Votes, Violence, and Networks:

1
Votes, Violence, and Networks Evidence from a
Field Experiment in Nigeria Pedro C.
Vicente Trinity College Dublin Research in
collaboration with Paul Collier and Marcel
Fafchamps CSAE University of Oxford TIDI
Launch February 27th 2009
2
Motivation
This election is a do or die affair. Feb. 10th
2007, President Obasanjo Elections could
further destabilise the violent, oil-rich Delta
region. Mar. 15th , 2007, The Economist,
headline Election breeds fear in Nigeria's
Muslim north. Apr. 12th , 2007, The Financial
Times, headline After rigged and violent local
polls, the opposition threatens a boycott of the
presidential race. Apr. 18th , 2007, The
Economist, headline Violence and fraud tarnish
Nigerian elections. Apr. 22nd , 2007, The
Financial Times, headline Rigging, violence and
intimidation were so pervasive and on such naked
display that they made a mockery of the electoral
process. Where voting did take place, many
voters stayed away from the polls. By the
time voting ended, the body count had surpassed
300. May 15th , 2007, Human Rights Watch
3
Main Research Questions

• Which are the key determinants of electoral
  violence (vs. ballot fraud or vote buying)?
• Can a NGO-conducted campaign against electoral
  violence help in undermining this phenomenon?
• Is violence diminishing voter turnout?
• Who are the candidates that are perceived as
  violence-prone?
• Are there network effects of the campaign
  against electoral violence? (reinforcement or
  externality)

4
Preview of Main Results

• Violence emerging in less competitive locations
  vote buying and fraud emerging in swing locations
• Clear effects of the anti-violence campaign on
  diminishing perceptions/experience of violence,
  increasing empowerment
• Violence causing a substantial decrease in voter
  turnout
• Incumbents benefiting from anti-violence message
  (terrorism story)
• Evidence of both reinforcement and externality
  network effects particularly for violence
  outcomes

5
Outline

• Model
• Background
• Experimental Design
• Fieldwork
• Anti-Violence Campaign
• Surveys
• Descriptive Statistics
• Econometric Results
• Concluding Remarks

6
2. Model

• Environment with swing and base voters, with an
  incumbent and a challenger
• swing voters can be intimidated not to vote
• at a cost losing support from ones base voters
• Such a world may have violence by each of the
  candidates and peaceful politics as possible
  equilibrium outcomes
• In Nigeria, the Incumbent is very powerful PDP
• has the money to bribe, the jobs to offer (vote
  buying)
• controls the electoral commission (fraud)

7

• Nigerian models implications
• when the Incumbent has a base advantage
• either intimidation by the Challenger or nothing
  happens (depending on the size of the base for
  the Challenger)
• when the Challenger has a base advantage,
• either vote buying or fraud may arise from the
  Incumbents side

8
3. Background

• Most populous African country with almost 132
• million inhabitants (2006 est.)
• Economy
• Ranks 201/233 countries in GDP pc (1400 USD PPP
  in 2005)
• Major oil-producing country (95 exports, 20
  GDP, 10th largest oil reserves in the world)
• Political History
• Independence in 1960 post-independence period
  dominated by military rule and instability (with
  numerous coups and internal conflicts)
• From 1999, civilian rule under democratic
  elections
• Elections in 1999, 2003, and 2007 marked by
  widespread violence and fraud vote buying also
  prominent

9

• The Elections of April 2007
• Four elections (president, federal assembly and
  senate, governor, state assembly)
• Presidential Election
• Obasanjo not running for a third term but
  appointing a protégé (YarAdua) through the
  rulling party (PDP)
• Opposition by Buhari (ANPP) second in 2003,
  and Abubakar (AC), former vice-president, in
  conflict with Obasanjo, and facing various
  accusations of corruption
• Focus on Presidential/Gubernatorial

10
4. Experimental Design

• Treatment provided by campaign against political
  violence by NGO (ActionAid International Nigeria)
  in randomized survey locations
• Targeted at a randomized set of households
• We compare
• Violence during 2006 vs. 2½ months before April
  2007 Round of Elections
• Voting intentions and actual reported behavior
  for the April 2007 Round of Elections
• Treated-Treatment, Control-Treatment, and
  Control-Control groups
• Measurement
• Before-after April 2007 Round of Elections panel
  surveys
• (Only for actual violence events) Independent
  local journalists

11
(No Transcript)
12

• Specifications
• 1) Effect of the Treatment on Violence
• where
• VC is violence and crime,
• i, l, t are subscripts for individuals,
  locations, and time (before / after)
• T is a binary variable with value 1 for treated
  locations,
• X is a vector of controls (demographic,
  attitudinal), potentially time-varying
• Y is a geographical fixed effect.
• 2) Effect of the Treatment on Voting
  (reduced-form)
• where
• V is strict voting behavior (intended-before and
  actual reported-after)

13

• 3) Effect of Violence on Voting
• where
• is instrumented by the treatment
• 4) Reinforcement network effects
• where
• O is an outcome variable, N is a network
  measure, and T is treated-treatment (1) and
  control-control (0)
• 5) Externality network effects
• where
• T is control-treatment (1) and control-control
  (0)
• All standard errors clustered at the enumeration
  area level

14

• Violence/Crime Outcome Variables (specification
  1)
• Perceptions of politically-motivated violent
  behavior
• in general, respondents area, from the top and
  from the bottom
• Actual political violence events Local Journals
• compiled by independent local journalists at all
  surveyed locations
• including descriptions of political violence
  events
• (sources town meetings, police)
• before and after AAINs campaign (2nd semester
  2006 to 2 weeks after last April elections)

