precinct closing and wait times in florida during the 2012 ... · precinct closing and wait times...
TRANSCRIPT
Precinct Closing and Wait Times in Floridaduring the 2012 General Election∗
*** DRAFT ***
Michael C. Herron† Daniel A. Smith‡
August 21, 2013
Abstract
Voting station congestion can be measured by late-closing precincts and long wait timesto vote. With this in mind we study Election Day precinct closing times in 43 Floridacounties and early voting wait times in one of Florida’s most prominent counties,Miami-Dade. Our analysis of the 2012 General Election covers 5,302 total ElectionDay precincts and all the early voting stations in Miami-Dade County. We showthat Election Day precincts with greater proportions of Hispanic voters in November,2012, had disproportionately late closing times and that precincts with many registeredDemocratic voters also tended to close relatively late. With respect to early votingwait times in Miami-Dade, we show that long wait times disproportionately affectedblack and Hispanic voters, and a natural experiment in Miami-Dade and Palm BeachCounties confirms that the final voters on the last day of early voting in these twocounties were disproportionately black, Hispanic, and registered Democratic. Votingplace congestion in the 2012 General Election, therefore, did not affect all Floridiansequally. This study, sponsored by Advancement Project and one of the first statisticalanalyses of observed closing and wait times across thousands of precincts in a politicallyimportant state, shows how the electoral environment in the United States continue toreflect racial disparities.
∗This paper draws heavily on Herron and Smith (2013a), a report commissioned by Advancement Projectand written by the authors of this manuscript. Katherine Culliton-Gonzalez of Advancement Project pre-sented the contents of Herron and Smith (2013a) at a meeting of the Presidential Commission on ElectionAdministration on June 28, 2013, and the authors thank Advancement Project for its generous support andMichael Martinez for comments on an earlier version of this paper.†Professor of Government, Dartmouth College. 6108 Silsby Hall, Hanover, NH 03755-3547
([email protected]).‡Professor of Political Science, University of Florida. 234 Anderson Hall, Gainesville, FL 32611-7325
1
Introduction
Lengthy lines at polling stations may be one of the more memorable, albeit unfortunate,
themes that many Americans associate with the 2012 General Election. Within the polit-
ically important state of Florida, there was extensive press coverage of Election Day lines
with the The Tampa Tribune reporting that, “[L]ines were especially long in Broward, Lee,
Miami-Dade and Orange counties, and voters in Lee and Miami-Dade cast ballots even af-
ter Republican challenger Mitt Romney had conceded to President Barack Obama.”1 Lines
during the early voting period in Florida were also prominent, and following complaints from
voters “who sometimes [had] waited seven hours to vote or who did not vote at all because
they could not wait for so long to do so” the Florida Democratic Party filed a federal lawsuit
on the last day of early voting.2 President Obama even commented on voting lines after the
2012 Election, saying during his post-election acceptance speech on the morning of November
7, “By the way, we have to fix that.”3
The 2000 Presidential Election and its controversial aftermath focused needed attention
on voting technology, and beyond hanging chads one legacy of that election is the Help
American Vote Act, a bipartisan election-reform effort passed by Congress in 2002 and signed
by then-President George W. Bush. The 2012 General Election may similarly lead to an
increased consciousness of the importance of voting lines, both their causes and consequences.
Indeed, half a year after this election two of Florida’s most populous counties, Broward and
Miami-Dade, decided to “reorganiz[e] hundreds of voting precincts with the goal of reducing
1See “In Florida, not every precinct is created equally,” Tampa Bay Tribune, November 16, 2012, availableat http://tbo.com/ap/politics/in-florida-not-every-precinct-is-created-equally-566822 (lastaccessed June 26, 2013).
2The lawsuit is described in “Democrats Sue to Extend Floridas Early Voting,” The NewYork Times, November 4, 2012, available at http://thecaucus.blogs.nytimes.com/2012/11/04/
democrats-sue-to-extend-floridas-early-voting/ (last accessed June 26, 2013).3See “President Obamas acceptance speech (Full transcript),” The Washington Post, Novem-
ber 7, 2012, available at http://articles.washingtonpost.com/2012-11-07/politics/35506456_1_
applause-obama-sign-romney-sign (last accessed June 26, 2013).
2
the long lines of voters that plagued last November’s presidential elections and embarrassed
the state.”4 At the national level, on March 28, 2013, President Obama signed an executive
order establishing a Presidential Commission on Election Administration whose goal is to
“promote the efficient administration of Federal elections and to improve the experience of
all voters.”5
Despite a wealth of voting line anecdotes in the vein of those above, there is at present
very little systematic information about many aspect of such lines and the broader subject of
precinct or voting station congestion. Consider, for example, the following questions. How
during General Elections do precinct line lengths evolve over the course of Election Day?
Are Election Day voting lines different than early voting lines? Are line lengths different in
predominantly white precincts compared to precincts with many black and Hispanic voters?
Finally, at what point do long lines deter people from voting and lead to what is known as
reneging? There are precious few answers to these important questions and many others
similar to them, and this hampers efforts to diminish precinct congestion and maximize the
ability of precincts to handle voters in an expeditious and fair way.
The reason that so little is know about precinct congestion—i.e., why we and others like
us cannot answer the above questions—is simple: a lack of data. We cannot, say, comment
on literature that describes how voter reneging rates—the rates at which voters get in line
to vote and then later depart without voting—vary by type of election, state, and county
because there is no such literature. Due again to a lack of data, there is similarly a dearth
of studies on the evolution of precinct lines, whether lines affect some voter types more than
others, and so forth.
4This quote is from “Counties reorganize precincts to help reduce voting lines,” Miami Her-ald, July 14, 2013, available at http://www.miamiherald.com/2013/07/14/v-fullstory/3499848/
counties-reorganize-precincts.html (last accessed July 15, 2013).5The text of the executive order is available at http://www.whitehouse.gov/the-press-office/
2013/03/28/executive-order-establishment-presidential-commission-election-administr (lastaccessed August 12, 2013).
3
With such data limitations in mind, our objective here is relatively modest. In particular,
we seek to characterize the nature of precinct congestion in one state, Florida, during the
recent 2012 General Election. While the analysis here does not extend to the entirety of this
state, it does cover 43 of Florida’s 67 counties and 5,302 of the 6,147 Election Day precincts
that existed in Florida on November 6, 2012.6
Why one state and why Florida in particular? We recognize that there are tradeoffs
between analyzing (a majority of) the counties in one state versus a selection of counties
or other geographical units across a variety of states. Nonetheless, as we describe shortly,
gathering precinct congestion data from even one state has proved to be complicated, hence
our coverage of approximately two-thirds of Florida’s counties. Given the dearth of data
across the United States on precinct congestion, the corresponding lack of scholarship, and
the present lack of standards that could in theory specify how congestion and in particular
voting lines should be measured, we believe that our focus on a single state is appropriate.
It is our opinion that a research design concentrating on one state, where we have some idea
of the set of potential confounds that might affect our results, is preferable to a design that
would entail casting a wide net across the United States. We hope that our Florida-based
research on precinct congestion will help establish criteria that assist other researchers in
broader efforts aimed at this subject.
Even with an arguably narrow focus on precinct congestion in a single state and in one
election, we have not been able to resolve fully the types of data quandaries that regularly
affect election administration research. While we would like to present here a study of the
correlates of Election Day voting lines in Florida, there is almost no data on Florida voting
lines available outside of anecdotes from press coverage and results from a small number of
public opinion polls. Indeed, in the course of our attempting to collect data on such lines
66,147 is the total number of precincts that existed in Florida as of December 31, 2012, according to theofficial Florida statewide voter file. We discuss this file and our use of it later. There were 5,950 precinctsin Florida with at least 20 registered voters as of the end of December, 2012.
4
across Florida counties, we quickly learned that, with only a few exceptions, neither county
nor precinct administrators in Florida appear to keep track of Election Day line lengths,
maximum or average wait times, much less the times when final voters were processed. Our
brief attempts to look for voting line data beyond Florida suggest that this appears to be the
case for counties across the United States as well. Nonetheless, despite an almost complete
lack of line data per se, we were able to gather from 43 Florida counties a set of what are
called precinct closing times. There are subtleties in what closing times mean, but the idea
is simple: they measure when a precinct finished processing voters and/or votes on Election
Day.
Throughout most of his paper, therefore, we focus on closing times, and we use these
times as proxies for voting lines which themselves proxy for precinct congestion. A Florida
precinct with a relatively late closing time on November 6, 2012—well beyond 7:00pm, the
official close of polls—must have had a line at the end of the day. We cannot of course know
whether the line to vote in said precinct was long at, say, noon. Indeed, a precinct with
a relatively late closing time on Election Day could in theory have had no line to vote for
most of Election Day but then suffered from a long line at the end. We make this point here
because we want to be open about the limitations of our closing time data. Nonetheless,
given the present dearth of data on precinct congestion, closing times are at present the best
available measure of precinct congestion.
A clearer understanding of disparate levels of precinct congestion—which is what we
are trying to achieve with this study—may open new avenues for state or federal oversight
of electoral processes across the country. In light of the Supreme Court’s 2013 decision
in Shelby County v. Holder, which invalidated Section 4 of the Voting Rights Act of 1965
(VRA), our analysis assumes even greater import as it provides an alternative measure of
a potential barrier voters face at the polls. In its 5-4 decision, the high court homed in
on the constitutionality of Section 4’s coverage formula—last amended by Congress in the
5
1970s—that in the majority’s opinion relied on outdated voter registration and turnout
figures to determine which jurisdictions had to seek federal preclearance before making any
procedural change to their voting laws. The Court was highly critical of Congress’s 2006
reauthorization of the VRA, which kept intact Section 4’s ostensibly antiquated formula,
finding that lawmakers failed to justify the formula’s distinction between covered and non-
covered jurisdictions with respect to the protection of minority voting rights.
