Another League Two Season Prediction
On Thursday I posted a set of forecasts based on Poisson modelling goals scored by teams in each match of the season. That’s a fearsome way to predict an entire season — 552 matches, loads of uncertainty. And it’s hard to factor in things like how the team has changed over the summer, and all of the kinds of factors that are generally believed to matter at this point.
There are a multitude of different ways of constructing a forecast, ranging from the purely quantitative, which my Thursday post was, to the purely judgemental — expert forecasts, like those by the excellent Edward Walker and D3D4Football on Twitter.
Expert forecasts have the advantage that a lot of information can be factored in — the mood around a club, the rumours, instability or otherwise, as well as the potential in signings made.
What I’ve done is somewhere in-between. I’ve taken from Soccerbase various bits of information — transfers for each team, squad details, managerial information, and historical results.
It’s very summary information — it’s not the detail in terms of who has stayed, who has gone in terms of identities, personalities, notions of player fit, mood in the dressing room, but it’s simple numbers. It’s:
A 0/1 variable for promotion and relegation, the mean age of the squad, size of the squad, mean appearances, mean goals, how experienced the manager is both overall and at the club, the number of players still at the club since last season, the mean number of seasons a player has been at the club, and last season’s position (in the full league ladder).
How do they affect a team’s final position? We can look at how changes in these numbers affect a team’s position using a linear regression. Essentially, it’s an equation of:
Position = a + b*promoted + c*relegated + …
In this expression, the letters a, b, c and so on give the relationship between league position and these numbers. In the following table, I’ve estimated the relationship:
Here, stars indicate how significant the relationship is. In other words, a lot of these things matter. A team promoted finishes a little lower down the table generally (a positive number is a higher position number, hence lower down the table), a relegated team finishes slightly higher up. Squad size and mean age are actually bad things, and experience doesn’t appear to matter. But the number of goals in a team matters a lot. An extra ten goals in the squad (on average across all players) is almost two league positions higher.
Team continuity matters, the more players continuing at the club between seasons, the higher a league position (ten players means nearly 3 league positions higher), and the longer players have been at a club the better a league position. Stability matters.
So, what does this mean for the coming season’s League Two?
Here’s the answer:
So, the model doesn’t factor in instability at ownership/board level of a club, and puts Swindon first — which seems unlikely given the instability at the club this summer. Salford at the top seems reasonable given their spending, and Bradford too.
At the other end, Leyton Orient absolute bottom is surprising, as is Forest Green in 18th and Exeter in 21st. Less surprising is Barrow (23rd), Hartlepool (22nd) and Mansfield in 20th.
Oldham, who spent much of last season in 17th, are 17th, unsurprisingly. This is slightly better than my Thursday prediction, where 23rd was most likely.
However, these are all not opinions — they are a regression equation, estimated on data. The data table with the numbers for each team will yield some light on why they are in particular positions.
They are also not facts. There’s a very high chance all 24 will be wrong. My Thursday forecasts showed that at this point, for most league positions there’s usually less than a 10% chance a team finishes there.
There’s just so much uncertainty ahead, but it’s all food for thought…
