Methodology

Pamelection is a civic prediction and citizen-voice platform for Haiti's 2026–2027 electoral cycle. For each race, participants can provide two independent signals: a citizen vote — “if the election were today, who would you vote for?” — and predictions — “how likely is each candidate to win or win a seat?”. These signals describe what participating people prefer and expect. They do not determine election results.

Every race collects two independent signals:

• Predictions (0–100%) One slider per candidate, answering “how likely do you think this candidate is to win?”. You can make predictions for as many candidates as you like; the sliders do not need to add up to 100%. In multi-seat races, such as Senate races where three candidates are elected, the question is instead “how likely is this candidate to win a seat?”. This is our probabilistic signal.
• Citizen vote (one pick) One choice per race, answering “if the election were today, who would you vote for?”. This is our preference signal — a simple measure of which candidate participating users would choose today.

They appear side by side because they measure different things. Someone might think a candidate has a 60% chance of winning while personally choosing someone else. Both signals can provide useful information about how people view a race.

A candidate's implied probability starts with the mean of all current predictions for that candidate, then is normalized so that the active candidates in a race sum to 100%. Citizen-vote shares are simple proportions of current picks. All aggregates use unweighted means at launch. If we ever introduce weighting, such as accuracy-based weighting, we will announce it here and in the changelog before it affects any published number.

The CEP has set two main dates for this electoral cycle:

• December 13, 2026 — Phase 1 First-round presidential, Senate, and deputy elections. Senate races elect 3 senators per département. Pamelection covers all 119 deputy circonscriptions; boundary details remain provisional until the relevant official electoral documentation is published. Any confirmed adjustment will be recorded in the changelog.
• February 21, 2027 — Phase 2 Runoff races and local elections, including municipal cartels, town delegates, CASEC and ASEC.

Haitians abroad cannot currently cast an official Haitian ballot from abroad. On Pamelection, diaspora participation is civic expression and forecasting, never actual voting. Participation on the platform is therefore not the same as eligibility to vote in the official election. Citizen-vote statistics are reported with their population definition clearly identified, including whether they represent participants in Haiti or the diaspora. Prediction aggregates can include participants worldwide and can also be separated by location.

Pamelection may allow participation by people who are not eligible to vote in the official election. This distinction is intentional. Platform participation is not an electoral ballot. When we publish a citizen-vote statistic intended to describe the population of people eligible to vote in Haiti, we use the relevant 18+ Haiti-based participant segment and clearly label it. Other platform participants may still contribute to predictions and other civic signals where permitted by the platform's participation rules.

Every race has a freeze time — by default, midnight before election day. After the freeze, predictions and citizen votes for that race are locked: nothing entered or changed after that moment is counted in the published pre-election aggregates. The results remain available as a record of what participants said before voting began. Freeze times are shown on each race page.

A first-round race is evaluated against the question “does this candidate win outright or advance to the runoff?” If a runoff is required, it opens as a separate Phase 2 race with its own predictions, freeze, and scoring. Predictions from the first round are not carried into the runoff, and the first-round score does not wait for the second round.

We do not create races for institutions or positions that citizens do not elect directly. These include the municipal assembly, the departmental council and assembly, the interdepartmental council, and appointed positions such as departmental delegates, judges, the prime minister, and ministers.

• Not an official voting channel. No response on Pamelection is a real ballot. Participation is civic expression, not official voting. Pamelection does not replace the electoral process or the procedures established by the CEP.
• Not a voter registration portal. Voter registration is handled by the CEP through its official channels. Pamelection may explain the process and provide tools to help people understand where and how to vote, but registration itself takes place with the relevant authorities.
• Not a scientific poll. Participation is voluntary and self-selected. Our data should therefore be understood as a snapshot of participating people, not as a statistically representative estimate of the entire Haitian population.
• Not a betting or wagering platform. No money, tokens, or financial stakes are involved. Prediction points are reputational only and are based on forecasting accuracy.
• Not a discussion forum. There are no comments, direct messages, or open-ended user posts. This is a deliberate design choice intended to reduce harassment, manipulation, and the spread of disinformation.

