Mexico’s citizen polling stations: randomly selected neighbors run the vote and count it

Mexico’s citizen polling stations (mesas directivas de casilla) are, by law, “órganos electorales formados por ciudadanos” (electoral bodies made up of citizens) that receive the vote and count it at each polling station, staffed by residents selected by lottery from the voter list and trained by the National Electoral Institute (INE).⁠¹ In the 2024-2025 extraordinary election for federal judicial posts, INE drew 13,455,088 citizens, trained 1,826,063, and filled 775,403 of 775,429 required posts at 83,974 polling stations, with 318,874 substitutions.⁠²

In plain words

In Mexico, the people who run your polling place are not government employees. They are your neighbors, picked at random.

Months before an election, the electoral institute draws a month of the year by lottery, which sets the starting point for a random draw of about one in eight people on each neighborhood’s voter list. Trainers visit them at home, explain the job, and train those who agree. Then a letter of the alphabet is drawn, and people whose last names come next in line get the main jobs: president, secretary, and counters.

On election day these citizens open the polling place, check voters against the list, and at the end count every ballot in front of party representatives. They write the results on an official form, give copies to the parties, and post the totals outside the door for anyone to see. Then the official counts are added up at district offices, where boxes can be recounted.

Defined terms

• Casilla. A polling station; by law, one per 750 voters or fraction in each electoral section, with sections capped at 3,000 voters.⁠¹
• Mesa directiva de casilla. The citizen board that runs a casilla: a president, a secretary, two counters (escrutadores), and three general alternates, with more members when federal and local elections coincide.⁠¹
• Insaculación. Selection by lottery of citizens to serve on casilla boards.⁠¹
• CAE and SE. Capacitadores asistentes electorales (electoral trainer-assistants) and supervisores electorales, temporary INE staff who visit, train, and support the drawn citizens.⁠²
• Acta de escrutinio y cómputo. The tally sheet completed at the casilla after counting.⁠¹
• PREP. The Programa de Resultados Electorales Preliminares, which captures, digitizes, and publishes tally sheet data as preliminary, “estrictamente informativo” (strictly informational) results.⁠¹
• Conteo rápido. A quick count: a statistical estimate of the result from a random sample of casillas, overseen by an expert committee.⁠^3,4
• Cómputo distrital. The official district-level count, where tally sheets are compared and, under set conditions, ballots are recounted.⁠¹

What happened

Mexico’s current framework is the General Law on Electoral Institutions and Procedures (LGIPE), published May 23, 2014, with its latest reform published June 2, 2026.⁠¹ It states that casilla boards exist “por mandato constitucional” and are charged during election day with ensuring “la libre emisión y efectividad del sufragio” (the free casting and effectiveness of the vote), guaranteeing ballot secrecy, and securing the authenticity of the count.⁠¹

In 2023, Congress amended several of the provisions on casilla integration and recounts; the LGIPE text now notes that the Supreme Court of Justice of the Nation (SCJN) declared those amendments invalid in an action of unconstitutionality notified June 23, 2023, and that the earlier text “recupera su vigencia” (regains force).⁠¹ The citizen selection procedure described below is that restored text.

The model has been used for successive federal elections. In the June 2, 2024 presidential election, INE’s quick count committee reported at 22:50 that night, from 5,651 of 7,602 sampled casillas (74.3 percent) covering all 300 districts, estimated turnout between 60.0 and 61.5 percent and vote ranges for each candidate, at a confidence level of at least 95 percent.⁠⁴ In 2025, Mexico held an extraordinary election for federal judicial posts under the same casilla model, with district counts from June 1 to 12 and turnout of about 13 percent of the voter list.⁠²

How it worked

Mexico’s selection procedure, as set out in LGIPE article 254, runs in two lotteries and two training stages:⁠¹

1. Month lottery. In December before the election, the INE General Council draws a calendar month; that month and the next are the basis for selection, using the voter list as of December 15.
2. First insaculación. From February 1 to 7, district boards randomly select 13 percent of citizens in each electoral section, never fewer than 50.
3. First training. Selected citizens are invited to a training course from February 9 to March 31.
4. Evaluation. District boards evaluate those trained “en igualdad de oportunidades” (with equal opportunity), “prefiriendo a los de mayor escolaridad” (preferring those with more schooling).
5. Letter lottery. In February, the General Council draws one of the 26 letters; citizens are ordered by paternal surname starting from that letter.
6. Second insaculación and assignment. By April 6, district councils select board members from eligible, trained citizens; by April 8, boards are formed and roles assigned by schooling; lists are published by April 10.

To serve, a person must be a Mexican citizen resident in the section, registered, holding a voter credential, in full exercise of political rights, and have taken the training.⁠¹ Parties and independent candidates may accredit representatives at each casilla and receive legible copies of the tally sheets.⁠¹ Citizens may also be accredited as electoral observers if they declare impartiality and no ties to any party.⁠¹ Each casilla must have voting screens that “garanticen plenamente el secreto del voto” with the legend “El voto es libre y secreto”.⁠¹

After counting, casilla presidents post the results outside the polling station, signed by the president and any representatives who wish.⁠¹ They deliver the sealed package to the district council immediately in the district seat, within 12 hours from urban casillas elsewhere, and within 24 hours from rural ones.⁠¹

