/cdn.vox-cdn.com/uploads/chorus_image/image/64008235/Rapid-33.0.0.jpg)
Change: inevitable, almost universally disliked. All I know is that change comes in ones, twos, or threes between the 8th and 90th minute. Over the course of two and a half seasons. At time of writing, the Loons have made 224 substitutions to turn losses into wins, wins into losses, and losses into slightly different losses. I looked at every one of them to see what patterns emerged from the Adrian Heath inspired rubble.
I started by building a data set of each substitution the Loons made in every MLS game. I excluded friendlies and Open Cup games as the data on those was harder to find reliably and the tactics in friendlies are different that in regular season games. Each entry consisted of the opponent, final score, time of sub, player on and off, score at time of sub, change in score between the sub and the final, and movement from sub. Midfielders were given an index of 0, attackers are 1, and defenders are -1. The difference between the index of a player on and the player off represents the mentality shift, i.e. Mason Toye, a striker, coming on for Brent Kallman, a defender, would be scored as 2, the inverse substitution would be -2.
Who is being subbed?
The first thing I was curious about was who exactly was being substituted. Defenders are rarely substituted, save for red cards or injuries, goalkeepers even more so. A defender generally doesn’t get subbed out for having a sub-par game, but attackers regularly do. They, along with midfielders, often run more than defenders and need more subs. That would explain why the most subbed off defender (to use that term loosely) was Alexi Gomez with eight, followed by Kallman and Miller with six apiece. The most subbed off Loons player ever? That would be none other than Christian Ramirez himself. Of the 44 games he started over the his year and a half as a Loon in MLS, he played 90 minutes just 23 times and was subbed off 21 times. He had six more appearances where he was subbed on.
Speaking of who is subbed on, the player that has been subbed on the most has been Abu Danladi with a whopping 31. His closest competitor is Mason Toye with 16 appearances off the bench. For reference, Rodriguez and Quintero have five and two apiece, so their contributions are really more or an all or nothing deal. Romario Ibarra is tied for 5th with 11 substitute appearances, owing to the congestion on the wings and at striker, but his speed did make him valuable late in games.
With Ramirez and Danladi topping the off and on charts, it would only make sense for them to be the most frequent pair replacing each other as well. Danladi replaced Ramirez six times and Ramirez replaced Danladi four times, bringing their total to 10 and narrowly pipping the combination of Toye on, Ramirez off, with eight. The most frequent pairs that didn’t involve Danladi or Ramirez were Warner for Ibson, Finlay for Schuller, and Warner for Martin, all occurring four times.
More importantly, why are they being subbed?
The rational move a team would make from a winning position would be bring off attackers for midfielders or defenders. A team that is losing would bring on positive reinforcements. Right? Right. That’s the easy part. The tricky part comes from the subs made when the game is tied. Do you sit and bunker and try to steal a point, or do you go all out and snatch a all three? Do you wait until the last possible moment to make your attacking sub and just bring on neutral subs to bide your time?
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/16313400/movement_by_sub_gamestate.jpg)
At first glance, the figures seem to make sense, defensive subs when you’re winning, offensive subs when you’re losing. All is right in the world.
The Loons have also chosen to bunker when drawing, behavior that would be consistent with a team willing to escape with just a point wherever they can get it. From this very limited sample of two and a half years, it looks like last year was a weird one from the standpoint of Heath’s decision making. A score of -.46 when drawing, but only -.25 when winning doesn’t make a whole lot of sense. That would go in line with looking for big wins, but also desperate to lock down a single point, not go for a win. The .45 score when losing would show that he’s more likely try to win or draw when losing than win when drawing. Bigger losses don’t mean much in the standings, aside from goal difference, so throwing everything forward in search of 1 point rather than 0 makes sense, but the chance to take 3 instead of 1 must be at least a little attractive.
Also important, when are they being subbed?
