LETTER: IRR polling was mispresented
Carol Paton evaluated polls as predictions, despite the institute stating clearly and upfront that they are not
Carol Paton fundamentally mispresented Institute of Race Relations (IRR) polling in her column (“More Losers Than Winners In the Battle Of the Pollsters”, May 14). In every poll the IRR stated clearly and upfront that it was “not a prediction”, and yet she proceeds to evaluate them as predictions.
That said, in reading the trends, polls do have predictive value. Our final poll (on May 4) modelled on the 2014 turnout, put the ANC on 54%, the DA on 23% and the EFF on 13% (with a 3% margin of error). Based on that, we said over the last few days the ANC was likely to push up and the EFF was likely go down. That is exactly what happened.
When we remodeled our numbers for the actual election turnout — 65% (which cannot be predicted, and no-one did) — we got the ANC at 57%, DA 23% and EFF 11%, an even closer approximation of the trend we identified.
That the IRR failed to model for the right turnout is something it is perfectly happy to take responsibility for, but that says nothing about the integrity of the poll itself.
Our polls accurately identified that the ANC was polling in the mid-50s, was significantly down in KwaZulu-Natal, had a serious problem in Gauteng, that the DA majority in the Western Cape was relatively safe, and that the ANC and EFF were locked in a battle for 5% of alienated ANC voters, and where they ended up would determine the result. All of which no-one else did.
To write all of that off is to profoundly undermine the value of polling to public debate. The public (and Business Day) seemed to generate great value from the IRR’s polls, which is ironic. Nevertheless, we will continue to provide that service fully confident that our methodology is sound and our insights helpful.
Gareth van Onselen
IRR head of politics