CALTECH/MIT
VOTING TECHNOLOGY PROJECT
A multi-disciplinary, collaborative project of
the California Institute of Technology – Pasadena, California 91125 and
the Massachusetts Institute of Technology – Cambridge, Massachusetts 02139
MEASURING THE IMPACT OF VOTING TECHNOLOGY ON
RESIDUAL VOTE RATES
DELIA GRIGG
CALTECH
Key words: voting technology, residual vote rates
VTP WORKING PAPER #37
August 2005
Measuring the Impact of Voting Technology on Residual Vote Rates
Delia Grigg
California Institute of Technology
Empirical Results
Summary
In the wake of the 2000 election, the importance of knowing the impact of voting equipment on
the number of uncounted ballots became evident. Using data from the 1988-2004 presidential
elections, this paper estimates the effects of voting technologies on residual vote rates using
several measurement techniques: a difference-in-differences estimator, fixed effects regression
models and a propensity score matching technique. The pattern of the results is robust to the
different methods. Paper ballots and lever machines produce the lowest rates of residual votes
followed by optically scanned ballots, direct recording electronic machines and punch cards.
Propensity Score Distributions
DREs
Optical Scan
Before Matching
6
5
5
5
Punch Cards
• 1988-2004 Presidential Elections
4
3
Density
2
2
2
2
Optical Scan
3
Density
Density
Density
3
Data
3
4
4
4
After Matching
6
After Matching
5
Before Matching
DREs
0
0
• Voting Technology
0.0
• Demographics
0.2
0.4
0.6
0.8
1.0
0.0
0.2
Propensity Score
• Data on approximately 12 of U.S. Counties
1
0
Usage of Voting Equipment in the 2004 Presidential Election, U. S. Counties
• Election Returns
0.4
0.6
0.8
1.0
0
1
1
1
Punch Cards
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
Propensity Score
Propensity Score
0.4
0.6
0.8
Propensity Score
Methods
• Residual Votes
* the fraction of total ballots cast for which no vote for president was counted.
Average Residual Vote by Machine Type and Year in U.S. Counties
1988-2004 Presidential Elections.
• Difference-in-differences
Exploit the natural experiment that occurs when counties change technology
Machine Type
Punch Card
Lever Machine
Paper
Optical Scan
Electronic
1988
3.5%
1.8%
2.7%
3.1%
3.6%
Counties
1992 1996 2000
2.7% 3.1% 2.6%
1.7% 2.2% 2.2%
1.9% 2.6% 2.2%
3.1% 2.4% 2.1%
3.8% 3.3% 2.4%
Estimated Effects
−80
−70
−60
−50
−40
−30
−20
−10
0
−80
−70
−60
−50
−40
−30
−20
−10
0
●
2004
2.0%
1.1%
2.2%
1.4%
1.6%
State, Year FE
Four time periods: 1988-1992, 1992-1996, 1996-2000, 2000-2004
●
Focus only on counties that switch from Punch Cards to Optical Scanners
State, Year FE
County, Year FE
DD = [Ê(Y1|OS) − Ê(Y0|OS)] − [Ê(Y1|P) − Ê(Y0|P)]
County, Year FE
County, Year FE w/ lag
Total
2.9% 2.4% 2.7% 2.3% 1.6%
• Fixed Effects
County, Year FE w/ lag
Matching
Percent of Population Using Technology
●
30
●
20
●
●
●
10
●
●
50
40
2. county and year fixed effects
●
• Propensity Score Matching
●
Years
2000
2004
−40
−30
−20
−10
0
Estimated Percentage Change in Residual Vote Rate
Switching from Punch Cards to Optical Scans
−80
−70
−60
−50
−40
−30
−20
−10
0
Estimated Percentage Change in Residual Vote Rate
Switching from Punch Cards to DREs
1988
Conclusions
●
Nearest-neighbor matching with replacement, using Matching package for R (Sekhon 2005).
• Matching estimates do not differ significantly from the linear models, suggesting that the linear
specification is appropriate.
Estimated average treatment effect (ATE): τ = E[YiOS − YiP]
• Difference-in-differences estimators theoretically appropriate, but not enough data to distinguish estimates from zero.
0
1996
−50
Propensity score estimated via logistic regression of Tit on Xit
●
1992
−60
●
●
●
●
1988
3. county and year fixed effects as well as lagged dependent variable
−70
●
30
40
●
−80
20
●
10
●
Matching
DD: 2000−2004
1. state and year fixed effects
Percent of Current Population
50
●
Punch Card
Lever Machine
Paper Ballots
Optical Scan
Electronic
Mixed
0
Percent of Counties
j
ln(F(Yit)) = αi + γt + Titλj + Xitβ + εit
60
60
Percent of Counties Using Technology
All specifications are variations on the following equation:
1992
1996
●
●
2000
2004
Years
Distribution of Voting Technologies across the U.S., 1988-2004
Multi-valued treatment – measured change from Punch Cards to each other technology individually
• Pattern of the results is consistent across estimators.
1.0