Wait times at the polls have never been included in the Section 4 formula, yet excessive
lines at polls may potentially disenfranchise some voters.7 ¡ Congestion at the polls is
effectively a “time tax,” as Mukherjee (2009) argues. Indeed, if wait times are understood
as a barrier to voting, it is important to document whether congestion at the polls has a
disparate impact on minority voters. Standard measures of minority voter participation, like
voter registration and turnout rates, may certainly serve as important indicators of racial
discrimination, but these indicators may belie the discriminatory effects of other state laws
or administrative decisions. Indeed, we believe that there is much merit to the point that
“increases in voter registration and turnout” is not “the whole story,” a point the minority
in Shelby took pains to emphasize.8
In part because of the Supreme Court’s recent action in Shelby, much of the analysis
that follows touches on the issue of race and ethnicity. We are interested in particular in
the question of whether certain racial/ethnic groups had to contend with precincts that had
relatively late closing times. The striking down of Section 4 encourages one to consider the
question of whether voting in a southern state like Florida is truly post-racial as of late 2012,
and to this end our data shed light on continuing racial/ethnic regularities that in the past
7See, for example, “Analysis: 201,000 in Florida didn’t vote because of long lines,” Orlando Sen-tinel, January 29, 2013, available at http://articles.orlandosentinel.com/2013-01-29/business/
os-voter-lines-statewide-20130118_1_long-lines-sentinel-analysis-state-ken-detzner (lastaccessed August 15, 2013).
8Opinions in Shelby County v. Holder are available at http://www.supremecourt.gov/opinions/12pdf/12-96_6k47.pdf (last accessed August 12, 2013).
6
have been used to motivate government regulation and monitoring of electoral practices.
Briefly, we find evidence of significant variation within Florida counties in precinct closing
times on Election Day 2012. Because key election features in Florida are for the most part
held constant within counties, the variance that we observe is presumably indicative of some
form of resource malapportionment or unanticipated surges of voters. With respect to race
and ethnicity in particular, we find strong evidence that Florida precincts with many Hispanic
Election Day voters had later closing times than precincts with many white voters. Moreover,
we show using actual voting line data from the eight-day early voting period in Miami-Dade
County that early voting stations with large numbers of non-white (i.e, black and Hispanic)
voters had longer wait times than stations with disproportionately many white votes. And
finally, a natural experiment that occurred in Miami-Dade and Palm Beach Counties during
Florida’s early voting period in 2012 shows clearly that the Miami-Dade and Palm Beach
early voters who were last in line to vote on Saturday, November 3, 2012, the last day of early
voting, were disproportionately black and Hispanic and tended to be registered Democratic.
The remainder of this paper is organized as follows. In the next section we discuss the
existing literature on voting lines and precinct congestion, and we explain how our results
contribute to this literature. We then describe how we collected our data on Florida precinct
closing and wait times and how these data should be understood. The next two sections
of the paper present Election Day closing time results for 43 counties in Florida and early
voting wait time results for Miami-Dade County, respectively. We end with suggestions for
research which cover a variety of open questions as well as data standards that could help
alleviate the current lack of data on precinct congestion.
7
Existing Literature on Precinct Congestion
There exists at present a small literature on precinct congestion. We have already noted that
this literature is hamstrung by a lack of data, and with respect to voting lines in particular
Stewart III (2013) summarizes the situation nicely: “[T]he study of waiting times is in its
infancy because measures of waiting to vote are primitive” (p. 10).
To circumvent a lack of data on lines and other aspects of precinct congestion some
scholars have chosen to rely on surveys. These surveys include voter self-reports and have
been conducted via telephone and the Internet. A reliance on voter self-reports in the matter
of precinct congestion is not ideal: whether voters are unbiased observers when it comes to
estimating how long they themselves waited in line before voting is unknown. Furthermore,
whether voters would admit to engaging in a socially problematic behavior like reneging is
similarly not clear. Nonetheless, large-scale surveys of voter experiences have the advantage
of being able to cover many jurisdictions across the United States, and survey data on
precinct congestion is certainly better than no data at all.
Stewart III (2013) is an excellent example of the value that surveys bring to the study
of precinct performance and voter wait times in particular. Focusing on the 2012 General
Election, Stewart III draws on two large-scale surveys, and offers several conclusions. One,
waiting times to vote vary significantly across the United States and some jurisdictions suffer
from line problems while others do not. Two, long voting lines tend to crop up in urban
areas and areas with large numbers of minority voters. Relatedly, Alvarez et al. (2009),
a survey-based study of the 2008 Genearl Election, finds that 70 percent of Election Day
respondents reported waiting fewer than ten minutes to vote and four percent, at least one
hour. Early voting lines in 2008, according to Alvarez et al., were in general longer than
Election Day lines, e.g., eight percent of early voting respondents reporting waiting at least
an hour to vote.
8
Also using survey data to measure wait times and precinct problems more generally
is Kimball (2013), who finds that urban voters across the country suffer from the longest
lines and that “urban voters contend with a disproportionate share of voting difficulties in
the United States” (p. 17). These findings are very similar to those in the aforementioned
Stewart III and Alvarez et al.. Claassen et al. (2008) use exit poll data in a study which
argues that overall voter sentiment about voting experiences is sensitive to the quality of
precinct administration and wait times in particular. Claassen et al. do not focus on the
causes of wait times per se but rather on the extent to which voters find lines troubling.
To the best of our knowledge, Spencer and Markovits (2010) is the only study that
measures voting line lengths with an eye on the factors that lead to long lines. In particular,
Spencer and Markovits focus on precinct performance during the 2008 California presidential
primary and highlight the substantial variance in voter processing rates across a set of
30 precincts. Most importantly from our perspective, they show that precinct voting line
lengths in 2008 fluctuated across Election Day and were longest around 6:00pm. Spencer
and Markovits also find an overall renege rate of 1.89 percent; this is the percentage of voters
who turned out to vote and then left a precinct without having voted. Moreover, Spencer
and Markovits observe a positive correlation between the number of people in line when a
voter arrived to vote and the probability that said voter reneged.
Reneging can have political consequences if the probability that a voter reneges is not
uniform across all voter types. This assertion is an underlying part of the analysis in High-
ton (2006), who argues that Franklin County, Ohio, precincts with relatively few voting
machines in the 2004 General Election suffered from low voter turnout, ceteris paribus. In
addition, Highton argues that a lack of voting machines in Franklin County cost Democratic
presidential candidate John Kerry in the neighborhood of 6,000 votes (this quantity would
not have been pivotal to Kerry’s 2004 loss in Ohio).
Our contribution to the extant literature on precinct congestion is as follows. As men-
9
tioned in the introduction, our analyses draw on observed election features, namely Election
Day precinct closing times in Florida and actual early voter wait times in one Florida county.
Our data do not rely on voter self-reports, which is fortuitous, but they do not have the scope
of the surveys used in Stewart III’s work on voting lines. We thus see our results as com-
plementing those in Stewart III and others, and we believe that the literature on precinct
congestion should reflect a mix of research that is based on observed quantities and research
that queries voters about their experiences.
Measuring Precinct Congestion
Precinct congestion has many facets. Our focus here is on a combination of Election Day
closing times across counties in Florida and early voting wait times in Miami-Dade County.
We now describe how we gathered the closing and wait time data used in the analyses that
follow, and in so doing we explain our data’s strengths and limitations. After explaining
how we assembled our data on precinct closing and wait times, we then discuss how we
integrate these data with Florida statewide voter files and early voting files, both of which
are compiled and published by the Florida Department of State (FDOS).
Closing Times in 43 Florida Counties
There are 67 counties in Florida, and elections within them are governed by county-level
Supervisors of Elections. All but the Miami-Dade County Supervisor of Election (SOE) are
elected, and most run as partisan candidates.9
To each SOE we sent a public records request for information on precinct administrative
capacity. In particular, we asked for a list of active precincts used during the 2012 General
9See the June 28, 2013, testimony of Penelope Townsley, Miami-Dade Supervisor of Elections, to thePresidential Commission on Election Administration at a hearing at the BankUnited Center, University ofMiami Coral Gables, Florida. The testimony is available at http://www.supportthevoter.gov/files/
2013/06/PCEA-Miami-Dade-Supervisor-Penelope-Townsley1.pptx (last accessed August 13, 2013).
10
Election, the times at which each precinct closed, the number of pollworkers per precinct, the
number of optical scan voting machines per precinct, and the number of electronic poll books
per precinct. Approximately one-third of our public records requests generated responses.
Some of the response we received provided machine-readable data on precinct closing times,
and others provided scanned images that we had to enter by hand. Some counties provided
all the information that we requested while others did not.
Beyond precinct-level data gleaned via public records requests, we were provided by Scott
Powers and David Damron of the Orlando Sentinel with a copy of a spreadsheet that the
newspaper assembled immediately following the 2012 General Election. Orlando Sentinel
staff members sent requests to the 25 largest counties in Florida asking for information on
precinct closing times, and we use the Orlando Sentinel data to supplement our data.10 When
we have data from both the Orlando Sentinel and from a public records request, we use data
from the latter. The only exception to this rule is Miami-Dade County. For reasons that
are unclear to us, the precinct closing times data we received from Miami-Dade via public
records request did not include closing times for several precincts mentioned in a report titled,
“Elections After Action Report – November 6 Presidential Election,” that was commissioned
by Miami-Dade County in the aftermath of the 2012 General Election.11 Insofar as these
precincts were specifically mentioned in the aforementioned report, and insofar as the report
notes that these precincts closed relatively late compared to other Miami-Dade precincts,
we believe we would be remiss not to include them. Hence, for Miami-Dade County we use
Orlando Sentinel data as opposed to data we gleaned from our requests. From our public
records requests and data from the Orlando Sentinel we have closing times data from 43 of
10The Orlando Sentinel data are the basis of “Analysis: 201,000 in Florida didn’tvote because of long lines,” Orlando Sentinel, January 29, 2013, available at http://
articles.orlandosentinel.com/2013-01-29/business/os-voter-lines-statewide-20130118_1_
long-lines-sentinel-analysis-state-ken-detzner (last accessed July 2, 2013).11The Miami-Dade report is available at http://www.miamidade.gov/mayor/library/
elections-after-action-report.pdf (last accessed July 2, 2013).
