The Impact of Ballot Type on Voter Errors

Paul S. Herrnson University of Maryland

Michael J. Hanmer University of Maryland
Richard G. Niemi University of Rochester

Studies of ballots have traditionally focused on roll-off, candidate order, and partisan advantage. This study is among
the first to assess the impact of ballots on individual-level voter errors. We develop new hypotheses by bringing together
theoretical insights from usability research and political science about the effects of ballots with and without a straight-party
voting option. By comparing voters’ intentions to the votes they cast, we are able to create two measures of voter errors:
votes unintentionally cast for the wrong candidate and unintentional undervotes. Voters generally make fewer errors of
both types when using a standard office-bloc ballot than when using an office-bloc ballot with a straight-party option, with
the number of wrong-candidate errors substantially exceeding the number of unintentional undervotes. Voters’ background
characteristics have a significant impact on their ability to vote without error. Our results offer a new perspective for
evaluating the use of the straight-party option.

T
he 2000 presidential election ushered in a new 2000 (Wand et al. 2001) and the apparent design flaw that
awareness of and concern for how Americans cast affected the congressional election in Sarasota County in
their votes. In reaction, Congress passed the Help 2006 (Frisina et al. 2008).
America Vote Act of 2002, and many states purchased Ballots constitute a front line of study for a number
new voting systems, computerized their voter registration of reasons. First, the ballot is the means through which
records, and improved their overall election administra- voters register their intentions, and it is the dominant
tion. Registration, voting systems, and administration, feature voters observe once they begin the voting process.
however, are only part of what voters confront when cast- Symbolically, it is more meaningful. As Beard pointed out
ing their votes. Whatever the specific mechanisms and a century ago: “The point of contact between the average
procedures in place, voters make use of a ballot. Whether voter and his government is the ballot . . . ” (1909, 590).
a series of names and images programmed on a touch Second, varieties of ballot layouts, options, and tasks rou-
screen or printed on paper, the way in which the ballot is tinely affect the number of votes cast (undervoting), the
formatted has the potential to influence voters’ abilities expression of voter attitudes (initiatives), and the rela-
to cast their votes as intended and perhaps even influ- tive advantage to candidates and parties (candidate or-
ence the outcome of an election. This point was forcefully der, straight-party options).1 Third, the impact of ballot
made by the butterfly ballot that caused such a stir in design may be conditioned by the mechanism by which
Paul S. Herrnson is Director of the Center for American Politics and Citizenship and Professor of Government and Politics, 3140 Tydings
Hall, University of Maryland, College Park, MD 20742 ([email protected]). Michael J. Hanmer is Research Director of the Center for
American Politics and Citizenship and Associate Professor of Government and Politics, 3140 Tydings Hall, University of Maryland, College
Park, MD 20742 ([email protected]). Richard G. Niemi is Don Alonzo Watson Professor of Political Science, University of Rochester,
Rochester, NY 14627 ([email protected]).
This work was supported in part by National Science Foundation grants EAI-0201634, IIS-0306698, and Carnegie Corporation of New York
grant D05008. We would like to thank Benjamin B. Bederson, Frederick G. Conrad, and Michael W. Traugott for their collaboration on the
larger project. We would also like to thank the voting system vendors, the participants in our study, the staff and undergraduate and graduate
research assistants at the Center for American Politics and Citizenship of the University of Maryland, and numerous undergraduate and
graduate research assistants at the University of Maryland, University of Rochester, and University of Michigan. For helpful comments we
wish to thank Daniel Biggers, Ozan Kalkan, Irwin Morris, and participants in the University of Maryland American Politics Workshop, and
the Rooney Center Visiting Scholar Speaker Series at the University of Notre Dame. We also wish to thank the five anonymous reviewers
and the Editor for their close reading of the article and thoughtful suggestions and guidance. All errors are our own. The data used in this
study can be found at www.capc.umd.edu/research/data archive/Impact of Ballots on Voter Errors AJPS.html.
1
On undervoting, see Kimball and Kropf (2005, 2008) and Lausen (2007); on initiatives, Smith and Tolbert (2004); on candidate order,
Krosnick, Miller, and Tichy (2004).
American Journal of Political Science, Vol. 56, No. 3, July 2012, Pp. 716–730

C 2012, Midwest Political Science Association DOI: 10.1111/j.1540-5907.2011.00579.x

716
IMPACT OF BALLOT TYPE ON VOTER ERRORS 717

voting is carried out, an interaction that has long existed We then compare two of the most common types of
(Reynolds and McCormick 1986) but has only recently ballots—a standard office-bloc ballot and an office-bloc
been studied. Fourth, debates regarding ballot features are ballot with a straight-party option—on two of the most
driven by political considerations. Allen’s statement that dissimilar voting systems currently in widespread use—
“state legislatures are perpetually tinkering their ballot a touch-screen system and a paper-ballot system with a
laws in one particular or another” (1906, 38) was recently precinct-based optical scanner (opscan). Because our re-
validated by debates over the straight-party ballot in search assesses the effects of different types of ballots and
Illinois and Michigan (Kimball, Owens, and McLaughlin voting systems on voter errors, we are able to provide gen-
2002). Finally, voters may be differentially affected by eralized support for our theory of voter interactions with
types of ballots or ballot-voting system combinations ballots. Utilizing a large-scale field study, we examine the
(Buchler, Jarvis, and McNulty 2004; Herrnson et al. 2008), extent to which voters cast ballots for unintended candi-
raising issues of fairness and equality. dates (or ballot measures), egregious errors that reduce
Political scientists such as Allen and Beard recognized support for voters’ preferred choices, increase support for
over a century ago that ballot-related issues influence po- the opposition, and are impossible to detect with tradi-
litical behavior and can complicate the lives of election tional data sources. Our design also allows us to separate
officials, politicians, and voters. As early as 1909, Beard intentional from unintentional undervotes, only the latter
called for additional research on the subject. Over 60 years of which are errors, and to consider the impact of voters’
later, Rusk echoed a similar sentiment in the context of the background characteristics on various types of errors.
emergent political behavior literature, including ballots We find that voters generally make fewer errors
among the “neglected . . . institutional properties [that] of both types when using a standard office-bloc ballot
provide the framework within which the effects of other than when using an office-bloc ballot with a straight-
independent variables must be judged” (1970, 1220). And party option, with the number of wrong-candidate errors
very recently, scholars (see, e.g., Fife 2010) have linked substantially exceeding the number of unintentional un-
ballot-related issues to the further development of Amer- dervotes. These results hold both for touch-screen and
ican democracy. Only in the past few years, however, have paper-ballot/optical-scan voting systems. We also find
these sentiments spawned systematic research efforts. that voters’ demographic characteristics have a signifi-
Recent election fiascos also have highlighted the fact cant impact on their ability to navigate the straight-party
that ballot design, along with voting technology and elec- ballot. Overall, our results offer a new perspective for
tion administration, can have a negative impact on public evaluating the use of the straight-party option.
opinion toward the political system. Real and perceived
shortcomings can encourage voters to view specific bal-
lots, certain types of voting equipment, an individual elec- The Impact of Ballots on Voter
tion, and even elections generally as failures, civil rights vi- Accuracy
olations, or fraudulent processes (Alvarez and Hall 2008;
Fife 2010). Voter accuracy has rarely been studied at the individual
In this study, we contribute to the literature by an- level. Indeed, it is impossible to do so in situ because of
swering the long-standing call for additional research on ballot secrecy laws and practical considerations involving
ballots and how individuals interact with them, focus- recording what happens in the voting booth. Yet, there
ing on a specific feature—the straight-party option. We are reasons to believe that ballot design can make a con-
address one of the most important, most difficult, and siderable difference. Several recent real-life cases indicate
least studied concerns—the effect of ballot design on the influence of ballot design on voter behavior.3 The
individual-level voting errors. This focus requires a differ- best-known case is that of the butterfly ballot used in
ent approach to hypothesis development and data collec- Palm Beach County, Florida, in 2000, in which statistical
tion and some uniquely constructed dependent variables. analyses convincingly indicate that a substantial num-
We first demonstrate the usefulness of a theoretical ber of voters, enough to alter the outcome, cast a ballot
perspective that brings together insights from political for Reform party candidate Patrick Buchanan when they
science and the field of “usability” (Helander, Landauer,
and Prabhu 1997; Laskowski et al. 2004; Nielsen 1994).2
ity.” Although we focus solely on accuracy and errors, the other
issues should be the subject of future research.
2 3
Usability studies (also known as human factors psychology and in- Ballots and ballot handling in the late nineteenth and early twen-
dustrial design) typically assess systems in terms of accuracy, errors, tieth centuries also provided a wide array of opportunities for error
user satisfaction, speed, “learnability,” efficiency, and “memorabil- and skullduggery (Blodgett 1889; Reynolds and McCormick 1986).
718 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

