Who Overvotes, Who Undervotes, Using Punchcards? Evidence from Los Angeles County

• Date: 01 March 2004
• DOI: 10.1177/106591290405700102
• Author: D. E. “Betsy” Sinclair,R. Michael Alvarez
• Published date: 01 March 2004
• Subject Matter: Articles

Who Overvotes, Who Undervotes, Using Punchcards?
Evidence from Los Angeles County
D. E. “BETSY” SINCLAIR and R. MICHAEL ALVAREZ, CALIFORNIA INSTITUTE OF TECHNOLOGY
In this study we examine over- and undervotes from the November 2000 General Election in Los Angeles
C o u n t y. Los Angeles County is the nation’s largest election jurisdiction, and it used a punchcard voting system
in that election. We use precincts as our unit of analysis and merge the 2000 election data with census data and
voter registration data; our dataset allows us to examine all of the countywide races in 2000 (including candi-
date and ballot measures). We use a multivariate statistical analysis employing negative binomial re g ression to
test hypotheses re g a rding the relationship between precincts’ political and demographic characteristics and over-
and undervotes. We demonstrate that both over- and undervotes vary systematically across precincts in Los
Angeles County, a finding that we argue has important implications for the re p resentation of political intere s t s .
In every election ballots for some races are not counted. p e rcent in presidential races from 1988-2000, followed by
These votes, commonly called the “residual vote” or
e l e c t ronic voting systems, optical scan systems, lever
“ ro l l o ff rate,” result from three diff e rent problems: over-
machines, and paper ballots (Caltech/MIT 2001).1 Further-
voted ballots (ballots where there are more votes cast for a
more, when studies have examined overvoting and under-
p a rticular race than allowed), undervoted ballots (ballots
voting rates for different voting systems, stark differences
w h e re there is no indication of a vote for a race), and spoiled
are found; some voting systems, like electronic and lever
or uncounted ballots (ballots that are not counted at all for
machines, can block overvoting—also, when certain voting
various reasons). As a result of the 2000 presidential election,
systems are used (like precinct-based optical scanning and
t h e re is now increased awareness of both overvoting and
newer “touchscreen” voting machines), voters can check for
u n d e rvoting, and increased concern that both might be
overvotes and undervotes before their ballots are counted.2
a ffected by the mechanism used to cast votes (called the
However, we are not concerned simply with the exis-
“voting system”), by diff e rences in voter sophistication, famil-
tence of residual votes, or which voting systems produce
iarity with the election process, or other voter attributes.
higher rates of overvotes or undervotes. Instead, our con-
Clearly, some voting systems generate a higher residual
cern is whether overvotes and undervotes occur randomly
vote rate than other systems, a fact noted in many studies
within a population of voters who use the same voting
(Ansolabehere 2001; Alvarez, Sinclair, and Wilson 2003;
system, here the “Votomatic” punchcard voting system.3
Knack and Kropf 2003; Posner 2001; Brady et al. 2001;
Thus, rather than examine residual voting rates a c ross diff e rent
Montgomery 1985; Shocket, Heighberger, and Brown 1992;
GAO 2001; U.S. House Committee on Government Reform
2001). Punch card voting systems generally have the high-
1 The terminology “voting system” used here refers to the physical means
est residual vote rates in recent elections, averaging 2.5
by which voters indicate their preferences in an election context, and
how those are then tabulated. Currently in the United States there are
four major types of voting systems. Paper ballots are voting systems
NOTE: We thank Conny McCormack (Los Angeles County Registrar
where the choices facing voters are printed on a sheet of paper, the voter
Recorder), and Brian Ikenaga and Vern Cowles (of the Los Ange-
indicates his or her preferences on the paper, and the ballots are then
les County Registrar Recorder Office) for their assistance with this
counted by hand. Lever machines are mechanical devices where voters
project. Some of the data we use in our analysis came from the
indicate their choices by pulling selected levers, and the votes are tabu-
California Statewide Database, made available by the Institute of
lated mechanically via a counter. Punch card voting systems employ
Governmental Studies at the University of California, Berkeley. We
computer-readable punch cards, of two basic types: “Votomatic” punch
also thank Catherine Wilson for her insights. Karen Kerbs and
card systems use a card that is inserted into a mechanical holder with a
Mary King Sikora provided administrative help. Earlier versions of
voting booklet, and the voter indicates his or her choices by punching
this research were presented at the 2002 Annual Meetings of the
certain holes out; and “Datavote” punch cards, where the choices are
Midwest Political Science Association, and at Stanford University’s
printed on the ballot, and voters indicate their preference by again
Department of Political Science’s American Empirical Seminar; we
punching out an appropriate location on the card. Last, there are direct
thank participants in both forums for their comments and cri-
recording electronic devices (“DREs”) and touchscreen type computer-
tiques. This research was supported by the Caltech/MIT Voting
ized voting systems; the DREs are essentially computerized lever
Technology Project with a research grant from the Carnegie Cor-
machines, while touchscreen systems are similar to bank automatic teller
poration; however the conclusions reached here reflect the views
machines. For further discussion of the differences between these sys-
of the authors and not necessarily the Caltech/MIT Voting Tech-
tems and their use in the United States, see Saltzman (1988) or Cal-
nology Project. This research was also supported by grants to
tech/MIT (2001).
2
Alvarez from the IBM Corporation through the University Match-
For detailed discussion of how these specific voting systems have
ing Grants Program, and from the Haynes Foundation.
worked in recent elections, see Alvarez, Sinclair and Wilson’s (2003) Cal-
ifornia’s analysis or Quirk et al.’s (2002) study of Illinois.
Political Research Quarterly, Vol. 57, No. 1 (March 2004): pp. 15-25
3 We define the “Votomatic” punchcard system below in more detail.
15