11
67 Florida counties.
Our experiences collecting closing time data from counties across Florida have made it
clear that the definition of precinct closing time is not standardized across counties. In the
process of contacting counties we encountered a variety of definitions for this term, i.e., the
time that a precinct’s last voted signed in to vote; the last time a vote was recorded; the time
when optical scan machines were powered off; and the time that results were reported to a
central location. With respect to Hillsborough County, for example, the Orlando Sentinel
spreadsheet describes the precinct times at which voting results were uploaded to a central
server administered by the Hillsborough SOE. In contrast, the Hillsborough County SOE
provided data to us that indicates when the county’s various optical scan machines, located
across many precincts, were shut down. We prefer machine shutdown times to data upload
times because it is our opinion that the former more closely approximate what closing times
are. Still, in our judgment the problem with differing definitions of closing times is a lack of
standardization as opposed to the existence of problematic definitions.12
Because of the lack of standardization across Florida in the definition of precinct closing
time—a subject to which we return in the conclusion—we do not compare closing times
across counties. Rather, we assess closing times only within counties. This applies to both
the closing time data that we received from various SOEs via public records requests and
the Orlando Sentinel data.
Early Voting Wait Times in Miami-Dade County
One exception to the dearth of voting line data can be found in Miami-Dade County. Very
fortuitously, Miami-Dade officials recorded approximate wait times at the 20 early voting
polling stations used across the county during the eight days of early voting prior to the 2012
12Even with respect to machine shutdown times there are subtleties. Many precincts have multiple opticalscan machines, and we cannot always be sure that closing times based on machine shutdown times reflectfirst machine shutdowns or the last machine shutdowns. This is the sort of issue that should be addressed instandards promulgated by the FDOS or perhaps the Florida State Association of Supervisors of Elections.
12
General Election. We employ the term “polling station” here when discussing early voting
in Miami-Dade County so as to distinguish the locations—“precincts”—where Miami-Dade
voters voted on Election Day and the locations—“polling stations”— where they voted early.
At a number of times during each early voting day, Miami-Dade officials recorded and
made public on the Internet the waiting times at the county’s 20 early voting stations. For
example, the first census of wait times on October 27, 2012, the first day of early voting, took
place at 7:28am. As of this moment, the wait time at Aventura Government Center, one of
20 Miami-Dade early voting stations, was reported to be one hour. At the Miami-Dade City
Hall, another such station, the corresponding wait time was 45 minutes. At the last census
of wait times on October 27, taken at 6:16pm, the Aventura wait time was four hours and
15 minutes while the Miami-Dade City Hall wait time was 50 minutes.
Across the eight days of early voting in Miami-Dade County, wait time censuses conducted
by SOE staff members did not take place at the same time. As noted above, the first census
on October 27, 2012, was made at 7:29am. On October 28 the first census was at 6:28am.
This particular census took place before early voting began, and all 20 early voting stations
had wait times of zero minutes. At 7:26am on October 28, however, wait times varied across
early polling stations with Aventura reporting three minutes and the Miami-Dade City Hall,
one hour and 15 minutes.
Given that wait time censuses were not taken at the same times on every day, we calculate
first- and last-reported wait times for each Miami-Dade early voting station and on each early
voting day. Since there are 20 stations and eight days of early voting, this exercise yields
160 first-reported wait times (generally around 7:30am) and 160 last-reported wait times
(generally around 6:30pm).
13
Florida Voter Files
Florida voter registration records are maintained in what are called statewide voter files, and
via a public records request to FDOS we acquired a Florida voter file dated December 2012.
Florida voter files are updated monthly, and each file contains demographic and participation
records on every Florida vote registered in the state as of the file’s promulgation. The
December 2012 file includes participation codes for elections through the 2012 General. This
means that the December 2012 file has codes that describe who among the registered voters
in Florida voted early in the 2012 General Election, who voted on Election Day, who voted
absentee, who did not vote at all, and so forth. The November 2012 voter file does not
contain participation records for the 2012 General Election, hence our use of the December
2012 file.
We acquired the December 2012 voter file in January, 2013. Subsequent to that we learned
via email communications with the FDOS on June 3, 2013, that this file contains errors
that affect the 2012 General Election participation codes for seven Florida counties. These
counties are, alphabetically, Broward, Duval, Lee, Leon, Palm Beach, Polk, and Sarasota.
We were also informed by a staff member at the FDOS that the 2012 General Election
participation codes were fixed as of the March 2013 Florida voter file. With this in mind we
acquired a copy of the March 2013 file and from this constructed a hybrid Florida voter file
that includes data from the December 2012 file (60 counties) and the March 2013 file (seven
counties). Polk County exemplifies the importance of using the March 2013 voter file when
examining 2012 General Election participation codes for the seven aforementioned counties.
According to the December 2012 voter file, that there were only 94 Election Day voters in
Polk. However, this figure is approximately 128,000 in the March 2013 file.
We note that Florida voter files consist of what are called extract and history files. The
former contain lists of voters, their addresses, and their demographics, and each voter in the
extract file (technically, one file per county) is associated with a unique voter identification
14
number. These numbers link to voter history files (again, one file per county) which contain
election participation codes. We use the December 2012 extract file in our hybrid because
we were told that the errors affecting the seven aforementioned counties affect only history
codes.13 Henceforth when we refer to a voter file we mean the hybrid voter file described
here.
From our hybrid voter file we generated a list of all precincts used in each of Florida’s
67 counties; there are 6,147 such precincts. We then merged this list of precincts with the
precinct closing times from our 43-county closing time dataset. This merge was not perfect,
and there exist voter file precincts for which we have no closing times and closing times
reported by counties that do not appear to correspond to actual precincts according to our
hybrid voter file.14 Furthermore, there are three Palm Beach County precincts in the Florida
voter file that had very small yet positive numbers of voters on Election Day but a total of
zero black, Hispanic, and white Election Day voters. We dropped these precincts from our
analysis.
Our coverage across Florida is described in Table 1. Among other things this table lists
for our set of 43 counties the number of precincts per county based on our hybrid voter file
as well as the number of precincts per county for which we have usable data, meaning usable
closing time data, usable racial demographics, and so forth.
Despite the complications in merging our voter file with county-level data on precinct
closing times, our coverage rates by county are very high. In Miami-Dade County, for
13The FDOS publishes what we have referred to as voter files, called VH01 files, and what the FDOS callselection recap files, so-called VH03 files. When we use the term “voter file” we are referring to a VH01 file.Our correspondence with the FDOS suggests that only four counties, as opposed to seven, had errors in theDecember 2012 VH01. However, we are concerned that counties listed as having VH03 errors also had VH01errors. To be conservative, we assume that any county known to have 2012 General Election participationcode errors in either the December 2012 VH01 or VH03 file is problematic. We replace 2012 General Electionparticipation codes for these counties with data from the March 2013 VH01.
14One might reasonably inquire as to why this situation exists. We cannot be sure of the explanation butsuspect that the root of this problem lies in coordination and communication difficulties between the FDOS,which maintains the Florida statewide voter file, and individual county SOEs.
15
Table 1: Summary of Analyzed Counties and Precincts
Existing Analyzed Percent Minimum Maximum Zero AverageCounty Precincts Precincts Coverage Delay Delay Delay Delay
Miami-Dade 801 789 98.5 1 420 0 72.4968Palm Beach 770 735 95.5 27 285 0 83.8902Broward 795 617 77.6 0 190 34 25.0551Hillsborough 347 347 100.0 1 167 0 30.1844Pinellas 299 298 99.7 7 120 0 41.0436Orange 233 227 97.4 4 318 0 85.7841Duval 198 197 99.5 0 170 1 34.8883Polk 167 163 97.6 4 194 0 39.1273Leon 136 129 94.9 14 137 0 40.7364Marion 127 127 100.0 6 144 0 26.5354Volusia 125 125 100.0 10 458 0 75.1600Lee 126 125 99.2 3 474 0 115.0160Brevard 166 116 69.9 11 216 0 62.6552Manatee 113 112 99.1 14 202 0 53.3304Pasco 111 109 98.2 0 308 1 41.1835Lake 104 102 98.1 5 84 0 20.3398Sarasota 100 99 99.0 5 197 0 36.0505Seminole 80 80 100.0 2 87 0 25.3500Escambia 79 79 100.0 0 33 1 11.3418Charlotte 79 78 98.7 0 0 78 0.0000Osceola 71 70 98.6 0 242 13 36.4143St. Lucie 65 65 100.0 17 381 0 96.1692Collier 62 61 98.4 0 305 44 8.2459Alachua 70 61 87.1 14 117 0 33.5902Clay 49 48 98.0 0 52 3 12.9792St. Johns 47 46 97.9 0 0 46 0.0000Bay 44 44 100.0 4 101 0 24.3182Indian River 37 37 100.0 6 56 0 17.6757Citrus 31 31 100.0 0 0 31 0.0000Martin 31 31 100.0 0 0 31 0.0000Sumter 24 24 100.0 9 162 0 46.0000Nassau 16 16 100.0 0 3 15 0.1875Taylor 14 14 100.0 0 0 14 0.0000Jackson 14 14 100.0 1 21 0 8.4286Bradford 14 14 100.0 0 0 14 0.0000Wakulla 12 12 100.0 0 0 12 0.0000Okeechobee 11 11 100.0 0 0 11 0.0000Union 11 11 100.0 0 0 11 0.0000Baker 9 9 100.0 0 15 5 5.0000Franklin 8 8 100.0 0 0 8 0.0000Hamilton 8 8 100.0 1 17 0 5.1250Holmes 8 8 100.0 1 16 0 8.1250Lafayette 5 5 100.0 0 0 5 0.0000
16
example, we have usable data on 98.5 percent of the precincts that exist in this county
according to the FDOS. Moreover, in many cases we have coverage for all of the precincts
in a county. The only outlier here is Broward County; we have usable data on 77.6 percent
of the precincts that exist in Broward according to the FDOS. Our efforts to rectify this
situation have not been successful.