intended to vote for Democrat Al Gore (Sinclair et al. column ballots and the straight-party option, in turn,
2000; Wand et al. 2001). In the 2006 congressional elec- were designed to lessen the harm to parties. They also
tion in Florida’s 13th district, the ballot was programmed were considered well-suited to a population with many
on a touch-screen system in such a way that it was the immigrant and uneducated voters because they facilitated
likely cause of more than 18,000 voters failing to cast a the selection of all of the candidates of one party in one
vote (Frisina et al. 2008); with just 369 votes separating the action. In discouraging voters from splitting their tickets
candidates, the layout of the ballot on the touch-screen or from failing to vote for offices below the top of the
system surely influenced the outcome. Likewise, Lausen ticket, straight-party options have been a success.6
(2007) shows that altering the design of the ballot for a
lengthy list of candidates in a Chicago judicial retention
election decreased dramatically the number of voters who Theory and Hypotheses
failed to vote. In New York State, long-standing anecdotal
evidence suggests that voting for multiple judicial candi- In contrast to the impact of straight-party options on
dates and ballot questions is problematic due to the ballot party voting, we know almost nothing about their effect
format on its lever machines. Recent systematic analysis on individual-level voter errors.7 We develop hypotheses
of full-face ballots bolsters this conclusion (Kimball and about such errors by integrating the political science and
Kropf 2008). usability literatures, bringing together insights regarding
An analysis of ballots by Niemi and Herrnson (2003) structure and function. The political science literature on
revealed numerous features of ballots that could lead to ballot design has focused on the structure ballots imposed
confusion and possibly error. One of their primary ob- on voters. That is, the focus has been on how the ballot me-
servations was that instructions for voting can be con- chanically translates the actions of the voter into the votes
fusing, especially as they relate to more complex tasks, that are recorded and the partisan advantages it provides.
such as voting for more than one candidate in a multi- As we discuss below, when a straight-party option is used,
candidate election or voting for an “exception” after us- it should reduce undervotes for partisan offices by auto-
ing a straight-party option (see also Campbell and Byrne matically recording votes in all partisan contests, thereby
2009). The order of listing candidates is another poten- advantaging candidates for lower office who belong to
tially confounding factor; rotating ballot order is done in the majority party. To this, we contribute insights regard-
some jurisdictions precisely because it is widely believed ing how ballot structure influences voter accuracy and
that candidates listed first are advantaged. Recent research the types of errors voters commit. Additionally, from the
also has demonstrated that ballot style can influence the usability literature we contribute insights regarding func-
number of votes that go unrecorded (see, e.g., Kimball tionality. That is, we draw on the main tenets of usability
and Kropf 2005). research: systems with simple, straightforward, end-to-
Ballots with a straight-party option, in effect in end designs, involving fewer steps, requiring little user
15 states in 2008, also increase straight-ticket voting memory, giving confirmation of one’s actions, and pro-
(Burden and Kimball 1998; Campbell and Miller 1957; viding system-based help are more effective than systems
Hamilton and Ladd 1996; Rusk 1970), diminish ballot that have added complexity, require individuals to re-
roll-off (Hamilton and Ladd 1996; Walker 1966), and member previous actions, do not provide assistance with
reduce overvoting (Kimball and Kropf 2005).4 Straight- cognitive tasks, and are inefficient (Helander, Landauer,
party options were introduced in the 1890s and early and Prabhu 1997). Some of the qualities that usability re-
1900s when U.S. political parties were strong and well search has demonstrated to improve human performance
organized (Burnham 1965), but were perceived as cor- 6
In some cases straight-party ballots have become victims of their
rupt and targeted by reformist impulses. The introduction own success. Illinois Republicans, for example, believed the use of
of official, state-administered ballots as replacements for straight-party ballots in their state made it too easy for Democrats to
party-distributed ballots that listed only one party’s can- vote for Democratic candidates all the way down the state’s lengthy
ballot. When they took control of the legislature and governor-
didates was intended to weaken party machines.5 Party ship, they removed the straight-party option from Illinois ballots
(Kimball, Owens, and McLaughlin 2002). See also the example of
4
The 15 states are Alabama, Indiana, Iowa, Kentucky, Michigan, Ohio in the 1940s (Walker 1966, 448).
New Mexico, North Carolina, Oklahoma, Pennsylvania, Rhode 7
Island, South Carolina, Texas, Utah, West Virginia, and Wisconsin Day and Becker (1984) came close to assessing error when they
(National Conference of State Legislators, http: //www.ncsl.org). showed from poll results that many Illinois voters in 1982 evidently
preferred the Republican James Thompson, while election results
5
Reynolds and McCormick (1986) argue that the effect of state- demonstrated a much larger vote for the Democrat Adlai Stevenson
administered ballots was actually to strengthen the two major par- III, apparently because many voters used a (Democratic) straight-
ties by reducing voting for independent candidates. party option.
IMPACT OF BALLOT TYPE ON VOTER ERRORS 719

