Allegations of Fraud in Mexico s 2006 Presidential Election

Allegations of Fraud in Mexico s 2006 Presidential Election Alejandro Poiré and Luis Estrada Presentation prepared for the 102nd APSA meeting Philadelphia, Penn. September 1, 2006 alejandro_poire@harvard.edu

Two types of claims by AMLO s coalition Unfair election Campaign spending, media access, incumbents intervention, etc. Rigged election Many abstract claims by the coalition Many specific, and often evolving ones We test each of the specific ones Key point: IFE count gives undue advantage to Calderón

The coalition s seven signs of electoral fraud 1. Unionized teachers were forced to vote for Calderón 2. AMLO votes were illegally annulled 3. PAN poll-station representatives conspired to rig the vote count 4. Non-random poll-station workers rigged the vote count 5. 55% of vote tallies had mistakes (72,197) 6. Too few recounts were authorized by IFE 7. Early results (PREP) were manipulated

A research strategy Build a model that adequately predicts vote outcomes and allegations of fraud Test for the presence of systematic effects of indicators of fraud on electoral outcome Y = β 0 + β 1 normal vote + β 2 vote rigging Ho: β 2 = 0

Model specification I Dependent variable: Calderón AMLO votes at the precinct level 130,777 cases mean = 1.86 median = 6 p25 = -73 p75 = 75 On average, 545.7 voters were registered per precinct, 319.56 voted Density 0.001.002.003.004-1000 -500 0 500 1000 Votes for Calderón minus votes for AMLO

Normal vote indicators Previous vote in the section on average, 2 poll stations per section PAN, PRD + PT + Convergencia, PRI+PVEM Regional effects North, Mexico City Incumbent effects (PAN, PRD) Demographics marginality of municipio, urban precinct Turnout effects

Fraud effects I 1. Teacher s union effect Congressional votes for Nueva Alianza in 06, 4.55% total, compared to only 0.96% for president 2. Null votes Indicator of null votes for president, controlling for null votes in the section in 2003 3. Presence of party representatives in poll stations PAN covered 79.9%, AMLO 79.5%

Fraud effects II 4. Non-random substitution of poll-workers on and before election day Substitutions from the queue on July 2 nd, and of previous substitutions, by polling station position: president, secretary, and vote counters. 5. Four types of math mistakes in PREP vote count: 1. Total voters arriving at booth total votes cast (6,739) 2. Total voters + leftover ballots total received ballots (56,507) 3. Total cast + leftover ballots < supposedly received ballots (8,740) 4. Total voters + leftover ballots supposedly received ballots (211) 5. Ballot stuffing: More votes cast than ballots actually received (Estrada & Poiré, n = 272)

Fraud effects III 6. Not enough recounts took place 1. Effect of opened precinct 7. PREP was manipulated 89.9% of precincts were published in the PREP, and 20% of these had some minor inconsistency 9.8% were not published for serious inconsistencies (total 3.2 million votes) PREP aggregate count was not random, totals for AMLO and Calderón should have crossed

Strong normal vote effects: Calderón did slightly better in PAN precincts than AMLO in coalition ones 73 more votes for Calderón for an increase from low to high previous PAN vote (p20 to p80) 60 more votes for AMLO for a similar increase in Coalition vote Calderón, Norte was worth + 52 votes per precinct For AMLO, Mexico City meant 68 more votes, but He lost 30 votes in all PRD incumbent states, while Calderón won 31 extra votes per precinct in PAN states Calderón benefited from high turnout His lead increases 13 votes from low to high turnout precincts (242 to 402 votes cast, p20 to p80) Previous PRI vote benefited AMLO, 20 votes (p20 to p80) AMLO did better in high marginality areas (7 votes) No identifiable urban effect

Effects of vote rigging, from low to high levels: 1. Votes for Nueva Alianza for Congress 2. Null votes for President 3. No AMLO rep., Yes Calderón rep. 3. No Calderón rep., Yes AMLO rep. 4. Pre-July 2 nd resampling of poll worker 4. Poll worker chosen from queue 5. Math mistakes 6. Ballot stuffing 7. IFE recount 1. Favors AMLO, 8 votes on average 2. Favors AMLO, 1.3 votes* 3. Favors Calderón, plus 16 votes 3. Favors AMLO, plus 21 votes 4. No significant effect on Calderón AMLO vote 4. Favors AMLO, plus 4.1 votes 5. Favor AMLO, 4.0 votes 6. No significant effect 7. No significant effect * Indicates result only robust at the 90% level.

And about the PREP Two tests were performed: 1. Did the difference between Calderón and AMLO in each precinct influence how long it took it to publish them? No, publication time responds as expected to structural determinants. 2. Did published precincts auto-correlate inversely with each other? It s actually the opposite. PREP precincts publication correlates slightly positively with previous precincts (good AMLO outcomes are followed by good AMLO outcomes, and viceversa)