"We were able to predict that voter turnout would be 75% of registered voters at 03:00 on 15 April after 18% of the voting district results were released," says Hans Ittmann, spokesperson of the CSIR team.
Contracted by the SABC, the CSIR’s decision support team worked in shifts around the clock from 14 April at the IEC headquarters in Pretoria to provide the public broadcaster with valuable information. With the aid of a statistical model, they accurately predicted the main trends of the elections, such as voter behaviour in the voting districts and provinces.
"The model is specifically designed to counter the bias in the order in which subsequent results come in. By segmenting the electorate we are able to predict the voting behaviour of voting districts, long before the actual results have come in," says Dr Jan Greben, project leader and designer of the model. This approach worked well in the 1999 and 2000 elections, and the current elections again showed some impressive predictions.
General trends identified by the team included that the traditional ANC strongholds lost only a small percentage of votes to the DA. In the voting districts where the DA and NNP were dominant, the DA gained many votes from the NNP.
"With the model we could zoom in and make observations on things like the fact that in Oudtshoorn, the New Labour Party got an extraordinary number of votes," Ittmann adds.
The data were useful to political analysts as well as journalists in their discussions around the elections and the team’s work received a great deal of television and radio coverage.
The CSIR team feels the model can be tweaked and refined before the next elections. Dr Chris Elphinstone explains that they would like to improve the model’s capabilities to automatically highlight interesting features in the forecast results. They feel the success of the project is significant and intend to present and publish academic papers on the subject in the near future.