvard Law Review 123 (2010): 1385. The authors use exit poll results and other techniques to discuss the implications of voting patterns by race for the future viability of Section 5.
2^ One might argue that deterrence is itself a good thing. But for reasons we explain, the expense and operational difficulty involved in exit polling make it an unattractive prophylactic.
29 Jasjeet Sekhon, "Data Troubles: Explaining Discrepancies between Official Votes and Exit Polls in the 2004 Presidential Election," May 24, 2006, 5, http://sekhon.berkeley.edu/ papers/SekhonExit.pdf. Note that the author of this piece is more sanguine than we are about the usefulness of exit polls as fraud-detection devices.
3° To be clear, we are referring here to manipulation of contests in the United States, where, as noted above, we suspect that the existing fraud is designed to be just great enough to tip the balance and thus difficult for an exit poll to detect. In newer democracies, according to media reports, fraud can be large and flagrant.
31 The figure comes from the 2008 Current Population Survey. See Thorn File and Sarah Crissey, "Voting and Registration in the Election of November of 2008," Report Number P20-562 (Washington, D.C.: U.S. Department of Commerce, 2010), 13, http://www.census.gov/ prod/20iopubs/p20-562.pdf.
32 By way of example, the City of Boston's precinct lines have not been redrawn for decades, leading to some polling locations that serve thousands of voters in contrast to other locations that serve a few hundred.
33 We have had to face the problem of what to do when our results appeared to demonstrate that our pollsters approached too many voters. (In fact, the presence of non-voters exiting the building interfered with our pollsters' attempts to count ever fcth voter.)
34 See, for example, Lonna Rae Atkeson, "Sure, I Voted for the Winner! Overreport of the Primary Vote for the Party Nominee in the National Election Studies," Political Behavior 21 (1999): 197. Overreporting for the winner could come from several psychological mechanisms: voters remember correctly who they voted for but lie about it because they do not want to admit that they backed a loser; voters forget who they voted for but remember who won, and confuse the two; or voters who voted for the winner remember more clearly who they voted for than voters who voted for the loser, so the former are more likely to respond to a post-election poll, to name a few possibilities.
35 Rachael Cobb, D. James Greiner, and Kevin M. Quinn, "Can Voter ID Laws be Administered in a Race-Neutral Manner? Evidence from the City of Boston in 2008," Quarterly Journal of Political Science 7(1) (2012): 1-33.
36 See, for example, Corey Dade, "South Carolina Sues Justice Department for Blocking Its Voter ID Law," NPR blog, http ://www.npr.org/blogs/itsallpolitics/20i2/o2/o7/i465362i4/ south-carolina-sues-justice-department-for-blocking-its-voter-id-law.
37 See Rick Brady, "A Critical Review of The Unexplained Exit Poll Discrepancy," March 28, 2005, http://web.archive.0rg/web/20060502035232/http://stonescryout.org/files/unexplained.pdf.
3** "Evaluation of Edison/Mitofsky Election System 2004," January 2005, 53, http://abcnews.go .com/images/Politics/EvaluationofE disonMitofskyElectionSystem.pdf.
141 (4) Fall 2012
21
LongLive 39 Barreto, Guerra, Marks, Nuno, and Woods, "Controversies in Exit Polling." e Exit Poll
4° Some of the best exit polling in the nation is done in Utah by the Utah Colleges Exit Poll. This poll ordinarily assigns at least eight pollsters per precinct, four in each of two shifts. See Mark Blumenthal, "What About Those German Exit Polls?" Mystery Pollster, December 19, 2005, http://www.mysterypollster.com/main/2004/12/what_about_thos.html.
41 Some research has suggested that respondents' answers or propensities to answer change according to, say, the age (or race, or other characteristic) of the interviewer/pollster. On the correlation between the age of the pollster and response rate, see Daniel M. Merkle and Murray Edelman, "Nonresponse in Exit Polls: A Comprehensive Analysis," in Survey Nonresponse, ed. Robert M. Groves et al. (New York: Wiley, 2002), 243, 250. For a discussion of the so-called Bradley or Wilder effect in a preelection poll prior to a racially charged election, see Steven E. Finkel, Thomas M. Guterbock, and Marian J. Borg, "Race-of-the-Interviewer Effects in a Preelection Poll: Virginia 1989," Public Opinion Quarterly 55 (1991): 313.
42 Patrick R. Cotter, Jeffrey Cohen, and Philip B. Coulter, "Race-of-Interviewer Effects in Telephone Interviews," Public Opinion Quarterly 46 (1982): 278 ; Darren W. Davis, "The Direction of Race of Interviewer Effects among African-Americans: Donning the Black Mask," American Journal of Political Science 41 (1997): 309.
43 For an example of one such firm's public opinion polling operation, see http://www .ibopezogby.com/services/public-opinion-polling/.
44 See http ://pewinternet.org/Trend-Data/Whos-Online.aspx.
45 See, for example, http://catalist.us/product.
46See, for example, George F. Bishop and Bonnie S. Fisher, "Secret Ballots and Self-Reports in an Exit-Poll Experiment," Public Opinion Quarterly 59 (1995): 568.
47 42 U.S.C. sec. 1973 - i973aa-6.
48 See Section 5 of the Voting Rights Act of 1965, codified as amended at 42 U.S.C. 1973c. See also, Northwest Austin Municipal Utility District No. 1 v. Holder (2009) (avoiding the issue of Section 5's constitutionality).
49 Bush v. Gore, 531 U.S. 98 (2000).
22
Dcedalus, the Journal of the American Academy of Arts & Sciences