A new MF Passing Rating that helps identify the best playmakers in the Scottish Premiership
written by: @TheGersReport
As we prepared for the launch of Modern Fitba, I decided to do a deep dive on some of the recent content being produced in the hockey analytics community. Honestly, my best ideas are so often "borrowed" from some of the great stuff being produced covering the NHL. One of the ideas that stuck out came from a Chase McCallum blog post from the now defunct site, Hockey Talk Blog.
The post has since disappeared from the world wide web, but the concept I really liked & was what McCallum called xG above Corsi (in the hockey world, Corsi means the same thing as a shot attempt).
The concept was simple, yet really effective. Take a player's Expected Goal totals & subtract it from the league average Conversion Rate on all shots. For example, let's apply it to the Scottish Premiership (because...why not). Alfredo Morelos had 18.25 Expected Goals last season off of 88 shots. The league average Conversion Rate on all shots (regardless of location) was 11.5%.
McCallum's approach was to take the xG total & subtract it by a player's total shot count multiplied by that baseline league average Conversion Rate. So applying Morelos' stats it would look like this: 18.25 - 10.12 (88*0.115) which equals +8.13. That number was the highest in the league last season. Alfredo Morelos was the very best in the league at consistently getting a high rate of quality chances that should have boosted his goal rate above that league average Conversion Rate of 11.5%. For the record, his Conversion Rate was 16%, noticeable lower than his Expected Conversion Rate of 22%. But, that's a different post for a different time.
The concept basically comes down to this, on average 11.5% of all shots were scored last season in the Premiership. If Morelos simply scored at that same rate he would have scored 10 goals given his volume of shots. His Expected Goals total was eight goals more than that. That's an indicator that he is a forward who was among the league best at getting on the end of dangerous shots more often than not.
If you're saying: why not just use their actual goal totals instead of xG...I invite you to read this from 2014 & embrace the fact that Expected Goals are more repeatable than actual goals (note: the guy who wrote about that is now running the analytics team for Toronto FC).
Ok....that was kind of a long intro, but I hope you get the concept. I do plan on coming back to McCallum's original concept & applying it to forwards, but I think it may actually be better served in identifying playmakers. Which players are setting their teammates up with chances that are consistently better than the league average?
Of all the shots that were set up by a pass last season, 11.6% of them were goals. What if we apply Expected Assists to see which players were the best at giving teammates a better chance of scoring than that league average Conversion Rate?
This is where the MF Passing Rating comes in (ummm...the MF stands for Modern Fitba....obviously).
The formula stays the same, take a players Expected Assist totals & see how much higher (or lower) it is when multiplying his Key Passes by that 11.6% Conversion Rate.
Below you'll find the 30 players who had the best MF Passing Ratings last season.
I'm not going to spend much time breaking down this list...but yes, this is further proof that Daniel Candeias had a great first season. He had 66 Key Passes (meaning his passing directly led to that many shots) & if his teammates finished at that league average rate of 11.6% that would translate to 7.66 goals. His Expected Assists total from last season was 14.06. So basically, his passing was setting his teammates up to DOUBLE the normal Conversion Rate on shots.
But, anyone who knows my work...knows we need to live in the world of per 90 minute averages. Why? The player who is third on this list, Scott Allan, played 1,564 LESS minutes than the guy in second. That translates to a difference of 17 matches (or basically half a season).
That's why I'm a per 90 kind of guy.
But, don't worry...we will look at some of the surprising names here (Roarie Deacon & Liam Craig...even Declan John) & oh...hey look it's Miles Storey. Also, I want to acknowledge the fact that "noted playmaker" Steven Anderson made the list...but let's save it for the next list.
Below you'll find the 31 players whose per 90 MF Passing Rating was the best in the league last season. Meaning, they were most likely to set up teammates with a consistent amount of quality chances that should be finished at a higher rate than that 11.6% number.
Why 31? Well...the player tied at 30 was recently featured on the site...so you know....validation.
Now, before we take a look at the top three & the rest of the pretty interesting names on this list.....ummm... can we talk about Martin Woods?
Woods was among the players not offered contracts by Partick Thistle as they drop down to the Championship next season. This despite the fact that he was actually one of the better playmakers playing for, arguably the worst team in the league.
