Following last week’s post detailing how I have created a betting a model using expected goals data, link below, I will be putting that into practice in an attempt to predict the outcome of the League Two play-offs.
https://eflnumbers.wordpress.com/2020/06/11/how-to-create-a-betting-model-using-expected-goals-data/
The season ended prematurely with Swindon Town, Crewe Alexandra and Plymouth Argyle filling the automatic spaces and a spot in League One next season. This leaves the next four placed teams, Cheltenham Town, Exeter City, Colchester United and Northampton Town to battle it out for the remaining promotion place via the dreaded play-offs where only one will be ultimately be successful.
Cheltenham Town
| # | Team | G | W | D | L | GF | GA | GD | Pts | Pts | xGF | xGA | xGD | xPts | xPts | xG# |
| 4 | Cheltenham Town | 36 | 17 | 13 | 6 | 52 | 27 | 25 | 64 | 1.78 | 45 | 41 | 4 | 51 | 1.41 | 9 |
Cheltenham Town grabbed 4th courtesy of their superior points per game finishing just ahead of Exeter City who achieved one point more from one fewer match. They may have ended up with 64 points and a +25 goal difference but my expected goals data suggests they over achieved this season. The model calculates an expected goal difference of just +4 with an over performance at both ends of the pitch scoring 7 more and conceding a whopping 16 fewer than expected. This ranks the Robins as the 9th best team this season.
Cheltenham’s form was generally strong throughout and they didn’t experience back-to-back defeats at any point. In fact they finished with the fewest defeats (6) of all 24 teams. The two stand out performances, as rated by expected goals, came at the Jonny-Rocks Stadium (Whaddon Road) in home games against Scunthorpe United (won 4-1) and Leyton Orient (won 2-1). The worst performance came in the very first game of the season, away to the aforementioned Leyton Orient (lost 1-0), where they created next to nothing.
Whilst form was strong throughout it’s noticeable how the bookends of the season achieved the greatest points hauls. The 11 games in the August – September returned 21 points and 3rd place before a slight slump in the winter months saw the Robins drop into the play-off places and just outside for a temporary period. A strong finish with a 6-1-1 W-D-L record from February onwards would have given the fans dreams of the automatic spots before the premature end.
Michael Duff has almost religiously played a 3-5-2 this season and it would be a surprise for this not to continue into the play-offs. Three other formations have been used, and the Robins have never been beaten in those matches but expected goals data would suggest it’s best to stick to the known.
Exeter City
| # | Team | G | W | D | L | GF | GA | GD | Pts | Pts | xGF | xGA | xGD | xPts | xPts | xG# |
| 5 | Exeter City | 37 | 18 | 11 | 8 | 53 | 43 | 10 | 65 | 1.76 | 54 | 44 | 10 | 56 | 1.53 | 5 |
Exeter City finished with the 4th highest points haul but were ultimately relegated a place due to points per game. In this case my expected goals data suggests they finished exactly where they should have. A goal difference of +10 perfectly matches the expected goals estimated, good enough for a 5th place ranking for the Grecians.
Exeter never won more than 3 games in a row but there was a steady stream of them which saw them finish with the most victories of the play-off quartet (18). There was a great three game performances, as measured by expected goals, early in the season in matches away to Carlisle (won 3-1), home to Leyton Orient (drew 2-2) and home to Port Vale (won 2-0). The worst performance, away to Forest Green, still resulted in a victory.
Highlighted above Exeter started the season strongly with 22 points in the opening 11 games and a long stint on top of the league before ending September in second place. 3 defeats across the next 8 games dropped the Grecians to 4th before a strong Christmas period and a W-D-L record of 6-3-1 returned them to 2nd and the automatic spot. The run didn’t continue and they must be ruing how the season ended with no win in four confirming it would be a play-off spot at best.
Matt Taylor started the first three games in a 4-4-2 formation and although this returned 7 points, the expected goals data didn’t point to the same domination and therefore it was not a surprise to see the team switch shape. A 3-4-1-2 was successfully used thereafter but was tinkered in favour of a 3-5-2/5-3-2 shape later on in the campaign. The expected goals data highlights the 3-4-1-2 to be the best option of the trio.
