Saturday, December 16, 2017

Raising their game or feasting on the weak (or, which teams do players do well against)

However good a player projection system is, it will always have certain assumptions built in which require some judgement on behalf of the model's creator. This might include how much weight to put on recent games versus historic data or how much to regress team or player conversion rates back to a league or historic average. One such assumption I do not currently factor into my own model is which kind of games a particularly player performs well in. For example, if Harry Kane accounts for 30% of his teams shots inside the box and Spurs are forecast for 10 SiB then his forecast will be three SiB regardless of who the opponents are. The strength of those opponents is of course somewhat baked into how we get to the 10 SiB projection in the first place, but no attention is paid to whether Kane has tended to over or under perform expectations against weaker or stronger teams, or whether he's struggled against teams who deploy three centre backs.

The data below takes the first steps to trying to factor that data in. Let's look at Harry Kane as an example:

I think there is a perception among some that Kane hasn't punished weaker opponents this year, which probably arose having failed to net against Burnley, Swansea and Bournemouth at home to start the season. In terms of expected goals though, we can see that he has actually excelled in all but one of his easiest fixtures (GW11 vs Crystal Palace). Generally his chart is exactly as an independent observer would expect - with him putting the sword to weaker opponents but having less potential success against stronger teams. In Kane's particular case there aren't any huge lessons here - you're not dropping him against anyone - but I do think this is a data point against necessarily captaining him without much thought to Spurs' opponent.

Let's check in with a couple of other players who have interesting profiles, then the visualization is at the bottom of the post for you to play around with.

Lukaku is probably the name most synonymous with feasting on weaker opponents, and this is indeed borne out in this data. In 10 fixtures which rank as easier than average, the United front man has averaged 0.8 expected goals per game, compared to seven trickier fixtures in which he's averaged just 0.3 expected goals per game. In real life this is a problem for Jose Mourinho, but in fantasy terms it's arguably beneficial to have someone who is predictably good against weaker sides and struggles against the top sides as he becomes easier to transfer in and out of your team (or captain). From GW19-24 United face only one fixture that ranks significantly harder than average in terms of xG conceded, making the Belgian a better target than I think most people realise.

Richarlison is generally one of the most interesting stories of the season so far, and his data here doesn't disappoint. The Brazilian has absolutely dominated in his sides tougher games, amassing 4.5 xG in 9 games against above average opponents (scoring 3 goals in the process). Yet, in eight easier games he's still netted twice but based on significantly worse underlying data (just 1.7 xG). This could potentially be a case of Richarlison excelling awat from home, where his pace can be better utilized in that inside forward role, where as at home he is perhaps getting isolated out wide as teams sit back and make Watford break them down. I don't know if this would encourage me to drop Richarlison in GW20 versus a shaky Leicester side, but it would definitely make me feel better about starting him during his away games over the holiday break (perhaps even at Man City depending on the strength of your squad).

This is someone I have paid attention to for the past few weeks after making a panicked transfer to bring him in when I had a big pile of cash to use on a forward (but not enough to snag Lukaku or Kane). His data isn't super interesting but it does tell a cautionary tale about relying on a single data point to make a decision. If you sort all forwards by Opta's xG then through 17 weeks you'll see Lacazette sat nicely in fifth place behind Kane, Lukaku, Aguero and Jesus i.e. right where his price tag says he should be. However, teasing the data out a bit shows that he amassed 3.0 of his 8.4 expected goals in a single fixture (GW15 vs Man Utd), without which his total for the season would be down with Okazaki and Abraham. Now, it isn't totally fair to start cherry picking games to remove from a player's season but it's worth remembering to dig into the data a bit more as we progress into the halfway point of the season and not be overly reliant on season totals on their face. On the positive side, this data shows that for the most part Lacazette has performed well in his easier fixtures, making him somewhat worth keeping in mind should Arsenal enjoy a couple of good fixtures and you need a short term fix.

I wanted to share this viz now to try and help with transfers over the busy period but will come back to it with any further players of interest I find and a corresponding viz for assist potential.