OUR STANDARD / SOURCES AND METHODS
Show your
working.
The rules we use before a number makes its way into a chart, share card or headline. What we know, what we don’t, and how we check.
01. Polls are not election forecasts.
Polls measure voter intention at the time people were interviewed. ElectionsX gives the fieldwork window, publication date, the population being described, and the pollster’s stated sampling margin. Differences in weighting, response options and filters can undermine naive comparisons.
Ipsos Wave 2 covered 12 September to 13 October 2026; 3,600 adults were interviewed, of whom 2,533 were in the registered-voter subsample. Its metro maximum sampling margins are ±5.89 percentage points for Johannesburg, ±7.8 for Tshwane, ±6.42 for eThekwini and ±5.95 for Cape Town. A maximum margin for one estimate is not automatically the correct test for a difference between two parties, two surveys or a poll and a vote count.
02. We don’t publish invented results.
Before results arrive, the ward explorer and live count remain visibly empty. Our pipeline saves raw source responses with timestamps and fingerprints. Raw data is not automatically treated as a published election result. A normalized feed must pass validation and editorial checks before it is displayed.
03. Provisional means provisional.
Incoming IEC counts can be revised. Every display must say whether values are provisional or final, when the underlying response was obtained, when it was reviewed and which official source it came from. We distinguish “voting districts counted” from votes counted. Early-reporting areas are not representative of the whole city.
04. Ballot types are kept separate.
South African local elections involve ward and proportional-representation voting, with an additional district ballot in relevant municipalities. A candidate’s ward support cannot be swapped for the party’s PR vote. Where a comparison crosses election cycles or voting district boundaries, we document the geographic and ballot crosswalk.
05. A council majority is not a political deal.
For a council of N seats, the smallest outright majority is floor(N / 2) + 1. Our coalition tool calculates that threshold from user-entered seat counts. It does not infer which parties are willing to cooperate, or how coalition agreements will work in practice.
06. Our future accuracy test is pre-committed.
Poll gaps will be calculated as actual valid-vote share minus the pre-election poll share in percentage points, at a carefully matched geographical level. We will discuss turnout, subsample uncertainty and apples-to-oranges comparisons. We will not claim that falling within a published margin of error validates a prediction.
07. No automatic election-night projection.
We may develop a model only after saving real result snapshots and back-testing it on held-out data. Until then ElectionsX will show counted-so-far figures without projecting the final outcome. The absence of a reliable forecast will be reported, not hidden.
08. Corrections are visible.
Datasets must retain source URLs, data versions and a correction log. If the source revises numbers, the change should be acknowledged, with the original preserved in the raw snapshot archive.