on different types of interfaces, such as web sites, apply tations for partisan contests and usability informs our
more to the standard office-bloc ballot, but others apply expectations for nonpartisan contests. The straight-party
more to the office-bloc ballot with a straight-party option option reduces the number of steps involved in the vot-
(Norden et al. 2008; Redish et al. 2010). The presence of ing process by automatically casting votes in all partisan
these qualities also varies according to the tasks voters races, thus eliminating voter oversights; this is why some
are expected to perform and the voting systems on which politicians like it and others do not. Due to this design
they are expected to perform them. Thus, from a usability feature:
perspective, we should not expect any one ballot, voting
system, or ballot-voting system combination to perform H2: We expect the number of undervotes for partisan
better in all areas. Moreover, some combinations of bal- offices to be lower on the office-bloc ballot with a
lots and voting systems are likely to produce more errors straight-party option than on the standard office-
of one type than another. bloc ballot.
Our first hypothesis relates to wrong-candidate errors
(i.e., voting for a candidate other than the one intended). This should be especially true in contests in which more
According to the usability literature, complex processes than one candidate is to be elected, as in multimember
that do not give confirmation of one’s actions increase state legislative districts. For the nonpartisan offices, us-
errors. The office-bloc ballot with a straight-party op- ability insights lead to the opposite prediction. Given the
tion adds complexity to the voting process with minimal lack of understanding of or attention to the limits of the
(at best) feedback. Conceptually, this ballot creates chal- straight-party option, combined with a lack of feedback:
lenges for voters who do not understand the idea of a
straight-party vote. Because we informed voters in the H3: We expect the number of undervotes for nonpartisan
orientation and via the voter booklet that they would offices to be higher on the office-bloc ballot with a
be casting a straight-party vote, and provided a standard straight-party option than on the standard office-
set of instructions on the ballot, we reduced the con- bloc ballot.
ceptual complexity somewhat; as a result, our error rate
might actually be lower than in a real-world setting. Op- Both the social science and usability literatures em-
erationally, this ballot creates challenges for those who phasize that background characteristics have an impact
do not understand how the feature works on a partic- on an individual’s knowledge of and ability to master var-
ular voting system.8 Complexity is added in that voters ious tasks. A basic tenet of usability is that systems have
confront an additional set of instructions and a process to be designed so that users of all types have a positive
that works differently than casting votes office by office. and successful experience (e.g., Helander, Landauer, and
Moreover, there is little, if any, feedback for voters. Even Prabhu 1997). Work in political science has for a long
those who read the instructions may not know what, if time called attention to inequalities in political knowl-
anything, will happen after they make a selection using edge and interest (e.g., Delli Carpini and Keeter 1996).
the straight-party feature; and after making the selection, The straight-party option was developed with the goal of
whether they were successful. In this instance, the greater providing political advantage, not universal accessibility
complexity of the straight-party option overrides the fact or political equality. Both literatures suggest that voters
that it requires fewer actions on the part of the voter. who are older, have less education, and have lower lev-
Thus: els of computer usage or relevant voting experience are
likely to commit more wrong-candidate errors and more
H1: We expect voters to commit more wrong-candidate
unintentional undervotes when using the ballot with the
errors when using the office-bloc ballot with a
straight-party option. Thus, we control for these variables
straight-party option than the standard office-bloc
as well as for race and ethnicity even though the evidence
ballot.
on the latter is mixed.9
Significantly, from the structural perspective, one error
using the straight-party option is multiplied into many
when the wrong party is selected. 9
The literature on the digital divide suggests that blacks are less
A second pair of hypotheses relates to unintentional likely than whites or Asians to be exposed to new technologies
undervotes. Here, political science informs our expec- (e.g., Mossberger, Tolbert, and Stansbury 2003). Brady et al. (2001),
Kimball and Kropf (2005), and others show that blacks are more
likely to cast residual votes, but Herron and Sekhon (2005) demon-
8
For evidence on both points, see Campbell and Byrne (2009) and strate that some of these “errors” are due to the strategic behavior
our findings below. of some black voters.
720 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

Methodology the first box on the ballot, labeled “STRAIGHT PARTY

VOTING,” presents the names of all of the parties whose
In the field study described below, we tested the abilities candidates are listed on the ballot with an oval next to
of voters to cast their votes as intended when using two it. To vote for every candidate affiliated with one party,
ballot formats with long histories of use: a standard office- the voter fills in the oval next to the name of that party.
bloc ballot, which lists all of the candidates for one office Thus, the voter’s first decision is whether to vote for all
together; and an office-bloc ballot with a straight-party of a party’s candidates. The next decision is whether to
option, which is similar in all respects, except that it allows override that initial decision by filling in an oval next to
an individual to cast all of his or her votes for candidates a candidate who is not affiliated with that party, which
running in partisan contests through one action. results in that candidate receiving the vote. Subsequently,
Each ballot was tested on the two most widely the voter can then proceed to the part of the ballot that
used types of systems: a touch-screen voting system (the lists nonpartisan elections and ballot questions. The vote
Diebold AccuVote-TS) and a paper-ballot system with tabulator will count all of the votes for partisan races
precinct-based optical scanning (the ES&S Model 100). as votes cast for the party selected in the straight-party
These voting systems are among the most dissimilar in voting bloc. The voter slides the ballot into the optical
use, and each uniquely conditions how a voter perceives scanner to complete the process.
the ballot and thus has the potential to influence the Consistent with our theory, our observations indi-
impact of ballot style on voter errors. Touch-screen sys- cated that some voters were confused by the straight-
tems, used by 31% of the voting public in 2008, feature party option, despite being instructed on how to use it.
an interface similar to those of ATMs used in banking. For example, some filled in the oval next to their intended
Paper-ballot/opscan systems, used by roughly 60% of all party and then filled in the oval next to the name of every
voters, are similar to multiple-choice forms used in testing candidate affiliated with that party. This did not result in
and by government and private industry (Election Data any errors, but it was extremely inefficient and showed
Services 2008). the voters’ lack of conceptual understanding, operational
knowledge, or both.
Voting on a standard office-bloc ballot on a touch-
Ballot Styles and Voting Systems screen system, such as the Diebold AccuVote-TS, is also
quite straightforward. Voters touch the screen to select
Voting on a standard office-bloc ballot using an opscan candidates from whichever office or offices appear on the
system is fairly straightforward. The voter fills in the oval first screen; they then instruct the system to advance to
or circle next to the name of each candidate or position the next office or offices by touching the appropriate tar-
on a ballot question he or she intends to support. When get area on the screen. After voters navigate their way
the ballot is completed, the voter slides it into a paper through all of the offices and ballot questions (if any),
feeder in jurisdictions that have optical scanners on-site. one or more review screens show how they voted in each
This provides feedback about overvoting (voting for more race and indicate whether any races were missed, includ-
candidates than is allowed for a given office).10 From ing when fewer than the allowable number of candidates
the paper feeder the ballot drops into a large ballot box were selected. If they wish, voters can change their voting
resembling a safe. Like other opscan systems, the Model decisions until they are satisfied with the entire ballot. At
100 has no review screen on which voters can check the that point, they touch the target area on the screen to cast
accuracy of their selections, though they can take a second the ballot.
look at the ballot before inserting it into the paper feeder.11 Voting on an office-bloc ballot with a straight-party
Voting on an opscan system using an office-bloc bal- option on a touch screen is similar. As is the case with
lot with a straight-party option is similar except that the opscan system, the first decision a voter makes is
whether to initially cast every vote in partisan contests for
10
Voters who cast overvotes are alerted by a warning tone and
candidates of one party. The system provides a box that
a brief text message (e.g., too many votes—county commissioner). lists the name of every party that appears on the ballot
They have the option of removing their ballot and replacing it with and instructs the voter that by touching the screen next
a new one (the prescribed procedure), erasing a filled-in oval where to a given party name, he or she can vote for all of the
they overvoted, or pressing a button that activates the computer to
accept the ballot as is, complete with overvotes. On the model used candidates of that party. Voters who choose this option
in our study, overvotes were reported one at a time, beginning with and touch the target area to advance are automatically
the error appearing nearest the top of the ballot. moved down the ballot to the first nonpartisan election.
11
Pictures of the voting systems are included in the online appendix. They can then vote for the nonpartisan elections and
IMPACT OF BALLOT TYPE ON VOTER ERRORS 721

