Previewing League Two, 2021–2022
On Saturday afternoon, another football season in England begins. Only two World Wars have got in the way of this late Summer/early Autumn tradition since 1888 when the Football League was founded. In 1921 the Football League expanded to 4 divisions with northern and southern third divisions, and in 1958 assumed its current structure, albeit with different names — four nationwide divisions.
So, in this 64th year of English Football’s League basement, what will happen? Well, aside from 552 matches taking place (though even that’s not guaranteed after the 2019/20 season), it’s very difficult to say much else.
Because none of those 552 matches have a guaranteed outcome. At most a team will have a 60–70% chance of winning a match, and even then such teams don’t always win. Nonetheless, we still want some kind of sneak peak into how things are going to turn out.
So, starting with the opening set of fixtures, I’ll talk you through one way to make some predictions about eventual season outcomes. Below is a table with a prediction of the outcome of all 12 of Saturday’s fixtures. On the left are the 3-letter Reuters codes for each team. My own team, Oldham Athletic, is OAT.
Following this, the expected goals each team will score is presented, and based on this, the most likely goalscorers on Saturday are Harrogate (1.8), Salford (1.8), and Newport (1.6).
This is generated from a statistical model of likely goalscoring. The model assumes that the number of goals scored by each team is a function of the strengths of the two teams (measured by Elo ratings), some measures of recent form (in match outcomes, goal scoring and goal conceding), and that is augmented by a variable that tells the model whether either team was promoted or relegated last season, a variable that describes the relative difference in tenure of each manager in the match, and a variable that tells the model if either the home or away coach is playing their first match.
That model then yields a set of probabilities of different possible scorelines. Historically, about 11% of matches have finished as 1–1 draws, and as such, it’s a very likely scoreline. Amongst the matches on Saturday, all have a likelihood of 1–1 of between 11% and 13%.
We can sum up all of these scoreline probabilities to get the total probability that each team wins, and we present that in the fifth and sixth columns. Hence Oldham are 34% to beat Newport, while Newport are 41.8% to win. For comparison purposes, the final two columns are average bookmaker odds for these outcomes, taken from Oddsportal.
What we can do then is take these modelled probabilities (forecasts), and generate match outcomes according to them for the whole season. That is, given Oldham have an expected number of goals of 1.41, we generate a draw from the Poisson distribution with a lambda parameter of 1.41, and take that as the number of goals Oldham scored in their match on Saturday.
One such iteration of doing that gave 2 goals for Oldham, and 0 for Newport. Both are fully consistent with expected goals of 1.4 and 1.6 respectively. Another iteration gave a 1–1 draw, unsurprisingly. We then do that for all matches each day, and update the information for the model — the new team strengths and goalscoring strengths shown. We do it for all matches of the season, iteratively, and that counts as a replication — a fully simulated Football League season.
If we create lots of these replications, then statistically speaking, we can start thinking about the observed frequencies of events occurring (teams to win promotion, get relegated, etc), as “frequentist” probabilities of the outcomes — forecasts.
We can be more refined than that — we can ask what is the probability that any given team finishes in a particular position? The answer, generally, is quite low, but you can see for yourself in the table below — each column is a team, each row a league position, and in each cell is the probability that team finishes there.
So Tranmere Rovers (TRR) have a 14% chance of finishing first — winning the championship. That’s not controversial since Tranmere were strong last season in their first season back down in League Two. That Sutton United (SUU), just promoted into the League from the National League, have a 20% chance of winning League Two is, however.
It’s here where I hoped that controlling for whether a team is promoted or not would reduce the likelihood of this. Sutton are strong in terms of Elo rating thanks to their promotion-winning campaign last season. However, playing against stronger teams week in week out in League Two will be a bigger test for them. Perhaps they will flunk. 80% of the time they won’t finish 1st, indeed, there’s a 30% chance they won’t make the playoffs. But they may do.
After all, Morecambe did much better than most expected last season. However, others more sensibly have Sutton to finish much more modestly this time out. Other sets of predictions, actual league position picks, are likely decided based on information processed intuitively by the pundit, or expert. A bigger range of information can be considered, but equally, that information may not be applied systematically across all 24 teams — one of the advantages of a statistical approach.
That distinction is perhaps most keenly felt for a club like the one I follow, Oldham Athletic. This is a club that has gone through a lot of managers in a short space of time — a mere 15 since 2013, the last time a manager managed more than a calendar year in the hotseat. In that time the club has undergone a change in ownership, albeit a seemingly messy one, a relegation, a huge turnover in playing (and non-playing) staff, has been served winding up orders, and placed under transfer embargoes by the Football League.
Last season they finished 18th, and although the current occupant of the hotseat, Keith Curle, has been in place since March, mid-season changes have been a regular feature of recent years at Boundary Park. Hence, understandably, Oldham’s most likely finish is 23rd — relegation, at 9.2%. But note that that means there’s a 90.8% chance they don’t finish 23rd.
Indeed, there’s a 2.2% chance they occupy one of the top three promotion places, and a 4.4% chance they finish 11th. In other words, a lot is possible, and there is a huge amount of uncertainty ahead of us over the coming season.
Below is a simpler table that summarises the likelihood of the three major possibilities for a team in League Two: promotion, the play-offs, and relegation.
So although Oldham have a 18% chance of relegation, they also have an 8.2% chance of making the play-offs, as well as that 2.2% chance of automatic promotion — that is, a 10% chance of a top-seven finish.
Very soon, everything will be post-hoc rationalised. After the event, it’s very easy to rationalise why everything happened exactly the way it did (and why that outcome fits with your opinion about how the world is). But it’s important to recognise that, objectively, there is a huge amount of uncertainty about the next ten months in the basement division of the English Football League.
