During the last 10 South African elections, specifically the 1999, 2004, 2009, 2014 and 2019 national elections, as well as the 2000, 2006, 2011, 2016 and 2021 municipal elections, the CSIR employed a statistical model to predict the election outcomes at various levels, such as national, provincial and municipal.

The CSIR's election prediction model relies on two core principles: the analysis of voter behaviour patterns and the sequence in which voting results are announced on election day. When combined, these two principles enable the team to group voters or voting districts based on their past voting behaviour, utilising a statistical clustering method.

When applied in previous elections, the model typically achieved a high degree of accuracy at a national level once approximately 5% of the results had been tallied. Due to the way the model works, the predictions become more stable and accurate as more voting districts are counted, ultimately converging to the final results once all voting districts have been declared.

“The CSIR’s election prediction model is not a polling system, but a model that uses statistical and mathematical analysis to predict election outcomes. It showcases how statistical clustering, and some mathematical algorithms can achieve good predictions from a small sample of results. The election prediction model operates on the basis of reducing the bias resulting from the ‘non-randomness’ of the incoming results that arise from the order in which the results are received.” says CSIR Chief Executive Officer, Dr Thulani Dlamini.

CSIR’s predictive modelling capabilities allow for this to be tailor-made for other predictive analysis work. As one of its kind in the continent, CSIR has the capability of assisting other countries that may require such a tool to support election transparency and engagement. 

“The organisation possesses robust capabilities in mathematics and statistics, which are applied to deliver precise results and can be customised for various forms of predictive analysis and forecasting” declares Dlamini.