ballot questions that remain on the ballot. Voters who appeared further down the ballot. The information in-
wish to vote straight-party with exception (i.e., override cluded their party affiliation, political and professional
their straight-party vote for one or more candidates) can experience, and positions on prominent issues that typ-
either change a vote by paging back through the ballot ically divide contemporary voters, such as tax cuts and
and changing their vote when they arrive at the specific government spending.13 This information enabled par-
election or by navigating to the race of interest by selecting ticipants to discern key differences among candidates and
that race when they reach the review screen. Should a voter make substantive choices based on factors that many vot-
not wish to use the straight-party option, the screen that ers consider when casting their ballots. It helped par-
offers this option also instructs them to move to the next ticipants become engaged at the outset, kept them at-
page, where they begin voting for the first partisan office. tentive and careful throughout the voting process, and
added realism to voter decision making and our simu-
lated election. The seriousness with which the partici-
Study Design pants took their participation in our voting study was
visible in two ways. First, in the course of conducting
Our field study resembles other political science field the study, we observed the participants referring to their
research and possesses properties common to usability voter guides while casting their votes.14 Second, when
studies. We assembled a demographically diverse group analyzing the data, we found that an extraordinary num-
of participants comprising the traits of the population ber of participants—more than 50%—took advantage of
of interest, asked them to perform a set of representative the opportunity to provide handwritten comments de-
tasks, and observed their performance of those tasks us- scribing their impressions of the voting process. Further
ing both obtrusive and unobtrusive measures.12 The data authenticity was created by asking voters to choose two
we analyze in this article focus primarily on voter errors. candidates for associate justice of the state supreme court
An expert review by usability specialists and a laboratory and two candidates for state representative, as these are
study helped us develop hypotheses about voter inter- multicandidate races in a number of states, and by provid-
actions with ballots and interpret the field study results ing voters with a summary paragraph for each of the ballot
(Herrnson et al. 2008). issues and asking them to circle “yes” or “no,” depending
The field study involved a diverse group of about on their preferences. Of course, providing substantive in-
1,540 participants recruited from university student bod- formation about candidates and issues, and employing a
ies, faculty, and professional and maintenance staffs; up- voter guide, takes time; in the interest of preventing the
scale and downscale shopping malls; senior citizens’ facili- simulation from becoming too long, voters were provided
ties; community centers; and professional office buildings with less information, comprising only party affiliation
in three states. The selection of states ensured that some (a standing voting cue) for some lower-level candidates,
respondents had been exposed to a straight-party option and instructed to vote for a preselected candidate for some
(Michigan) and some had not (Maryland and New York), offices. Voters were instructed to circle the names of all
as well as a variety of voting systems. After receiving a of the candidates and issues for whom they intended to
brief orientation to the project, participants were asked vote.15
to read a voter guide (or information booklet), one for use The ballot we used was realistic in content and re-
with a standard office-bloc ballot or one for use with the sembled a ballot that individuals would encounter in the
office-bloc ballot with a straight-party option. The guide
resembled the nonpartisan sample ballots mailed to vot-
ers by election officials in some localities and published 13
The voting guides are reproduced in online Appendix Figure 1.
in many local newspapers. Voters frequently circle their
14
choices on voter guides and take them into the voting This personal observation also enabled us to exclude data col-
lected from the few individuals we observed who were not serious
booth. By recording their voting decisions on a guide and about their participation in the study.
using it at the polling place, voters are able to cast their 15
Research by Everett, Byrne, and Greene (2006) found no dif-
ballots without relying entirely on memory. ference in error rates based on whether participants were able to
The guides included a paragraph of descriptive in- choose freely or were given a set of candidates for whom to vote
formation for the candidates for president and for can- (undirected and directed voting) or on the use of real or fictional
candidates. More recent research by the same team (Byrne, Greene,
didates for some state, local, and nonpartisan offices that and Everett 2007) found that fewer errors were produced when
voters were directed, suggesting that results “based on directed vot-
ing may be underestimating the real error rate” (176). The latter
12
For a description of the respondents, see online Appendix results were not available at the time of our tests, so we used a mix
Table 1a. of directed and undirected voting.
722 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

voting booth.16 It consisted of 18 offices and four bal- Finally, participants concluded the study by complet-
lot questions and contained the usual features, including ing a questionnaire to ascertain background information,
partisan and nonpartisan sections; elections for federal, such as their prior voting experience, age, race, educa-
state, and local offices; an uncontested race; and races tion level, and other relevant information. Background
with two candidates to be elected. The candidates in- information, the voter guides (which recorded voter in-
cluded men and women whose names represented various tentions), actual votes cast, and reactions to the voting ex-
ethnic backgrounds and were associated with the Demo- perience using each system were matched using a unique
cratic, Republican, and other contemporary parties. The name cast for a write-in vote and some identification
ballot questions were taken from actual ballots. numbers.18
Once they had completed the voter guide, partici-
pants were directed, in random fashion, to vote on one of
the voting systems. Written instructions on how to vote Dependent Variables
were posted at each system, and research assistants, play-
ing the role of election judges, were available to answer Our dependent variables measure two different types of
voters’ questions and help those who appeared to have dif- voter errors—wrong-candidate errors and unintentional
ficulty completing a ballot. Once voting was completed, undervotes. Wrong-candidate errors occur when a voter
voters filled out a questionnaire recording their reactions. intends to cast a vote for a specific candidate (as indicated
Then, they voted on another voting system as determined by his or her selection on a voter guide) and mistakenly
by the randomization process. The number of partici- casts a vote for another candidate. Unintentional under-
pants for each type of ballot was roughly the same; each votes occur when a voter intends to cast a vote for a can-
participant used the same type of ballot throughout the didate for an office and fails to vote for any candidate for
study. Because we were only able to obtain one voting that office. The latter measure contains one component
system of each type and the systems needed to be re- of the “residual vote” used by the CalTech/MIT Voting
programmed in order to change ballot styles, we were Technology Project (2001), which combines the number
unable to randomly assign participants to the two ballot of overvotes, undervotes, and spoiled ballots into a sin-
types. However, both ballots were deployed in the same gle aggregate-level measure. The residual vote has been
locations where possible or in similar locations to control an important starting point for comparisons of the per-
for the possibility of systematic bias. Although we had formance of different voting systems and ballot designs
little reason to worry about bias resulting from differ- across different jurisdictions and types of voters (e.g.,
ences in the types of respondents assigned to each ballot Brady et al. 2001; Hanmer et al. 2010; Tomz and Van
type, to alleviate potential concerns we used a propensity Houweling 2003). One of its major advantages is that it
score matching procedure to preprocess the data prior can be used in studies of actual elections. Our measure
to implementing our statistical tests (see, e.g., Morgan of undervotes has the benefit of separating unintentional
and Winship 2007). Specifically, we used kernel matching undervotes (committed in error) from intentional under-
to balance the individual characteristics across the two votes (committed deliberately). Given that undervotes are
types of ballots, with the matching done separately for often intentional (Herron and Sekhon 2005), distinguish-
each voting system.17 Individuals were matched across ing between the two types is important to measuring
the two ballot types based on measures of education, age, and understanding errors resulting from voter interac-
race, computer usage, partisan strength, experience with tions with ballots. Another improvement associated with
a similar voting system, and location. The balance across our measures of voter errors is that they are based on
individual characteristics was good before matching and individual-level data and thus avoid the ecological fallacy.
improved after matching (see online Appendix Table 1b). The metrics in this study enable us to assess the in-
teraction of ballot designs, voting systems, and citizens’
traits on their abilities to cast their ballots as intended.
16
There were, in fact, two ballots, but they differed only in whether Most important, our research design makes it possible
they contained a straight-party feature. We designed our ballot after
reviewing ballots used between 1998 and 2000 that were collected 18
from all 50 states. Abbreviated versions of the ballots are reproduced All write-in names were of the same length and relatively simple.
in online Appendix Figures 2 and 3. Voters who failed to enter a name that could be matched with a
booklet (about 10%) were excluded from the analysis because there
17
Kernel matching is one of the approaches recommended by was no way to determine whether they voted accurately. Errors on
Morgan and Winship (2007, 109). Matching was implemented the write-in for those who could be matched are strongly related
using software provided in Leuven and Sianesi (2003). We ob- to errors on other parts of the ballot; thus, eliminating the cases
tained substantive results that were similar when trying the other with the worst errors on the write-in almost surely leads to slight
approaches. underestimates of the error rates.
IMPACT OF BALLOT TYPE ON VOTER ERRORS 723

