The Literary Digest holds an interesting place in the history of politics, thanks to its role in the 1936 US Presidential election. For several elections before it had been conducting a mass poll that had allowed it to successfully predict the result of the election, which obviously helped to gain it a lot of attention and sales. In 1936, it did the same thing, sending out over 10 million surveys to voters, and receiving more than 2 million back, which gave it the confidence to predict the election result. The result of their poll was clear: Governor Alf Landon of Kansas was going to defeat incumbent President Franklin Roosevelt.
As we know, that wasn’t just wrong, it was badly wrong. Roosevelt won the election in one of the biggest landslides the US has ever seen, and the Literary Digest, which was already doing poorly in the face of the Depression, went out of business two years later. Meanwhile, George Gallup had used a poll of just 5,000 people and predicted the result of the election much more accurately (though not completely accurately – he missed the size of the Roosevelt landslide).
Gallup’s success came from something we take as routine now – rather than aiming to cover as many people as possible, his poll had taken a sample from the population. In trying to cover as many people as possible and sending their samples to names they had from their subscriber records, phone directories and car registrations, the Literary Digest had failed to sample across the whole of the population, as the poor were unlikely to fall into any of those three categories and were much more likely to vote for Roosevelt than Landon.
What’s important to note here is that the Literary Digest’s methods had worked before and successfully predicted the result of previous Presidential elections, hence their confidence in calling the 1936 result from their data. What they’d missed was the effect of the Depression on both their sample and voting patterns. A large group of people were excluded from their sample because of their poverty, and because of that poverty that group had a very different voting behaviour.
Which brings us to Liberal Democrat Voice. They’ve been conducting regular surveys of members of their forum (which you have to be a Liberal Democrat member to join) and publishing the results on the site for a while. Now, while this is a sample of Lib Dem members, it’s not a randomly chosen sample but a self-selecting one, especially skewed towards those who like to talk and read about politics on the internet. Now, they regularly claim that when tested against other surveys of Lib Dem members their poll is generally accurate, and thus they refer to the poll as a survey of ‘Lib Dem members’ not ‘our forum members’ in headlines, but we’ve now got a strongly testable prediction to see just how accurate a representation it is.
As many of you will likely have noticed, voting in the Liberal Democrat Presidential election finished yesterday, and Lib Dem Voice published the results of their latest survey, asking how people would vote in that. That gave a result of 52% of first preference votes for Daisy Cooper, 30% for Sal Brinton and 18% for Liz Lynne. Unfortunately, there’s no George Gallup in this scenario, who’s done a survey using a different methodology, so it may turn out that they’ve got the result right. However, to me, it looks like a very big hostage to fortune that might well have oversampled a particular type of party member whilst missing out a large chunk who will vote in the election.
We shall see when the result comes out, but there might be a few nerves at LDV Towers while they await it…