The Drunkard’s Search: Analytics, #confidence, and Observing the Unobservable

“There is the story of a drunkard, searching under a lamp for his house key, which he dropped some distance away. Asked why he didn’t look where he dropped it, he replied ‘It’s lighter here!’. Much effort […] in behavioural science itself, is vitiated, in my opinion, by the principle of the drunkard’s search” – Abraham Kaplan (1964)

One of my favorite memes of Soccer Analytics TwitterTM is attributing anything strikers do out of the norm to #confidence. It’s a truly awful cliche that announcers fall back on far too easily, and is deservedly mocked. On the other hand, confidence is a very real thing, especially for athletes as seemingly mercurial as world-class strikers. I remember Fernando Torres at Chelsea on a breakaway with only a single defender and keeper between him and a goal. “Old Fernando Torres” would have done an ankle-check on the defender and slotted the ball in the upper left corner of the net, but unconfident Fernando Torres sort of flicked the ball up into the defender’s arm hoping for an incredibly soft penalty (which wasn’t given). This was clearly a case of a formerly world class striker lacking confidence. I saw the same thing last weekend in the Swansea game where Gomis had a one-on-one opportunity where he slowed to a crawl, clearly over-thinking what his next move was to the point where a defender caught him and tackled the ball away. Contrast this against a confident striker in another game, given the same opportunity, making a single quick move, shooting, and scoring an easy goal. There’s a middle ground between announcers attributing everything to some mystical force, and analytics types ignoring it because we have no way of measuring it.

The biggest weakness of behavioral research is that it struggles to identify the “intangibles.” All we can see are outcomes, and often the process ends up being ignored. In political science, up until Lau and Redlawsk (and later Lodge and Taber) we spent a lot of time studying how people vote but not nearly enough studying how they decide to vote the way they do. Votes are observable, the process that gets people there isn’t, so in classic drunkard fashion that’s where political scientists looked.

Trey Causey has been railing against (American) football analytics’ focus on win probability models, but there’s a reason we do this (beyond gambling). Wins are observable, and models can be assessed fairly easily given a large enough sample. Soccer analytics likes Expected Goals models because again the outcomes are observable, making it another relatively easy analysis task.

As someone whose major contribution (at least not during transfer windows) is win probability, I agree with Trey that we need to take the next step and build on win probability models to improve decision-making. This is difficult, in no small part because so much of what we do is hidden from sight, all we have are observable implications of unobservable concepts. This is much trickier, and may be another way to move the ball forward in soccer analytics.

Using #confidence as an example, let’s assume for a minute we can’t survey players and ask them how confident they are. Then how can we measure confidence? The next step would be to ask “what does a confident player look like?” Off the top of my head, and in no particular order, confident strikers might do the following things:

  • Shoot rather than pass1
  • Take low ExG shots more frequently than unconfident players
  • Take more first time shots
  • Shoot rather than passing, even when a potentially easy assist presents itself
  • Move toward goal more than away from goal
  • Shoot more frequently when defended

There are undoubtedly more, but I feel that this is a good start and shows how maybe we can start to go beyond the things we can observe and start to investigate things that were previously thought to be unobservable.



  1. Some of these may overlap with various ExG models. Individual components may be useful, or the aggregate measure might be. Either way it’s worth thinking through as a thought experiment.

One thought on “The Drunkard’s Search: Analytics, #confidence, and Observing the Unobservable”

  1. No background in behavioural research (and thanks for giving me something else I have to start from 0 on) so excuse me if I’m missing something but couldn’t this just end up recording a natural side effect?

    Theoretical cycle: 1. A player scores a goal off a low xG chance. He feels more confident because of this and starts making better decision.

    2. Now confident player has a run of games where he scores. Striking the ball cleanly from good positions as thanks to being calm.

    3. Player misses a series of high xG chances. His confidence is shaken so he tries to force things making bad decisions that leads to worse play.

    4. Player scores of a low xG chance…

    Now to me that could be a few months out of the career of any striker ever. Confidence was an important part but the measurable effects were already there to show it. And that low confidence run have little effect on his overall performance that season.

    Another player could go those months without scoring stay confident, keep making the right decisions and because luck is involved have the same results.

    All of which is a long way of asking what do you think there is in creating some kind of #confidence metric?

Leave a Reply

Your email address will not be published. Required fields are marked *