to determine directly whether an individual casts a vote and one for unintentional undervotes. As noted earlier,
for the intended candidate or for some other candidate. these analyses were done after preprocessing the data via
This is important because voting for the wrong candi- matching techniques. Because our dependent variables
date is worse than failing to vote in a race altogether, as are counts, with an expectation for overdispersion, we use
one candidate inadvertently loses a vote and an opponent negative binomial regression (Long 1997).21 We include
receives it. As a practical matter, this error is especially measures of the age, education, race, computer use, ballot
egregious because the erroneously selected candidate is type, and previous experience with the type of voting
most likely to be a candidate listed on the ballot imme- system being used. We also include controls for strength
diately before or after the desired candidate and often of partisanship (which functions as a surrogate for voter
that candidate’s major competitor. Finally, we combine interest in and knowledge about politics), state (where we
wrong-candidate errors and unintentional undervote er- capture whether a voter would have been exposed to a
rors to create a measure of overall voter errors that is straight-party option or a particular voting system), and
not clouded by the inclusion of intentional undervotes or the order in which the individual voted on the particular
deficient in its omission of wrong-candidate votes.19 voting system (to control for the possibility of learning
Although our research has the advantage of allowing effects).
for the study of individual-level differences in interac-
tions with ballots, it is not without shortcomings. Among
them is that we were able to introduce only a limited Results
number of treatments. For example, due to reliance on
manufacturers to provide and program the voting sys- We begin with a test of our initial hypotheses in the context
tems, we could test only one (albeit the most widely used of standard races where the task is simply to cast a vote
and simple) form of straight-party option and only one for one candidate. Our first two hypotheses suggest that
ballot-marking system (fill-in-the-oval).20 These and re- voters using the ballot with the straight-party option are
lated issues (such as the use of a convenience sample) more likely to cast votes in all partisan races than voters
result in findings that have strong internal validity, but using the standard office-bloc ballot (H2), but more of
less external validity than the findings of aggregate-level those votes are likely to be cast for candidates other than
voting studies. Another limitation is we did not evaluate the ones voters intend to support (H1).
variations in the time it took to vote across the two bal- The results provide strong support for our hypoth-
lot types. Since election officials might support straight- esis regarding wrong-candidate errors. For the two races
party voting due to its potential to reduce the amount of at the top of the ballot, the wrong-candidate error rates
time it takes to vote, future research should examine the on the ballots with a straight-party option are more than
relationship between errors and time spent voting. three times as large as those on the standard office-bloc
ballot (see Table 1). For example, on the touch-screen
voting system, 4.1% of the voters made wrong-candidate
Statistical Techniques errors on the ballot with a straight-party option when
voting for president compared to 1.2% of voters us-
After providing an overview of the data and testing some ing the standard office-bloc ballot. The results for the
basic hypotheses with descriptive statistics, we perform Senate election are similar. It is noteworthy that in the
two multivariate analyses: one for wrong-candidate errors race for president across both ballot types, and both voting

19
We do not test for errors that can be committed on only one 21
system, such as voting for more candidates than allowable, making Other election studies use a similar approach when ordinary least-
stray marks on or tearing a paper ballot, screen freezes caused by squares regression (OLS) is not appropriate (Long 1997), such as
dragging a finger down the screen, or other equipment failures. when assessing residual votes generated by small numbers of voters
(e.g., Kimball, Kropf, and Battles 2006). The same concern holds
20
We chose to test an office-bloc ballot over a party-column (or row) for wrong-candidate errors and unintentional undervotes. First,
ballot because most DREs are only programmed with an office-bloc the distributions of both types of error have a lower bound of 0 and
ballot. Also, an examination of ballots in all 50 states in 1998 or are strongly skewed, with most of the observations clustered at 0.
2000, prior to the adoption of DREs, showed that the office-bloc Second, we expect that errors across the ballot should not be treated
ballot was much more common, as only 10 states used a party- as independent events, as those who make one error (e.g., due to
column/row ballot. The ballot we tested provides one straight-party problems of understanding the ballot or voting system) are likely
option that applies to all partisan races; others allow for more (e.g., to make additional errors. Thus, the nature of our data violates
Oklahoma), and the North Carolina straight-party mark excludes some of the most basic assumptions of OLS. As expected, the alpha
presidential electors. Another ballot-marking system involves con- estimated from each of the negative binomial regression models is
necting arrows to indicate vote choice. We allowed voters to scan positive and statistically significant, supporting our concern with
paper ballots for errors, an option some jurisdictions do not allow. overdispersion.
724 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

TABLE 1 Type and Percentage of Errors for Races at the Top of the Ballot, by Ballot Type and Voting
System
Touch Screen (Diebold) Paper Ballot/Opscan (ES&S)

President Straight-party Office-bloc Prob.a Straight-party Office-bloc Prob.a

Wrong candidate 4.1 1.2 0.013 5.6 1.8 0.005
Undervote 1.0 0.1 0.075 0.2 0.0 0.318
Total Error Rate 5.1 1.3 0.003 5.8 1.8 0.003
U.S. Senate
Wrong candidate 3.6 1.5 0.072 4.2 1.2 0.010
Undervote 1.0 0.3 0.234 0.2 1.3 0.017
Total Error Rate 4.5 1.8 0.032 4.4 2.5 0.129
(N) (507) (634) (503) (635)
a
p-value (two-tailed) is for the difference between straight-party and office-bloc ballots.
Notes: Analysis done after preprocessing the data via kernel matching. All undervotes are unintentional undervotes. The total of wrong-
candidate and undervote error rates does not always add to the total error rate due to rounding.