---

16
POLITICAL RESEARCH QUARTERLY
voting systems, our approach here is to examine them across
and the Fourteenth Amendment to the United States Consti-
precincts in one election jurisdiction to see if they are ran-
tution. Such cases have been filed recently in Illinois, Geor-
domly distributed over the population of voters, holding the
gia, Florida, and California; some of these cases (for example
type of voting system constant.
in California) have resulted in the prohibition of some
This is a critical question for students of elections. While
p u n c h c a rd voting systems.4 F u rt h e rm o re, the recent federal
t h e re is a vast literature on voting behavior, too large to easily
election re f o rm bill (the “Help America Vote Act”) contains a
summarize here, this literature has largely ignored what hap-
number of provisions that may lead to the elimination of
pens when a citizen enters the voting booth and casts a ballot.
many current punchcard voting systems in the United States.
Instead, re s e a rch in this area usually assumes that a voter’s
Our study uses data from Los Angeles County, Californ i a .
p re f e rences are directly and accurately translated into ballot
We examine over- and undervote rates for a series of election
choices, and hence are converted certainly into election out-
contests in the 2000 election. We integrate into our database
comes. If error is introducted into the voting process, either
p re c i n c t - b y - p recinct information on voter registration, race
by the mechanical or electronic device that re c o rds and
and ethnicity, gender, and the use of non-English ballots. We
counts ballots, or by diff e rences in how certain voters interact
test hypotheses re g a rding partisanship, race and ethnicity,
with the voting device, then election re t u rns may not accu-
and language use, on over- and undervoting. Early re s e a rc h
rately reflect voter pre f e rences. Errors and uncounted votes
by Steifbold (1965) provides an analytic framework for gen-
also make it possible that in a close election, the election out-
erating hypotheses about the types of voting behavior that
come might be affected by imperfections in the voting
can lead to over- or undervoted ballots (which Steifbold
p rocess. This, combined with the possibility that uncounted
called “invalid” ballots). The first type of voting behavior
or uncountable votes might be more likely to come from cer-
leading to over- and undervoted ballots Steifbold (1965:
tain types of voters, implies that the voting process itself
406) called ballots cast by “apathetic invalids,” “who
might affect the quality of political re p re s e n t a t i o n .
respond(s) to the general electoral mobilization because it is
The most pressing question for our study, then, is
the thing to do, but who remains actually indiff e rent to the
whether certain groups of voters are more likely to cast
political system. Whether out of ignorance, negligence or
residual votes, in jurisdictions using punchcard voting sys-
c ro s s - p re s s u res,” they over- or undervote. The other gro u p
tems. Recent studies have found that residual votes tend to
Steifbold called the “highly politicized invalid,” “who knows
be...