That Confounded Double Pivot: Stats on the Central Midfield Debate

Perhaps nothing on this site, besides the great Barbecue-Grill Massacree of aught-13, has been as tensely debated as the problems in the double pivot for Tottenham Hotspur. Why was Scott Parker assigned to play as a box-to-box midfielder when he clearly does not have the skills to handle that role anymore? Parker's poorly considered runs then left Moussa Dembele, a far more dangerous on-ball central midfielder, stuck behind him as an almost purely defensive player. They never seemed to jell. It was an obvious step back from the fluid double pivot of Dembele and Sandro which had graced our midfield for much of the first half of the season.

If there's one criticism of Andre Villas-Boas that landed hardest, it's this one. Why didn't he assign Parker to hold and give Dembele more freedom to run? Why didn't he try out another pairing in midfield, particularly one involving England's Xavi, Tom Carroll? Was he too beholden to veteran players? Was he unable to see what seemed obvious to everyone on the internets?

In this article, I'm going to suggest another possible answer. Maybe he looked at the results and saw they were just as good. I introduced a plus-minus method of player analysis a month ago, looking at Spurs' performance with and without Gareth Bale. In this discussion, I'm going to use the same basic method to consider how the club played with different central midfield pairings. I've gone through the season and categorized midfield groupings as "Dembele-Sandro", "Dembele-Parker", and "Other." This isn't perfect, because I don't have time-series numbers and I have to roughly assign whole matches to one bucket or the other, even if a player was subbed off in the 70th minute. (I assign a match to the first two buckets if the two played together for over 60 minutes.) One thing that works out rather nicely for analysis--if less nicely for Tottenham's season--is that due to the fragility of the players involved, there were equal numbers of games with each of those midfields. 12 with Dembele-Sandro, 13 with Dembele-Parker, and 13 with something else.

Plus-Minus Statistics for Tottenham Midfields

As I said, the statistics do not show a significant decline from the Dembele-Sandro pivot to the Dembele-Parker pivot. I list for each group my typical power rankings statistics, with league average scaled to 100.

With Dembele paired with either Sandro or Parker in the double pivot, Spurs were an elite team, nearly the equal of the Manchester giants. With any other central midfield, Spurs were no better than West Bromwich Albion.

Now, before I get into the questions that these numbers raise, I want to consider one important confounding factor. Hugo Lloris. As Jonathan Wilson argued back in February, the replacement of Brad Friedel in goal changed not just how Tottenham played defensively, but altered the whole Spurs club front to back. Lloris' presence was nearly as important for creating the opportunities for Gareth Bale to run rampant as it was for allowing the back line to play the high pressing game that AVB wanted. (And by my analysis last month, the numbers bore out Wilson's observations.) Hugo Lloris, of course, took over the keeper job in November and only missed one league game from then through the end of the season. Scott Parker was injured the first half of the season, meaning that he only played one game with Brad Friedel in goal. So there is a bit of Lloris-induced air in the Dembele-Parker pivot numbers.

This effect, however, is seen mainly in the "other" numbers. Spurs only played three matches with the Dembele-Sandro pivot and Friedel in goal--and that included the win at Old Trafford--so the plus-minus numbers do not change too much. Since there were only four total matches where the first-choice central midfield pairings played with Friedel, I am not listing those numbers in the table. There just isn't enough data.

So the other central midfields weren't that bad, at least, when they had Hugo Lloris running the game from the back. Still, the gap is large. Dembele-Sandro and Dembele-Parker are roughly even, and everything else is quite a bit worse.

Theories and Speculations

I think this is a counter-intuitive result that demands explanation. I don't have any single explanation I'm happy with, though I don't think this is an effect that can be explained by just one cause. Here are some possible causes, along with some discussion.

• Maybe this is just random chance. 12 and 13 matches aren't huge samples. Maybe what we saw was entirely real, and if they had played 100 matches it would have become clear. I don't really like this explanation--especially when the plus-minus numbers have nicely conformed to our expectations and observations with regard to Lloris and the third-choice midfields. But random variation can never be entirely ruled out.
• Maybe this is entirely real and entirely an effect of Parker and Dembele working well together in central midfield. I do think that it's notable that both (a) AVB strongly preferred Dembele and Parker to any other midfield once Sandro was injured, and (b) by the numbers, Spurs played just as well with their second-choice midfield as with their first, but much worse with any other midfield in there. We should at least pause and consider that he saw something that we missed. I'm not sure what that would be--perhaps Moussa Dembele was particularly effective in that more defensive role, such that Parker could affect the game as a runner without Spurs losing much defensively when he lost track of his other responsibilities. I would be much happier if I could come up with a good explanation of how Dembele and Parker played together better than we thought. But this explanation deserves real consideration.
• Maybe this is entirely real, but an effect of something other than Dembele and Parker playing well together. Perhaps we're seeing here the knock-on effects of Gareth Bale changing games from a central role. If other clubs were forced to commit extra midfielders to tracking Bale in central areas, that may have opened up the center of the pitch for Dembele and Parker to be effective together, even if they weren't an ideal pairing. Maybe it's just a function of the club jelling together--though that explanation doesn't work terribly well, since Spurs were clearly worse with Dembele out at the end of the season, too. I'm guessing other people have theories here which might serve as further possible explanations.
• Maybe the numbers are missing something. While the goals scored and expected goals scored numbers are pretty consistent for each subset, the goals conceded and expected goals conceded numbers diverge significantly. Based on just real goal conceded (adjusted for schedule difficulty), I have the Dembele-Sandro pivot rated as 56 Def-, Dembele-Parker rated as 80 Def- and Other at 107 Def-. Opposing teams were producing more big chances, shots on target, and shots in the box against Dembele-Sandro, but they weren't converting them. It is plausible that the statistics aren't finely grained enough here, and defensive errors from the Dembele-Parker midfield produced "bigger" big chances and better shots for Spurs opponents. I'm hesitant to accept this in full, but it's certainly plausible.
• Even if the above explanation were entirely right, and the underlying stats were just misleading, the Dembele-Parker pivot was still very effective. Using just G/GA numbers, adjusted for schedule, I have a Team+ rating for Dembele-Sandro as 158, Dembele-Parker 140, and Other 108. So "the underlying stats aren't working" can't serve as a full explanation of the effect.

I should also note that the criticism of AVB usually also included calls for Tom Carroll to play in midfield. Since he never got a real run-out for a full match, I can't evaluate that counter-factual.