Colchester United
| # | Team | G | W | D | L | GF | GA | GD | Pts | Pts | xGF | xGA | xGD | xPts | xPts | xG# |
| 6 | Colchester United | 37 | 15 | 13 | 9 | 52 | 37 | 15 | 58 | 1.57 | 51 | 39 | 12 | 56 | 1.53 | 6 |
Seven points adrift of the first pair was Colchester United who finished in 6th, ahead of Northampton Town courtesy of a superior goal difference. Akin to Exeter City, the expected goals performance (+12 xGD) closely aligns with the actual goals performance (+15 GD). Ultimately this was enough to be ranked as the 6th best team in the league.
Colchester United started the season with 2 points from their first 4 games and must have been thinking it was almost season over but for a 16 game unbeaten run, the longest of the play-off four, later in the campaign. Their was a clear winner for performance of the season, as calculated by expected goals, coming in the 3-1 home win against Stevenage. There weren’t many bad days though.
The poor start saw Colchester drop as low as 21st but they managed to end September in 9th. A constant supply of points across the middle part of the season saw a peak of 4th on Boxing Day. The U’s were 5th by end of January but will be disappointed by the end of the campaign finishing with a W-D-L record of 3-0-4 in the last 7 games.
Jon McGreal has typically used a 4-2-3-1 throughout the whole season, and has only started matches in three different shapes, the fewest of the play-off teams. It’s interesting that a 4-4-2 has only been used 5 times but has picked up 10 points in those matches. This has reduced the number of goals scored but has made them a much solid proposition defensively backed up by the expected goals data.
Northampton Town
| # | Team | G | W | D | L | GF | GA | GD | Pts | Pts | xGF | xGA | xGD | xPts | xPts | xG# |
| 7 | Northampton Town | 37 | 17 | 7 | 13 | 54 | 40 | 14 | 58 | 1.57 | 51 | 52 | -1 | 50 | 1.36 | 12 |
A 7th place finish and the remaining play-off spot went to Northampton Town. The Cobblers finished with a +14 GD but this is considered to be a large over performance based on the expected goals data, which estimated a -1 xGD across the season. This was largely due to the team conceding 12 goals fewer than suggested. The model ranks Northampton Town as the 12th best team in the league and therefore the weakest of the play-off quartet.
Northampton were another who started slow with 1 point from the first 3 matches. The Cobblers were beaten 13 times in all, the most of the play-off teams, which meant they ultimately didn’t draw very often to achieve the 7th place. There will be encouragement in the fact that one of their best performance, as measured by expected goals, was against a fellow promotion rival in the 2-0 victory against Exeter City. There was also the odd stinker though, notably the 4-0 drubbing at Crawley Town.
Northampton have had the trickiest journey in securing their play-off spot, by September they were still in the bottom half in 13th. Two defeats at the start of October saw them drop further to 18th before a very impressive 10-4-2 W-D-L record until the end of January rose them to 6th spot and into the play-offs. In the end it was another case of a disappointing February – March period, the poorest of the play-off hopefuls, where they just about managed to hold onto 7th.
Keith Curle started the season with a 4-2-3-1 but a poor start ultimately led to a reshuffle with Northampton using the most formations (8) of the four play-off teams. It’s interesting the most popular formation and the one used at the end of the season, a 3-4-1-2, was not used until mid November. They conceded few than one goal a game whilst using this set up but the expected goals data implies they were fortunate not to concede more.
Exeter City (5th) v Colchester United (6th)
The play-offs kick off on Thursday evening with Colchester United hosting Exeter City at the JobServe Community Stadium before the return leg at St James Park four days later.
The teams have faced each other twice in the shortened season. Firstly on the 29th December 2019 in an open 2-2 draw with Ryan Bowman’s penalty late into the second half earning the visitors a point. Exeter had the better of the chances and arguably should have taken all three points though Colchester were content to sit back at 2-1 up.
