The difficulty is that you need to have a minimum number of fixed sums bet on the football and the coupon’s odds will change significantly. An overall win is thus more likely, because you know the minimum you can risk is higher than before. This may explain why bookmakers can set lower odds for general elections than on the National League (leaving aside the case of the EU referendum which we did not examine, and which is a pretty tough case).

An alternative interpretation is that election outcomes have higher ‘error’ because they are less reliable. This interpretation is strengthened if the errors in the results are exactly known. For example, the ‘chaos theory’ described by social scientists Haidt and Campbell asserts that if your voting preference is randomly shuffled then your probability of voting for a party with an overall majority, or winning a seat, is roughly 1/101. In other words, if the entire country (or a small fraction of it) makes a one-vote decision – choosing which candidate to vote for – then the probability that an individual in the population will end up with a majority is 1/101.

But imagine that in the middle of the election there are 2,000 people who have no idea which party they want to vote for, let alone which candidate to vote for. Let us assume they all voted in a random way. The size of the voter-base could be represented by a game of Tetris, which has a unique solution for every set of pieces. The total number of pieces would be 1,001, and the number of ways to arrange those pieces is 1,000,000. The actual number of voters is therefore just 200 million. Since you cannot construct a puzzle with an equal number of pieces, the chance that each voter can be in exactly the right position to create a Tetris-like block is 1 in a million, or a small fraction of 1%. What the statistics show, then, is that when a lot of people vote as randomly as possible, with no uncertainty, there is virtually no chance that any particular block will be created.

If, however, someone holds a special knowledge of which parties will win and which candidates will gain seats, then the game of Tetris is transformed: a block can be constructed; the likelihood of a single winning block is much greater, and the size of the electorate is much larger. In this situation, the probabilities of winning and losing are nearly identical. Because of this, the probability of winning is much lower than for normal random-number generation. There are, in fact, two probabilities. First, the probability of winning based on random-number generation; and second, the probability of winning based on the knowledge of the winning party’s nominee. The first probability is just a fundamental property of the game of Tetris. The second probability is unique to the “contest” – to the election in general.

In this contest, each voter is permitted to place 100 blocks. In total, there are 100 million voters.