15

• Empowerment against political violence
• Perceptions
• Postcard variable
• Sympathy for unlawfulness
• Crime
• perceptions
• experience

16

• Network Variables (specifications 4-5)
• Social Interaction
• How frequently do you calmly chat about the day
  events with the following individuals or members
  of their households?
• Family
• Are the following individuals relatives of
  yours (i.e. members of your family)?
• Geographical Proximity
• Enumeration area maps (both using coordinates
  and reported distance between interviews by the
  enumerators)

17
5. Fieldwork

• Survey interviews (face-to-face) conducted in
  coordination with Afrobarometers (AB) Nigerian
  partner (PSI)
• Campaign conducted by ActionAid International
  Nigeria (AAIN)
• Whole project coordinated and supervised-in-field
  by the authors

18

• Panel Surveys/Journals conducted in Pre-election
  survey submitted to
• 24 enumeration areas (each with average approx.
  500 households) chosen within ABs representative
  sample of 275 EAs (in all 37 states of Nigeria)
• Campaign (Feb. 13 - 27) conducted in
• 12 out of these 24 surveyed locations
• Pre-election survey (Jan. 20 - Feb. 3) responded
  by
• 1200 subjects/households (50 per EA)
• run as an oversample of the AB 2006 survey, with
  minor questionnaire adaptation
• Post-election survey (May 22 - Jun. 5) responded
  by
• 1149 subjects/households (panel, 96
  re-surveyed)
• 300 oversample respondents in treatment
  locations (25 per EA)

19

• Anti-Violence Campaign
• Main message, crucially uniform countrywide
• Main NO TO ELECTION VIOLENCE
• Vote Against Violent Politicians
• Means
• Distribution of materials targeted primarily at
  panel-surveyed households
• T-shirts (3,000, i.e. 1 for every 2 households
  on average), Caps (3,000), Hijabs (1,000)
• Posters (3,000), Stickers (3,000), Leaflets
  (5,000)
• Road shows (using jingles in 3 languages,
  Yoruba, Hausa, Pidgin English)
• At least one Community Meeting at each of the 12
  locations
• At least one Popular Theatre representation at
  each of the 12 treated locations (same script,
  available upon request, used in all)

20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
Town meetings and popular theatres
24
Distribution of materials and roadshows
25

• More on
• http//www.iig.ox.ac.uk/research/08-political-viol
  ence-nigeria/default.htm/

26

• Sampling Process
• 1. ABs sampled EAs chosen randomly weighting by
  the number of households (representative)
• 2. Non-random component 2 states in each of the
  main 3 regions of the country (Southwest,
  Southeast, North) chosen by looking at recent
  history of violence
• In order to put the emphasis on studying
  violence, while keeping diversity
• 3. 2 treated EAs in each state chosen randomly,
  within ABs EAs
• 4. Control EAs chosen by choosing the closest EAs
  from ABs list, in the same state, of the same
  type (Large Urban, Small Urban, Rural)
• 5. Households and corresponding respondents
  within a census area chosen randomly using ABs
  standard techniques
• 6. Conditions for sampling within a household
  18 years, residence in EA
• 7. (only for oversample) not approached by
  campaign

27
6. Descriptive Statistics
28
(No Transcript)
29
(No Transcript)
30
7. Econometric Results

• A Model of Electoral Violence, Ballot Fraud, and
  Vote Buying
• Swing locations more prone to fraud and vote
  buying less prone to violence
• Locations where PDP is the incumbent see less
  violence from PDP/Opp. (terrorism) and more vote
  buying from PDP/Opp (money matters) see more
  ballot fraud (incumbents business)

31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34

• Randomization
• Treated and control locations/respondents seem
  not to be statistically different

35

• Effects of the Campaign on Violence Outcomes
• Clear effects on perceptions (general and
  location-specific)

36

• Effects of the Campaign on Violence Outcomes
  (conted)
• Clear effects on perceptions behavior-changing
  (postcard)

37

• Effects of the Campaign on Violence Outcomes
  (conted)
• Crime decreases too (not as clearly)
• Mainly on perceptions

38

• Effects of the Campaign on Violence Outcomes
  (conted) (JOURNALING)
• Campaign leads to a decrease in the intensity of
  violence

39

• Effects of the Campaign and Violence on Voting
  Behavior
• Increased turnout (9 for the treatment, 10 for
  each unit of security)

40

• Effects of the Campaign on Voting Behavior
  (Conted)
• Harming Abubakar (AC) (presidential)
• Benefiting PDP/Incumbents (gubernatorial)
• Main story benefiting Incumbent harming others

41

• Effects of the Campaign on Voting Behavior
  (Conted)
• Benefiting PDP (gubernatorial) mainly Oyo,
  Rivers, Plateau
• Moves from and to abstention

42

• Network effects
• Mainly through family connections
• On Perceptions of violence/empowerment not on
  voting behavior

43
8. Concluding Remarks

• Political violence emerges in less
  electorally-contested areas
• This is in marked contrast with ballot fraud and
  vote buying
• A NGO-conducted campaign was highly effective in
  changing perceptions, self-reported experience
  with violence, and margins of related behavior
• Violence seems to dramatically decrease voter
  turnout
• Violence seems to be coming from non-incumbent
  groups
• Consistently with a terrorism notion
• Networks seem to be playing a relevant role in
  reinforcing and spreading the direct message of
  the campaign