You are on page 1of 59

The impact of election fraud on government

performance

Abigail Peralta
Texas A&M University
Rowan University seminar
February 1, 2018
Motivation

• Poor government performance remains a significant problem in


many countries (Olken and Pande, 2012)

• Bureaucratic red tape and corruption is well-documented (Svensson,


2003; Freund, Hallward-Driemeier, Rijkers 2015)
• can inhibit economic activity through effect on firms (Mauro, 1995;
Fisman and Svensson, 2007; Asiedu and Freeman, 2009)

1
Corruption matters to firms

Cross-country World Bank Enterprise surveys, Freund, Hallward-Driemeier, Rijkers (2015)


2
Corruption matters to firms

Cross-country World Bank Enterprise surveys, Freund, Hallward-Driemeier, Rijkers (2015)

3
Why does corruption matter to firms?

Cross-country World Bank Enterprise surveys, Freund, Hallward-Driemeier, Rijkers (2015)

4
..because corruption is associated with longer waiting times

Cross-country World Bank Enterprise surveys, Freund, Hallward-Driemeier, Rijkers (2015)

5
Why we expect elections to matter

• Elections are a means for citizens to incentivize better performance


(Key, 1966; Barro, 1973; Ferejohn, 1986)

• Electoral accountability is associated with corruption and


performance (Adser, Boix, and Payne, 2003).
• Voters reward politicians for anti-poverty programs (Labonne, 2013).
• Exposing corrupt behavior of incumbents negatively affects their
electoral outcomes (Ferraz and Finan, 2008).
• Reelection incentives decrease corrupt behavior (Ferraz and Finan,
2011).

6
Why we expect elections to matter

• Elections are a means for citizens to incentivize better performance


(Key, 1966; Barro, 1973; Ferejohn, 1986)

• Electoral accountability is associated with corruption and


performance (Adser, Boix, and Payne, 2003).
• Voters reward politicians for anti-poverty programs (Labonne, 2013).
• Exposing corrupt behavior of incumbents negatively affects their
electoral outcomes (Ferraz and Finan, 2008).
• Reelection incentives decrease corrupt behavior (Ferraz and Finan,
2011).
• But electoral pressure can also have perverse effects (Park, 2016)

7
But elections can fail to put pressure on government

• Election fraud prevents citizens from holding ineffective or corrupt


politicians accountable.

• Understudied because:
• previous lack of election fraud measures
• settings where there is demonstrable variation in election fraud
• Ferraz and Finan (2008 and 2011), and Fujiwara (2015) studied
Brazil, where there is little evidence of election fraud −− > no
opportunity to test if election fraud affects government performance.

8
Research Contribution

This paper provides the first causal evidence that election fraud
affects government performance.

9
Overview

• Research Question: Does reducing election fraud improve


subsequent government performance?
• Setting: Mayoral Elections in the Philippines from 2001-2013
• Nationwide switch from manual to automated election system in
2010 reduced election fraud.
• Election fraud measure: 1) list from police, and 2) digit-based test
• Government performance measure: building permits
• Preview of findings: election reform –> reduced election fraud –>
improved government performance

10
Background on election reform

• Synchronized national and local elections are held every three years.
• 1634 towns, headed by mayors

• Manual election system (in use until the 2007 elections)


• ballots counted by hand

• Automated election system (in use beginning in the 2010 elections)


• ballots counted by voting machines

11
Manual ballot vs Automated ballot

Genuine automated election ballot has code in UV ink.

12
Precinct level scanners transmit to two types of servers

Source: Rappler.com

13
Suggestive evidence that government performance improved

Manual Election Period Automated Election Period Difference


Waiting time 132.4 106.7 -25.7***
Bribe request incidence 0.30 0.24 -0.06
Number of procedures 28.25 30.5 2.25**
Data: World Bank Enterprise Surveys (2009 and 2015) and Doing Business Surveys (2008 and 2011)

14
Suggestive evidence that government performance improved

Manual Election Period Automated Election Period Difference


Waiting time 132.4 106.7 -25.7***
Bribe request incidence 0.30 0.24 -0.06
Number of procedures 28.25 30.5 2.25**
Data: World Bank Enterprise Surveys (2009 and 2015) and Doing Business Surveys (2008 and 2011)

• But a pre-post comparison is potentially problematic.


