HomeProductProofBlogAboutContactFAQBook a demo
Public test

We published the call before the vote

A referendum is the rare research question with a public, dated, unarguable answer. So we ran one, in the open, before the result existed.

What we predicted

The initiative would not pass

Our synthetic population of Swiss voters was run on the ballot question ahead of the federal vote. The simulation returned a rejection, with no majority at either level required for the initiative to carry.

What happened

The initiative was rejected

The outcome matched the simulation. Published polling in the run-up had pointed the other way, which is precisely what made the test worth running in public rather than in a slide.

Synthetic electorate, run before the vote

Agent-level positions, aggregated to a national call

The two thresholds an initiative must clear
Threshold 1

Majority of voters

Not reached

Threshold 2

Majority of cantons

Not reached

Result

Rejected

Matching the direction our population returned before the vote, and not the direction of the published polling.

Why this test and not a claimed accuracy figure

Vendors in this category quote accuracy percentages. We do not, because the number moves with category, question wording, and study design, and because a figure nobody can check is not evidence. A referendum cannot be reinterpreted after the fact. It is one question, one date, one answer, decided by the whole population rather than a sample.

Before the vote

Population built

Agents weighted to the federal electorate: age, region, language, household composition, and turnout propensity, drawn from public statistical sources.

Before the vote

Ballot question asked

The official wording put to the synthetic population, with the media context of the campaign period layered in.

Before the vote

Call published

The predicted outcome recorded and timestamped ahead of the result, alongside the published polling that disagreed with it.

Vote day

Outcome confirmed

Rejected, with no majority at either level. The simulation matched the direction of the national result.

What this does and does not prove

It proves a calibrated synthetic population can track a real electorate on a question with a hard answer, without any client data and without a panel. It does not prove every category behaves like a referendum. Consumer choice is noisier, and we say so. The honest test for your organization is a parallel run on one of your own past studies, where you already know the answer.