The various delay columns in Table 1 represent closing delays past 7:00pm based on
counties’ methods of determining closing times. A precinct that was reported to have closed
ten minutes line might have closed its doors at 7:00pm yet uploaded its data ten minutes
later. Hence one should not compare across counties in Table 1, and absolute delays are
not nearly as important as variability in delays within counties. Indeed, one can compare
minimum, maximum, and average delays within individual counties, and Table 1 highlights
significant variability.
For example, in Palm Beach County, closing delays ranged from 27 to 285 minutes with
an average of almost 84 minutes. In Hillsborough County the range of closing days is one
minute to 167 minutes. The questions pursued shortly turn on explaining this variance and
in particular considering whether precincts with relatively long days (i.e., late closing times)
had particular racial or party profiles.
Early Voting Files
Our final source of data consists of what are called county-level early voting files. As the
eight-day 2012 early voting period in Florida progressed, the FDOS published a list of
every registered Florida voter who voted early. These lists (one per county) specified the
station at which each early voter cast his or her vote, the date on which said vote was cast,
and said voter’s unique voter identification number. The presence of these identification
numbers allows us to link official early voting files with the hybrid voter file that we discussed
above. Since the latter contains data on registered voter race/ethnicity, we can calculate,
17
say, fraction Hispanic of all early voters at each early voting station in Florida on each early
voting day. Fractions like these are particularly useful in the context of our Miami-Dade
early voting analysis.
Precinct Closing Times in Florida
We pursue two questions with respect to Election Day precinct closing times in Florida. First,
what racial/ethnic types of Election Day voters voted at precincts with relatively late closing
times? Second, what aggregate party affiliations were associated with late-closing precincts?
The first question is important in light of the racial history of Florida and the ongoing debate
in the United States on the role of race/ethnicity and voting rights; earlier we commented
on this debate as reflected in the recent Supreme Court decision in Shelby County v. Holder.
The second question is important insofar as it approaches the matter of vote choice. Ballot
secrecy notwithstanding, to the extent that aggregate precinct partisanship is related to
closing time, one should be concerned that phenomena like reneging due to long lines may
not be politically neutral.
The Racial/Ethnic Correlates of Precincts with Late Closing Times
Consider a single Florida county, say, Orange. This county, located approximately in the
middle (north-south) of Florida and home to the city of Orlando, has 227 precincts in our
dataset as noted in Table 1. Each of these precincts is associated with both a closing time
and a racial/ethnic triple of fraction black voters on Election Day, fraction Hispanic voters,
and fraction white voters. These three fractions sum to almost one for all Orange County
precincts (some registered voters in Florida do not provide race/ethnicity, and some provide
a race/ethnicity that is neither black, Hispanic, nor white), and we plot closing delays beyond
7:00pm and these racial/ethnic triples in a ternary diagram. The Orange County ternary
18
Figure 1: Race/Ethnicity and Closing Times in Four Florida Counties
(a) Orange
White Hispanic
Black
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●
●
● ●
●
●
●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●●●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●
●●
● ●●
●
●●
●
● ●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●● ●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●●●
●
●● ●● ●
●●
●●
●
●
●
●
●
●
●●
(b) Osceola
White Hispanic
Black
●●●
●
● ●
●
●
●
● ●●●
●●
●●
●
●
●
●●
●
●
●●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●●●
●
●●
●
●
●
●●
●
●●
●
●●
● ●
●
●
●●
●●
●
(c) Hillsborough
White Hispanic
Black
● ●●●
●
●
●● ●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●●
●
●●
●
● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●
●●
●
●●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●●
●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
● ●●●●
●
●
●
●●
●●
●
●
●
●
●
●●
●●
● ●
●
●●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
● ●
●
●●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●●
●
●
●
●
●
●
● ●
●
●
●● ●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
(d) Miami-Dade
White Hispanic
Black
●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●●
● ●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●●
●
●●● ●●
●●
●
●
●
●
●
●
●
●
●
●●●
●
●
● ●
●●
●
●●
● ●●● ●●
●
●
●● ●
●
●
●
●
●
● ●
●
●●●●
●
●
●●
● ● ●●●
●●●●
●
●
●●●
●
●
●●●●●
●
●
●
●
●
●
●●● ●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●
●
●
●
●
●●
●
●
●
●●●
●
●●
●
●
●
●
●
●
●
●●● ●
●
●
● ●● ●
●
●
●
● ●●●●
●
●●
●
●
● ●●
●
●
●●
●
●●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●●
●●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●● ● ● ●●●● ● ●●
●
● ●●●●●
●
●●
●●
●
●
●
●
●
●
●
● ● ●●●
●●
●
●
●
●
●●●
●
●
●
●
●
●● ●●● ●●●●
●
●
●●
●
●●
●
●
●●●
●
●
●
●
●●●
●
●
●
●
●
●
●● ●●
●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●● ●●●
●●
●●
●
●
●
●
●●●●
● ●●
●
●●
●●
●
●
●
●
●
● ●●
●●
●
●●
●●● ●
●
●
● ●●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●●
●
●
●
●●
●
●
●●
●
●●
●●
●
●
●
●
●
●
●●
●
●● ●
●
●
●
●
●
●
● ● ●●●●
●
●
●●
●
●
●
●●
●
● ●●● ●●
● ●●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●●
●● ●●● ●●●
●
●
●
●
●
●
●
● ● ●
●
●
●
●
●●
●●
●
●
●
●
●
●
● ●
●
●● ●
●
● ●
●
●
● ●●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●●
●●
● ●
●
●●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●● ●●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●●
●
●
●
●●
●● ●
●●
●
●
●
●● ●
●
●
●●
●
● ●●
●
●
●
●
●
●●●
●●
●
●
●●
●
●●
●
●●● ●●
●
●
●
●
●
●
●
●
● ●●●● ●
●
● ●●●●●
●
● ●●●
●
●
●●
●
●
●
●●●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●●
● ●
●●
●
●
Note: each dot represents an Election Day precinct and the racial/ethnic composition of theprecinct’s Election Day voters. Shading is proportional to closing times within county withdarker dots connoting later times.
19
diagram is in Figure 1a.
Within this figure and in the other three ternary diagrams in Figure 1, each dot represents
an Election Day precinct. Moreover, a dot’s location connotes the associated precinct’s
racial/ethnic distribution of Election Day voters. For example, a precinct with a distribution
of Election Day voters that was exactly one-third black, one-third Hispanic, and one-third
white would be represented by a dot precisely in the middle of a ternary diagram; a precinct
whose Election Day voters were completely Hispanic would be located on a Hispanic vertex;
and, a precinct that was split between black and white voters with no Hispanics at all would
be located on a line connecting black and white vertices.
The greyscale shading of the precinct dots in Figure 1’s four ternary diagrams represents
closing time delays, and darker colors indicate later times after 7:00pm. It is important to
note that the colors of the dots in Figure 1 cannot be directly compared across counties.
This is because an individual ternary diagram’s dot shading scheme is determined by making
comparisons of precinct closing times within a county. We have already noted that Florida
counties do not have identical standards for reporting precinct closing times, and because of
this we cannot in general compare closing times across counties. What is important in Figure
1, then, is not absolute closing times themselves but rather the within-county variance in
these times.
With respect to Orange County, Figure 1a makes it clear that the whitest precincts in
this county (based on fraction of Election Day voters) did not suffer from late closing times.
One can see this in the fact that the dots nearest the white vertex in Figure 1a are not
particularly dark. In contrast, the precincts nearest the Hispanic vertex in the figure have
some of the darkest dots in Orange County. In general, Figure 1a shows that the distribution
of Election Day precinct closing times in Orange County was not uniform across precincts
insofar as precincts with large numbers of Hispanic voters tended to have later closing times
than precincts with large numbers of white voters. And, precincts used heavily by black
20
voters did not close particularly late.
Figure 1b displays a similar ternary diagram for Osceola County, a county adjacent to
Orange but one that is much smaller. Osceola County is one of the most Hispanic (largely
Puerto Rican) and fastest growing counties in the state. The darkest dots in the Osceola
ternary diagram are relatively close to the Hispanic vertex, and this implies that Osceola
precincts with many Hispanic Election Day voters had relatively late closing times. In this
way Orange and Osceola Counties are similar. One way in which they are different is that
Osceola County contained no precincts which had a large number of black Election Day
voters.
There are no obvious racial patterns in Figure 1c, which describes Hillsborough County,
formerly a Section 5 Voting Rights Act jurisdiction in Florida. There is certainly little
evidence that black voters in Hillsborough disproportionately voted at late-closing precincts
on Election Day 2012. However, Figure 1d shows that the distribution of precinct closing
times in Miami-Dade County was not uniform across racial/ethnic groups. To wit, the
darkest dots in this figure are located closest to the Hispanic vertex.
Hispanics voting in precincts that had relatively late closing times is a general finding
in our data. Up to this point, though, we have only illustrated this claim with a series of
ternary diagrams. Shortly, though, we present more systematic evidence which generalizes
our finding that largely Hispanic precincts tended to have relatively late closing times.
Beyond race/ethnicity, another important voter characteristic is party registration. Ac-
cordingly, Figure 2 contains ternary diagrams akin to the racial/ethnic ternary diagrams we
just presented. Here, though, triangle vertices indicate aggregate party registration fractions
for Election Day voters. At present the three dominant party affiliations in Florida are
Democratic, No Party Affiliation (NPA), and Republican. A precinct, for example, that was
one-third Democratic, one-third Republicans, and one-third NPA would thus be located in
a ternary diagram’s centroid. As before, dot shading in Figure 2 is based on precinct closing
21
times and darker shaded precincts are those that closed relatively late.
One simple generalization of the patterns in Figure 2 is that precincts with many reg-
istered Republicans voters on Election Day tended not to close late. One can see this in
the locations of the black and otherwise dark dots in the four ternary diagrams in Figure 2.
These dark dots are not close to Republican vertices, and Figure 2b’s depiction of Osceola
County is a clear example of such a Democratic bias in late closing times.
