OPINION: How close were the CSIR result predictions?

As in previous years, SABC News once again teamed up with the Council for Scientific and Industrial Research (CSIR) to forecast election results during the 2026 Local Government Elections.

The CSIR, one of the Science Councils under the Department of Science, Technology and Innovation, has a longstanding relationship with the SABC dating back to the 2004 elections. Since then, the CSIR has continuously refined its algorithms to ensure greater accuracy in its predictions. Beyond forecasting outcomes, these models also help the public make sense of vast amounts of election data and better understand emerging voting patterns.

The CSIR prediction model uses the official early results released by the Electoral Commission (IEC) as a sample to predict the final results.

The CSIR prediction model uses the official early results released by the Independent Electoral Commission, as a sample to predict the final results. These predictions are updated continuously as more information becomes available.

CSIR figures for both national and local government elections are remarkably accurate, even when new parties emerge or major parties lose support. CSIR claims that its early predictions are within 1 to 3 percentage points of the final national support levels for most leading parties.

The model starts making predictions once as few as 5 per cent of votes have been verified. But because of the risk of unexpected support levels, particularly at a local level, these predictions aren’t made public until close to 15% of votes have been verified. The system works only with official IEC verified results, and is based on two principles: the analysis of historical voting behaviour patterns and an understanding of how voting district results are received and reported during the counting process. According to their website, “By combining these principles, the model groups voting districts with similar characteristics through statistical clustering techniques, allowing it to generate projections before all voting districts have reported their results.”

So how good has the CSIR model been this year?

As an example, in overall votes in North West, with just 14% of votes officially verified, the CSIR predicted the ANC would garner 44%, with the DA getting 14% and the EFF 11. With the count now at 69%, more than two thirds done, the ANC has 44% and the DA 14, exactly as predicted. The EFF is slightly higher for now, on 12.5%.

There’s been a lot of interest in the eight metros, with only one, the City of Cape Town, not heading for a coalition. One of the biggest changes is in Mangaung, which the ANC controlled after 2021.  With 24% of votes in, the CSIR suggested the ANC support would fall to 37%. Now that 85% are tallied, the ANC support lies at 31.6%, below the DA at 36.4%. But the CSIR model takes into account the voting patterns from the areas still outstanding, and in Mangaung, all the outstanding wards are in areas with strong ANC support. That means despite what the figures say right now, the CSIR predictions are probably going to be close to the final numbers.

Even before voting took place, experts predicted that all three Gauteng metros – Ekurhuleni, Tshwane and the City of Johannesburg – would end up with hung councils. The election results have done little to change those opinions. The success of ActionSA was foreseen early on by the CSIR model, and the party currently lies 3rd  in Joburg and Ekurhuleni and  2nd in Tshwane.  The impact of the MK Party was also factored in.

The only unforeseen circumstance before the count began was the rise of the SA Communist Party (SACP), who gained enough votes to win some council seats.

The CSIR is pleased that its predictions have stood the test of time. They will continue to work on the model in the lead-up to the next National and Provincial Elections, when South Africans will go to the polls again in 2029.

Graphic by Sibusiso Biyela – SABC Digital News