• need to identify a suitable comparison

15
Difference-in-differences research design

• Treatment group: towns that historically experienced high levels of


election fraud.

• Comparison group: towns that historically experienced low levels of


election fraud.
• Main results exclude towns adjacent to high-fraud towns.
• Results are similar for full sample.

• Intuition: Previously high-fraud towns experience a greater reduction


than towns already low-fraud.

16
Identification

• Identifying assumption: Absent the switch to automated elections,


historically high-fraud towns would have experienced changes in
government performance similar to what other towns experienced.

• Tests of identifying assumption:


1. Examine whether high-fraud towns and low-fraud towns trended
similarly during the manual election period.
2. Allow towns in different regions of the country to follow different
trajectories.
3. Account for town-specific time trends.

17
Data for Measuring Election Fraud

• Starting point: list of “consistent election hotspots” from the


Philippine National Police

18
High-fraud areas in the Philippines

Source: Philippine National Police, using 2001, 2004, 2007 elections


19
Data for Measuring Election Fraud

• Starting point: list of “consistent election hotspots” from the


Philippine National Police → validate using forensic measure of
election fraud

• Election fraud measure: uniformity of the observed distribution of


last digits of vote totals (Dlugosz and Muller-Funk, 2009; Beber and
Scacco, 2012)
• advantage: limited data requirement
• vote totals for each candidate in local (mayoral) elections, for the
period 2001-2013
• people favor zero as the last digit (Boland and Hutchinson, 2000)

• This measure has been used and validated against a recount-based


measure of election fraud (Weidmann and Callen, 2012)
validation of fraud measure

20
Data for Measuring Government Performance

• Number of approved building permits in each city/town, 2006-2015


• useful proxy for local economic activity (Berman, Felter, Kapstein,
Troland, 2015)
• directly controlled by mayors

• Importance of building permit processes to economic growth (World


Bank, 2011):
• biggest regulatory impediment to doing business in the area
(Singapore Business Federation, 2009)
• Harder and more expensive in the Philippines relative to neighboring
countries (World Bank, 2011)

21
Variation in application processes across towns and over time

2008 2011
Number of Procedures Average waiting time Number of Procedures Average waiting time
Caloocan 29 135 31 109
Cebu 31 83 36 92
Manila 24 203 26 169

Data: World Bank Doing Business Surveys, 2008 and 2011

Mayors have control over this process.

22
Corruption is associated with even longer waiting times in the Philippines

Data: World Bank Enterprise Surveys, Philippines 2009 and 2015


Association persists even after controlling for firm and worker characteristics and degree of interaction with government
officials.

23
Corruption is associated with even longer waiting times in the Philippines

• Firms asked for bribes wait 50 percent longer.


• Less red tape or corruption, or a combination of both, can lead to
more building permits getting through the application process.
24
Generalized difference-in-differences

ln(BPit ) = β(hotspoti ∗ automatedelectionst ) + ci + ut + εit

where:

• BPit is the total building permits approved in town i in year t


• hotspoti is an indicator for whether a town was high-fraud
• automatedelectionst is a dummy variable for 2010 and later
• β – coefficient of interest, captures the effect of reducing election
fraud
• ci are town fixed effects
• ut are year fixed effects

25
RESULTS-MEASURING FRAUD

• examine distribution of last digits of vote totals from mayoral


elections

• no evidence of election fraud: uniform distribution of last digits

26
Hotspots experienced more election fraud

27
Hotspots experienced more election fraud

28
adjacent towns
Hotspots experienced more election fraud

29
adjacent towns
Differences disappear after switch

30
Differences disappear after switch

31
RESULTS - IMPACT ON GOVERNMENT PERFORMANCE

• Did the reduction in election fraud cause government performance


to improve?