systems, substantially more voters made candidate selec- complete all of the votes for the partisan office in one
tion errors—the more severe of the two types of error— action, without having to recognize the new instruction
than unintentional undervotes. regarding the ability to vote for two (H2). As anticipated,
With respect to our hypotheses regarding undervotes, voters using the straight-party option on the opscan sys-
the results indicate that voters committed few of them tem cast fewer undervotes than those who used the stan-
but that the straight-party option generally did not reduce dard office-bloc ballot. Though there also were fewer un-
their frequency. An exception is found in the Senate race dervotes when voters used the straight-party ballot on the
for those using the opscan system; here, the percentage touch-screen system, the difference was neither substan-
of unintentional undervotes was considerably smaller on tively nor statistically significant.
the straight-party ballot than on the standard office-bloc The results for the nonpartisan election for associate
ballot. It is worth noting that, despite the dissimilarities justice, where the straight-party option is not operative,
in the interfaces of the two voting systems, for the most help us to better understand the straight-party option’s ef-
part the error rates for the paper-ballot/opscan system fect when more than one candidate is to be selected (H3).
were similar to those for the touch-screen system. This In the partisan two-to-be-elected race, the straight-party
is somewhat surprising given the public criticisms raised option automatically chooses two candidates of the same
about electronic voting systems (e.g., Alvarez and Hall party; in the nonpartisan race, the voter must follow the
2008), but it is consistent with research showing that vot- instructions and deliberately select two candidates. Com-
ers rate various touch-screen systems as favorably, if not paring the two contests demonstrates that voters using
more so, than opscan systems (de Jong, van Hoof, and the straight-party option commit fewer undervotes when
Gosselt 2008; Herrnson et al. 2008, chap. 3). the straight-party mechanism is operative (in the race
When further complications are added to the voting for state representative) than when it is not (in the race
task, the error rates and types of errors can change dra- for associate justice). While the undervote rates on the
matically. Table 2 presents the results for races in which standard office-bloc ballot for both races were similar, on
two candidates were to be elected: the race for state repre- the ballot with the straight-party option the percentage
sentative, a partisan contest; and the race for state supreme of undervotes in the nonpartisan race was much higher,
court associate justice, a nonpartisan race. In both races, by a factor of almost three, than in the partisan race.22
voters were asked to “Vote for no more than TWO” can- The final set of descriptive statistics examines over-
didates. Only voting for the second candidate is shown— all error rates across the ballot. The analysis includes
i.e., the candidate most likely to be overlooked.
For the partisan race, we again expected that voters 22
While the comparison is complicated by the fact that the election
would cast fewer undervotes on the ballot with a straight- for associate justice is further down the ballot, the larger number of
undervotes relative to wrong-candidate errors in the justice election
party option than on the standard office-bloc ballot in comparison to the state representative election demonstrates that
because the straight-party option enables the voter to the location of the election on the ballot does not drive the result.
IMPACT OF BALLOT TYPE ON VOTER ERRORS 725

TABLE 2 Type and Percentage of Errors for Races with Complex Tasks (“Vote for no more than
TWO”), by Ballot Type and Voting System
State Representative (partisan; second choice, when voting for two)
Touch Screen (Diebold) Paper Ballot/Opscan (ES&S)

Straight-party Office-bloc Prob.a Straight-party Office-bloc Prob.a

Wrong candidate 3.2 2.0 0.299 3.8 1.6 0.050
Undervote 1.6 1.9 0.780 2.2 4.5 0.058
Total error rate 4.7 3.9 0.576 6.0 6.1 0.928
State Supreme Court Associate Justice (nonpartisan; second choice, when voting for two)
Straight-party Office-bloc Prob.a Straight-party Office-bloc Prob.a
Wrong candidate 4.5 2.6 0.165 2.0 0.9 0.234
Undervote 4.5 2.9 0.201 6.2 3.6 0.074
Total error rate 9.1 5.5 0.054 8.2 4.4 0.029
(N) (507) (634) (503) (635)
a
p-value (two-tailed) is for the difference between straight-party and office-bloc ballots.
Notes: Analysis done after preprocessing the data via kernel matching. All undervotes are unintentional undervotes. The total of wrong-
candidate and undervote error rates does not always add to the total error rate due to rounding.

TABLE 3 Errors across the Ballot, by Ballot Type and Voting System
Touch Screen (Diebold) Paper Ballot/Opscan (ES&S)

Straight-party Office-bloc Prob.a Straight-party Office-bloc Prob.a

Mean number of errors 0.9 0.6 0.009 1.0 0.7 0.008
Mean number of votes for the wrong candidate 0.7 0.4 0.004 0.8 0.4 0.000
Mean number of undervotes 0.2 0.2 0.803 0.2 0.3 0.497
Partisan offices 0.07 0.08 0.826 0.04 0.10 0.017
Nonpartisan offices 0.10 0.08 0.478 0.19 0.16 0.589
Percentage of voters who made at least one error 30 30 0.968 35 30 0.153
(N) (507) (634) (503) (635)
a
p-value (two-tailed) is for the difference between straight-party and office-bloc ballots.
Notes: Analysis done after preprocessing the data via kernel matching. All undervotes are unintentional undervotes.

18 offices and 20 votes (two offices were races where two candidate was a larger problem than unintentional un-
were to be elected).23 The number of errors per ballot is dervoting, a result that holds for both the touch-screen
seemingly small, averaging just over one-half to one per and opscan systems. We find modest support for H2, that
ballot (see Table 3), but they are consequential. As sug- undervotes on partisan contests are lower on the ballot
gested by H1, the error rates were significantly larger on with a straight-party option.24 However, contrary to H3,
the office-bloc ballot with the straight-party option. As undervotes on the nonpartisan portion of the ballot do
with the specific races shown above, voting for the wrong not differ by ballot type.
When viewed another way, the magnitude of the er-
23
Given complexities in the instructions for voting for the U.S. ror rate is considerably more striking. Depending on the
House of Representatives, Secretary of State, and Attorney General, ballot type and voting system, between 30% and 35% of
we exclude these races from the analysis. For the latter two, the
instructions precluded the possibility of certain types of errors on the voters cast a ballot with at least one error (see the bot-
both ballots. For the U.S. House race, we asked only those voters tom row of Table 3). The error rate was the same across the
using the straight-party ballot to go back and change their vote, so
we cannot separate the effect of the task from that of the ballot. The
24
layout of the ballots makes it highly unlikely that votes for other Voters were expected to have (or were assigned) preferences in
races were influenced by this additional task. Our observations all but one partisan office. Where they lacked such preferences, the
from the field test corroborate this. straight-party option would likely have a greater effect.
726 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

FIGURE 1 Predicted Number of Wrong-Candidate Errors across the Ballot by Ballot Type,
Voting System, and Individual Characteristics

Notes: Results are based on estimates from online Appendix Tables 3 and 4. Calculations were made by manipulating the
values for the ballot while holding all other values at their observed values (see Hanmer and Kalkan 2011). All differences
between the straight-party ballot and the office-bloc ballot are statistically different at p < 0.05 (two-tailed). The basic pattern
of higher errors on the straight-party ballot holds for the control variables (not shown).
IMPACT OF BALLOT TYPE ON VOTER ERRORS 727

FIGURE 2 Predicted Number of Unintentional Undervotes across the Ballot by Ballot

Type, Voting System, and Individual Characteristics

Notes: Results are based on estimates from online Appendix Tables 5 and 6. Calculations were made by manipulating the
values for the ballot while holding all other values at their observed values (see Hanmer and Kalkan 2011). None of the
differences between the straight-party ballot and the office-bloc ballot are statistically different at p < 0.05 (two-tailed).
Similar conclusions are drawn for the control variables (not shown).
728 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

