Soccermetrics and Scottish Football – a follow up

A couple of months ago, I wrote a blog about the rise of soccermetrics in Scottish football, and how I wasn’t sure how much value they added. To be honest, my tone was a bit snarky. I’d like to say it wasn’t deliberate, but I think I’d snapped a bit after reading yet another tweet shared onto my timeline about how stats ‘proved’ Barrie McKay and James Tavernier were Rangers’ best players, and I went off a bit, at the wrong targets. I should have been criticising the credulity of people that were displaying their confirmation bias rather than the guys who spend a lot of time analysing data, and I apologise for that.

That said, I still think it’s important to critique, audit, and monitor soccermetrics to ensure their validity. I’m a big fan of looking at football as objectively as possible, as humans are terrible at processing stimuli – we’re far too emotional. Conversely, sometimes we can get a little too objective, and little bit of interpretation of data is often helpful. It’s getting the balance right that’s the trick. A lot of Rangers supporters were, and still are, fans of James Tavernier and Barrie McKay, and would regularly retweet posts from metrics accounts that highlighted when they had high Expected Assist (xA) scores, which seemed to me the type of confirmation bias rampant on social media. I have to be open here – I’m not a massive fan of either player, despite recognising each is talented, and so I convinced myself that there was an issue with the xA model.

For most of the 2016-17season, both Tavernier and McKay had far higher xA than actual assists. McKay ended the season with 0.15 assists per 90, and Tavernier 0.20. In context, there were 8 Rangers players who assisted more regularly, and four who racked up 2nd assists more frequently. Combining goals per 90 and assists per 90 for the entire squad, McKay was ranked 10th, and Tavenier was ranked 12th. This discrepancy between expected and actual assists was often put down to the paucity of the strikers who were receiving the key passes, although I maintain that by rights, if Tavernier and McKay were more creative than the rest of the team, they should have had more assists per 90 than the rest of the team, regardless of who was playing up front. Funnily enough, four strikers that Rangers offloaded during the summer have had impressive starts at their new teams, at comparable levels of the game (although sustaining early form over the course of the season is always hard.)

Player

Goals per 90 2016-17

Goals per 90 2017-18 (to date)

Joe Dodoo

0.44

2.05

Joe Garner

0.33

1.00

Michael O’Halloran

0.00

1.20

Martyn Waghorn

0.33

1.14

 

In the summer of 2017, new Rangers manager Pedro Caixinha made a few changes to the Rangers squad. Most pertinently to this post, Barrie McKay left to join former manager Mark Warburton at Nottingham Forest, and Daniel Candeias was signed from Benfica. The Portuguese midfielder has had an immediate impact, contributing 1 goal, 3 assists, and 4 secondary assists after 8 games in all competitions, and has hopefully ameliorated fears that the departure of McKay would make Rangers less creative (McKay managed 6 goals, 12 assists, and 10 secondary assists in 46 games in all competitions last season.)

Candieas isn’t a straight replacement for McKay though, given the Scotsman played on the left at Ibrox, and the Portuguese works on the right. Pedro Caixinha not only changed much of the playing squad over the summer, he tweaked the formation as well, from the nominal 4-3-3 favoured by Warburton (realistically more like a 2-3-5) to something more akin to a traditional 4-4-2 (although operating more like a 2-5-3.) Under Mark Warburton’s stewardship, Rangers tended to line up with a slightly asymmetrical formation, with Barrie McKay and Lee Wallace teaming up on the left and Tavernier tending to mostly have the right flank to himself for the best part of 2 seasons. There’s some data to back this up – the Rangers Report’s Controlled Zone Entries metric measured Tavernier’s tendency to get forward from right-back. This season Rangers have lined up with Wallace and Windass on the left, and Candeias in front of Tavernier on the right. The Englishman’s xA figures appear to have dropped off this season, which is consistent with having an out-and-out right midfielder playing in front of him, and limiting his chances to get into the final third. It should be noted that his goals per 90 and assists per 90 are actually up on last season at this stage, and his xA (at least before the international break) is more tolerably closer to his actual assist figure.

This does raise a little doubt in my mind again about expected assists models. Why was Tavernier’s xA and xG so high for most of last season, despite his actual A and G scores being lower? Was it due to the paucity of the strikers, who are now netting regularly elsewhere? I think this is down to an issue of quantity. Expected Goals and Expected Assists are designed to attempt to measure the degree of difficulty of an attempt on goal, something a lot of us would like to know. But when you start aggregating xA and xG over the course of a season, you begin to lose a bit of that hard-earned granularity. As you can probably tell, I’m still a little bit sceptical about xG and xA, as I think they can be decontextualized by aggregation. I like that the Backpass Rule factors in the number of attempts that have generated the xG figure – for me at least, this makes the figure far more intuitive, and I’d be interested to see the same thing done regularly for xA.

For example, we know that James Tavernier scored 1 league goal last season despite taking 45 shots, meaning that of those Rangers players that scored a goal, he had the worst conversion rate. The average xG figure per shot seems to average about 0.2, so Tavernier could easily have reached 5.4 xG over the course of the season just from his 27 shots off target. I suspect something similar happened with his xA tally – it was bolstered by aggregation of lots of poor quality passes. While Tavernier’s xG+xA seems to have dropped off a little, Rangers’ goals-per-game has increased, from 1.47 last season to 2.2 this season, so the team’s not showing any signs of a downturn in creativity. Of course, it’s early days yet as only 5 league games have been played, and perhaps Alfredo Morelos’ 1.35 goals per 90 stat shows he is a better striker than last season’s crop. Of course, I could have written this entire piece to try and justify what is primarily me subjectively not rating a pair of footballers. But I do think the data lends credence to my theory here, even as much as it lends itself to other people’s contrary opinion.

Advertisements

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 )

Google+ photo

You are commenting using your Google+ 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 )

Connecting to %s