In the first installment of The Hinrunde Review, the findings suggested that Dortmund are capable of challenging the very best in the league (Bayern) on their day, but their inconsistency means that they are ultimately some way off both Bayern Munich and RB Leipzig. But what causes Dortmund’s inconsistency?
Using data taken from Understat, FiveThirtyEight, and WhoScored, I will analyze what causes Dortmund’s inconsistency on the pitch, and consider the underlying explanations for their failures. I will argue that the prime suspects are the dreaded “mentality problem” and Lucien Favre’s conservative approach to game management.
In order to do this, I will first look at performance issues in the first half of this season, before broadening my scope and considering Favre’s tenure as BVB manager. Finally, if only because an entire article dedicated to BVB’s issues is depressing, I will finish on an optimistic note.
Taking a Lead and Keeping It
It feels like Dortmund have made a habit of squandering advantages, and when looking through the Hinrunde, it is clear that is the case. Dortmund have taken the lead over their opponent a total of 13 times, but they average only 2.15 points from winning positions. They have won only 8 of those 13 games, drawing 4 and losing 1. In comparison, Bayern average 2.36 points, Leipzig average 2.83, and remarkably, Borussia Monchengladbach have not dropped a single point from a lead.
Gladbach also lead the average points table from deficits, averaging 1.29 points from losing positions, joint top with RB Leipzig. Dortmund are 4th, winning an average of 1.14 and Bayern are 5th, winning an average of 1 point.
The Foals appear to be incredibly ruthless, and Leipzig are not far behind them. Being able to maintain an ironclad grip on leads, while also giving yourselves some hope from deficits, is an impressive feat. Dortmund, on the other hand, are incredibly brittle when they’re in the lead. If Dortmund held on to winning positions at a similar rate to Leipzig, they would currently be top of the table.
Performances in Different Game States
In order to dig a little deeper, I compared the performances of the top four teams in different contexts: when drawing, when winning or losing by one goal, and when winning or losing by two or more goals.
Compared across different states, all four teams look reasonably similar. However, Dortmund perform considerably worse than their rivals when they are leading by a single goal. Figure 1 plots expected goal difference per 90 (xGDiff/90) during the different game states.
When winning by one goal, Dortmund’s xG90Diff/90 is -0.63. Compared with the rest of the top four, the issue becomes even more glaring. When up a goal, Bayern’s xGDiff/90 is 2.01, Leipzig’s is 1.96, and Gladbach’s is 1.05.
Things don’t look that much better when considering actual goal differences per 90, as shown by Figure 2. When up a goal, Dortmund goal difference is 0, while Bayern’s is 1.33 per 90, Leipzig’s is 2.83 per 90, and Gladbach’s is 2.67 per 90. When the other three teams take a lead, they also take control of the game, ramp up the pressure, and make things harder for their opponent. When Dortmund take a lead they become much more vulnerable.
To further compound the issue, the timing of goals seems to suggest that Dortmund are worse in the second half of games, right after half time (46-60mins) and at the end of games (76mins+). In each of these periods, Dortmund have conceded 6 goals, which is more than any other period. These two periods in games are the only points at which BVB’s xGDiff is negative too, with the former being -1.14 and the latter being -0.17.
Errors and their Effect on the Outcome of Games
The way Dortmund perform under certain conditions and at different points in games seems to suggest that they have a tendency to capitulate when they can least afford to. Performing poorly and conceding late in games, especially when leading by a goal, is hugely costly. On top of this, Dortmund also commit a huge amount of errors, the vast majority of which occur in the second half of games.
Using data from WhoScored on individual errors leading to shots or goals, I compared the number of errors committed by Dortmund players and their rivals. I used a welch’s t-test to test whether the average number of errors Dortmund make is different from the average number of errors made by the rest of the top four in the last two seasons, producing statistically significant results. This means that the number of errors Dortmund made is significantly different to those made by their rivals, rather than simply being the result of normal variance or chance.
