let’s not forget about accuracy henry e. brady class of 1941 monroe deutsch professor of political...
TRANSCRIPT
Let’s Not Forget About Accuracy
Henry E. BradyClass of 1941 Monroe Deutsch
Professor of Political Science and Public PolicyUniversity of California, Berkeley
Accuracy and Security
• Accuracy: Votes are properly recorded according to voter’s intention
• Security: Votes are not changed because of misfeasance or malfeasance – mistakes or fraud
A Word About Security
• Note:– Risk = Threat X Vulnerability
• Many discussions only consider vulnerabilities and not threats.
• But it takes both to create a risk.
Measuring Accuracy
• Residual votes:– Number of ballots minus number of votes in a race– Example: 100 ballots & 97 votes so 3 residual votes
• Residual votes are sum of:– overvotes (ballots with more than one mark for a
contest, thus invalidating the vote) and – undervotes (ballots with no mark for a contest, thus
not counting as a vote)
• Residual vote rate-- Residual votes over number of ballots-- Example: 3/100 = 3%.
Cause of Residual Votes
• Not all residual votes are “errors” – sometimes people choose not to vote in a race: They intentionally undervote or overvote.
• But for “top-of-the-ticket” contests, residual votes should be low – around ½% to 1%.
• If two jurisdictions are similar except for voting systems and one has significantly lower residual vote rates than the other, then the difference is the performance of voting systems (technology, poll workers, administration).
Table 1 – Average of Lowest Residual Vote Rate For Each County among Eight Propositions by Polling Place Voting System Types
Type of System Model Mean Over
Counties
Number of Counties
Mean weighted by
Ballots
Number of Ballots
Direct Record Electronic
Accuvote-TS 1.00% 2 0.88% 322,048
AVC Edge 2.09% 11 0.92% 1,047,960
Hart E-slate 1.12% 1 1.12% 634,588
iVotronics 1.07% 1 1.07% 35,215
Total 1.81% 15 0.98% 2,039,811
Optical Scan—Precinct Count
Accuvote-OS 0.82% 15 0.84% 1,523,657
Eagle 2.38% 3 2.82% 364,198
Total 1.08% 18 1.22% 1,887,855
Optical Scan—Central Count
Ink-A-Vote 3.52% 1 3.52% 1,661,675
M100,550,650 0.69% 7 0.64% 739,621
Mark-A-Vote 1.60% 5 1.17% 263,703
Total 1.26% 13 2.49% 2,664,999
Punch-Card Datavote 1.01% 8 0.93% 330,168
Total 1.01% 8 0.93% 330,168
Mail Mail 0.69% 2 0.83% 1,936
Total 0.69% 2 0.83% 1,936
Total All Systems
1.29% 56 1.62% 6,924,769
Comments and Caveats
• Data are from preliminary canvas• Residual rates are by polling place voting
system – but many votes are absentee• Two types of averages in the table:
– Simple county averages– Averages weighted by number of ballots
• Two counties omitted:– Monterey (used two types of systems)– Butte – Instructive story
Further comments
• Most rates less than 1.2%
• High ones due mostly to two counties:– Los Angeles – 3.5% -- Ink-A-Vote– San Francisco – 4.0% -- ES&S Optech Eagle
• Only three other small counties over 1.6%:– Sutter – 4.5% (Mark-A-Vote)– Mariposa – 6.5% (Sequoia AVC Edge)– Mono -- 8.5% (Sequoia AVC Edge)
Lowest Residual Vote Rate for Each County
8.75
8.13
7.50
6.88
6.25
5.63
5.00
4.38
3.75
3.13
2.50
1.88
1.25
.63
Figure 1: Histogram of Residual Vote Rates
for 57 California Counties 35
30
25
20
15
10
5
0
Std. Dev = .01
Mean = .01
N = 57.00
LA,SF,Sutter Mariposa Mono
What’s Going on In Los Angeles?
• Consider November 2004 data
• Strong positive relationship between:– Percent minority and residual vote
• But this seems to be the result of strong negative relationship between: – Percent high school graduates and residual
vote
Pres. Residual Rate by % Minority for Cities in LA County
Preliminary Data Collected From LA County Website in Nov. 2004
% Hispanic or Non-White
110%
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Resid
ual R
ate
for
Pre
sid
ency,
Nov.,
2004
5.0%
4.5%
4.0%
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
.5%
0.0%
WhittierWestlake Village
West Hollywood
West Covina
WalnutTorrance
Temple City
South Pasadena
South Gate
South El Monte
Signal Hill
Sierra Madre
Santa Monica
Santa Fe Springs
Santa Clarita
San Marino
San Gabriel
San Fernando
San Dimas
Rosemead
Rolling Hills Estate
Rolling Hills
Redondo Beach
Rancho Palos Verdes
Pomona
Pico Rivera
Pasadena
Parmount
Palos Verdes Estates
Palmdale
Norwalk
Monterey Park
Montebello
Morovia
Maywood
Manhattan BeachMalibu
Lynwood
Los Angeles
Long Beach
Lomita
LawndaleLa Verne
La Puente
Lancaster
La Mirad
Lakewood
La Habra Heights
La Canada/Flintridge
Irwindale
Inglewood
Huntington Park
Hidden Hills
Hermosa Beach
Hawthorne
Hawaiian Gardens
GlendoraGlendale
Gardena
El Segundo
El MonteDuarate
Downey
Diamond Bar
Culver City
Cudahy
Covina
Compton
Commerce
Claremont
CerritosCarson
Calabasas
BurbankBeverly Hills
Bell Gardens
Bell Flower
Bell
Baldwin ParkAzusa
Avalon
Artesia
Arcadia
Alhambra
Agoura Hills
Pres. Residual Rate by Percent HS Graduation
for Cities in LA County in 2004
Per Cent High School Grad
110%
100%
90%
80%
70%
60%
50%
40%
30%
20%
Resid
ual R
ate
for
Pre
sid
ency,
Nov.