two ballot types when the touch screen was used, but on system, voters who are 75 and older cast more undervotes
the opscan system the percentage of ballots with at least with the straight-party option than the standard office-
one error was greater on the ballot with the straight-party bloc ballot. Perhaps the most interesting result is that
option.25 regardless of voter characteristics, the number of unin-
The results in Figure 1 (generated from negative bino- tentional undervotes was almost identical across the two
mial regressions) predict the number of wrong-candidate ballot types on the opscan voting system. However, not
errors across the ballot by voter characteristics.26 First, too much should be made of this apparent equality across
as expected, voters over the age of 75 make substantially groups because, as our earlier results show, unintentional
more errors on the straight-party ballot than on the stan- overvotes occur less frequently than wrong-candidate er-
dard office-bloc ballot.27 For example, on the opscan sys- rors, especially among blacks and the least educated.
tem the mean number of wrong-candidate errors on the
standard office-bloc ballot for those aged 75 and up was
0.82, while voters of that age group are predicted to av-
erage almost 1.5 errors on the straight-party ballot. The Conclusion
results are similar for the touch-screen system.28
Also as expected, when the least educated use the Elections in the first decade of the twenty-first century
straight-party option the average number of candidate have focused attention, to an extent unprecedented since
errors is high, regardless of the voting system used. Error the adoption of the secret ballot a century earlier, on the
rates across ballot type also differ by voters’ racial or eth- methods citizens use to vote. Although recent media cov-
nic backgrounds. Voters of all races make more errors on erage has focused heavily on voting systems, it has become
the ballot with the straight-party option than the stan- clear over the course of a decade that the entire electoral
dard office-bloc ballot, but non-Hispanic blacks make process, from registration through vote counting, is in
considerably more errors when using the ballot with the need of thorough examination. Our research has focused
straight-party option. Voters who never used a computer on one of the least studied parts of the process—how
tend to have high error rates in general, and as expected, individuals interact with the ballot.
these voters committed more errors when they used the Using a field study that allows us to match voter
straight-party option. intent with votes cast, we demonstrate that voters who
The differences across ballot types for unintentional use standard office-bloc ballots make fewer candidate-
undervotes are generally smaller and less consistent across selection errors and fewer total errors than those who
individual characteristics and voting systems, proba- use ballots with a straight-party option. Voters who use
bly because so few undervotes occurred (see Figure 2). a straight-party option have fewer unintentional under-
Nevertheless, the results show that with a touch-screen votes for partisan offices, especially when multiple can-
didates are to be selected, but they commit as many or
25
The error rates are surprisingly high. But in an experiment using more errors of omission in nonpartisan races. Our re-
college students, who are used to taking standardized tests, and
paper ballots, the error rate was 1% and 14 of 126 ballots (11%) sults do not permit estimation of the number of errors in
had at least one error. See Everett, Byrne, and Greene (2006). any given election because errors will vary from time to
26
Because negative binomial regression provides coefficient esti- time and from one jurisdiction to another depending on
mates that are difficult to interpret directly, we focus on the more factors unique to each contest. However, the surprisingly
substantively interesting predicted counts generated by the mod- large number of errors and the greater number of and
els. Moreover, since the negative binomial model is nonlinear, the
estimates of the predicted number of errors for a change in a given patterns of errors among those who use a ballot with a
independent variable depend on the value of all of the independent straight-party option suggest that this ballot feature is fre-
variables in the model. We report the model results for Figures 1 quently misunderstood and misused, especially by older
and 2 in online Appendix Tables 3–6. While one can report factor
or percentage changes, which are not sensitive to the value of the
and less educated voters.
other independent variables, they are less interesting and poten- Our results highlight the importance of individual-
tially misleading. level research that records voting intentions in order to
27
Calculations were made by manipulating the values for the ballot learn whether new ballot designs or voting systems im-
while holding all other values at their actual values (see Hanmer and prove voters’ abilities to cast their votes accurately. The
Kalkan 2011). To assess the statistical significance of the effect of findings also have implications for public policy beyond
the ballot on the number of errors committed, confidence intervals
were constructed using stochastic simulation (see Herron 1999). their ballot-specific results. The relatively high incidence
28 of wrong-candidate errors among older, less educated,
For each voting system, the differences between the ballot with
the straight-party option and the standard office-bloc ballot are and, evidently, among black voters who use a ballot with a
statistically significant at p < 0.05 (two-tailed). straight-party option has repercussions for civil rights and
IMPACT OF BALLOT TYPE ON VOTER ERRORS 729

voter education. They demonstrate that although perhaps Burnham, Walter Dean. 1965. “The Changing Shape of the
not intended, ballot design—like voter identification and American Political Universe.” American Political Science
Review 59(1): 7–28.
registration laws, the allocation of voting equipment, and
the financing of campaigns—can increase the political Byrne, Michael D., Kristen Greene, and Sarah P. Everett. 2007.
“Usability of Voting Systems: Baseline Data for Paper, Punch
disadvantages of traditionally underrepresented groups. Cards, and Lever Machines.” Proceedings of the SIGCHI
More generally, the findings suggest that making more Conference on Human Factors in Computing Systems, San
voter education and assistance available at precincts where Jose, CA.
numerous senior citizens and less educated persons vote Caltech/MIT Voting Technology Project. 2001. Voting: What Is,
could help equalize the voting experience. We hope ad- What Could Be. http://vote.caltech.edu/drupal/node/10.
ditional light is shed on this issue through future studies Campbell, Angus, and Warren E. Miller. 1957. “The Motiva-
tional Basis of Straight and Split Ticket Voting.” American
that investigate other aspects of usability, such as effi- Political Science Review 51(2): 293–312.
ciency and satisfaction.
Campbell, Bryan A., and Michael D. Byrne. 2009. “Straight-
Challenges that voters face when they go to the polls Party Voting: What Do Voters Think?” IEEE Transactions on
involve more than voting systems; they also involve bal- Information Forensics and Security 4(4): 718–28.
lots. Given that most state legislatures already have pur- Day, Richard, and Kurt M. Becker. 1984. “Preelection Polling
chased new voting systems, it seems reasonable that legis- in the 1982 Illinois Gubernatorial Contest.” Public Opinion
lators and reformers turn their attention in the future to Quarterly 48(3): 606–14.
another fundamental component of the voting process— de Jong, Menno Joris van Hoof, and Jordy Gosselt. 2008. “Vot-
ers’ Perceptions of Voting Technology: Paper Ballots versus
the ballot itself. Ballot features designed for the hand- Voting Machine with and without Paper Audit Trail.” Social
marked, hand-counted paper ballots introduced at the Science Computer Review 26(3): 399–410.
end of the nineteenth century lead to complexities and Delli Carpini, Michael, and Scott Keeter. 1996. What Americans
errors when used on modern touch-screen and paper- Know about Politics and Why It Matters. New Haven, CT:
ballot/optical-scan voting systems. Simplifying ballots Yale University Press.
and configuring them to suit the operations of specific Election Data Services. 2008. “National Voting Equipment
voting systems could go a long way toward reducing voter Report.” Washington, DC: EDS.
errors and improving elections. Voter education pro- Everett, Sarah P., Michael D. Byrne, and Kristen Greene. 2006.
“Measuring the Usability of Paper Ballots: Efficiency, Effec-
grams and informed on-site assistance could have similar tiveness, and Satisfaction.” Proceedings of the 50th annual
effects. Given the centrality of elections to representative meeting of the Human Factors and Ergonomics Society, San
democracy, such efforts are warranted. Francisco.
Fife, Brian L. 2010. Reforming the Electoral Process in America.
Santa Barbara, CA: Praeger.
Frisina, Laurin, Michael C. Herron, James Honaker, and Jeffrey
References B. Lewis. 2008. “Ballot Formats, Touchscreens, and Under-
votes: A Study of the 2006 Midterm Elections in Florida.”
Allen, Philip L. 1906. “Ballot Laws and Their Workings.” Polit- Election Law Journal 7(1): 25–47.
ical Science Quarterly 21(1): 38–58.
Hamilton, James T., and Helen F. Ladd. 1996. “Biased Ballots?
Alvarez, R. Michael, and Thad E. Hall. 2008. Electronic Elections: The Impact of Ballot Structure on North Carolina Elections
The Perils and Promises of Digital Democracy. Princeton, NJ: in 1992.” Public Choice 87(3/4): 259–80.
Princeton University Press.
Hanmer, Michael J., and K. Ozan Kalkan. 2011. “Behind the
Beard, Charles A. 1909. “The Ballot’s Burden.” Political Science Curve: Clarifying the Best Approach to Calculating Pre-
Quarterly 24(4): 589–614. dicted Probabilities and Marginal Effects from Limited De-
Blodgett, James H. 1889. “Suffrage and Its Mechanism in Great pendent Variable Models.” Working paper, University of
Britain and the United States.” American Anthropologist 2(1): Maryland.
63–74. Hanmer, Michael J., Won-ho Park, Michael W. Traugott,
Brady, Henry E., Justin Buchler, Matt Jarvis, and John Mc- Richard G. Niemi, Paul S. Herrnson, Frederick G. Conrad,
Nulty. 2001. Counting All the Votes. Survey Research and Benjamin B. Bederson. 2010. “Losing Fewer Votes: The
Center and Institute of Governmental Studies, Univer- Impact of Changing Voting Systems on Residual Votes.”
sity of California, Berkeley. http://ucdata.berkeley.edu/ Political Research Quarterly 63(1): 129–42.
new web/countingallthevotes.pdf. Helander, Martin G., Thomas K. Landauer, and Prasad V.
Buchler, Justin, Matthew Jarvis, and John E. McNulty. 2004. Prabhu. 1997. Handbook of Human-Computer Interaction.
“Punch Card Technology and the Racial Gap in Residual New York: Elsevier.
Votes.” Perspectives on Politics 2(3): 517–24. Herrnson, Paul, Richard G. Niemi, Michael J. Hanmer,
Burden, Barry C., and David C. Kimball. 1998. “A New Ap- Benjamin B. Bederson, Frederick G. Conrad, and Michael
proach to the Study of Ticket Splitting.” American Political W. Traugott. 2008. Voting Technology: The Not-So-Simple Act
Science Review 92(3): 533–44. of Casting a Ballot. Washington, DC: Brookings.
730 PAUL S. HERRNSON, MICHAEL J. HANMER, AND RICHARD G. NIEMI