I recently talked about Relative Shot Creation Passing stats which takes some of the key passing stats & compares a player's rates only to the other midfielders & wingers on his team. Why? Because it's not really fair to compare a player like Woods, from a bad team, to a Stuart Armstrong or Daniel Candeias. Celtic had 402 more shots than Partick last season (I shit you not), so of course a player like Woods will not create as many shots as someone like Callum McGregor...it's not a fair comparison. But, if you compare Woods to the peers on his team then you can compare his Relative Shot Creation Passing stats to players on better teams. For example, Woods has a Relative Key Passes per 90 rate of +1.03 - which means he averaged a full key pass more per 90 than the rest of the midfielders on his team. That rate is one of the best in the league & somewhat similar to Armstrong's +1.46 rate (despite the fact that Woods averaged 1.72 Key Passes per 90 compared to Armstrong's 3.18).
Now the purpose of the MF Passing Rating is to not only look at volume of Key Passes but also the quality of the chances created. Woods' Relative Expected Assists rate is +0.23 per 90, which is similar to Stuart Armstrong (+0.27), Scott Allan with Hibs (+0.25), & Elliot Frear (+0.27). I've argued all season long that Armstrong & Allan were the best playmakers in the league & have known that Frear was sneakily right there with those two...but there's Woods as well. Honestly, even though he's 32-years old he still could be a very valuable asset to a team in the bottom half of the Premiership.
OK...Martin Woods tangent over....let's return to the countdown.
Here's the list again along with some quick observations:
If Elliot Frear gets more minutes next season, I think you have a real candidate to be among breakout players of the year. He's at the top of the charts here, largely because he is among the league's best players at setting up teammates with shots in very dangerous areas. His Relative Scoring Chance Key Pass rate of +0.63 is topped only by Armstrong & Allan. Scoring Chances are kicked shots from the heart of the box & kicked/headed shots from within the six yard box. This is where the majority of the goals come from, but are also among the most difficult shots to create. Few do it better than Frear did last season. He only played a little more than 1,200 minutes but Motherwell would be smart to pencil Frear's name in the starting eleven each & every week next season.
Candeias, Allan & Armstrong rounding out the top five with Woods & Frear isn't surprising. Armstrong will now look to see if he can replicate that kind of production with Southampton, while you wonder if Scott Allan will get the chance to replace Armstrong in the Celtic lineup. Candeias will get a chance to finally find some stability in his career & be among the key influencers in Rangers attack next season.
Surprised to see Liam Craig & Roarie Deacon so high? Me too (kind of). As I was tracking the data last season, I knew they were having pretty good seasons but seeing them higher than the likes of Kieran Tierney & Chris Cadden is interesting. Craig only played in 48% of the available minutes last season which is a bit of a surprise given how much St Johnstone struggled to create chances. His Relative Scoring Chance Key Passes was +0.33, which was similar to Ryan Christie's & better than James Forrest. Deacon is the more intriguing player here given his age (26) & given the fact he is entering the final year of his contract with Dundee. His Relative Expected Assists rate was similar to Ryan Christie, Chris Cadden & Jordan Jones while being better than the likes of James Forrest & John McGinn. That's good company. Smart teams will be keeping a keen eye on Deacon in the first couple of months this season because Dundee will likely be in a position to sell Deacon this January given that will be their last chance to cash in on him.
What's also very interesting on this list is the non-playmakers who are setting up above-average chances at a consistent rate. Steven Anderson was mentioned in the previous list & then there's Steven Caulker of Dundee, Ross County's Liam Fontaine, & 6'6" Harry Souttar (also from Ross County). Nobody is calling these center backs playmakers, however - you can see their value in the opponents penalty box given the data here. Anecdotally, I'm assuming most of their Expected Assists came as target men on set pieces - as they knocked the ball down for a teammate in a dangerous area. This is an effective approach & the evidence from the MF Passing Rating is that they are definitely creating dangerous chances from these kinds of plays.
When Hearts signed 35-year old Steven MacLean, one of the arguments in support for the signing was his link-up play. Ummm...he's missing from this list & you can also see why I suggested Billy McKay was actually a better option.
I recently pointed out that Ali Crawford would be an excellent replacement for David Milinkovic. Again, the stats support that claim.
There are some notable names who did not make the top 30. Here are a few & where they ranked:
Stephen O'Donnell (37th), Greg Docherty (38th), James Tavernier (42nd), Don Cowie (48th), Ryan Edwards (52nd), Callum McGregor (56th), Tom Rogic (70th), James Forrest (71st), Josh Windass (86th), Olivier Ntcham (88th), Jason Naismith (92nd), John McGinn (99th), Steven MacLean (111th), Niall McGinn (118th), Kenny McLean (124th), Moussa Dembele (126th), & Youssouf Mulumbu (140th).