2004
5.0%
4.5%
4.0%
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
.5%
0.0%
City Size
gt 1000K
200-1000K
100-200K
50-100K
30-50K
20-30K
10-20K
5-10K
lt 5000
Total Population
WhittierWestlake Village
West Hollywood
West Covina
WalnutTorrance
Temple City
South Pasadena
South Gate
South El Monte
Signal Hill
Sierra Madre
Santa Monica
Santa Fe Springs
Santa Clarita
San Marino
San Gabriel
San Fernando
San Dimas
Rosemead
Rolling Hills Estate
Rolling Hills
Redondo Beach
Rancho Palos Verdes
Pomona
Pico Rivera
Pasadena
Parmount
Palos Verdes Estates
Palmdale
Norwalk
Monterey Park
Montebello
Morovia
Maywood
Manhattan BeachMalibu
Lynwood
Los Angeles
Long Beach
Lomita
Lawndale La Verne
La Puente
Lancaster
La Mirad
Lakewood
La Habra Heights
La Canada/Flintridge
Irwindale
Inglewood
Huntington Park
Hidden Hills
Hermosa Beach
Hawthorne
Hawaiian Gardens
GlendoraGlendale
Gardena
El Segundo
El MonteDuarate
Downey
Diamond Bar
Culver City
Cudahy
Covina
Compton
Commerce
Claremont
CerritosCarson
Calabasas
BurbankBeverly Hills
Bell Gardens
Bell Flower
Bell
Baldwin Park Azusa
Avalon
Artesia
Arcadia
Alhambra
Agoura Hills
What Causes these Relationships?
• Consider Fresno’s “experiment” of going from Votomatic style punchcards in 1996 to optical scan with precinct count in 2000
• The people voting remained essentially the same, only the voting system changed.
• Hence, difference in residual rates must be due to voting system.
Residual Votes in Fresno County
with Votomatic Punch in 1996
Percent Minority Voting Age Population in Tract
100%90%80%70%60%50%40%30%20%10%0%
Perc
ent
Pre
sid
ential R
esid
ual V
ote
s
8%
7%
6%
5%
4%
3%
2%
1%
0%
Residual Votes in Fresno County
with Optical Scan Precinct in 2000
Percent Minority Voting Age Population in Tract
100%90%80%70%60%50%40%30%20%10%0%
Perc
ent
Pre
sid
ential R
esid
ual V
ote
s
8%
7%
6%
5%
4%
3%
2%
1%
0%
Comparing Los Angeles with Other Counties
• Compare residual vote rates with adjoining Counties.
• If counties are similar, then those with higher residual vote rates have voting systems that are performing badly.
• Are counties similar?
Table 2: Residual Vote Rates for Los Angeles and Adjacent Counties
County Residual Vote Rate November 2004
Residual Vote Rate November 2005
Kern 1.51% 0.51%
Los Angeles 2.02% 3.52%
Orange 1.74% 1.13%
San Bernardino 1.06% 0.61%
Ventura 0.93% 0.72%
Note: In November 2004 the residual vote rate is for the presidential race. In November 2005, the residual vote rate is the lowest among all eight propositions.Sources: California Secretary of State’s Web Page and Los Angeles Counties web page.
Maybe Los Angeles is Just Different?
• Again consider November 2004 data.
• Look at precincts at LA border and compare them with contiguous precincts in adjoining counties.
• People should be similar, but voting systems are different.
• If difference in residual votes, then due to voting systems.
Conclusions
• Some voting systems more accurate than others, • Residual vote rates can identify problems with voting
systems,• Better data, such as breaking out undervotes and
overvotes by absentee, early-voting, and polling place voting in each precinct, would be very useful for assessing accuracy,
• The state of California should, after adjusting for differences in voters across counties, regularly prepare a “Voting Performance Report Card” to identify systems with low residual vote rates and those with high rates,
• Accuracy should be given as much weight as security in considering the performance of voting systems.