Defend Truth


Can we trust all the SA 2024 election opinion polls?


Melanie Verwoerd is a former ANC MP and South African ambassador to Ireland.

As we get closer to election day, more opinion polls are released. With the media keen for news, it is a great way for an organisation to get press coverage. Of course, the more controversial or outlandish the results are, the more attention these polls tend to get, but can we trust them?

Recently, in a period of seven days, three polls hit the headlines. Ipsos put the ANC at 40.2%, the DA at 21.9%, the EFF at 11.5% and MK at 8.4%. The second poll, commissioned by Iqbal Survé’s Independent Media, had the ANC at 18%, MK at 17%, the DA at 13% and the EFF in the lead with 19%. It was only after reading the small print that it became apparent that they only polled voters in Johannesburg, Durban and Cape Town — three metros where the ANC is doing particularly badly. 

A few days later, Gareth van Onselen tweeted about the election-tracking done by the Social Research Foundation: “The ANC has been absolutely surging the last two weeks. Up to 45% as of yesterday on the 56% turnout model and no signs of slowing. If it continues this momentum, 50% becomes a distinct possibility.”

So, how to make sense of these different results? As a political analyst, I study all the polls and, apart from ensuring that the poll is representative of class, race, gender and the rural/urban divide, there are questions I ask when evaluating each one. 

First, it is important to know when the research was done. Polling results are only snapshots of voter sentiment at the time the fieldwork was done. Naturally, what was happening at that time could influence voter sentiment either positively or negatively. If, for example, there was Stage 6 load shedding when the research was done, one would expect to see a negative impact on the ANC numbers and an upturn for the opposition parties.

Second, it is important to know who did the research. Big agencies, such as Ipsos and MarkData, have proven their salt over many years and despite some objections by political parties, tend to be reliable. The Brenthurst Foundation (set up by the Oppenheimer family) and the Social Research Foundation also do regular polling, but questions have been raised about their proximity to the DA. 

According to Independent Media, its poll was done by African Innovation Research South Africa (Airsa). The Airsa website lists its address as 122 St Georges Mall, Cape Town (which formerly housed Independent Media) and its phone number doesn’t work. This would suggest that it was an internal poll done by the newspaper while trying to dress it up with an academic-sounding name. This raises serious questions about the legitimacy of the poll as borne out by the results (it also claimed the EFF would win the Western Cape).

This brings me to the “how”. Agencies use different polling methods. For example, MarkData and Ipsos tend to do face-to-face polling, with mock ballot papers and big sample sizes of between 2,500 and 3,000 respondents. Brenthurst and the Social Research Foundation, on the other hand, usually do telephone polling, with much smaller sample sizes. 

Size matters when it comes to polling and, as a rule, the bigger the sample size, the better. We also know that in South Africa phone polls tend to under-poll rural — and particularly ANC — voters. However, phone-polling is quicker and cheaper, and can be done more frequently, as in the case of the very useful daily tracking by the Social Research Foundation. 

When it comes to newspaper coverage, it is also important to look at the margins of error, which should be declared by the pollsters. It is usually between 1% and 4%, which means that the results can go either up or down by that percent.

So, recent “ANC support plummets” headlines after the release of the Ipsos poll might be true if they refer to the election results of 2019, when the ANC got 57%, but are misleading when comparing Ipsos polls over the last year, which have differed by only one to 2 percentage points — well within the margin of error. 

Of course, such a sensational headline will get more “eyeballs” than an “ANC holds steady at 40%” headline, but it can cause unnecessary anxiety or kindle false hope among voters and international investors, especially when there are still a few weeks to go before election day.

Experience has taught us that it is only two to three weeks before an election that polls become more accurate. So, even though they may be interesting and fun to look at for political nerds like me, I would wait a while before ordering your celebration party supplies or buying more antidepressants. DM

Read more in Daily Maverick: Elections 2024


Comments - Please in order to comment.

  • William Kelly says:

    So where’s the data? Tell us which polls predicted what and when in the last three elections via some actual analysis? That at the very least removes some element of what is essentially just your opinion at present.

    • Kenneth FAKUDE says:

      Melanie did quote the origin of the polls and results, they are available for verification question is are they reliable? If ever you differ with any information she gave out you can bring your data and point out what she got wrong.
      Its very difficult to attack someone using data without presenting data for correction unless you have personal beef with the person.
      If you don’t have disputing data just leave her alone.

    • Andrew Newman says:

      It would be very useful if they cannot display the polls to at least provide links to the polls.
      Daily Maverick used to be good at that. One of the main advantages of Daily Maverick.

    • Reasonable Observer says:

      Agree. Given the headline I expected to see some reference to how polls have fared in the past.

      Very interesting topic. Potentially. But all we really got is potential reasons why pollsters could be under reporting on the ANC.

  • Kanu Sukha says:

    Thanks for the very salutary and smart concluding sentence ! HOWEVER should there not be a third choice between the two extremes … to just continue as if nothing has changed ?

  • Geoff Coles says:

    So Melanie is a ‘political analyst’, ex ANC MP no less with, as always, a dig at some pollsters because of their DA connection.

  • robby 77 says:

    I’m not sure what the point of having proximity to the DA is trying to achieve. Does it mean you will produce bias poll results just for the heck of it, and in the process make your polling company look stupid?

  • Agf Agf says:

    What qualifies Melanie as a “political analyst”? As an ex ANC MP? As an ex ANC appointed ambassador? As a columnist whose pro ANC stance is clear to all readers? I think not.

    • Niek Joubert says:

      I think she is feeling increasingly uncomfortable with the obvious demise of her corrupt political party, which even she cannot defend anymore.

  • Jeff Pillay says:

    What a lame article from a political analyst. Think most are aware of what was written. What was the take away. This close to an election a crediable political analyst will throw out his/her predictions & justify why not worried about the final outcome.

  • Reasonable Observer says:

    With all due respect, I do not want to come across as too harsh.

    But another person’s job looks simple if you know little about the discipline. And in that case it’s best not to be too liberal with advice or (very superficial) criticsm from a distance. This is true of stem fields too, which is why hard maths and applied stats that we’re talking about here, aren’t everyone’s cup of tea.

    All the how’s and pitfalls you mention and where the sampling is done, are factored into polls, and different respondents’s responses are weighed according to demographics they likely represent.

    They don’t just simply go and ask 3,000 sort of representative respondents. Statistical modelling from sampling is a well developed field in mathematical sciences. It is extremely naive to think you’re the first person to think of all these factors. The pollsters have thought of that as well as of many other factors that can create inherent bias, that this piece does not even touch upon, which is really 101 stuff for them. And there are methods deriving rural outcomes based on polling respondents in metros, from known historical correlations. It’s not a detail they “missed”. Sampling have practical constraints and they have to make calls on what is material vs the cost. Even these decisions tend to be quantified to see which assumptions are material, and which ones we can live with.

    Which is why the polls were incredibly close. This piece attempting to cast doubt, on the other hand, aged poorly.

Please peer review 3 community comments before your comment can be posted

MavericKids vol 3

How can a child learn to read if they don't have a book?

81% of South African children aged 10 can't read for meaning. You can help by pre-ordering a copy of MavericKids.

For every copy sold we will donate a copy to Gift of The Givers for children in need of reading support.