32
Building permits approved, hotspots versus other towns

33
Estimated difference in building permits

34
Reducing election fraud increases building permits

Dependent variable: Log(building permits) (1) (2) (3) (4) (5)

Hotspot*automated election 0.166** 0.166** 0.149** 0.145** 0.167**


(0.0706) (0.0706) (0.0665) (0.0737) (0.082)

Hotspot*year before automated election -0.0163


(0.0657)

Observations 7070 7070 7070 7070 7070


Town and year fixed effects x x x x x
Control for population x x x x
Region-by-year fixed effects x x x
Town-specific time trends x

full sample

35
Reducing election fraud increases building permits

Dependent variable: Log(building permits) (1) (2) (3) (4) (5)

Hotspot*automated election 0.166** 0.166** 0.149** 0.145** 0.167**


(0.0706) (0.0706) (0.0665) (0.0737) (0.082)

Hotspot*year before automated election -0.0163


(0.0657)

Observations 7070 7070 7070 7070 7070


Town and year fixed effects x x x x x
Control for population x x x x
Region-by-year fixed effects x x x
Town-specific time trends x

full sample

36
Reducing election fraud increases building permits

Dependent variable: Log(building permits) (1) (2) (3) (4) (5)

Hotspot*automated election 0.166** 0.166** 0.149** 0.145** 0.167**


(0.0706) (0.0706) (0.0665) (0.0737) (0.082)

Hotspot*year before automated election -0.0163


(0.0657)

Observations 7070 7070 7070 7070 7070


Town and year fixed effects x x x x x
Control for population x x x x
Region-by-year fixed effects x x x
Town-specific time trends x

full sample

37
Reducing election fraud increases building permits

Dependent variable: Log(building permits) (1) (2) (3) (4) (5)

Hotspot*automated election 0.166** 0.166** 0.149** 0.145** 0.167**


(0.0706) (0.0706) (0.0665) (0.0737) (0.082)

Hotspot*year before automated election -0.0163


(0.0657)

Observations 7070 7070 7070 7070 7070


Town and year fixed effects x x x x x
Control for population x x x x
Region-by-year fixed effects x x x
Town-specific time trends x

full sample

38
Reducing election fraud increases building permits

Dependent variable: Log(building permits) (1) (2) (3) (4) (5)

Hotspot*automated election 0.166** 0.166** 0.149** 0.145** 0.167**


(0.0706) (0.0706) (0.0665) (0.0737) (0.082)

Hotspot*year before automated election -0.0163


(0.0657)

Observations 7070 7070 7070 7070 7070


Town and year fixed effects x x x x x
Control for population x x x x
Region-by-year fixed effects x x x
Town-specific time trends x

full sample

39
Suggestive evidence that standards were not loosened

Manual Election Period Automated Election Period Difference


Waiting time 132.4 106.7 -25.7***
Bribe request incidence 0.30 0.24 -0.06
Number of procedures 28.25 30.5 2.25**
Data: World Bank Enterprise Surveys (2009 and 2015) and Doing Business Surveys (2008 and 2011)

40
Potential mechanisms

• Did the reduction in election fraud change who got elected?


• Little evidence that incumbents became less likely to win.

• Did the reduction in election fraud affect incumbent vote shares?


• Incumbents win by narrower vote margins after the reform.

41
Reducing election fraud led to narrower incumbent victory margins

Dependent variable: Incumbent victory margin (1) (2) (3) (4)


Hotspot*automated election -0.102* -0.0967* -0.118** -0.125
(0.0513) (0.0553) (0.0541) (0.437)
Election year before automated elections -0.0589
(0.0824)
Observations 2882 2882 2882 2882
Town fixed effects x x x x
Region-by-year fixed effects x x x
Town-specific time trends x

42
Conclusion

• In the Philippines, switching to automated elections significantly


reduced election fraud.

• The reduction in election fraud results in better government


performance:
• 15 percent increase in number of building permits approved annually
• effect may work through electoral pressure

end

43
Research Agenda

• Factors that affect government performance


• The impact of election fraud on government performance
• Does electoral pressure lead to better government performance?

