Predicting the 2016 Local Government Election Results

As soon as 5% of the counting process has happened the CSIR will be able to predict who will win in the Municipal Elections.

The race for votes in the 2016 Local Government Elections is underway but it will be a while before we know the final results. In the meantime the Council for Scientific and Industrial Research (CSIR) will be predicting who will win by analysis of statistical data.

CSIR Analyst, Hans Ittman, explains that as soon as the results start coming in they use that information to forecast how the rest of the people will vote. The CSIR can already start their predictions once 5% of the counting process has happened.

“The predictions are based on a statistical methodology that we call statistical clustering. We divide the voting population into a number of groups and currently we’ve been using about 20 clusters where each cluster consists of more or less of the same kind of people with the same behaviour pattern,“ says Ittman.

Ittman, a lieutenant in the South African Air Force from 1970-1973, qualified as a pilot in 1968. He has degrees in BSc(Hons) and MSc (both majoring in Operations Research) as well as an MBA with a thesis on the scope and impact of decision support systems on organizations.

Ittman explains that these predictions are based on a very specific model developed especially for the 2016 Municipal Elections,

“Our main aim in the race for votes is to provide predictions at a national level but we also focus on a provincial level. We then look at a number of the main metros and have also selected a number of interesting municipalities, for example, Rustenburg is a metro that people are talking about,  but there are a number of those.”

Another CSIR Analyst, Zaheed Kimmie, holds a PhD in Mathematics from the University of Cape Town and a Masters in Public Health from Harvard University. Kimmie says that the experts working on the CSIR predictions consist of experienced mathematicians, engineers and computer scientists. He says that you need training in applied mathematics and statistics to be able to construct a model to understand the data,

“Once you’ve written down the model on paper you need the computer scientists to actually make it works in practise with real data in real time. The important thing here is that we’re not sitting months after the elections saying that this is what we would have predicted. We’re doing it on the fly, before anyone else knows what the end result will be. We’re making a stab at saying what the final result will be.”

Registered professional Geographic Information Science practitioner Peter Schmitz holds a PhD in Geography and a BSc (Hons) degree in Geoinformatics. Schmitz is a principal researcher at the CSIR Built Environment and is active in geospatial analysis and forensic geography. He explains that in the past elections they were a bit out with some of the smaller parties however with the bigger parties they were within 1% of accuracy with their predictions.

He adds. “It will be interesting to see how the model is going to forecast especially the effect of EFF. Hopefully we will be a bit more accurate with EFF than we have been in the past because we now added the party to the process from the past 2014 elections.”

“Personally I feel it’s going to be the era of coalitions and there won’t be an outright majority in one of the big metros but that’s my own gut feeling. There’s going to be some serious negotiations that need to go on and I think this is a new phase in South African politics that we don’t know, we’re not used to coalition politics like the Europeans are.”

Preliminary results are expected by the weekend and the Independent Electoral Commission must determine final results within seven days.


SABC © 2016