The ins and outs of making better predictions on both teams to score in the EFL

Oldham scored and conceded in 31 games last season, more than anybody else in the EFL. How often will they do it this season? Preston scored and conceded in 15 games last season, fewer than anybody else still in the EFL. How often will they do it this season?

One way of answering those sorts of questions is described in a book called Noise by Daniel Kahneman, Olivier Sibony and Cass R Sunstein. Kahneman is a winner of the Nobel Prize in Economics.

There is more than one type of noise. Kahneman, Sibony and Sunstein focus on what they call unwanted variations in judgements. That is to say, different people giving different judgements on the same thing, or the same person giving different judgements on the same thing at different times. One of their mantras is: “Wherever there is judgement, there is noise, and more of it than you think.”

Judgements include predictions. Another type of noise can be found in the data people use to make predictions. Take betting. We want to know what a team will do next. They might not perform as they performed before, and how they performed before might not be fully revealed by the data. They might have got more or less than they deserved. And what they get next might be more or less than they deserve next.

Kahneman, Sibony and Sunstein describe a four-step process for making predictions. It can be used in a lot of betting situations. I will illustrate it with the examples of Preston in the Championship and Oldham in League Two.

Step one. To start, you need one piece of specific information. We have that for Oldham and Preston. Last season Oldham scored and conceded in 31 games, Preston scored and conceded in 15 games.

Assume for the moment that the specific information you have gives perfect predictions of what you want to know.

If the number of times a team scored and conceded in one season was a perfect predictor of how often they would score and concede in the next season, our predictions for this season would be that Oldham will score and concede in 31 games and Preston will score and concede in 15 games.

This can be called taking an inside view. We are considering only information specific to Oldham and Preston.

The second step requires us to take an outside view. To do this you will need one piece of general information. I will give it to you in a moment.

Pretend now that you know nothing about Oldham and Preston. How often do you think they will score and concede this season? The only thing you can do is use general information. How often typically do a team score and concede?

In just over half of all EFL games both teams score. EFL teams play 46 games in a season. The average number of games in which a team score and concede in a season is 24. Or at least it was over the past 26 seasons, 1995-96 to 2020-21. (I ignored season 2019-20 in League One and League Two because then teams played fewer than 46 games).

If we knew nothing about Oldham and Preston the best predictions we could give for this season would be that Oldham will score and concede in 24 games and also that Preston will score and concede in 24 games.

Step three. This is the hardest. The authors say: “Your task here is to quantify the diagnostic value of the specific data you have, expressed as a correlation with the outcome you are predicting. Except in rare cases, this number will have to be a back-of-the envelope estimate.”

If one thing predicted another perfectly the correlation would be 1.00. If the first thing did not predict the second thing at all the correlation would be 0.00. I can give you one piece of advice from what I have learned through painful experience over many years. Your estimates almost always will be too high. A lot of information is less predictive than we imagine.

Fortunately we do not have to rely on guesswork about the predictive value of specific data on both teams scoring in the EFL. The correlation between how often an EFL team score and concede in one season and how often they score and concede in the next season is about 0.10. Put that another way: the number of times EFL teams score and concede in one season explains about one-tenth of the variation in the number of times they score and concede in the next season.

I say this having studied every pair of EFL seasons from 1995-96 and 1996-97 to 2019-20 and 2020-21. I calculated the correlation between how often teams scored and conceded in the first season and how often they scored and conceded in the second season. I ignored teams who played in a higher or lower division in the second season, and any other teams in a pair of seasons that included 2019-20 in League One or League Two.

The fourth step brings together the previous three. We move from the outside view toward the inside view. How far we move depends on our estimate of the predictive value of the inside view. I suggest that for Oldham and Preston we should move one-tenth of the distance between the outside and inside views.

Our outside-view predictions for Oldham and Preston were that this season they will score and concede in 24 games. Our inside-view prediction for Oldham was that they will score and concede in 31 games. The gap between 24 and 31 is seven, and one-tenth of that is 0.7. Round to a whole number: one. Our best prediction of all for how many times Oldham will score and concede this season is 24 + 1, which is 25.

Our inside-view prediction for Preston was that this season they will score and concede in 15 games. The gap between 24 and 15 is nine, and one tenth of that is 0.9. Rounding to a whole number here also gives one. Our best prediction of all for how many times Preston will score and concede this season is 24 - 1, which is 23. You can see that the most appropriate prediction for how many times a team will score and concede in a season will never vary far from 24.

Of course this season has already started. Oldham and Preston have played four EFL games. Oldham scored and conceded in two and Preston scored and conceded in two. We still do not know what will happen in the remaining 42 games. The reason for following this four-step process is not that your final predictions will always be right but that over time they should be wrong by smaller amounts than other predictions.

Not got a bet365 account? Sign up today and get up to £100 in bet credits

Up to £100 in Bet Credits for new customers at bet365. Min deposit £5. Bet Credits available for use upon settlement of bets to value of qualifying deposit. Min odds, bet and payment method exclusions apply. Returns exclude Bet Credits stake. Time limits and T&Cs apply.

CLAIM OFFER HERE

MORE FREE BETS

Today's top sports betting stories

Follow us on Twitter @racingpostsport