Herron, Michael C. 1999. “Post-Estimation Uncertainty in Lim- Sinclair, Robert C., Melvin M. Mark, Sean E. Moore, Carrie A.
ited Dependent Variable Models.” Political Analysis 8(1): Lavis, and Alexander S. Soldat. 2000. “An Electoral Butterfly
83–98. Effect.” Nature 408: 665–66.
Herron, Michael, and Jasjeet S. Sekhon. 2005. “Black Candidates Smith, Daniel, and Caroline J. Tolbert. 2004. Educated by Ini-
and Black Voters: Assessing the Impact of Candidate Race tiative. Ann Arbor: University of Michigan Press.
on Uncounted Vote Rates.” Journal of Politics 67(1): 154–77. Tomz, Michael, and Robert P. Van Houweling. 2003. “How
Kimball, David C., and Martha Kropf. 2005. “Ballot Design and Does Voting Equipment Affect the Racial Gap in Voided
Unrecorded Votes on Paper-Based Ballots.” Public Opinion Ballots?” American Journal of Political Science 47(1): 46–
Quarterly 69(4): 508–29. 60.
Kimball, David C., and Martha Kropf. 2008. “Voting Technol- Walker, Jack L. 1966. “Ballot Forms and Voter Fatigue: An Anal-
ogy, Ballot Measures, and Residual Votes.” American Politics ysis of the Office Bloc and Party Column Ballots.” Midwest
Research 36(4): 479–509. Journal of Political Science 10(4): 448–63.
Kimball, David C., Martha Kropf, and Lindsay Battles. 2006. Wand, Jonathan N., Kenneth W. Shotts, Jasjeet S. Sekhon, Walter
“Helping America Vote?” Election Law Journal 5(4): 447–61. R. Mebane Jr., Michael C. Herron, and Henry E. Brady.
Kimball, David C., Chris Owens, and Matt McLaughlin. 2002. 2001. “The Butterfly Did It: The Aberrant Vote for Buchanan
“Straight Party Ballot Options and State Legislative Elec- in Palm Beach County, Florida.” American Political Science
tions.” Conference on State Politics and Policy, Milwaukee, Review 95(4): 793–810.
WI.
Krosnick, Jon A., Joanne M. Miller, and Michael P. Tichy. 2004.
“An Unrecognized Need for Ballot Reform: The Effects of
Candidate Name Order on Election Outcomes.” In Rethink-
ing the Vote, ed. Anne N. Crigler, Marion R. Just, and Edward
J. McCaffery. New York: Oxford University Press, 51–74.
Supporting Information
Laskowski, Sharon J., Marguerite Autry, John Cugini, William
Killam, James Yen. 2004. Improving the Usability and Ac- Additional Supporting Information may be found in the
cessibility of Voting Systems and Products. Washington, DC: online version of this article:
National Institute of Standards and Technology.
Lausen, Marcia. 2007. Design for Democracy. Chicago: Univer- Online Appendix Figure 1 shows the voter information
sity of Chicago Press. guides for the standard office-bloc ballot and for the
Leuven, Edwin, and Barbara Sianesi. 2003. “PSMATCH2: office-bloc ballot with straight-party option.
Stata Module to Perform Full Mahalanobis and Propensity
Online Appendix Figures 2 and 3 show portions of the
Score Matching, Common Support Graphing, and Covariate
Imbalance Testing.” http: //ideas.repec.org/c/boc/bocode/ standard office-bloc ballot and the office-bloc ballot with
s432001.htm. straight-party option (in their paper ballot versions).
Long, J. Scott. 1997. Regression Models for Categorical and Lim- Online Appendix Pictures 1a and 1b show the ES&S
ited Dependent Variables. Thousand Oaks, CA: Sage. Model 100 optical scanner without a ballot and with a
Morgan, Stephen L., and Christopher Winship. 2007. Coun- ballot being scanned, respectively.
terfactuals and Causal Inference: Methods and Principles for Online Appendix Pictures 2a and 2b show the Diebold
Social Research. New York: Cambridge University Press.
AccuVote-TS opening screen and ballot with votes cast,
Mossberger, Karen, Caroline J. Tolbert, and Mary Stansbury.
2003. Virtual Inequality beyond the Digital Divide. Washing-
respectively.
ton, DC: Georgetown University Press. Online Appendix Table 1a presents the background char-
Nielsen, Jakob. 1994. Usability Engineering. San Francisco: Mor- acteristics of the study participants across the three loca-
gan Kaufmann. tions.
Niemi, Richard G., and Paul S. Herrnson. 2003. “Beyond the Online Appendix Table 1b presents the background char-
Butterfly: The Complexity of U.S. Ballots.” Perspectives on acteristics of the study participants as well as their char-
Politics 1(2): 317–26. acteristics, for each voting system, before and after the
Norden, Lawrence, David Kimball, Whitney Quesenbery, and matching process.
Margaret Chen. 2008. Better Ballots. New York: Brennan
Center for Justice. Online Appendix Table 2 presents the coding of the vari-
Redish, Janice, Dana E. Chisnell, Sharon J. Laskowski, and Svet-
ables.
lana Lowry. 2010. “Plain Language Makes a Difference When Online Appendix Tables 36 present the statistical esti-
People Vote.” Journal of Usability Studies 5(1): 81–103. mates used to generate Figures 1 and 2 in the text.
Reynolds, John F., and Richard L. McCormick. 1986. “Outlaw-
ing ‘Treachery’: Split Tickets and Ballot Laws in New York Please note: Wiley-Blackwell is not responsible for the
and New Jersey, 1880–1910.” Journal of American History
72(4): 835–58.
content or functionality of any supporting materials sup-
plied by the authors. Any queries (other than missing
Rusk, Jerrold G. 1970. “The Effect of the Australian Ballot Re-
form on Split Ticket Voting.” American Political Science Re- material) should be directed to the corresponding author
view 64(4): 1220–38. for the article.