Now, this is only one experimental rating that really acts as an entry point to further analysis. So let's break down the biggest surprises on this list of also-rans: Tavernier, Forrest & John McGinn.
Tavernier had 7.863 Expected Assists last season from 51 Key Passes. If we apply that 11.6 conversion rate from shots set up by a pass that would translate to 5.916 goals. That becomes a MF Passing Rating of +1.95 which meant he made the top 30 on the first list, but when broken down to a per 90 average his rating was +0.05 (for context, the next best full back was O'Donnell [+0.06] & the best was Kieran Tierney with a +0.10 rating). So the margins aren't huge but Tierney had four less Key Passes than Tavernier but 0.48 more Expected Assists. Ultimately, the difference isn't extreme & this would be a case in which I'd want to look at 2-3 seasons worth of data to see what the trends would be then.
James Forrest had 9.17 Expected Assists from 73 Key passes. His overall MF Passing Rating was +0.70 (9.169-8.468), which basically means when you add up all those chances he created, he set teammates up to score at a league average Conversion Rate. Of course, he did set up plenty of quality chances - but he also set up a lot of low quality chances. Part of this is a by-product of Celtic's willingness/eagerness to take long range shots, but that doesn't explain why Armstrong, Tierney, Scott Sinclair & Leigh Griffiths made the top 30 for per 90 MF Passing Ratings.
One of my goals next season, is to categorize the Key Passes players set up to see what trends emerge. Maybe a large percentage of his Key Passes did lead to long range efforts? Again, this is an example of how data can help influence what performance analysts look at while examining match footage. Maybe Forrest's numbers were inflated by high volumes of so-so chances?
One more thing on Forrest. So much of the debate over the Premiership's Team of the Year came down to that right wing spot. Most of the conversations centered on Forrest v Daniel Candeias. Let's see again how Candeias' rating broke down. He had 14.05 Expected Assists from 66 Key Passes. Notice right away that Candeias had seven fewer Key Passes, but five more Expected Assists. Now, THAT is a noticeable difference. For the record, Candeias' MF Passing Rating was a +6.40 (14.051-7.656). Candeias' Rangers teammates were TWICE more likely to score from his Key Passes than the league average rate.
John McGinn had 7.64 Expected Assists from 64 Key Passes. His overall MF Passing Rating was +0.22 (7.643-7.424) which translates to a per 90 rating of +0.01. Ultimately, he set up lots of shots (basically the same as Candeias & Forrest) but they were mostly average chances. His Expected Assists per Key Pass was 0.119, which is essentially identical to the league average Conversion Rate on all shots from Key Passes (0.116). This is not saying McGinn is a bad playmaker...it's highlighting how truly average his playmaking was last season.
Compare McGinn's numbers to Scott Allan's who had 8.49 Expected Assists, includes his time with both Dundee & Hibs, from 48 Key Passes. Allan had nearly one full Expected Assist more than McGinn, despite having 16 less Key Passes. Also, remember he played 1,511 less minutes than McGinn (which translates to nearly 17 matches). His MF Passing Rating was +2.92, which translated to a per 90 average of +0.16 (the third best rate in the league).
Stats used in this post come from my season long tracking of Scottish Premiership data. Next season, Modern Fitba will work off a unified set of stats.
Earlier, I mentioned Scott Allan's Relative Expected Assist rate of +0.25 with Hibs being among the best in the league. The best rate? Scott Allan with Dundee, where he had a rate of +0.44.
For more reading on Expected Assists, I recommend this 2015 OptaPro blog post by Sam Green.
I can't stress enough...this is just one experimental rating. No player rating is ever 100% accurate, but what it is - is a tool to identify deeper performance analysis. As a recruitment tool, it could help flag players to be scouted more. Internally, clubs could use it as ONE aspect of their own player evaluation as they make line-up decisions.
The next step will be to look at data from multiple seasons to see what trends & players emerge from a deeper set of data.
This was written under the influence of Pierced Arrows, Vietnam, CSNY, & Jets to Brazil.
Ok...now for the fun part....with no further commentary...the 30 players with the worst MF Passing Ratings (again, presented as a per 90 average).