A critical analysis of political polls in South Africa
Part 1: Misrepresenting polling numbers
If you want to manipulate or rig an election, surveys (or ‘polls’) of political sentiment are a crucial tool.
First, in order for outcomes to be credible they need to align with polls. If surveys show that 60% of voters support party X and then that party gets only 30% of the vote, the public may be suspicious. Similarly, if party Y has 1% support in the polls and then wins 15% that may also generate suspicion. What this means is that if you want to get away with outright rigging of election results (changing party Y’s votes from 1% to 15%) you need to first control or influence the polling numbers. Direct manipulation of election results is an extreme scenario, but there are other scenarios in which a similar logic applies.
A second consideration relates to voter beliefs and behaviour. Polling results do not just describe the views of actual or potential voters: they can influence whether people decide to vote, and how the decide to vote. For example, a voter who thinks that their party could win an outright majority by a thin margin may be more likely to vote because they believe their vote will make a significant difference. Alternatively, an undecided voter who believes a small or new party is rapidly gaining support may be more likely to vote for that party – in a form of ‘herd’ effect.
For these and other reasons, the credibility of polling and interpretation of polling numbers if fundamental for a functioning democracy. Neither is the case in South Africa at the moment. Some of the issues I will discuss in this series of posts are:
deliberate misrepresentation of polling numbers
questionable backgrounds of poll interpreters
questionable backgrounds of those conducting polls
questionable practices and patterns in commissioning polling
lack of credibility of polling methods and data disclosures.
To illustrate, let me start with a case where a senior newspaper editor dramatically misrepresented the results of a (seemingly) independent poll, conveniently laying the ground for a narrative that favoured the political biases of that editor, her newspaper and a range of other interests.
These events occurred back in 2022. On the 14th of August 2022, Ferial Haffajee1, an editor at Daily Maverick, published an article in which she asserted that “the ANC is collapsing as a majority party” and that “poll by Ipsos suggests that if an election were held tomorrow, the ANC would get 42%...second poll, published by Rapport at the weekend, has the ANC at 38%”. These claims were misleading at best, outright lies at worst.
A savvy reader would have noticed the obvious problem: in the numbers Haffajee reported for the Ipsos poll, the largest opposition parties would only receive support from 28% of the population. That leaves 30% of the votes unaccounted for. (If the ANC got 42% and opposition parties got 28%: 42% + 28% = 70%). What was going on?
The answer is simple: Haffajee either deliberately misrepresented the meaning of these polling numbers, or is so grossly incompetent she did not understand them. The Ipsos poll percentages need to be interpreted relative to those likely to vote. If one excludes those who said they would not vote, who were not registered or who refused to answer, the ANC actually had 53.16% of the likely votes. A further decline from its 2018 national vote percentage of 57.5%, but not a loss of its majority or a ‘collapse’.
In an article some weeks later, Haffajee wrote about alternative interpretations. But that article was still misleading as it presented her analysis as equally reasonable: it was not. The assumptions required to justify the following claim are so strong that they were completely implausible. One is left with two interpretations of the original article: gross incompetence, or gross bias. Either one of them is very concerning from a senior journalist.
Incorrect interpretations of a similar kind have been made by other journalists and the average reader would likely not go through the statistics closely enough to realise they are being misled. That is harmful to democracy. But there are many other problems with the polling that has been reported since August 2022, and the pollsters, journalists and commentators analysing it. That will be the subject of my next post.
I used to have a great deal of respect for Haffajee: I considered her one of the best editors in the country when she was editor of the Mail and Guardian. Unfortunately, over a number of years, Haffajee has become biased and unprofessional in her commentary and reportage, regularly misreporting on a range of crucial national issues in a way that consistently serves certain political agendas and biases.