Doubt cast on election poll data

As Ipsos projections are queried, it says it mainly tracks the proportional representation vote

The accuracy of statistics provided by the only active opinion polling company in South Africa is being disputed. Ipsos’s weekly election polls indicate that the ANC will suffer a substantial drop in the August 3 poll, but the company has conceded that these cellphone surveys are less accurate than its half-yearly face-toface polls.

Ipsos’s most recent face-to-face poll in Gauteng shows the ANC winning close to 50% of the vote across Tshwane, Johannesburg and Ekurhuleni.

Ipsos has also commissioned a special weekly opinion survey from a sample of 3 000 people, conducted by cellphone on Mondays and Tuesdays in Johannesburg, Tshwane and Nelson Mandela Bay. The survey controversially predicts the ANC will lose power to the Democratic Alliance in Nelson Mandela Bay and Tshwane.

The Paris-based company has admitted that its weekly updated election outcome polls could be off the mark by as much as 6%.

But it could be closer to 10%, with undecided voters holding the swing votes, according to South African big-data firm Municipal IQ.

The country’s political parties have maintained a dismissive attitude towards the forecast.

The representivity of the sample group and Ipsos’s methodology have come under scrutiny.

Municipal IQ says the projections from the cellphone surveys don’t add up. It said that, although trends over the past three elections con- firm ANC support is in decline, they do not point towards a sharp drop in support for the ANC.

“Unless there is a catastrophic happening around the political party, like a split, we just don’t anticipate that 20% of voters will suddenly stop voting for the ANC.

“Combined opposition support is rising, yes, but there is nothing indicating that support for opposition parties, either alone or together, is surging,” Municipal IQ director Kevin Allan told the Mail & Guardian this week.

But Ipsos, is sticking to its data. It said that, because cellphone penetration is greater than 90% in each of the three metros surveyed, the information is reliable.

“It is a concern, but because we do it every week, there is less of a concern. Obviously the sample group of people we recruited [is] representative of the group we are looking for. Most of them feel quite fine to tell people who they vote for,” said Ipsos public affairs manager Mari Harris.

ANC communications head Khusela Sangoni-Khawe brushed off the Ipsos forecasts and said the party conducts its own internal polls.

In addition to the weekly survey, Ipsos conducts a wide-ranging poll every six months, regardless of whether it’s an election year. Though the number of people who answer the questionnaire telephonically each week is about 1 510, the total sample of 3 000 is also interviewed separately, face to face.

This survey, dubbed “the people’s poll”, was concluded in May and estimates ANC support at about 50%, DA support at 23% and 7% for the Economic Freedom Fighters.

Harris believes the information emanating from these interviews is the most reliable, saying: “This survey is … done face to face in the homes of the people in their home languages.”

Apart from this, Municipal IQ has questioned the representivity of each ward in the metros. “You’ve got to assume as well that those 3 000 people represent the different wards in a balanced way. How many people per ward and voting district are represented? There’s a lot there that could reduce the accuracy,” Allan said.

Ipsos admitted that not all wards are represented and conceded that its polls are primarily a reflection of the proportional representation part of the vote. “That’s what makes local government elections very difficult. We are actually just measuring the [proportional representation] vote, because we are not saying: ‘Which ward councillor will you vote for?’ ” Harris said, adding that the company has attempted to factor in the increasing youth vote, as well as the expected turnout of first-time participants in the election.

Municipal IQ’s concern is shared by another big data management company, Afrobarometer. Last year it released the outcome of a 10-year study, which showed that 73% of young people indicated an interest in politics, compared with 60% in 2004. “The formula Ipsos would use to ensure that this new section of the electorate is adequately represented is quite important,” said Afrobarometer spokesman Sibusiso Nkomo.

Another crucial determining factor will be the undecided vote. The total percentage of people who are undecided hovers around 10% across all Ipsos surveys — something that is perfectly normal, said Harris. “They are always a big group. Even in the face-toface studies, they are about 10%. These are people who voted ANC in 2014. That group has two choices: either vote for the ANC again or vote for an opposition party, which I don’t think will happen.”

But Municipal IQ’s Allan is not convinced. “The number of undecided voters, even in their tiny little poll, is huge. It’s around 11% to 15%. So if you say the ANC gets 35%, and you slot in the undecided, it’s suddenly close to 48% — which is in line with the actual voter behaviour,” he said.

Ipsos is by no means a new kid on the election polling block. It is currently contracted to do polling on the United States presidential election for international news wire Reuters. It predicted South Africa’s 2011 and 2014 polls to within a margin of accuracy of 1.6%. Despite reservations shared by surveying companies around the world about cellphone polling, Harris said the Ipsos forecast is “on the money”.

“This poll is representative of people with cellphones in [the metros]. The margin of error for Tshwane is between 1.6% and 3.7%, in Johannesburg around 2%, and in Nelson Mandela Bay between 2.5% and 6%.”

The outcome of next week’s local government elections has been one of the hardest to predict in South Africa’s young democracy. As ANC heavyweights crisscross the country for “victory” rallies, the DA campaign is laying it on thick with SMS reminders of government scandals.

If the accuracy of the only available election polling data is under fire, the result is impossible to forecast.