At the district count, the council compares each package’s tally sheet with the copy held by the council president; if they do not match, if there are evident alterations, or if the sheet is missing, the casilla is recounted in the open.⁠¹ A recount is also required when there are evident errors, when null votes exceed the margin between first and second place, or when all votes went to one party.⁠¹ If the district margin between first and second is one percentage point or less and the runner-up’s party requests it, all casillas in the district are recounted.⁠¹

Two fast information channels run in parallel. The PREP publishes digitized tally sheet data as preliminary results.⁠¹ The quick count estimates the outcome from a random sample whose methods are “definidos con anticipación y son aprobados por el Consejo General”;⁠³ in 2024 an 11-member technical committee of statisticians signed the presidential estimate.⁠⁴

Evidence: measured findings

INE’s numeralia for the 2024-2025 judicial election gives the full citizen staffing funnel:⁠²

Several ratios follow from these official figures (this site’s arithmetic): about 54 percent of drawn citizens were notified, about 13.6 percent were trained, and roughly one designated board member was replaced for every 2.4 appointed.

The 2024 presidential quick count reported, from 5,651 of 7,602 sample casillas, 95 percent confidence ranges of 58.3 to 60.7 percent for Claudia Sheinbaum Pardo, 26.6 to 28.6 percent for Bertha Xóchitl Gálvez Ruiz, and 9.9 to 10.8 percent for Jorge Álvarez Máynez, with turnout between 60.0 and 61.5 percent.⁠⁴ These are statistical estimates released on election night, not official results; official results come from district counts.⁠¹

For the 2024-2025 judicial election, the numeralia reports between 84,266 and 83,974 tally sheets computed per contest and turnout of 12.93 to 13.02 percent of the voter list across the six judicial contests.⁠²

Failures and criticisms

Mexico’s model has costs and critics, and the record shows tensions.

Scale and attrition. Reaching 13.5 million people to seat about 775,000 requires tens of thousands of temporary staff.⁠² The 41 percent substitution rate in 2025.⁠² means many final board members were not the first assignees, which can weaken the randomness that is part of the model’s appeal and shift reliance to alternates and to whoever is available on the day. This review found no official study of how substitutions affect accuracy.

Schooling preference. The law prefers citizens with more schooling and assigns roles by schooling⁠¹ Supporters see this as practical, since the president and secretary complete legal forms; critics could argue it makes boards less representative of their sections. Both views are interpretations; the law’s text is the only fact here.

Political contestation. The 2023 amendments to casilla and recount provisions, later invalidated by the SCJN,⁠¹ show that the rules governing citizen boards are themselves politically contested. The LGIPE’s latest reform, published June 2, 2026,⁠¹ continues that pattern; this page does not assess it.

Low-turnout elections. The 2024-2025 judicial election used the full casilla apparatus for turnout of about 13 percent.⁠² Whether the cost of a full citizen board structure is justified for such elections is a policy debate this page does not settle.

What citizen boards cannot fix. Citizen counting protects the count at the station. It does not by itself address vote buying, pressure on voters outside the polling place, or unequal campaign conditions. See intimidation.

Competing interpretations

Distributed custody as an integrity safeguard. Supporters argue that when hundreds of thousands of randomly chosen, trained neighbors count ballots in public, in front of party representatives, and post the results on the door, manipulating a national result would require corrupting an implausible number of independent people, and any discrepancy can be checked against posted sheets and party copies.⁠¹ Layered checks (posted results, party copies, PREP, quick count, district recounts) make the system hard to rig quietly.

Expensive, and dependent on an institution. Others argue the model’s integrity rests heavily on the electoral institute that draws, trains, and supports citizens, so it is only as independent as that institution. The attrition figures.⁠² suggest the lottery is the start of a long selection process shaped by availability, schooling, and staff effort.

A civic education program in disguise. A third view emphasizes that training more than a million citizens per election⁠² is itself civic education: people learn how votes are handled by doing it. This is a plausible interpretation; this review did not find an official evaluation measuring that effect.

Lessons

These are interpretations drawn from the record.

1. Randomness distributes trust. Selecting poll workers by public lottery makes capture by any single group difficult⁠¹
2. Publish at the lowest level. Posting results outside each station lets anyone check sums from the bottom up.⁠¹
3. Separate preliminary from official. The PREP is “estrictamente informativo,” the quick count is a statistical estimate, and district counts are official.⁠^1,4
4. Write recount triggers in advance. Mexico’s rules (mismatched sheets, null votes exceeding margins, one-point district margins) are set before the vote.⁠¹
5. Measure the funnel. Publishing drawn, trained, designated, and substituted counts lets the public see where the model strains.⁠²

Application to this framework

Mexico’s citizen polling stations connect to the My Vote Forward framework’s interest in sortition and distributed responsibility. The framework’s governance principles ask who controls each step and how that control is checked; Mexico answers for the count by handing it to randomly selected citizens observed by competitors.

The case also informs the framework’s published releases approach. Any release of public priorities could follow Mexico’s pattern of preliminary versus official numbers, statistical estimates with stated confidence and sample coverage, and fixed rules for corrections, all declared before results exist. And if the framework ever involved citizens in reviewing questions or methods, a lottery with published selection and attrition statistics, like Mexico’s, is one tested pattern (see question design).

Unresolved questions

• How do substitution rates vary by region and over time, and do they affect counting errors or recount rates?
• How often do district councils recount casillas, and for which triggers? This review did not find a consolidated national statistic.
• Does serving on a casilla board change participants’ trust in elections or later participation?
• How did the June 2026 reform change the casilla and recount rules, if at all?
• Could the selection funnel be designed to keep more of the randomness while managing attrition?