Of course, all this is predicated on the idea that subs are on the field long enough to make an impact, and aren’t just brought on to waste a few seconds and interrupt the flow of the game.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/16313471/average_sub_time_and_type.jpg)
In games the Loons are winning, three sub are all squeezed into the space of about 15 minutes and are on the whole, defensive. Again, no surprises there. During losses, the first two subs come in the 60th and 70th minutes, not too different than the standard, run of the mill tactics there, but it takes until almost the 80th minute before any drastic changes are made, and even then there’s only about 15 minutes left to make an impact. Long enough to score a goal, yes, but something that could have come much sooner.
The pattern of subs during draws looks eerily similar to the pattern for wins. That sure looks a lot like a coach treating draws like wins, and if they aren’t throwing in the towel with a defensive sub around the 65th minute, they surely are with another defensive sub at the 83rd minute. For a team fighting for the playoffs this year, and needing every point they could get the past two, trading a shot at 3 points for not even a guarantee of 1 isn’t playing your odds well.
What about when it works? Or doesn’t...
Out of the 84 games played, there have been 41 games total where the Loons have taken points, 27 wins and 14 draws. Of those 27 wins, only eight have come from games the where a substitution was made before the Loons went up. In six other games the Loons rescued a point to turn a loss into a draw. Total, that makes 14 games out of 84 where subs were used to change the game. If only including games with points that’s 14/41, or 34% of point winning games where someone outside of the starting 11 was on the field at a crucial time to help the team gain points.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/16313584/points_gained.jpg)
On the flip side, there have been ten games that have been, according to the model, mismanaged and subs have cost the Loons points. In both samples, only the worst case scenario was counted. If a sub was brought on at 1-0 Loons, another at 1-1, and they lost the game 2-1, it would go down as -3 points.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/16313587/points_lost.jpg)
It total, the Loons have gained 41 points from substitutions and lost 22. In a vacuum, that looks pretty good. Subs generally come in late in the game and if Minnesota United isn’t dropping a lot of points because of subs late in games, then the subs are doing their jobs, right? That would be true if the Loons overall record was positive. Their total record so far is 27 wins, 14 draws, and 43 losses. The implication of this is that games are being lost early. 41 points gained from subs is good, but there are so many more opportunities where they haven’t made an impact because the game is too far gone.
So now what...
In the end it comes down to draws. The substitution behavior during draws has been overwhelmingly negative to try to lock down a single point. Minnesota United has been conceded goals while drawing late in games while trying to lock the game down with defensive substitutions and it hasn’t been working. When drawing you can either go on to win(+2), remain drawing(±0), or lose(-1). Defensive substitutions try to avoid the last option and the Loons have been bad at it. This extreme risk aversion has failed more times than not an invites pressure onto a tired, leaky defense.
Essentially, a late tied game is just a shorter game, 10-20 minutes instead of 90. Shorter games allow more to chance and the underdog to prevail, “Any given Sunday...” and “If we play them 10 times they might win 9” kind of thing. If the Loons really are going to stay above the playoff line this year and host a playoff game in MLS’s most glorious stadium, draws are not going to cut it. They’ll need to be a bit more bold and a bit more brave in their remaining 20 games to close out the season, and be ruthless when the opportunities present themselves.
Caveat about Miguel Ibarra:
Batman...Utility Belt...Utility Player...?
One potential issue with this analysis could be Miguel Ibarra. Ibarra could have fallen into any category due to his versatility, but was included as a midfielder. Ibarra coming on for Cronin would be an attacking change, yet would register no change in the model, Ibarra on for Finlay, a neutral change depending on how the game is going, would register -1. Ibarra has taken part in 34 substitutions(14 on, 20 off) and the average movement resulting from those changes came out to be about 0, which seems right given how he’s been used. A couple other players had that issue as well, like Alexi Gomez, Jose Leiton, Hassani Dotson, and Eric Miller, but they made up a much smaller portion of the dataset.
If you want a look at my data set and the Jupyter Notebook I used, it can be found here.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/16313489/all_players.jpg)