Ahead of the municipal elections on 3 August, eNCA has partnered with Ipsos to measure voter sentiment in Nelson Mandela Bay, Tshwane and Johannesburg. The SA Communist Party recently voiced its concerns about these polls, insisting that the sampling method used was not accurate.
So just how reliable are the eNCA/Ipsos polls? Emerging market economist at Nomura, Peter Attard Montalto, says:
|“We are very sceptical of them – both in their construction and in their usefulness as a predictor.”First there’s the construction issue: “The sample of 1,400 in earlier polls rising to 1,500 across three cities, two gender buckets, four races, four age buckets, four income buckets and four answers (three parties and then a range of other responses from don’t know to unregistered) mean there are 128 bucket combinations per answer per city to be weighted or less than one respondent (0.91 rising to 0.98) per bucket per answer per city. This makes proper weighting difficult.”Attard Montalto says South Africa naturally has a problem because of different racial groupings being more important than in other countries as traditionally a guide to voting preference.
“However, it means there is serious volatility in the individual demographic bucket support levels that IPSOS publishes with up to 10-15pp swings in support week by week for individual party support in given buckets seen. Overall, we think the sample size should be some 2-3x as big to be fair (ie 1,000 per city).”
He adds that to be fair to IPSOS, it does say the margin of error is somewhere between 2.5% and 5.7%, “but that just reinforces our point.”
Cell-phone polling has also been utilized “and is more likely to overweight wealthier voters than poorer and younger rather than older – exacerbating weighting problems from a small sample size.”
Furthermore, previously published polling results from IPSOS on political participation of the youth suggest it may be using too higher turnout rate for the youth (around 75% of the youth said they would turn out to vote when asked vs previous polls after national elections suggesting real turn out may be as low as 30%).
“We believe the political parties utilize much lower turn-out assumptions for the youth. Even zooming into say the ‘50+’ age bucket in Johannesburg we see a marked increase in support for the DA from 32.5% to 47.1% in the past six weeks which seems highly unlikely to us.”
“So we take polling results with a large pinch of salt – investors should not simply read them as result predictors,” he says.