Polling Unpacked. By Mark Pack. Reaktion Books. 2022.
This is written by the founder of PollBase and traces the history of opinion polls from various ‘straw polls’ in the nineteenth century that were conducted to test public opinion, to the development of what we now know as modern opinion polling. This was pioneered by the US Literary Digest in the early twentieth century, before the game-changing entrance of George Gallup into the sphere, who recognized that the total number of people polled was far less important than how representative these people were.
Opinion polls – as developed over recent decades – have a somewhat mixed reputation, though tend to be more accurate than people often think, at least when the margin for error (typically +/- 3 percent) is taken into account. And of course, sometimes voting systems – especially First Past The Post (FPTP) – throw up vagaries of their own. Trump won the US presidency in 2016 even though he lost the popular vote by a very similar margin to that predicted by the polls, as in 1951 and February 1974 the party with the most votes in the UK did not get the most seats and form the government, which means that even when voting intention polls are broadly right they can sometimes appear ‘wrong’.
But some polling does genuinely go awry, including when Cameron secured a narrow overall Tory majority in 2015 when the polls had predicted a dead heat with Labour and then with the underestimation of the Labour vote in 2017. There are a number of reasons for polling errors and they can be complex, including the wording of key questions, which can provide startlingly differing results based on the smallest nuances (polling on single issues rather than overall voting intention is notoriously problematic for this reason alone). Arguably the most common voting intention issue, however, is the difficulty of getting accurate, weighted national samples when the demographics of society are changing fast. There are also significant allied issues like some types of voters becoming more inclined to turn out (or not) than others, some being disproportionately reluctant to give their real voting intention to pollsters saying ‘don’t know’ or ‘won’t say’ instead, differential willingness among particular party supporters to even participate in polls in the first place, and people being sampled in polls who are not actually on the electoral register for various reasons. Significant polling errors are usually caused not by one factor like this, but by more than one of them working in concert to skew the result.
A new type of polling has recently arisen to address some of these issues for FPTP elections called Multi-level Regression with Poststratification (MRP). This aims to generate constituency-by-constituency predictions from a national poll sample that normally wouldn’t be considered large enough for these purposes. It is based on demographic information in particular seats being used as the basis for projecting a result informed by what the national sample has been indicating about certain types of voters and their probability of voting in particular ways (eg, white men over 65, having say, a 55 percent probability of voting Conservative). Its track record so far – both in 2017 and 2019 in the UK – has been pretty good, including YouGov’s detailed MRP prediction of parts of the Red Wall going Blue, but it is early days.
Pack is clearly an expert and does a good job at unpacking all this (pun intended). How useful polls are is another issue of course, and Pack looks at this too. Echo chambers can have a particular but limited use and many argue that polling is given far too much prominence. There can also be a commercial aspect and a distorting element to polling, especially in the United States where skewed ‘paid for’ polling by political parties and groups is more common. In some countries, polling is banned completely during election campaigns lest it be deemed to influence the result.
Will opinion polling have a use of sorts in socialism? Possibly, though direct democracy and decision-making is likely to feature much more heavily and so the need for polls may well lessen. And hopefully, to paraphrase the late Tony Benn, we will be busy creating signposts rather than following proverbial weather vanes.
Dave Perrin
No comments:
Post a Comment