Pollsters jostle in the race to get it correct
Which pollster will win the May 8 elections? The answer is almost as keenly awaited as the election results, with vastly divergent predictions sometimes even greater than 10 percentage points on what the outcome will be.
In part this is because a greater proportion of the electorate are more in flux than before as traditional voting choices are questioned due, more probably, to the extent to which disappointed voters are rethinking their choices.
Both the ANC and DA have had a torrid five years wracked by in-fighting and internal division, while smaller parties are arousing more interest than before. The time at which the survey is conducted has been more sensitive to change than before, illustrating these shifting loyalties.
But the bigger reason for the divergent polls this time is the arrival of a new outfit on the political market research field.
The Institute of Race Relations (IRR) has set out to become an authority in the area and has been polling more frequently and quite differently from rival Ipsos, which was previously unrivalled in its frequency and reach. To arrive at its best estimate (all pollsters insist they are not giving a prediction when actually they are), the IRR does two things that not all opinion polls do. It models the likely turnout and it assigns voters who would not say or have not decided to what it deems the most likely preference. In this way, it arrives at quite different outcomes.
The IRR’s head of politics and governance, Gareth van Onselen, says this is done by asking respondents several questions woven through from the 80question survey, from how favourably the various parties are viewed to leadership approval. If a person’s preference simply cannot be discerned they are not assigned.
In the most recent survey which was in the field over the past two to three weeks the ANC stands at 51% on the national ballot, assuming 71.9% of voters turn up on the day. This is considered a plausible turnout
following recent trends of a gradual decline in turnout, which in 2014 was 73.4%.
This is vastly different from the Ipsos survey also conducted in April, which put the ANC on 61% based on a 71% turnout.
The estimates for the DA on the national ballot (when the above turnouts are factored in) are also vastly different, with Ipsos estimating 15.2% and the IRR pegging it at 25%.
The most shocking divergence is the prediction of what will happen in Gauteng. Here we can compare the IRR’s poll with the poll conducted on behalf of the Gauteng ANC by Wits professor David Everatt.
Like Ipsos, Everatt does not assign undecided voters as he says it cannot be easily and accurately done. Everatt pegs the ANC in Gauteng at a respectable 56%. The IRR says that its survey found it to be on a paltry 39% based on a 70.4% turnout. In this scenario the DA is on 39% and the EFF at 12%.
Van Onselen does attach a cautionary to the numbers he released this week because, he says, in the 14 days before voting day big parties put smaller ones under “the squeeze” and grow their share of the vote in the immediate period leading to the election. The ANC in Gauteng is therefore likely to grow at the expense of the EFF, which Van Onselen says is in competition for the 5%-8% share of the electorate that has been shifting between them.
In the Western Cape, the IRR April poll has the DA only just hitting 50% on a 71.9% turnout, having lost 10% of its share of the vote to the ACDP and the Good party. The fight will be to squeeze these parties to win an overall majority.
The IRR, on the one hand, and the Ipsos and Everatt polls on the other differ in broader methodology. Ipsos and Everatt conduct face-to-face interviews with people in their homes in their language of choice, selected randomly in a household sample, which is weighted demographically.
The IRR does its polling telephonically, using cellphones as sampling points in a list which is generated off every cellphone in the country. It then also gets weighted demographically.
Everatt insists that this is not a properly randomised sample, while Van Onselen says that it is and that the owner of the cellphone is called back eight times if they are not available at first before the interviewer abandons the call and randomly selects another cellphone.
In addition to the pollsters there are those who have built models based on past election and by-election results and likely turnout from which a range for each party can be predicted.
Dawie Scholtz, who has built a solid reputation doing this, earlier this week told News24 his “best guess” outcome puts the ANC at 58% on the national ballot, followed by the DA with 23% and the EFF with 10%.
By the time 5% or so of the results are in, Scholtz reckons he will be able to make an accurate estimate of the outcome.