YouGov's latest poll projection has the country heading for a hung parliament. This is certainly the most eye-catching forecast in the election so far, aimed to fuel lumpen-commentariat reaction about the closing gap between the Tories and Labour.

After the 2015 general election, the polling biz ended up with egg – in fact, a full-size battery farm – on its face, and it didn’t do much better with last year’s US presidentials or the EU referendum. Part of the problem lies in guesstimating outcomes from poll numbers, given the idiosyncrasies of the electoral system. YouGov says it has about 75 respondents per UK parliamentary constituency – far too few for robust predictions at that level. But that’s not the whole story in the UK, where polls haven’t been much good even at estimating the popular vote in the country as a whole.

Like credit rating agencies, pollsters such as YouGov are private businesses, with an interest in talking up their own performance. When the information they dispense proves inaccurate, business demands inaccuracy about that, too. YouGov’s CEO Stephan Shakespeare boasts that YouGov’s new MRP model (‘multiple regression and post-stratification’; it’s explained here by ‘YouGov’s chief scientist’ Doug Rivers) for predicting the election outcome ‘is something we publicly tested during last year’s EU campaign and it always had “Leave” ahead’. Not quite. In a tweet on the eve of the referendum, YouGov’s freelance wonk Benjamin Lauderdale said that the result would be within the 48-52 per cent range – but didn’t say who’d win. 'Estimates are now within 0.1% of 50-50’. A pseudo-exit poll YouGov did after voting ceased at 10 p.m. on 23 June 2016 reported leave at 48 and remain at 52 per cent – though, in fairness, that didn’t use MRP (re-analysis of the data using MRP had the result as a dead heat).

Modelling is the point where voodoo enters the system. Raw data consists of raw responses to pollsters’ questionnaires, which somehow have to be parlayed into a prediction. The main gaps concern demographics, and the relation between expressed intention and action. Explaining MRP, Rivers tells us that ‘YouGov has estimated the number of each type of voter in each constituency,’ which prompts the question what types there are. What type are you? Maybe the type that doesn’t vote, or gives flippant answers to pollsters. No doubt there are numbers to scale that, too.

Pollsters have an interest in being seen as generally accurate. But they are also, even willy-nilly, intervening in as well as observing the electoral process. Shakespeare stood as the Conservative candidate for Colchester in 1997 and was formerly a member of the Socialist Workers’ Party; he founded YouGov in 2000 with Nadhim Zahawi, the Tory MP for Stratford-on-Avon before the dissolution. Perhaps they see their role in present circumstances as that of warning against complacency. People who think the election is a done deal tend to vote with their arse. More significantly, pollsters have an interest in making elections seem tighter than they are. So do the print and e-media, as do the political parties. All have more attention paid them if ‘the race is too close to call.’ Newspaper coverage of poll 'findings' is a free advertising platform for YouGov and its ilk. Its latest poll, with the Tories on 42 to Labour's 39, has got splashy airplay, and that's what matters. The numbers rest on highly optimistic assumptions about how many young people, who strongly favour Labour, are going to vote. But no publicity is bad publicity. YouGov's business goes well beyond political polling, and election campaigns help to flog it.

What would one predict on the basis of all this? That firms like YouGov will tend to underestimate the Tory vote. Rivers and Lauderdale are both scientists of a sort: they're trained in political science – a discipline which, as the late Ben Barber once said, avoids politics while failing to achieve science.