As argued in one of my previous blogs, building on the “Decision Rule” from Myers, Premier League managers could increase their chances of turning around a game if they would make earlier substitutions. As can be seen here managers on average implement their first, second and third substitutions in the minutes 62. 72 and 81 respectively.
Mourinho does no deviate much from that average (62, 73 and 82), but did something interesting last month: two to one down in the Manchester derby he decided to throw in two fresh players at half time. Jesse Lingard came on for Marcus Rashford and Ander Herrera was replaced by Henrikh Mkhitaryan. Interesting. It follows the logic of making player changes early when losing, but this was a but rigorous. It does not seem to happen that often that a manager half way decide to change 2 of their 10 outfield players. It did not help in this case though. The score remained the same: 2-1 for the visitors. (Although you could argue they lost the first half 2-1 and after the changes drew the second half 0-0 and thus improved.)
In 1520 games (involving 3040 teams) in the Premier League seasons 2012-13 until 2015-16 a double substitution was made 70 times. Interesting enough, but also quite logically maybe, none of those 70 double substitutions was made in a winning game state. Alan Pardew, quite used to finding himself in a negative game state, was responsible for 9 of those 70 double changes.
Here is the distribution with the effects on the results and game state (win, draw or loss):
As you can see from the above table the double substitution led to an increase in game state in a fair amount of cases, 17 of the 70 (24%) to be exact. When brining on two fresh players at half time when drawing, the final result delivered a win in 7 of the 15 games. When losing by 1 goal at the break and doing a “Pardew” this gave 24% (7/29) of success. And even when 2 goals down after 45 minutes there were 3 teams that came back to get at least a point after leaving two starters in the dressing room at half time.
But how does this compare to similar situations? Are those percentages higher or lower than when not making the double sub? Let’s see:
Without further analysis we could compare the effect of the double sub for three different half time results. Over the last four season the only comeback from 2 goals down came after making a double substitution at half time (Southampton beating Liverpool 3-2 in March 2016). And the percentage for improvement in game state (14%) is higher in our test group than in the standard games (9%), but sample size is very low here. For a -1 result at half time the contrary is visible (24% vs 30%), which leaves an ever greater doubt over the previous case. In the ‘drawing’ situation there seems to be an advantage for the ‘double sub’ group again (47% against 35%), but again, small sample size.