There are many race/ethnicity and party affiliation ternary diagrams that we could have
shown, and the ones we have displayed thus far illustrate typical patterns that obtain across
many Florida counties. However, to make our results more compelling, we now present
estimates from a tobit regression model of closing delays on precinct aggregates. In this
model the dependent variable is the number of minutes after 7:00pm that a precinct closed.
This variable has a minimum value of zero since by law every Florida precinct was open until
at least 7:00pm, and thus our use of a tobit model.
Our tobit model includes county fixed effects, and these effects reflect two factors: dif-
ferent closing time reporting standards and differences across counties that are independent
of standards. With respect to the former, a county can have a relatively large and posi-
tive intercept if, say, precinct closing times in the county are based on data upload times
as opposed to machine shutoff times. It would not be appropriate to interpret such an in-
tercept in a critical way. Notwithstanding standards, however, some counties may operate
less efficiently than others and thus may have had precincts that closed relatively late on
Election Day 2012. This would also be reflected in a large and positive intercept. Due to the
lack in Florida of a common definition of precinct closing times, we cannot separate in our
estimated county fixed effects the effects of different definitions of closing times and actual
problems that affected a county’s precincts. Hence, in what follows we do not comment
on counties with large or small estimated fixed effects. Finally, while we have data on 43
counties, we include in our tobit model only counties that had at least 80 precincts and only
22
Figure 2: Party Registration and Closing Times in Four Florida Counties
(a) Orange
Democratic Republican
NPA
●
●
●
●
●
●
● ●
●●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●●
●
●
●
●
● ●
●●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●
●●
●●
●
●
●
●
●
●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
(b) Osceola
Democratic Republican
NPA
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●●
●
●
●
●●
●
●
●
●
●
●
●●●
●
●●●
●
●
●
● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●●
●
●
●
●●
●
(c) Hillsborough
Democratic Republican
NPA
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●●
●●●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
● ●
●
●●●●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●●
●
●●
●
●
●●
●
●●
●
● ●
●
●
●●
●
●
● ●
●●
●
●
●
●●
●
●
●
●
●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●●●
●
●
●
● ●
●
●
●
●●
● ●●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
● ●
●●
●
●●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
(d) Miami-Dade
Democratic Republican
NPA
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
● ●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●
●●
●●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●● ●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●●
●●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●●
●●
●
●
● ●
●●
●
●●
●
●
●●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●
● ●
●
● ●
●
●
●●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●●
●
●
●
●
●
●
●
●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●●●
●●
●
●●
● ●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●
●
●
● ●●
●●
●
●●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●●
●●●●
● ●●
●
●
●● ●
●●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
● ●
●
●
●
●
●●
●
●
● ●
●
●
● ●
●
●
●●
●
●●
●
●
●●
●
●●
●●● ●
●
●
●●
●
●
●●●
●
●●
●
●●
●
● ●
●
●
●
●
●
●●
●●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●●
●●
●●
● ●
●●
●●
●
●
●
Note: each dot represents an Election Day precinct and the party registration of theprecinct’s Election Day voters. Shading is proportional to closing times within county withdarker dots connoting later times.
23
those precincts that had at least 20 registered voters. This leaves us with 18 counties and
4,931 total precincts (recall that the entire state of Florida has 6,147 precincts).15
Table 2: Tobit Regression Results for Precinct Closing Times
Estimate Standard Error z score p-valueFraction black -7.60 5.46 -1.39 0.16Fraction Hispanic 49.51 4.90 10.10 0.00Fraction Democratic 20.25 7.44 2.72 0.01Fraction NPA 45.89 9.25 4.96 0.00Constant 44.02 5.34 8.24 0.00Broward -49.61 4.64 -10.68 0.00Miami-Dade -17.28 5.03 -3.44 0.00Duval -27.08 5.29 -5.12 0.00Hillsborough -41.16 4.85 -8.49 0.00Lake -41.72 6.05 -6.90 0.00Lee 50.03 5.76 8.68 0.00Leon -21.04 5.78 -3.64 0.00Manatee -8.84 5.91 -1.50 0.13Marion -35.05 5.75 -6.10 0.00Orange 13.10 5.18 2.53 0.01Palm Beach 15.65 4.35 3.59 0.00Pasco -24.11 5.96 -4.04 0.00Pinellas -23.91 4.90 -4.88 0.00Polk -27.64 5.42 -5.10 0.00Sarasota -26.32 6.11 -4.31 0.00Seminole -43.86 6.51 -6.74 0.00Volusia 7.90 5.78 1.37 0.17Log(scale) 3.80 0.01 382.36 0.00
Note: based on 5,111 precincts of which 36 have censored delays. Above the dashed line areprecinct aggregrates; below are county fixed effects. Brevard County is the base case forcounties, and Fraction white and Fraction Republican are reference categories forrace/ethnicity and party affiliation, respectively.
Tobit regression results are in Table 2, and they have several notable implications. First,
the Hispanic finding that we saw earlier in various ternary plots is apparent in the table.
15We ignore small counties because their fixed effects are barely identified. Furthermore, some of the verysmall counties in Florida had no variance in closing times and a tobit model fixed effect for such a county isnot identified.
24
To be precise, the Hispanic estimate in Table 2 is positive and easily statistically significant
(z ≈ 10.10). This means that, within the 18 counties analyzed in Table 2, precincts with
large numbers of Hispanic Election Day voters closed disproportionately late. Given the
county-fixed effects in Table 2, this finding cannot be explained by different definition of
closing time.
We do not find evidence that such a finding holds with respect to black voters, and in
fact the opposite is true: the black estimate in Table 2 is negative albeit not significant
at conventional confidence levels (z ≈ −1.39). This indicates that heavily black precincts
appear not to have been inconvenienced by late-closing precincts on Election Day. To some
extent this result is surprising given the considerable controversy surrounding the passage
of House Bill 1355, a piece of election-reform legislation that became law in Florida in mid-
2011. We have more to say about Table 2’s black result below, and see Herron and Smith
(2013b) for a general discussion of House Bill 1355 and a study of its effects on minority
voter registration in Florida.
Table 2 contains two party variables, and the omitted group is Republican. Because both
the Democratic and NPA estimates in the table are positive and significant (z ≈ 2.72 and
t ≈ 4.96, respectively), it follows that precincts with large numbers of Republican voters
tended to close relatively early on Election Day. Such a political bias in closing times could
be problematic and even have subtle effects on vote shares if late closing times are associated
with long lines, especially if lengthy line lead to a reneging.
We want to emphasize the motivation for the tobit model in Table 2. With the model
we addressed the question, did voters of a particular racial/ethnic or party affiliation vote in
precincts that had late or early closing times, ceteris paribus? We have shown that Hispanic
voters in particular seem to have been disadvantaged in terms of where they voted in Florida,
and this is notable in light of ongoing debates about voting rights and the VRA in particular.
Democratic and NPA voters along with young voters also voted in late-closing precincts.
25
Of course we are not making a causal argument here about Hispanics, Democratic voters,
and so forth. Rather, we suspect in accordance with Spencer and Markovits (2010) that
precinct features and resources are associated with closing times and that the distribution
of these resources is key to variability in closing times.
One very important point to keep in mind about the closing times analyzed in Table 2
is that voters in Florida choose when they want to vote. Namely, registered Floridians can
vote on Election Day, vote absentee (no excuse needed), or vote early. Different types of
voters tend to vote at different times, e.g., blacks are heavy users of early voting (Herron
and Smith, 2012) and Republican voters often vote absentee.16 This means that the Election
Day wait times we observe are confounded by the fact that some voters may have selected
out of Election Day voting while others aimed to select in.
This is an important point for the following reason. We offered earlier a positive perspec-
tive on the black finding in Table 2, but one might be concerned that this finding reflects the
fact that many black voters chose to vote early in the 2012 General Election. These voters
could not have been in line on Election Day because they had already voted. Accordingly,
the positive black estimate in Table 2 may reflect the fact that precincts with many black
Election Day voters were also places that had large contingents of black early voters.
We see evidence of this confound in the two panels of Figure 3. In particular, Figure 3a
plots for Florida precincts the black Election Day voting rate against the black early voting
rate.17 Note the negative slope in the regression line that is superimposed on the precincts
depicted in Figure 3a. The implication here is that precincts with more early voters had
fewer election day voters. This statement is not tautological: it is possible that precincts
16See, for example, “Error and Fraud at Issue as Absentee Voting Rises,” New YorkTimes, October 6, 2012, available at http://www.nytimes.com/2012/10/07/us/politics/
as-more-vote-by-mail-faulty-ballots-could-impact-elections.html?pagewanted=all&_r=0 (lastaccessed August 13, 2013).
17To calculate the black early voting rate by precinct we divided the number of black early voters in aprecinct by the number of registered black voters. Black Election Day voting rates were calculated similarly.
26
with more early voters also had more election day voters and correspondingly fewer absentee
voters. Nonetheless, the more that a precinct’s black voters voted early, the fewer blacks in
line on Election Day. This is only part of the story since a plot (not shown) like Figure 3a
for Hispanic voters shows an identical pattern.
