CRISPIN HULL: Itâ€™s tough being an opinion pollster
By Crispin Hull
Published Tuesday 19 January 2016
PRIME Minister Jim Hacker in the Yes, Prime Minister series said of the British public, “I am their leader, I must follow them.”
Something similar was said by the French 19 century democrat Alexandre Auguste Ledru-Rollin who said, “There go the people – I must follow them, for I am their leader.”
The quotes describe the main dilemma for political leaders in a democracy. To what extent do they lead and to what extent do they follow the people’s will – the democratic mandate.
Of course, immediately after an election the winning party can say that whatever policy it put at election time is part of the mandate. However, what happens when something new arrives that demands political action. How can a politician read the mood of the people?
Opinion polls – both public and those done by political parties privately – must play a major role. Often “poll-driven” policy-making is derided, as if the view of the people is a poor substitute for “good” policy that the people would disfavour.
But either way politicians at least need to know what they are up against if they don’t want to follow their voters’ wishes.
You might imagine that with technological advances opinion polling would become easier. But to the contrary, it is getting more difficult.
A couple of weeks ago, I wrote about research into people’s attitude to immigration and population done by Ipsos Australia for Sustainable Population Australia.
It was not done by the standard random telephone-survey method. Rather it was an internet-based survey. That no doubt would give rise to immediate suspicion, until you look at it a bit more closely.
As it happens, by pure co-incidence I was invited this week by an Ipsos researcher to take part in an internet-based survey on media-consumption habits. I agreed, partly through curiosity and partly because I consider it a democratic duty to take part in political and social surveys – otherwise how would decision-makers get informed.
I know it was pure co-incidence because the invitation came via the landline, and that number is not publicly available.
Once you do one internet survey you are invited to take part in others. If you like you can go into a draw for cash prizes.
All sounding a bit fishy? Well consider this: latest (late 2014) figures show that nearly a third of Australian adults live in households with no fixed phone line, and rising. Only about 15 per cent of households have no internet access, and falling.
Further, population researcher Katharine Betts says that many people with fixed lines screen their calls. They let callers go to voice mail and only return the calls they want to – usually ignoring strangers, marketeers and opinion pollsters.
People are sick of being annoyed and sick of phone interruptions at meal times.
By 2004, response rates had fallen to 17 per cent. Now it is as low as 2 or 3 per cent.
It’s tough being an opinion pollster.
There are two major tests for the credibility of opinion polls. One is at election time when nearly everyone votes. Pollsters that predict results accurately, particularly the primary vote, can sell themselves more easily to commercial clients.
And with compulsory voting, the test is more rigorous.
Essential Research in Australia, using an internet panel technique, predicted the primary vote in the 2010 election more accurately than any other poll.
The other test is to match the sample with census data. So if the sample has the same age, sex and geographic profile as the population as a whole at the census, the pollster can be confident (95 to 99 per cent) that the opinion will be accurate to within a few percentage points, depending on the sample size.
Some pollsters crib a bit, by asking “randomly” selected people their age, sex or postcode and if they already have their quota of those people, they move on. It means the sample will match census data, but it will not have been selected randomly.
In these days of declining landline respondents, internet surveys have another advantage – cost. Once you have got a willing respondent (through random telephone calling) they can be used more than once. Further, the pollster does not have to go through the time-consuming exercise of asking all the questions.
Internet respondents can also stop and start as they choose, making it more convenient.
Moreover, an internet survey can be quite long and complicated, and if people get coupons or cash prizes as a reward they can be quite attentive.
That said, my guess is that busy people on reasonable incomes simply would not bother unless they were sufficiently curious.
So those things would introduce biases.
But all samples are biased. For example, a large number of young people are not polled in phone surveys. It used to be that wealthier people (and therefore more educated people) tended to have a telephone and poorer people did not. These days, perhaps more education technically savvy people dispense with their landlines.
Some researchers have concluded that the larger sample sizes in internet surveys -- made possible because they are much cheaper -- more than offset problems with the randomness of the sample.
It may also be that having attentive people doing the survey makes them more reliable than people on the end of the phone just wanting to get it over and answering without much thought – especially if a distraction arises during the call.
Obviously, self-selecting internet polls, especially on media sites, are notoriously unreliable, often reflecting the political stance of the media in question. But internet panel surveys are not self-selecting.
One thing we know for sure – opinion pollsters want to get it right. They have a commercial interest in doing so.
For example, a car-manufacturer client might pay good money to know what colour of cars people prefer so they don’t produce a lot of cars of the wrong colour. A polling company that has been accurate at election time will be more trusted by those commercial clients.
So it is small wonder that polling organisations are seeking new ways to measure opinion accurately as the use of landlines becomes ever less reliable. It is not a good idea to dismiss internet surveys out of hand.