The reverse fixture on the 25th January 2020 also ended in a draw. Second time round the game was a little tighter with the teams cancelling each other for a deserved stalemate with the best chance falling to Colchester’s Callum Harriott shortly before half time.
Finishing 5th and 6th in both the actual table and my expected goals table, as well as two tight games this season indicates this is a close match-up. To calculate the expected match probabilities I use a rolling 46 game season of data which ordinarily would be a complete season but in this instance uses a handful of games from the season before.
A poisson distribution estimates the following probabilities for the number of combined goals scored across the ties.
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | |
| Exeter City | 10% | 22% | 26% | 21% | 12% | 6% | 2% |
| Colchester United | 7% | 19% | 25% | 22% | 14% | 7% | 3% |
My model has Colchester United as favourites to qualify for Wembley, a reversal of the bookmakers odds, and is the suggested selection for this match up.
| Modelled Probability | Bookmakers Odds | Bookmakers Probability | Difference | |
| Exeter City | 44.7% | 5/6 | 54.5% | -9.8% |
| Colchester United | 55.3% | 11/10 | 47.6% | 7.7% |
Cheltenham Town v Northampton Town
Thursday’s action concludes with Cheltenham Town travelling to Sixfields to face Northampton Town with the reverse tie also four days later at Whaddon Road.
There’s been two heads to heads this season with the first also happening on the 29th December 2019 with the teams playing out an interesting 1-1 draw. Conor Thomas scored early for Cheltenham and they effectively shut up shop thereafter, likely to be the same tactic on Thursday should history repeat itself. Jordan Turnbull equalised just before half time and the hosts continued to push into the second half but didn’t manage to find a way to score a second.
Cheltenham Town hosted the return on 25th February 2020 securing a 2-1 win identical to the estimated expected goals data. It was again the visitors who scored early but this time the home team levelled a lot earlier. Both teams had minor chances until Luke Varney’s close range goal won all three points.
This looks the weaker match according to the expected goals with neither performing well enough for a play-off spot in my table. Cheltenham took 4 points from the head-to-heads this season but the match data implies there was very little between the teams with both deserving to win the home ties. As per the other play-off I will calculate the expected match probabilities using a rolling 46 game season of data which ordinarily would be a complete season but in this instance uses a handful of games from the season before.
A poisson distribution estimates the following probabilities for the number of combined goals scored across the ties.
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | |
| Cheltenham Town | 6% | 18% | 24% | 22% | 15% | 8% | 4% |
| Northampton Town | 7% | 19% | 25% | 22% | 14% | 7% | 3% |
My model has Cheltenham Town as slight favourites to qualify for Wembley, mirroring the bookmakers odds, but no value for a betting selection.
| Modelled Probability | Bookmakers Odds | Bookmakers Probability | Difference | |
| Cheltenham Town | 52.4% | 5/6 | 54.5% | -2.1% |
| Northampton Town | 47.6% | 21/20 | 47.6% | 0.0% |
Outright
Given how the expected goals data rates the four teams it is no surprise that it thinks the winner of the Exeter v Colchester United game would be a 60-65% favourite to win promotion against the other two teams.
The model thinks Colchester United are most likely to go up and are estimated to have a 1 in 3 chance to do so. The bookmakers are a little more pessimistic at 3/1 (or a one in 4 chance) and therefore these are the suggested value play.
| Modelled Probability | Bookmakers Odds | Bookmakers Probability | Difference | |
| Cheltenham Town | 17.6% | 3/1 | 25.0% | -7.4% |
| Exeter City | 29.0% | 9/4 | 30.8% | -1.8% |
| Colchester United | 34.9% | 3/1 | 25.0% | 9.9% |
| Northampton Town | 18.5% | 10/3 | 23.1% | -4.6% |
Recommended Bets
Colchester United (11/10) to qualify for Wembley
Colchester United (3/1) to win promotion
EFL Numbers 