Theories on defensive stats

I’ve been looking at defensive statistics for individual players for several weeks now and, while I’m still very much at a basic level, there are some things that I think I’ve noticed that I feel I should put down on virtual paper.

Defensive stats on an individual player level are quite difficult because you’re trying to measure something invisible, essentially how many goals does this person prevent out of the number of chances that they have to prevent goals. Not only is the thing we’re measuring invisible, but many of the indicators are hard to quantify. Off the ball movement is huge, as it can put the defender in the right place to shut down an attack before it properly develops.

However, off the ball movement data isn’t collected at the moment in football, and even if it was I imagine it would be hugely difficult to use. You wouldn’t just need the defender in question’s off-the-ball heatmap, you’d presumably need the positional data of the player on the ball at any given time, as well as various attackers and probably the rest of the defensive line too. That’s about six players’ positional data you would need to have and be able to analyse just for a single defender.

Decision making would also be difficult to judge, I imagine, from statistics alone. Firstly, because it would probably involve deciphering the positional data discussed above. Secondly, though, there would have to be a way to accurately apportion blame for an opposition attack without being too distracted by the success of that attack. If a defender makes several poor decisions in a match but, luckily, they go unpunished that doesn’t necessarily make them better than a player who makes one mistake that is punished by a goal. Alternatively, though, the one mistake that led to an opposition goal could have been a HUGE error in judgement, whereas the several instances of poor judgement by the first defender may have just been small ones.

There is also the issue of how to adjust a defender’s defensive score for how dominant his own team is. Obviously, a defender who only stops a few attacks but they’re the only attacks they have to deal with has performed better than a defender who only stops a few attacks while his team was on the back foot for the whole match.

I know a few people are using possession to adjust for dominance however a team may have a lot of possession but do little in the way of attack with it, which would unfairly harm the opposition defenders’ scores (as their team had little possession meaning their scores are reduced, however they may still have had very little to do). This is why I use shots on target and shots blocked as well as possession, as I feel this gives a better indication of how dominant, in an attacking sense, teams are. I don’t think it’s perfect, but I do think it’s better than only using possession.

A potential problem is that I work out the player’s score for each match and then take an average of those, rather than taking an average of every stat from the whole season together, if that makes sense. I’ve not thought about this too much, and I’m it’s a while since I did any proper maths, so I’m not exactly sure whether taking a lump end-of-season stat would create different figures than stats for each match, but I guess it might be a possibility.

A similar-ish problem occurs when players play in varying positions throughout the season. This is an advantage of doing things match by match because it means, in my case for example, I only input the data when players played at centre-back. If I did it match by match indiscriminate of their primary position for that match then the data could be skewed (for example David Luiz played about half his time in central midfield last season, so if I took every minute he played on the pitch and put that on a central defender radar chart template then it wouldn’t be an accurate indication of his centre-back performance). This is a potential issue because 1) playing in position ‘x’ might naturally give you better or worse stats in certain areas than playing in position ‘y’ (as a crude example, you’ll score more goals up front than in goal) and 2) because players might perform to different standards or play in different ways depending on what position they are in.

Those are all my thoughts for the moment, and I’d genuinely love to hear what people think. Either comment below or get me on Twitter @ETNAR_uk

Advertisements

2 thoughts on “Theories on defensive stats

  1. Suggestions on defensive stats

    I totally agree with you that measuring defensive perfomance is very difficult. The reason is that, as you claim, there aren’t any useful off-the-bal stats. Running distance and number of sprints tell us HOW MUCH a player ran, but not WHERE he did it and if it suited the tactical concept.

    The lack of off-the-ball stats is particularly an issue in defensive metrics: During an opponent’s possession, a player NEVER touches the ball. Only at the END of a possession a defending team logs a touch, so that only these events (tackles, interceptions) are logged. That’s different for offensive metrics: Not only the END of a possession is logged (shot, turnover, unsuccessful pass), but also what happens during the attacking sequence (dribble, successful pass, final third pass, forward pass, key pass etc.).

    Although it doesn’t solve the “off-the-ball problem”, I’d propose to start logging more actions during an opponent’s possession such as…

    -How often did the players press onto the opposing player carrying the ball? And in doing so, how often did they force the opponent to a pass backwards, how often did they allow a pass forward?
    -How often was the ball played past them (without intercepting the pass/blocking the shot)?

    Perhaps, these additional actions can be categorised as “AT the ball, but not on the ball”. The player who makes pressure on a player in possession doesn’t touch the ball, but he is not OFF it.

    Surely, these metrics wouldn.t consider “off-the-ball” qualities such as help-defending and following tactical rules. We still have to hope that the following assumptions is correct: “A player with great positioning and defensive effort will get into situations in which he can tackle, clear etc. more often.than another one with weaker positioning and less effort.” Though, this won’t be correct for at least two examples.

    -(I’m not sure if this is a valid example): Some players might have such a great positioning that they reduce/nullify the danger before the ball is played to the player/space he marks. I know that was always said about Paolo Maldini. I’d be very interested in if he really recorded less defensive actions (TICAD?) than other defenders on elite level.
    -Some players might be so aggressive (in terms of pressuring opposing players). This characteristic will lead to more tackles and interceptions, which will make him look a world-class defender. But due to his aggressivity, he is often out of position, and the opposing team can exploit this space. In his case, less aggressivity (=> less tackles+interceptions) would be more.

    As noticeable, a lot of work is still to do for defensive metrics, but I’m glad that a blogger like you writes about them.

    Reply
    1. Mark Thompson Post author

      Thanks.

      Those possible ‘at the ball’ metrics are interesting. I feel like they’re partly to do with style, though, rather than definitive ‘worth’ eg more passes played past them probably indicates a high line, but it could also mean poor positioning. Even if they are more indicative of style I think that could still be useful because you might be able to adjust (somehow, I don’t know how it’d work) a defender’s base numbers depending on the style of the back line they’re playing in.

      I agree with both your points and it’d be really interesting to see Maldini’s stats broken down like we have for current seasons. I wonder if you can think of any defenders similar to him who don’t register many tackles but are undoubtedly good?

      I think another problem is that it depends on your defensive partner. Someone like Maldini might not tackle much but may be the main reason why the opposition have so few good shots but their partner might be less effective but make a couple more tackles and clearances, so they’d appear better. Similarly, with a more aggressive CB their rashness might go unpunished because of their partner who might clear things up. Were it not for their partner, they’d score much worse because they’d be giving away more opportunities.

      I think I’ll have a better idea this coming season of whether overly rash players are unfairly rewarded by my system. I’ve done all the work on last season’s stats this summer so I’m not too sure on players’ styles as I’ve not watched them recently (and I don’t think the World Cup is a hugely great indicator for those that were there). The only player I know for sure who’s rash/aggressive is David Luiz and he’s right down at the bottom of the list mostly because he misses an awful lot of tackles (missed over half he went in for when playing at CB). I *think* that aggressive players tend to miss more tackles etc and so are penalised for that in my system but that’s more something that I assume than that I know. There’s also no measure of missed/failed interceptions on the StatsZone app which I use to get my stats which could be helpful in, essentially, punishing overly rash or aggressive players

      Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s