In the end, none of the pollsters got it right, and some did more badly than others.
With such widely varying predictions for the election and vastly different methodology, it was always going to come down to a race between the pollsters.
Least wrong was Ipsos, one SA’s oldest market-research companies, which incidentally does some of the fieldwork for the ANC’s internal polling.
Ipsos’s result in its last poll done in April came within range of the final result, taking the margin of error into account. On a turnout of 71% — generally assumed to be most likely — on the national ballot Ipsos predicted 61% for the ANC; 19% for the DA and 11% for the EFF. With a margin of error of 2% this put it in the ballpark of the final outcome, which was ANC 57.5%, DA 20.77% and EFF 10.79%.
Ipsos has by far the most resources in the market to conduct research. It has a larger sample than other pollsters (2,834) randomly drawn from all eligible voters in which every adult has an equal chance of being selected. It conducts face-to-face interviews with people in their homes in their home language. The larger sample gives it a relatively small margin of error.
The curve ball for all the pollsters was the low turnout. In the last national election the turnout was 73.4% and both Ipsos and the Institute of Race Relations (IRR) thought it reasonable to assume a moderate drop, in line with the trend of the past 20 years. The turnout dramatically alters the projection of the outcome, as the IRR found out.
The IRR, which entered the political market research game for the first time this election, aims to become an authority in the field, an endeavour that has not started well. Interviews are conducted by cellphone, with a smaller sample — 2,375 — drawn from a database of every potential cellphone in existence in the country, from which a probability sample is drawn. Like the Ipsos sample, it is demographically representative. The margin of error is 3%.
On the national poll, the IRR’s last survey carried out days before the vote took place put the ANC on 54%, the DA on 23% and the EFF on 13%. Taking into account the margin of error, the IRR just made it into the ballpark on the national poll. But in its poll of Gauteng things went badly awry.
In its final poll of Gauteng the IRR projections had the ANC on 41%, the DA on 33% and the EFF on 14%. This was against a final outcome for Gauteng of ANC 50.19%; DA 27,45% and the EFF 14,69%.
Gareth van Onselen, who designed the poll and is the IRR’s head of governance and politics, wasn’t the only one to get Gauteng badly wrong. While Ipsos nailed the Gauteng ANC outcome with a projection of 50,19% a third poll — conducted by Wits professor David Everatt for the Gauteng ANC — was also far off the mark.
Though the ANC doesn’t release its internal polling to the public (neither does the DA), Gauteng Premier David Makhura made sure that the ANC’s internal poll was strategically put into the public domain, mainly to counter the IRR, which the ANC believed was putting out surveys with a political agenda.
The last Everatt poll, conducted using the same face-to-face methodology as Ipsos, projected the ANC to be at 56%, the DA at 17% and the EFF at 11% in the province. In addition, it reported that 12% of those surveyed would not say or did not know who they would vote for.
This is a critical difference with the IRR polling methodology which allocates participants who say they don’t know or won’t say to political parties on the basis of their responses to other questions. In the lead-up to the actual vote Van Onselen and Everatt argued strongly for and against this approach.
As it turns out, with both so wide of the mark, it is not possible to make a call on whose methodology was superior.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.