Figure 3: Evidence for Black Shifts to Early Voting
(a) Black Election Day and Early Voting Rates
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●●
●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
● ●●
●
●
●
●
●
●
●
● ●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●●
●● ●●● ● ●
●●● ● ●
●
●●
●● ●● ●●
●
●●●
●
●
● ●
●
● ●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Black Early Voting Rate
Bla
ck E
lect
ion
Day
Vot
ing
Rat
e
(b) Black and Hispanic Early Voters
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●● ●● ●
●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●●
●●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ● ●
●
●
●●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
● ●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
● ●
●●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●●
●
●● ●
●●●
●
●
●●●●
●●
●
●
●
●●●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ● ●● ●●●
●●
●●
●●● ●
●
●●
●
●● ●
●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●●
●●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●●
●
● ●●
●●
● ●●●● ●
● ●●●
●
●●
●
●
●
●
●
● ● ●●●
●●
●●
●
●●
●
●
●
●●
●
● ●
●
●●● ●●
●
●
●
●
●●
●
●
●
●●●
●●●●
●
● ●
●
● ●
●
●
●●
●
●
● ●● ●
●
●
●
● ●●
● ●●● ●
●
●●
● ●●
●
● ● ●● ● ●●●
● ● ● ●●● ●●
●● ●
●
●
●
● ●●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●●●
● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●●●●
●●
● ●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●● ●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●● ●
●●
●
● ●●● ●
●
●
●
●
●
●
●
● ●
●●
●
●
● ●
●
● ●
●●
●
●●●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●●
●
●●
● ●
●
●
●
●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●●
●
●
●
●
●
●
●
●
●●●●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
● ●
●●
● ●
●
●
●
●
●●
●●
●●
●
●●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●● ●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●●●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
● ●
●●●
●
●●
● ●●
●●
●
●
●●
●●
●●
●● ●●●●
●
●
●
●
●
●●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
● ●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●●●●
●
●
●●
●
●
●
●●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●●
●
●
●●
●
●
●
●●
●● ●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●
●
●
●●
●
●●
●
● ●●
●
●
●
●
●
●●
●
●
●●
●
● ●
●
●
●
●
●●
●
●
●
● ●
● ●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●●
●●
●
●
●
●
●
●
●
●
●●
●●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●●
●●
●
●●
●
●
●
●
●
●
●
●
●
●●●
●
●●●●
●
●
●
●
●
●●●
●
●●
●
●●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
● ●
●
●●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●
●●
●●
●
●
●
●
●
●
●
●
●●
●
● ●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●
●
●●
●
●
●
●●
●
●●
●
●● ●●● ●●● ●● ● ●
●
●●
●
●
●
●
●
●
●
●●●
●
●
●
●●
●
●
●
●●
●●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●●●
●
●
●
●
●
●
●
●
●
●
●●
●● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●●●
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●●
●●
●
●●●●
●
●
●
●
●
● ●
●●
●
●
●●
●
●
●●
● ●
●
●
●
●
●●● ●
●
●
●
● ●
●●●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●●●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●●
●
●
●
●●
●●●●●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●●●
●●
●●
●
●
●
● ●●●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
● ●●●●
●●●
●
●
●
●●
●● ●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●●
●
●
●
●●
●●
●●
●●
●
●
●
●●
●
●●●●●●
●
●
●●
●● ●
●
●
●
●
●
●
●●
●
●
●
●
●●●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
● ●●
●●●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
● ●
●
●●
●
●
●
●
●
●
●
●●
●●●●●
●●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●●●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●●
●
●●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●●
●●
●
●
●
●
●
●●
● ●●● ●
●
●●●
●●
●●●
●
●
●
●●
●
● ●
●
●
●●
●
●●
● ●
●●
●●●●
●●●
●
●
●●●
●
●
●●
●●
●
●●
●● ●
●●●● ● ●
●
●●●●●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
● ●
●
●
●
●
●●
●
●●
●
●
●
●
●● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●●●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●●●●
●●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●
●●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●●
●
●
●●●
●
●
●
●
●
●●
●
●
●●●
●●
●●
●
●
●
●
●
●●
●
●●●●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●● ● ●●
●
●
●
●
●
●
●
●
●
●●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.00 0.05 0.10 0.15 0.20 0.25
0.00
0.05
0.10
0.15
0.20
0.25
Precinct Fraction Hispanic Early
Pre
cinc
t Fra
ctio
n B
lack
Ear
ly
Note: each dot represents a Florida precinct, and dot sizes are proportional to the numberof registered voters per precinct. The left panel contains a superimposed regression line, andthe right panel, a 45 degree line.
Although both black and Hispanic voters displayed similar patterns in tradeoffs between
early and Election Day voting, one difference between these two groups is that, on average,
black voters make up larger fractions of precincts. To see this, consider Figure 3b. This
figure plots the fraction of black early voters divided by total precinct registration against
the fraction of Hispanic early voters divided by total precinct registration. More dots are
above the 45-degree line in Figure 3b, and the dots above the line extend higher on the black
27
axis (vertical) than the Hispanic axis (horizontal).18 This is evidence that black early voters
constituted larger portions of precincts than did Hispanic early voters. Moreover, we can
calculate the average difference between the black early voting fraction in Figure 3b and the
Hispanic early voting fraction. This difference is approximately 0.0234 with s = 0.0726. A
simple difference in means test based on the 5,889 precincts in the figure implies that the
difference has a z statistic of almost 25. Therefore, black early voters on average made up
more of a precinct in 2012 compared to Hispanic early voters. As such, when black voters
voted early, there were fewer eligible voters available on Election Day. This is of course true
for Hispanic voters as well, but the effect is greater for black voters. Thus, blacks shifting
away from Election Day voting to early voting may explain the negative and non-significant
black estimate in Table 2. We will provide related evidence shortly that black voters had
long early voting lines, a conclusion that complements what we have just shown.s on Election
Day.
A follow-up model to that in Table 2 is one that focuses on why some classes of voters
cast ballots in late-closing precincts. Is the limiting factor voter check-in machines and/or
procedures, vote tabulating machines, pollworkers, language assistance, voting booths, or
something else? This matter is beyond our present analysis, and we return to it in the
conclusion when discussing suggestions for future research.
Early Voting Wait Times in Miami-Dade County
We now to turn to early voting wait times in Miami-Dade County, and as noted earlier we
have these times because Miami-Dade election officials surveyed the county’s 20 early voting
stations during Florida’s eight day early voting period in 2012. In what follows we describe
patterns in 160 first-reported wait times (20 polling stations across eight days of early voting)
18The axes in Figure 3b are truncated at 0.25 so as not to distort the scale of the figure. There areaccordingly some precincts in Florida not pictured in the figure.
28
and 160 last-reported wait times. In so doing we often refer to what are called early voting
station/day combinations where such a combination is either a first- or a last-reported wait
time on a given early voting day.
Figure 4: Race/Ethnicity and Wait Times in Miami-Dade Early Voting Stations
(a) First-Reported Wait Times
White Hispanic
Black
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
(b) Last-Reported Wait Times
White Hispanic
Black
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Note: each dot represents a Miami-Dade County early voting station on one of eight earlyvoting days. The large dots denote the last day of early voting. Shading is proportional tofirst-reported wait time with darker dots connoting longer waits.
For each Miami-Dade polling station and for each early voting day, i.e., for each sta-
tion/day combination, we calculate fraction black, Hispanic, and white of the combination’s
early voting pool. For example, we calculate fraction black, Hispanic, and white for the first
day of early voting at Aventura Government Center. Then, we plot our 160 first-reported
station/day combinations in a race/ethnicity-based ternary diagram that bears similarity to
the ternary diagrams we have previously discussed. See Figure 4a. In this plot, dot color is
proportional to first-reported wait time, and darker colors connote longer wait times. Fur-
thermore, dot size is larger for the 20 wait times that are associated with the last day of
29
early voting, Saturday, November 3, 2012. We highlight this day in particular because press
reports in Florida noted that there appear to be exceptionally long lines on this day.
The ternary plot in Figure 4a has several implications. First, it shows that non-white
individuals were heavy users of early voting in Miami-Dade County during the 2012 General
Election. The is evident in the lack of dots close to the white vertex in Figure 4a. Indeed,
as is made clear by the clusters of dots near the black and Hispanic vertices in the figure,
a significant number of Miami-Dade station/day combinations were very heavily black and
others, heavily Hispanic.
Second, the color distribution in Figure 4a is not uniform among the 160 pictured sta-
tion/day combinations. The dots closest to the white vertex (and the word “close” should
be understood here in a relative way) are relatively light-colored. However, the darkest
dots in the figure are associated either with heavily black station/day combinations, heavily
Hispanic combinations, or combinations that are roughly split between black and Hispanic.
We do not know the exact times on which any individual Miami-Dade early voter voted.
Because of this, we cannot ascertain precisely which early voters in Miami-Dade had to
contend with first-reported wait times. Thus, while Figure 4a is consistent with the con-
jecture that black and Hispanic early voters had longer early voting wait times than white
votes, strictly speaking the figure implies only that early voting station/day combinations
with disproportionately many blacks and Hispanics tended to have longer first-reported wait
times.
The patterns we observe for first-reported wait times are present in last-reported wait
times as well. See Figure 4b for the latter. The dot shadings in this figure cannot be
compared to the shadings in the earlier Figure 4a, and this is because dot shadings in our
diagrams describe relative wait times as opposed to absolute times.
Regardless, what we see in Figure 4b is that the early voting station/day combinations
with the longest last-reported wait times (these are the darkest dots in the figure) are not in
30
the vicinity of the figure’s white vertex. As with first-reported wait times, the longest early
voting wait times in Miami-Dade County occurred at station/early voting day combinations
that were heavily black, heavily Hispanic, or, roughly speaking, a combination of black and
Hispanic.
We now consider the distribution of party affiliation and early voting wait times. Recall
that each registered Florida voter can choose a party of registration, and the three most
common choices in late 2012 were Democratic, NPA, and Republican. Figure 4 contains
two ternary plots that describe early voting station/day combinations and party composi-
tion. In both plots, dot shadings for the 160 pictured combinations are lighter (darker) for
combinations with relatively shorter (longer) wait times.
Figure 5: Party and Wait Times in Miami-Dade Early Voting Stations
(a) First-Reported Wait Times
Democratic Republican
NPA
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●●
● ●●
●
●
●
●●
●
●
●●
●
●
●
●●
●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
(b) Last-Reported Wait Times
Democratic Republican
NPA
●
●
●
● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●●
●
●●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●●
●
●
●
Note: each dot represents a Miami-Dade early voting station on one of eight early votingdays. The large dots denote the last day of early voting. Shading is proportional to lastreported wait time with darker dots connoting longer waits.
The two ternary plots in Figure 5 have similar features, and without loss of generality
31
we focus on first-reported wait times and party affiliation. With this in mind, Figure 5a has
two notable implications. First, early voting station/day combinations across Miami-Dade
County had a disproportionate number of Democrats. This is consistent with evidence in
Herron and Smith (2012), who show that early voting across Florida during the 2008 General
Election was heavily used by voters registered as Democrats.