Having established that the number of errors Dortmund made is statistically significant, I used an ordered logistic regression model to test the impact that errors had on the probability of Dortmund results, including goals scored and NsXG as control variables. This simply tests whether errors have a statistically significant (not occurring by chance) effect on the number of points Dortmund win, holding the goals scored and NsXG constant. The findings suggest that errors have a statistically significant and substantively large negative effect. This means that the large number of mistakes that Dortmund make don’t go unpunished and are not negated by performances at the other end of the pitch. Instead, they significantly increase the probably of losing. Using the regression results, I also computed the predicted probability of match outcomes based on the number of BVB errors. All else being equal, a Dortmund mistake reduces the predicted probability of BVB winning from an average of 0.86 to 0.59. The second error, however, has a massive impact, reducing the predicted probability from 0.59 to 0.15.
Finally, I plotted the impact that errors have on the probability of BVB match outcomes, depending on the number of goals Dortmund score in the same game, presented in Figure 3.
Unsurprisingly, the more goals Dortmund score, the more likely they are to win. The probability of a loss is highest when they’ve scored 1 goal, the dark blue line, while the probability of a win is highest when they’ve scored three goals, the red line. The effect that an error will have on the probability of a draw is highly dependent on the number of goals they score. For example, when Dortmund score one goal, a single error increases the probability of a draw, because that error may lead to an equalizing goal, but multiple errors significantly decreases the probability of a draw because this is likely to lead to a loss instead.
While it may not be hugely surprising that errors have a significant impact on match outcomes, I think the magnitude of the effect they have on Dortmund’s probability of winning games, and the fact the number of errors they make seems to be significantly greater than their rivals, is particularly noteworthy.
A Glimmer of Hope? Improvements When Playing in a 3-4-3
Despite all this, Dortmund have improved in recent weeks, and those improvements seem to coincide with the switch to a 3-4-3. I analyzed the impact that the formation change has had on Dortmund’s performances, despite a very small sample, and while the errors haven’t disappeared, there are some real signs for optimism.
When lined up in a 4-man defense (typically a 4-2-3-1, but not exclusively), Dortmund averaged an xGDiff/90 of 0.21, compared to 1.05 in the new formation. They have also increased their key passes and shots on goals in a 3-4-3. Perhaps most notably, Dortmund have produced a much higher NSxG in recent weeks. The mean NSxG when playing in a 3-4-3 is 2.46, while in formations using a 4-man defense it is 1.6.
It is hard to draw too much from this due to the very limited sample size, but I think that there are good football reasons to believe that these results may continue. In a 3-4-3 Dortmund are able to get more attacking and creative players in the game. The formation affords the wing-backs more opportunity to provide width in attack, puts Brandt at center stage to start pulling strings, and gives Reus, Sancho, and Hazard a great deal more freedom to play interchangeably in the final third. So far that has produced more shots, more goals, a higher xG and higher NSxG, all while conceding around the same amount. If Dortmund can’t stop making errors, perhaps they can just score so many goals the errors don’t matter anymore?
Dortmund have a frustrating habit of orchestrating their own downfall. They make errors that allow opponents back into the game when they should be piling on the pressure and seeing out results. The findings suggest that Dortmund are vulnerable when they possess a one goal lead, especially in the second half of games. This in and of itself is obviously a real problem, but in this light, Favre’s typically conservative approach to game management looks particularly damaging. Instead of looking to finish opponents off, Favre seems content to try and defend small advantages for long stretches, despite the reality that Dortmund are particularly bad at doing so.
The errors are not a new thing. Dortmund committed 15+ errors in the three seasons preceding Favre as well, which suggests that he is not to blame for this particular problem (though he hasn’t fixed it either). However, his conservative game management makes Dortmund much more vulnerable when errors do occur. His conservatism has become a bigger problem this season, in part because Dortmund were playing so well last season that it papered over the cracks, but they still conceded more goals and they still threw away a number of leads. A conservative approach to managing games is clearly inappropriate for a team that makes so many mistakes. We know BVB can score goals. Why not go for broke?
With that said, there is a glimmer of hope. Obviously we can’t draw too many conclusions from a small handful of games, however, it is at least promising that performances when playing in a 3-4-3 have been so good. It at least justifies exploring the 3-4-3 further.
Next, we will look back through the seasons to get some perspective on this season’s performances, to see how much of this is on Favre, and whether he has significantly improved the team’s performances in his time as manager.