Sceptics say the best way of predicting tomorrow’s weather is to say it will be the same as today. Predictive models struggle to beat that pedestrian approach, which one could label a rolling null hypothesis.
Today, Wednesday, pollsters are predicting how voters will behave tomorrow, Thursday 12th December, election day. That ought to be fairly easy. Polling has been carried out every month, and more intensively over the last 6 weeks of the campaign. Among those people willing to divulge their opinions, and likely to vote, the pattern is pretty clear: probably some sort of Conservative majority over all the other parties combined.
This is the stage at which pollsters are looking carefully at their predictive models, aware that an accurate result will gain them commercial advantage in the years to come. Are you part of a business anxious whether consumers will buy your new product? Why not test it by giving the contract to the pollster who got the General Election right?
Naturally, pollsters would like some wiggle room. Yes, they will be tested on their final prediction, but if necessary, they can look back a week or two at their previous predictions, and tell some sort of post hoc story: “we got it right the week beforehand, but gave a little too much emphasis to factor Y. We have learned our lesson, and will include Factor Y in our survey work on your product”.
More wiggle room comes from the dual nature of their predictions: calculating the popular vote (reasonably easy) and calculating how that will translate into seats won in 650 constituencies under the harsh first-past-the-post system (much harder). If something is good enough for horse-racing, it should be good enough for judging politicians, but one vote can make a difference, and some margins in the 2017 were no more than 20 votes.
The YouGov MRP model is based on the largest poll, involving 105,612 interviews. This is a good number, given that most polls are in the 1,500-2,000 range. They say of their model:
It works by modelling the relationship between respondent demographics (age, gender, class or past vote) and their vote, and how this changes in different types of seat (region, marginality or incumbency). This model is then applied to the demographics and political circumstances in each seat, projecting results in each of the 632 constituencies in Great Britain.
In fact, this is the familiar Multilevel Regression and Post-stratification approach to produce estimates for states and congressional districts in US elections, and parliamentary constituencies for UK elections.
The idea behind MRP is that we use the poll data from the preceding seven days to estimate a model that relates interview date, constituency, voter demographics, past voting behaviour, and other respondent profile variables to their current voting intentions. This model is then used to estimate the probability that a voter with specified characteristics will vote Conservative, Labour, or some other party. Using data from the UK Office of National Statistics, the British Election Study, and past election results, YouGov has estimated the number of each type of voter in each constituency. Combining the model probabilities and estimated census counts allows YouGov to produce estimates of the number of voters in each constituency intending to vote for a party. In 2017, when we applied this strategy to the UK general election, we correctly predicted 93% of individual seats as well as the overall hung parliament result.
Without labeling them as 95% confidence limits, they admit that their predictions about the number of seats won by the major parties covers a wide range, and there could be no majority (Hung Parliament) or a larger majority Conservative majority. Could be. To get an outright majority, the winning party must get 326 seats, since none of the other parties are willing to help the Conservatives, the front-runners.
YouGov’s latest and final general election MRP model shows the Conservative Party headed for an overall majority. Predicted vote shares in our final poll:
Conservatives 43%, 339 seats
Labour 34%, 231 seats
Scottish National Party (Scotland only) 41 seats
Liberal Democrats 12%, 15 seats
Brexit Party 3%, 0 seats.
Conservative overall majority of 28.
If this turns out to be true, the Conservatives would win 339 seats (22 more than they took in 2017) and a vote share of 43%, their best performance since 1987. The actual number could range from 311 to 367 seats. However, drilling down into the detail, applying general results to individual constituencies is an error-prone procedure, even before tactical voting is factored in. The Labour vote is strengthening later than it did last time, but is still on an apparent upward trend. The lower estimate of the projected Conservative majority may turn out to be closer to the actual result.
Please note that a failure to predict a specific constituency does not invalidate the overall prediction. An intelligence test can predict who will do well on average, but there will be individual exceptions.
So, that is the overall picture about what will happen tomorrow. Other predictive sites (Electoral Calculus) have come to slightly different conclusions, suggesting a Conservative majority of 46.
For those of you who think that we are being manipulated by outside forces, this election is a test case. You have a chance to tell us, with your inside knowledge, what the actual outcome will be. Let me hear your predictions before 22 hours, on Thursday 12th December.
For the record, I have no such insight. The few people I know who could claim some inside knowledge believe that the Conservatives will not obtain a working majority, and that we will be in the limbo of a hung Parliament again.