Second, the most Republican of the station/day combinations in Figure 5b are not darkly
colored. In contrast, the darkest dots in the figure are much closer to the Democratic vertex.
There are some dark dots located approximately in the center of the Figure 5b; these connote
station/day combinations that were roughly split with voters registered Democratic, those
registered Republican, and those without party affiliation. But, Figure 5b highlights clear
skew in the distribution of dark dots and in particular a lack of dark dots near the Republican
vertex.
Table 3 contains tobit regression results for last-reported Miami-Dade early voting wait
times. The dependent variable in the tobit model is wait time in minutes, and the smallest
possible wait time is zero minutes. We see in Table 3 longer wait times on the station/day
combinations that had many black and Hispanic voters; note that white fraction is the
omitted category in the early voting tobit. The Democratic estimate in Table 3 is not
significant at conventional confidence levels, but the NPA estimate is; fraction Republican
is omitted. The tobit model in Table 3 includes fixed effects for the 20 early voting stations
in Miami-Dade County, and Aventura Government Center is omitted as a base category.
We do not know many details about resources for these 20 voting stations, but what is
particularly notable in Table 3 is the variability in fixed effects. Some early voting stations
had last-reported wait times that were almost five hours longer than the base category,
ceteris paribus.19
19A tobit model for first-reported wait times has similar results (available from the authors). The onlyqualitative differences between first-reported tobit results and those in Table 3 are in fraction black (z = 1.83,less significant with first-reported wait times) and fraction Democratic (z = 2.09 more significant).
32
Table 3: Tobit Regression Results for Last-Reported Miami-Dade Early Voting Wait Times
Estimate Standard Error z value p-valueFraction black 859.57 183.45 4.69 0.00Fraction Hispanic 944.56 179.06 5.28 0.00Fraction Democratic 241.30 197.36 1.22 0.22Fraction NPA 1027.02 197.51 5.20 0.00Constant -702.05 126.73 -5.54 0.00City of Miami - City Hall -31.93 39.97 -0.80 0.42Coral Gables Library -67.66 60.31 -1.12 0.26Coral Reef Library -18.92 33.61 -0.56 0.57Florida City - City Hall -198.78 36.43 -5.46 0.00John F. Kennedy Library -285.20 112.46 -2.54 0.01Kendall Branch Library -37.48 58.89 -0.64 0.52Lemon City Library -222.99 65.64 -3.40 0.00Miami Beach City Hall -23.93 39.61 -0.60 0.55Miami Lakes Public Library -255.16 72.03 -3.54 0.00Model City Library (Caleb Center) -306.11 76.92 -3.98 0.00North Dade Regional Library -242.49 83.12 -2.92 0.00North Miami Public Library -208.48 73.10 -2.85 0.00North Shore Branch Library -109.94 41.20 -2.67 0.01SOE Main Office -331.30 81.82 -4.05 0.00South Dade Regional Library -141.82 41.08 -3.45 0.00Stephen P. Clark Center -263.07 42.10 -6.25 0.00West Dade Regional Library -219.94 110.05 -2.00 0.05West Flagler Branch Library -363.14 102.91 -3.53 0.00West Kendall Regional Library -210.29 73.53 -2.86 0.00Log(scale) 3.93 0.06 69.80 0.00
The four ternary diagrams in Figures 4 and 5 and the regression results in Table 3 are
consistent with the conjecture that non-white, Democratic, and NPA-registered voters had
longer early voting wait times than their racial/ethnic and party counterparts. With respect
to black voters in particular, they are also consistent with the idea that many of these voters
selected out of Election Day voting which alleviated late-closing precincts times at the cost
of longer early voting wait times.
Like our closing time results, these conclusions rely heavily on aggregate data, and from
these data we cannot make individual-level conclusions that are completely certain. This
33
is because we do not know the precise times when Miami-Dade early voters actually voted,
who was in line at noon at each station, and so forth. The caution that we are emphasizing
here reflects the risk of ecological fallacy and the fact that we cannot be sure of the races,
ethnicities, and party affiliations of the voters who were in line at the beginning or end of
each early voting day in each early voting station. It turns out, however, that a fortuitous
natural experiment occurred in Miami-Dade and Palm Beach Counties, and this experiment
provides us with some unequivocal information about the types of voters who were last in
line on one day of early voting.
The source of the natural experiment in Miami-Dade and Palm Beach Counties is as
follows. In these two counties, some early voters who were in line on Saturday, November 3,
ended up voting after midnight on Sunday, November 4, i.e., on the day after they initially
got in line. These voters had arrived at their early voting stations before the polls closed at
7:00pm on November 3, but the extensive lines that they faced delayed their actual voting
until November 4. Thus, each November 4 early voter waited at least five hours to vote. In
Palm Beach County it was reported that early voted ended at 2:30am on the morning of
Sunday, November 4. There were a total of 573 early voters on this Sunday, of which 431
voted in Miami-Dade County and 142 in Palm Beach County.20
As an aside, there were also 332 early voters who voted in Orange County on Sunday,
November 4. The explanation for this collection of voters, though, has nothing to do with
voting lines on November 3. Rather, a bomb threat on November 3 at an Orange County
early voting center, Winter Park Library, led to the center’s being evacuated. After an
emergency motion was filed in state court by the Florida Democratic Party, a judge ordered
20For details on the early voting lines, see “Early Voting Ends With Long Lines, LongWaits,” CBS News Report, November 4, 2012, available at http://miami.cbslocal.com/2012/11/04/
early-voting-ends-with-long-lines-long-waits (last accessed November 5, 2012). For the PalmBeach closing time, see “Another election, another legal tangle in Florida,” Sun Sentinel, November 5,2012, available at http://www.sun-sentinel.com/news/broward/fl-florida-voting-mess-20121105,
0,6901020,full.story (last accessed November 6, 2012).
34
that early voting at Winter Park Library be held on Sunday, November 4.21
Because the November 4 early voters in Miami-Dade and Palm Beach Counties have
November 4 voting dates in the two counties’ respective early voting files, we can distinguish
these voters from the rest of the voters who lined up on November 3 to vote in Miami-Dade
and Palm Beach Counties. This allows us to address the matter of the types of voters in
Miami-Dade and Palm Beach Counties who were last in line on November 3. Voters who
arrived at an early voting station after 7:00pm were not allowed to join early voting lines;
this means that all November 4 were in line as of 7:00pm on November 3.
Table 4 lists the three race/ethnicity groups considered here along with the percentage
(“Voting Percent”) that each group constituted in the November 4 early voting electorate.
The table also lists the percent (“Registration Percent”) that each race/ethnicity group made
up of the combined registered voter pool in Miami-Dade and Palm Beach Counties as of the
2012 General Election.
Table 4: Sunday, November 4, Voting in Miami-Dade and Palm Beach Counties
Race/Ethnicity Count Voting Percent Registration PercentBlack 251 43.8 16.7Hispanic 182 31.8 35.5White 70 12.2 41.4
Note: percents do not sum to 100 because only three racial and ethnic groups are listed.
The implication of Table 4 is straightforward: black and Hispanic voters were dispropor-
tionately the last in line to vote in Miami-Dade and Palm Beach Counties on November 3 and
then on into November 4. While blacks made up approximately 43.8 percent of November
4 early voters, they constituted only 16.7 percent of registered voters in these two counties.
21For details on the Orange County bomb threat, see the Reuters report “Early voting extendedat Florida site hit with bomb scare” available at http://www.reuters.com/article/2012/11/04/
us-usa-campaign-florida-idUSBRE8A30GG20121104 (last accessed November 5, 2012). See as well, ABCNews Story, “Judge Orders 1 Fla. County to Extend Early Voting,” November 4, 2012, available athttp://abcnews.go.com/US/wireStory/fla-democrats-early-voting-extended-county-17636899#
.UJiUqYb4Img (last accessed November 6, 2012).
35
In contrast, white voters (12.2 percent) were disproportionately under-represented among
November 4 voters compared to their status as the largest racial group (41.4 percent) in the
combined Miami-Dade and Palm Beach Counties.
To get a sense of the magnitudes of the voting percentages in Table 4, consider Figure
6, which for the entire early voting period in Florida plots the racial/ethnic compositions of
early voters in the combination Miami-Dade and Palm Beach Counties.
Figure 6: Composition of Early Voting Electorate by Race, Miami-Dade and Palm BeachCounties
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Date
Com
posi
tion
Ear
ly E
lect
orat
e
10/27 10/28 10/29 10/30 10/31 11/01 11/02 11/03 11/04
●
●●
● ● ●●
●
●
●●
●● ●
●
●
●
●
●●
● ● ●●
●
●
●
●
●
●
African AmericanHispanicWhite
According to Figure 6, not only were minority voters—here, black and Hispanic voters—
disproportionately likely to vote on Saturday, November 3, the last official day of early
voting, but black voters in particular were disproportionately likely to vote toward the end
36
of this day (on into the next day). Indeed, the percentage of Miami-Dade and Palm Beach
County black voters who voted on November 4 skyrocketed compared to November 2 and
3. Thus, not only did long voting lines disproportionately affect minority/non-white early
voters on November 3, but among those early voters who were still in line on November 3,
the worst affected appear to be black voters. One interesting question is the extent to which
long lines at early voting stations, in Miami-Dade, Palm Beach, and elsewhere, led some
voters to abandon their plans to vote early. To the best of our knowledge, there is at the
time of this paper’s writing no data that bears directly on this question.
It is possible that the black and Hispanic voters who voted very early on November 4
arrived very late, yet prior to 7:00pm on November 3, at their early voting stations. We do
not know early voter arrival times (these are not known in general), but we can check the
wait times at the five Miami-Dade early voting stations that had early voters on November
4 in order to see if they had a surge of voters at the end of the voting day. These wait times
are described in Table 5.