We state these limitations clearly because they are inherent to a platform based on voluntary participation. Our numbers describe the participating Pamelection community, not the Haitian electorate as a whole. They should be understood as a live measure of preferences and expectations among participants, not as a scientific forecast of the population. Participation may also differ by age, location, political interest, internet access, and other factors. Motivated actors may attempt to manipulate participation or results; our anti-abuse protections and audit practices are documented separately. Small segments can also be statistically noisy, which is one reason public segments are hidden below 25 participants.

We never display individual answers — to anyone. Public data is aggregated using predefined segments such as national, département, diaspora country, and age band. Any segment with fewer than 25 participants is hidden as “segment hidden”, helping reduce the risk of identifying individuals. Only predefined segments are exposed. We do not publish arbitrary combinations such as “women 18–24 in Nippes”. This helps reduce differencing attacks, where overlapping groups can be compared to infer individual responses. The same privacy principles will apply to any future public API. If you delete your account, your answers are removed from future aggregates; historical aggregates that were already published remain immutable, subject to our Privacy Policy.

When a race is resolved and the CEP's definitive results are published, we calculate a candidate-level Brier score for each participant:

1. For each day from your first prediction in the race until the freeze, we use your latest prediction for each candidate. That prediction carries forward until you change it.
2. Each prediction is converted from a percentage to a probability between 0 and 1 and compared with the official outcome. For a single-winner race, the elected candidate has an outcome of 1 and every other candidate has an outcome of 0. In multi-seat races, such as Senate races electing 3 candidates, every elected candidate has an outcome of 1 and every non-elected candidate has an outcome of 0. The prediction question in those races is “how likely is this candidate to win a seat?”.
3. Candidates you never predicted are treated as 0 (“will not win”). This prevents participants from improving their score simply by predicting only the candidates they consider most likely to win.
4. The mean squared error across all candidates gives that day's Brier score. The mean across your active days gives your final score for the race. Points = (1 − Brier) × 1000. Lower Brier scores indicate more accurate predictions; higher points indicate better performance.

Days before your first prediction are excluded from your average. To appear on the accuracy leaderboard for a race, however, your first prediction must be at least 7 days before the freeze. This requirement gives participants enough time to demonstrate sustained forecasting rather than joining immediately before the deadline. These scoring rules are fixed before a race is resolved and will not be changed retroactively.

Election results can sometimes be contested. When that happens, we mark the race as disputed and pause score computation until the CEP publishes its final ruling. We never declare winners ourselves. If official results change following a ruling, the affected race and scoring record are updated according to the published rules, and the change is recorded in an audit log.

Candidate colors in charts are assigned independently of political-party colors. A fixed system is used so that a candidate's visual identity remains consistent across the platform and does not imply political affiliation or preference.

The platform interface is available in French and English, while Kreyòl is an important part of our accessibility strategy. We are producing Kreyòl audio versions of key guides that can be shared through WhatsApp and made freely available to radio stations. Spoken Kreyòl can reach people beyond the limits of a written interface. If you would like to contribute your voice or review scripts, contact us through the About page.

• Candidate lists Candidate information is based on official candidate lists published by the Conseil électoral provisoire (CEP). When official publications are ambiguous or incomplete, candidates are not published until the information has been verified through our process.
• Official results Official results are transcribed from publications by the CEP. Each result entry is recorded in an audit log. If a race is disputed, scoring remains paused until the CEP issues its final ruling.
• Constitutional and legal references Descriptions of elected offices, responsibilities, eligibility, and terms of office are based on the applicable Constitution, electoral legislation, electoral decrees, and official CEP publications. Official legal sources take precedence over Pamelection's interpretation.
• Corrections When we identify an error in our data or interpretation, we document the correction publicly rather than silently changing the information.

• 2026-08-21 Methodology revised for clarity and consistency. Clarified the distinction between platform participation and electoral eligibility; clarified multi-seat prediction semantics; expanded the candidate-level Brier scoring explanation; clarified leaderboard eligibility; strengthened limitations and source-verification language; simplified implementation-specific sections.
• 2026-08-20 Deputy coverage updated: all 119 circonscriptions covered, with boundaries provisional until the relevant official documentation is published. Scoring trigger clarified as definitive CEP results. Clarified that unverified candidates are never published. Kreyòl policy clarified as audio-first. Editorial consistency fixes.
• 2026-08-08 Initial public version. Added freeze rules, runoff semantics, aggregate formula, Brier scoring specification, diaspora participation, anonymity rules, and limitations.