Nco Dube | 30 October 2026

South Africa produces a remarkable number of polling experts whenever somebody’s party loses support. Suddenly everyone wants the sampling frame. Rural coverage becomes a national concern. Funding must be investigated. Let the next poll show a recovery and the forensic unit closes shop. The same supporters are now forwarding the graph as a verdict from the people.

As 4 November approaches, there is plenty to choose from. Ipsos’s September release put the ANC on 31% and the DA on 28% in Johannesburg. The eNCA/MarkData survey released on 18 October put the DA on 35% and the ANC on 21.8%. October’s Social Research Foundation and Common Sense findings put them on 36% and 22%. Depending on which graph arrives in the family WhatsApp group, the ANC is either still ahead or being comfortably beaten.

In eThekwini, Ipsos recorded MK support at 55%, against 43% in the SRF/Common Sense survey. News24/SABI put the DA on 32% and the ANC on 24% in Tshwane. These are different surveys with different assumptions and interview periods. They demand investigation. Instead, the abundance of polling makes it easier to shop for a result and dismiss whatever contradicts it.

This habit has emotional roots. Party allegiance can carry family history, racial experience and years of personal sacrifice. People have defended these organisations at funerals, around dinner tables and through scandals they would never tolerate from an opponent. An unfavourable poll threatens more than an electoral prediction. It suggests that other people are walking away from something in which we have invested a part of ourselves. Discrediting the survey spares us that uncomfortable reckoning.

Political psychologists call it motivated reasoning. In US experiments, Gabriel Madson and Sunshine Hillygus held the source and methodology of a poll constant while changing its reported results. People judged polls more favourably when those results agreed with their existing views. The research had not become better. It had become more agreeable. I think that distinction explains much of our own polling debate.

Education offers no automatic escape. Research by Ozan Kuru, Josh Pasek and Michael Traugott found that the tendency to discredit unwelcome poll findings was particularly evident among politically knowledgeable respondents. Knowing the terminology can make someone more resourceful in defending a preference. The supporter demanding methodological perfection from an unfavourable poll may understand every objection being raised. The dishonesty lies in suspending those objections when the party gains two points.

For campaign leaders, selective scepticism also serves a practical purpose. A bad poll ought to prompt questions about weak candidates or disappointed voters. Attacking the researcher changes the subject. Supporters who might have asked what the leadership intends to fix are instead enlisted to fight a conspiracy. The organisation escapes examination while its most loyal members congratulate themselves for seeing through the enemy. That can become a habit of political management. Every warning is hostile. Every critic has an agenda. Each setback confirms that powerful forces fear the party. Leaders who organise their supporters around permanent suspicion can survive an extraordinary amount of failure. They only need to keep the argument focused on who is supposedly attacking them.

Then there are polls that should never enter the argument as research. On 29 October, Africa Check rated as fake the IPRG graphic claiming 43% ANC support in Tshwane. The graphic lacked a sample size and an account of how the data were collected. Africa Check could find no evidence of an established, transparent South African research institution behind the name.

The 43% claim had been promoted by the ANC’s Kgoši Maepa on 26 October. He presented the supposed pollsters as independent university professors and political-science professionals. Academic credentials cannot substitute for evidence of the research. A politician asking voters to trust such figures should be able to name the researchers and explain how they reached their finding. That duty applies before the graphic is shared.

Fabricated polls exploit the very loyalty that makes inconvenient research so easy to dismiss. A campaign-friendly number arrives dressed as science and supporters do the distribution. By the time it is debunked, the claim has already acquired the familiarity that comes from repetition. The willingness to believe has done the work that evidence should have done.

There are good reasons to question a poll. A claim that rural voters were missed, for example, can be tested against the sample breakdown. The published eNCA/MarkData questionnaire on Johannesburg’s water management supplies one. Its positive response linked an assessment of performance to deserving political support. Its negative response described the leadership as hopeless and called for a different mayor and ruling majority. Those answers combine two judgements that a careful questionnaire should separate. A resident might think water management is poor without wanting to replace the mayor, or think it tolerable while preferring another party.

That wording deserves an explanation. It deserves one even from a poll whose headline result pleases us. The useful criticism identifies what is wrong with the question and how it could distort the answer. Calling the entire exercise rigged because the ANC is behind supplies neither an explanation nor evidence. Equally, the DA’s lead does not excuse weak questions elsewhere in the survey.

Misreading can also happen without a defective question. In the same MarkData survey, Herman Mashaba drew 23.4% as the preferred mayoral candidate, while ActionSA recorded 10.3% in party voting intentions. Turning his personal rating into support for his party would more than double its apparent position. These figures answer different questions. An attractive graphic can quietly lose that distinction, especially when the larger number suits whoever is sharing it.

The science requires us to keep track of what was measured. A sample estimates the views of a defined population. Its reported precision depends on how people were selected and how the data were adjusted. A margin of sampling error describes one source of uncertainty; it cannot certify the wording, coverage or turnout assumptions. Scientific credibility comes with reasons that can be examined. It cannot be conferred by a party’s communications team.

Timing helps explain why apparently competing results need not describe the same electorate at the same moment. Ipsos’s September release used interviews conducted in June and July. MarkData interviewed during August and September. The SRF/Common Sense fieldwork ran from 18 September to 12 October. News24/SABI used a 52% turnout model; MarkData used 48.8%. Those choices can affect the findings. They deserve the same scrutiny whichever party benefits.

But uncertainty is also being abused. A narrow lead may be statistically inconclusive. That does not make every lead meaningless, or allow a party to erase any result it dislikes by invoking the margin of error. Nor can certainty be manufactured by selecting the most flattering estimate from several surveys. The critic has to do the work of showing why one estimate deserves more confidence than another.

A simple discipline would expose much of this nonsense. Decide what makes a poll credible before checking the party percentages. Keep the same concerns about funding, rural coverage and question wording when your party leads. Then swap the party names mentally. If that changes your assessment of the research, your political preference has probably been doing the assessing.

There is no requirement to become politically indifferent. I can want a party to win and accept evidence that it is losing. Indeed, accepting that possibility gives a supporter something useful to demand from the leadership: an account of what has gone wrong and a convincing response. Denial leaves the party free to repeat its mistakes while its members spend the campaign quarrelling with researchers.

South Africans voting on 4 November will still need working water systems and functioning municipalities after the celebrations end. They gain little from organisations trained to treat bad news as an enemy operation. When supporters demand perfection from pollsters and accept excuses from their leaders, they are giving those leaders permission to fail.