Table 5 shows that wait times in the five Miami-Dade early voting stations were relatively
long throughout November 3. They were, roughly speaking, later toward 7:00pm, but Table
5 does not contain evidence that these stations had slack prior to 7:00pm and then were
inundated with a crush of voters. Indeed, these five stations operated with lines throughout
the entire day on Saturday, November 3. Table 5 contains data from Miami-Dade County
only because, to the best of our knowledge, no wait time data on Palm Beach County early
voting stations exist.
Finally, with respect to party registration there is a Democratic bias among November 4
early voters in Miami-Dade and Palm Beach Counties. On November 4—more precisely, in
the early morning hours of the day—there were 485 Democratic voters, 207 NPA voters, and
180 Republican voters. Percentages are approximately 54, 23, and 20, respectively (percent-
ages do not sum to 100 because a small number of November 4 early voters were members
37
Table 5: November 3 Wait Times in Miami-Dade Early Voting Stations with November 4Voting
Time of Day Aventura Miami Lakes North Miami South Dade West Kendall7:25 1:15 1:30 1:00 3:00 4:009:19 4:00 2:30 2:00 5:00 5:0010:29 3:30 3:00 2:00 3:00 4:0011:09 4:10 3:00 4:00 5:00 4:3012:00 4:10 3:00 4:00 5:00 4:3012:20 3:00 3:00 4:00 3:00 5:001:09 3:00 3:00 4:30 5:00 4:002:08 5:00 5:30 4:00 5:00 5:003:54 3:00 6:00 5:00 5:00 6:004:21 5:10 5:00 3:00 5:00 5:004:37 6:00 6:00 3:00 6:00 5:005:30 6:00 6:00 4:00 5:00 6:006:42 4:00 6:00 2:30 6:00 5:30
Note: this table shows wait times on Saturday, November 3. Early voting station names areshortened in the table. Aventura is Aventura Government Center; Miami Lakes is MiamiLakes Public Library; North Miami is North Miami Public Library; South Dade is SouthDade Regional Library; and West Kendall is West Kendall Regional Library.
of third parties). Across Miami-Dade and Palm Beach Counties, registration fractions by
party were approximately 44 percent Democratic, 25 percent NPA, and 29 percent Repub-
lican. Hence, the extreme end of the early voting lines in Miami-Dade and Palm Beach
Counties were disproportionately Democratic and disproportionately non-Republican.
The natural experiment we have described is informative but, like all natural experiments,
has its limitations. One could in theory argue that November 3 early voters in Miami-Dade
and Palm Beach County were fundamentally different than voters on other early voting days.
Perhaps black and Hispanic voters on November 3 arrived late in the afternoon to vote while
in contrast these types of voters voted earlier in the day on November 2. We know of no
such information that would sustain this assertion, but nonetheless it is possible. Similarly,
one could conjecture that Democrats voted lated late in the day on November 3 but early in
the day on October 28, the first day of early voting. Again, we know of no reason to think
38
that such a conjecture holds. Rather, the natural experiment caused by extensive lines in
Miami-Dade and Palm Beach Counties on November 3 has made it clear that, in the one
place where it is knowable, racial minorities and voters registered Democratic made up a
disproportionately large share of the early voters who were last in line to vote.
Conclusion
This paper is to the best of our knowledge one of the first statistical analysis of observed
precinct closing times, and to a limited extent early voting wait times, that covers a major
state—Florida—in an important election—the 2012 General. Both poll closing and wait
times are measures of precinct congestion, and our findings contribute to a small but devel-
oping literature on the distribution of congestion and the types of voters who have to deal
with congested voting stations. Here we have focused on race/ethnicity and party-based ag-
gregates in our study of the types of Florida voters who were subjected to congestion in the
2012 General Election, and given the primitive state of the literature on precinct operations
this is appropriate, in our opinion. An obvious extension to our results would be one that
considers the causes of the late closing and long wait times that we have identified, and to
this end our analysis is hopefully but one of the first to study patterns in observed closing
times.
We have shown that Hispanic and Democratically-registered voters in Florida who voted
on Election Day 2012 tended to use precincts that closed disproportionately late. And, our
early voting analysis of precinct wait times, limited primarily to Miami-Dade County due
to data availability, showed that black and Hispanic voters disproportionately voted in early
voting stations that had long lines. These findings are based on aggregate data, which may
not be ideal, but at present aggregate data are essentially the only data available with which
scholars can assess observed measures of precinct congestion. Our analysis has benefitted
39
from a natural experiment that occurred in Miami-Dade and Palm Beach Counties, an
experiment which shows unequivocally that the voters last in line in these counties on the
last day of early voting in 2012 were disproportionately minority (black and Hispanic) and
Democratic. This latter statement is not subject to critique on the basis of ecological fallacy.
In the shadow of the Supreme Court’s recent overturning of Section 4 of the Voting Rights
Act, our findings show that undue congestion at the polls may effectively deny or abridge the
right to vote of people of color. The differential impact that congestion had on thousands
of minority voters in in the 2012 General Election in Florida, both during early voting as
well on Election Day, may likely interest several parties in the post-Shelby landscape with
regard to the weakened VRA. Interested parties may include members of Congress keen on
revamping outdated Section 4 geographical targeting and preclearance criteria, voting rights
activists pursuing Section 2 anti-discrimination litigation to address racially biased voting
practices, and the United States Justice Department which can use Section 3 of the VRA to
bail-in a jurisdiction with the aim of preventing persistent infractions of voting rights.
The greatest limitation that studies of precinct congestion face is lack of available data.
The research design and findings herein suggest that several improvements can help alleviate
the data problem in the future and in so doing lead to research results which improve the
voting process in the United States.
Our first suggestion is to standardize the definition of precinct closing time. We suggest
that precincts record the precise time when the last voter processed completes his or her
voting process, meaning, has his or her ballot counted by a machine. In Florida this would
generally mean the last time that a voter has a ballot read by an optical scan voting machine.
This time, we believe, is an intuitive definition of closing time. If, say, all counties in
Florida were to adopt this definition and make public their closing times, then scholars and
election administrators could easily compare closing times within and across counties so as
to better understand what leads to late-closing precincts. This could in theory lead to a
40
better allocation of election resources. Furthermore it could assist county SOEs in efforts
aimed at developing best practices as they pertain to precinct congestion.
A second suggestion is for precincts to estabish a regular schedule for monitoring lines.
For example, precincts could count the number of people in line to vote at the top of every
hour. These data could be made public, either during an election or thereafter, and such
data would shed light on waiting line evolution and the types of precincts that tend to have
long lines. It is important that all precincts count their lines simultaneously. If scholars knew
line lengths at regular intervals on Election Day then they could rather easily characterize
the evolution of lines throughout the day.
Our third suggestion is to resolve the aggregate data problems that affect some of the
analyses we have described. These problems could be addressed as follows. Before each
election, the Florida Department of State could designate a few select precincts wherein
some type of a line monitor who would collect or write down the name of every voter (or
a randomly selected set of voters) who joined a voting queue. A monitor on the back side
of a precinct would request names of voters as they left. With timing monitors like these—
and various electronic systems may be feasible as well, i.e., voters send a text message to
a provided number upon entering a line and then a text message upon leaving—it would
be possible to determine at an individual-level whether certain types of prospective voters
were waiting for a long time to vote. The Florida Department of State could also randomly
select, say, 10,000 voters to monitor during the run-up to Election Day with the idea being
that these voters would be surveyed about their waiting periods and so forth either during
or after a general election’s voting period.
This study shows that congestion at the polls in the 2012 General Election was not
uniformly distributed across Florida but rather had a disproportionate impact on people of
color, especially Hispanic voters. Our evidence writ large casts doubt on the claim that there
is no longer racial discrimination in the matter of voting in the United States. The results we
41
have discussed here are based on precinct closing and wait times, and we have been forthright
about the limitations in these data. In what is now the post-Shelby era, policymakers seeking
measures which are sensitive to the presence or absence of racial/ethnic discrimination in
elections might begin by looking critically at patterns in voting lines and identifying the
voters most affected by them.
42
References
Alvarez, R. Michael, Stephen Ansolabehere, Adam Berinsky, Gabriel Lenz, Charles StewartIII and Thad Hall. 2009. “2008 Survey of the Performance of American Elections.” Reportfrom the Caltech/MIT Voting Technology Project.URL: http://www.vote.caltech.edu/sites/default/files/Final%20report20090218.pdf
Claassen, Ryan L., David B. Magleby, J. Quin Monson and Kelly D. Patterson. 2008. “”AtYour Service: Voter Evaluations of Poll Worker Performance.” American Politics Research36(4):612–634.
Herron, Michael C. and Daniel A. Smith. 2012. “Souls to the Polls: Early Voting in Floridain the Shadow of House Bill 1355.” Election Law Journal 11(3):331–347.
Herron, Michael C. and Daniel A. Smith. 2013a. “Congestion at the Polls: A Study ofFlorida Precincts in the 2012 General Election.” Report commissioned by AdvancementProject.URL: http://b.3cdn.net/advancement/f5d1203189ce2aabfc 14m6vzttt.pdf
Herron, Michael C. and Daniel A. Smith. 2013b. “The Effects of House Bill 1355 on VoterRegistration in Florida.” Forthcoming, State Politics & Policy Quarterly.
Highton, Benjamin. 2006. “Long Lines, Voting Machine Availability, and Turnout: TheCase of Franklin County, Ohio in the 2004 Presidential Election.” PS: Political Scienceand Politics 39:65–68.
Kimball, David C. 2013. “Why Are Voting Lines Longer for Urban Voters?” Paper preparedfor presentation at the Annual Meeting of the Southwestern Social Science Association,New Orleans, LA.URL: http://papers.ssrn.com/sol3/papers.cfm?abstract id=2255009
Mukherjee, Elora. 2009. “Abolishing the Time Tax on Voting.” Notre Dame Law Review85(1):177–246.
Spencer, Douglas M. and Zachary S. Markovits. 2010. “Long Lines at Polling Stations?Observations from an Election Day Field Study.” Election Law Journal 9(1):3–17.
Stewart III, Charles. 2013. “Waiting to Vote in 2012.” Forthcoming, Journal of Law andPolitics.
43