Polls apart

By Heidi Dietzsch

As we head into the ANC’s elective conference in December, political opinion polls are again top of mind. The outcomes of the conference will depend on the ANC’s complex decision-making machinery, starting at the branch level and working its way through provinces, its various executive structures, and to the votes of delegates who attend. If only we could anticipate the outcome by polling!

At the best of times, polling is difficult. The value of opinion polls has been sharply questioned after failing to foresee British voters’ decision to leave the European Union and the election of Donald Trump as US president.

Polling is a cousin of market research. They are a huge business around the world, with political campaigns pumping enormous funding into them in an effort to gain a strategic advantage over opponents. Increasingly others have used them too, such as hedge funds that try and anticipate election outcomes in order to inform trading strategies.

Before scientific polling there was straw polling, akin to a piece of straw being held in the air to determine in which direction the wind blows. These early straw polls were usually conducted by journalists who interviewed a large but unrepresentative sample of the population. The first of these impromptu interviews took place in the US in 1824. Sometimes exit polls were conducted where voters are asked whom they voted for as they leave the voting stations.

In 1932, George Gallup correctly predicted, in the first scientific political survey ever conducted, that his mother-in-law would win Secretary of State in Iowa. This triumph led him to establish the American Institute of Public Opinion in 1935, known today as Gallup Incorporated. Gallup himself is considered the father of public polling.

But it was the 1936 “depression era election” where he made his breakthrough, forecasting victory for Franklin Roosevelt. An influential national magazine, The Literary Digest, incorrectly predicted that Alf London would win.

Market research and polls are similar in some ways, but there are a few important distinctions. While a poll is general, a market research survey will be more detailed, requiring respondents to provide personal comments and views. Polls usually consist of a single multiple-choice question with two or three answer choices, while surveys are complex with many different types of questions. Data collected through surveys require a great deal of analysis and comparison studies, while data gathered through polls can be used almost immediately.

But why do polls often fail dismally to accurately predict outcomes? In particular, what went wrong for pollsters during the 2016 US election and British referendum?

One factor could be the gap between rational behaviour and emotional behaviour. How people feel in the privacy of a voting booth might differ from how they will react to a polling question when they feel that a justifiable, sensible and politically correct answer is required.

Jon Cohen, chief research officer at SurveyMonkey, an online survey portal, reckons young voter turnout is likely one of the major factors in the discrepancy between polls and the final outcome of the “Brexit” referendum, where the British voted to leave the EU. “The remain campaign was heavily dependent on support among younger voters and they simply didn’t show up,” Cohen told CNBC.

Anthony Wells, research director of YouGov, an international internet-based market research firm, explains that polls undercount voters who are hard to reach. Poll fieldwork periods are usually stretched over three days. Wells suggests that the period should rather be six days in order to include hard-to-reach voters. 

There is also the phenomenon of the “shy voter”. Through social media, people have learnt that serious repercussions can follow when they express opinions contrary to the mainstream discourse. Therefore, voters may tell pollsters that they are undecided or will vote for the socially acceptable option, while planning to vote for a more controversial candidate on election day.

Other factors threw the British and US pollsters off kilter: in times of political instability, voters are making their final decisions later, changing their minds more often and are unsure of whether they will vote at all.

Pollsters were, however, able to save face when they were spot on with the outcome of the landmark French presidential elections at the beginning of 2017. So, what did the French pollsters do right? For one thing, they proved to be far more skilled at assessing the anti-establishment political atmosphere. There is, however, a good reason for that. They have more than 40 years’ experience in dealing with the far-right National Front. Therefore, they are used to taking the “shy National Front” voters into account.

Unlike in the US, online polls are the norm in France and people are much more inclined to admit far-right sympathies in an internet poll than on the telephone.

French pollsters also attribute their success to better demographic modelling, especially in reaching first-time voters, senior citizens, the less-well educated and rural residents. They also focused on turnout, accurately assessing that abstention rates would be lower than in previous elections.

Bruno Jeanbart, director of French polling firm OpinionWay, does, however, admit that the French system is much easier to read than the US electoral college system. In the US it is necessary to poll nationally as well as state-by-state, while one only needs to count the total votes for each candidate in France.

Such considerations are even more complex in SA, particularly at the early stage of the ANC’s elective conference. The delegates who attend vote by secret ballot, so even when nominated by branches to vote according to certain instructions, there’s no way of knowing if delegates do. Those delegates are drawn from provinces as well as the party’s three leagues (women’s, youth and veteran’s). The ANC’s national executive committee members, the top six officials and provincial executive committees also can vote. Any branch with at least 100 members can nominate a delegate. The ANC has so far confirmed there will be 4 723 branch delegates and there are another 525 delegates to be made up from the leagues and office holders. Unlike the US primaries and the UK’s party conferences, it is extremely difficult to anticipate how the ANC’s complex structure will translate into leadership choices. While there have been a few efforts at polling research among branches, the results have been far from conclusive.

There is no such thing as a political crystal ball and pollsters are not clairvoyants. There are many variables and every election is a fresh challenge with unique parameters and issues. However, one thing is clear – when the pollsters do get it right, they are lauded almost as much as the winning candidate. But when they are wrong, their critics have a field day.