Opinion polls and their diverging results
“Opinion polling is a science; running an election campaign is an art.” So said ANC elections head Fikile Mbalula in a conversation we had last week about opinion polls.
Mbalula might find, though, if he were to run an opinion poll himself it would be just as much of an art as his campaign, if the vast differences in methodology in two recent ones are anything to go by.
The most hard-fought battle of the election will be Gauteng. In the 2014 election the ANC got 53.6%, the DA 30.8% and the EFF 10.3%.
Last week’s poll by the Institute for Race Relations (IRR) pinned the ANC and EFF very differently. The sample is drawn from a database of every cellphone number in SA, is demographically representative and includes only registered voters. It put the ANC on 41.6%; the DA on 32.4%; and the EFF on 18.2%.
But while this looks an extreme outcome, it is not what the IRR predicts the election outcome to be. The methodology used in this poll, conducted in February, also models the likely result according to the turnout. On a 71% turnout of Gauteng-registered voters (con- sidered a plausible scenario given a slightly declining trend from previous elections) a less dramatic picture emerges: the ANC gets 47%; the DA 37%; and the EFF 11%.
The IRR poll also allocates undecided voters. This is done by asking those who are undecided a series of proxy questions and allocating them a preference. Unless done this way, says survey designer Gareth van Onselen, undecided voters distort the result and a realistic picture of actual voter support cannot be drawn.
A very different poll, with different methodology and findings, was conducted by the Gauteng ANC, designed and analysed by Wits academic David Everatt.
Though conducted in December – when the ANC had not yet been hit by Bosasa and loadshedding and so voters were slightly more kindly disposed towards it than in February – the results are so vastly different that despite the time difference they are remarkable.
Fifty percent of registered voters in Gauteng would vote ANC, the DA would get only 14% and the EFF 9%.
When those who said they did not intend to vote were excluded from the count, the ANC’s share jumped to 54%. “This is our best guess as to where the ANC in Gauteng will be,” says Everatt.
The ANC survey has a sample size almost three times that of the IRR and interviews people in person, in their homes, in their own language.
The random sample is selected from census and other Stats SA demographic data.
Everatt is scathing of the IRR sample, which he says can simply not be random enough given multiple Sim cards held by individuals.
Telephone interviewing has also been criticised in the UK for getting the national election result badly wrong. One reason is that those who agree to be interviewed tend to be more enthusiastic about politics and so not entirely representative.
Everatt is also opposed to allocating undecided voters, which he says is nothing more than the guesswork of the pollster.
What the two surveys share broadly is the analysis that over the second Zuma period a large portion of the black electorate – about 10% – shifted away from the ANC. These voters are now returning with the election of Cyril Ramaphosa as president.
The IRR also shows that trend was somewhat dampened in 2019, with Bosasa and loadshedding. They differ fundamentally, though, on what is going on now. The IRR says the fortunes of the ANC and EFF are intertwined.
Where the ANC declines the EFF rises, and vice versa.
The EFF is the only party that can take votes off the ANC.
The Zuma era gave rise to a pool of about 10% of black voters floating between the ANC and the EFF. About 5% of that pool is still available to either, Van Onselen says.
Everatt says this is profoundly wrong. The EFF has flat-lined and is not appealing enough to voters, who are largely conservative. The real dog fight of the election will be between the EFF and the DA for second place.
The surveys have implications for voters contemplating a strategic vote.
For many, the EFF and its antimarket policies is bogeyman No 1 and the IRR survey would indicate it is best to vote ANC and keep the EFF out.
The ANC survey, on the other hand, suggests it is more important to vote DA to ensure it remains the official opposition. That decision could be what we can call the art of voting.
Opinion polling is a science; running an election campaign is an art
Fikile Mbalula