• Effects of government policies


• Moving to floodplains: How perverse incentives from the National
Flood Insurance Program affect population flows
(with Jonathan Scott)

• Development
• Does conflict exposure increase in-group bias? Evidence from
experiments in the Philippines
(First author, with Natalia Candelo, Sun-Ki Chai, Debbie Gundaya, Catherine
Eckel, Katerina Sherstyuk, and Rick Wilson)

44
Closely related research

• Does increasing electoral pressure make politicians exert more effort?

• Philippine gubernatorial elections


• Major responsibilities include evacuating for tropical cyclones.
• Requires only coordinating resources provided by the national
government
• Strategy: Compare outcomes in provinces whose governors are eligible
to run for re-election to provinces whose governors are ineligible, at
the start of their terms relative to the last year of their terms.
• Main result: Increased electoral pressure results in more thorough
evacuations.

45
Governors facing increased electoral pressure evacuate more...

46
...relative to governors not facing increased pressure.

47
Electoral pressure results in more thorough evacuations

(1) (2) (3) (4) (5) (6)


Eligible to run for 0.232** 0.225** 0.227** 0.204* 0.219* 0.201
re-election and approaching (0.109) (0.109) (0.112) (0.119) (0.112) (0.120)
end of term

Observations 472 472 472 472 472 472


Province and year FE x x x x x x
Control for HH population x x x
Control for distance of storms x x x
Governor FE x x

for every 100 people affected, an additional 20 are evacuated when provincial
governor is under increased electoral pressure

48
Conclusion

• Results show that governors respond to increased electoral pressure


by evacuating more people in preparation for storms.

49
Conclusion

• Results show that governors respond to increased electoral pressure


by evacuating more people in preparation for storms.
• More thorough evacuations imply increased effort because resources
are already provided.

49
Conclusion

• Results show that governors respond to increased electoral pressure


by evacuating more people in preparation for storms.
• More thorough evacuations imply increased effort because resources
are already provided.
• Takeaway: Electoral institutions matter.

49
THANK YOU
aamperalta@tamu.edu

50
Last-digit test for election fraud validated against audits

Weidmann and Callen (2012), using Afghanistan data

51
Back
Adjacent towns seem to be in between high and low-fraud

- potentially because there is some sharing of resources used to commit fraud


Back
52
Relationship between bribe requests and waiting time

(1) (2) (3) (4) (5)


Firm is asked for bribe 0.624*** 0.622*** 0.462** 0.452** 0.428**
(0.159) (0.160) (0.200) (0.195) (0.196)

Observations 173 173 119 117 105


Sector fixed effects x x x x
Controls for firm and managerial characteristics x x x
Control for worker productivity x x
Controls for firm visibility and interaction with government officials x

53
Relationship between bribe requests and waiting time

(1) (2) (3) (4) (5)


Firm is asked for bribe 0.624*** 0.622*** 0.462** 0.452** 0.428**
(0.159) (0.160) (0.200) (0.195) (0.196)

Observations 173 173 119 117 105


Sector fixed effects x x x x
Controls for firm and managerial characteristics x x x
Control for worker productivity x x
Controls for firm visibility and interaction with government officials x

54
Relationship between bribe requests and waiting time

(1) (2) (3) (4) (5)


Firm is asked for bribe 0.624*** 0.622*** 0.462** 0.452** 0.428**
(0.159) (0.160) (0.200) (0.195) (0.196)

Observations 173 173 119 117 105


Sector fixed effects x x x x
Controls for firm and managerial characteristics x x x
Control for worker productivity x x
Controls for firm visibility and interaction with government officials x

Less red tape or corruption, or a combination of both, can lead to more building
permits getting through the application process.

55
Results including all towns

Dependent variable: Log(building permits) (1) (2) (3) (4)

Hotspot*automated election 0.120* 0.119* 0.133* 0.113


(0.0644) (0.0642) (0.0720) (0.0780)

Hotspot*year before automated election 0.0563


(0.0665)

Observations 16340 16340 16340 16340


Town and year fixed effects x x x x
Region-by-year fixed effects x x x x
Control for population x x x
Town-specific time trends x

Back 56

You might also like