Can the polls ever be right?
Sat, 30 Jul 2016 09:50:49
As August 3 looms the results of studies predicting the results have been released. These are generally based on “scientific” surveys conducted with a cross section of the adult population.
Unfortunately these polls lend little certainty as to what to expect once the polls close. Telephone surveys by Ipsos for eNCA routinely show the DA leading in each of the metros they survey (Tshwane, Johannesburg and Nelson Mandela). In all of these cities the DA is expected to get more votes than the ANC but will still fall short of a simple majority.
The methodology used by Ipsos is largely untested. It is based on interviewing random survey numbers. By contrast ‘tracking’ by ANN7 show that the ANC is leading in those three metros (albeit without a simple majority). They also show the ANC leading with a simple majority in Ekuruhleni. (ANN7 has not yet released predictions for the other cities). When compared with the Ipsos poll ANN7 show significantly lower levels of support for the EFF. The method used by ANN7 is unclear and does not rely on a random selection of potential voters. As such the study is not ‘replicable’ – a key determinant of reliability.
To some extent the pollsters cover their backs by indicating a high level of ‘undecided voters’. This suggests that should most of the undecided decided to vote for a single party the results will differ significantly from that predicted. Ipsos indicates that 12 to 16% of voters are undecided. ANN7 shows that only 4 to 8% of voters are ‘undecided’. However it is improbable that the undecided will swing to a single party.
The differences in the polls are surprising as the usual weak point of polls is not their ability to determine party support levels. The vast majority of voters already know what their preference is. The weak point of polls is in determining whether or not the supporter of a particular party will actually vote. The vast majority of respondents (typically over 70%) indicate they intend to vote in the upcoming election. However we know by comparing previous surveys and elections that far fewer people actually cast their ballots on the day. This is particularly the case for local government elections. Asking whether you intend to vote is a bit like asking you if you intend being socially responsible. Respondents tend to give a polite answer (“yes I will vote”) rather than a realistic one. The kicker is: If the support patterns of voters who actually vote differ from that of people who said they intend to vote but do not then the polls will be wrong. These two populations are, in terms of political behaviour, defined by their differences and an error in the polls is assured.
In the end party supporters who are losing enthusiasm for it are more likely to stay at home on the day, go shopping or have a braai. This factor plays a far more important role in election outcomes than whether or not an undecided voter makes up his/her mind as to which party to vote for and then go and vote. Generally the alternative to voting for your usual party is abstaining rather than “defecting”.
Ultimately pollsters should be guided by the question that should accompany all unexpected findings – is it more likely that a) the world has changed or b) my math or method is wrong?
Michael O’ Donovan