Using Stats to Decide on the Best Goalkeeper for GW1

 Introduction

I saw a twitter thread posted by a user by the name of FPLWheeler earlier this week which used data from the past 7 Fantasy Premier League seasons in order to determine who is the best budget keeper for the upcoming campaign. In his thread, Wheeler concluded that there are '3 stats that are more significant than any other', when determining the best 4.5m goalkeeper for FPL: Non-Penalty expected goals conceded per 90 (npxGC/90), Post-shot expected goals conceded minus goals conceded per 90 (PSxGC-GC/90) and Post-shot expected goals conceded divided by shots on target conceded (PSxG/SoT). He then looks back at the highest scoring GK in each of the last 4 FPL seasons, and presents their data for each of these categories; producing an average which he then uses as a bench mark in order to rank the 4.5m GKs for FPL 24/25.

I absolutely love this idea, and wanted to explain and expand on this idea in a slightly longer form, but I give full credit to Wheeler for the original idea and thread. I will be using data from fbref (who get their data from opta), rather than Wheeler's data, so if there are some discrepancies, it could be due to slightly different data sets.

How are Goalkeepers awarded points in FPL?

There are a number of ways that a goalkeeper can score points in FPL, here is an exhaustive list:

- Playing up to 60 mintes : 1 point

- Playing 60 minutes + : 2 points

- Scoring a Goal : 10 points

- Assisting a Goal : 3 points

- Keeping a Clean Sheet: 4 points

- Saving 3 Shots : 1 point

- Saving a Penalty : 5 points

- Bonus Points : 1-3*

They can also lose points for the following:

- Conceding 2 goals : -1 point

- Receiving a Yellow Card : -1 point

- Receiving a Red Card : -3 points

- Scoring an Own Goal : -2 points

- Missing a Penalty : -2 points

Clearly from this list there are some ways that FPL Goalkeepers frequently score points, and some which happen very rarely, for example scoring a goal. Therefore, before deciding on which stats to use in order to decide on the best GK for FPL, we should rank the ways that a GK scores points in order of likelihood. In order to do this, I have used data from the top 5 highest scoring FPL goalkeepers in each of the past 5 seasons.

Clearly Goals and Assists should be discounted immediately, the two stats which are evidently the most important are Clean Sheets and Saves. The second thing to mention is that, as Wheeler did, I will be removing Alisson from all further analysis, as his 20 clean sheets and only 75 saves is clearly an outlier, and will skew the data. I know that all of this may seem very obvious, but it's important to set the scene and check that our preconceptions are correct.

So based on this information, the average highest scoring goalkeeper in recent FPL history has kept 13.75 clean sheets and made 134.25 saves per season. (Please note that bonus points for goalkeepers are mainly awarded bonus points for saves and clean sheets, which is why I haven't mentioned them as much.) 

Now lets look at Wheeler's thread to see if that helps us in predicting which GK will achieve these two target stats in the upcoming season.

Metrics used by Wheeler

- npxGC/90 is, according to Wheeler, the 'most important stat in predicting clean sheet points', as it represents how good a team is defensively, by measuring the quality of chances that a team gets on average per 90 minutes against the goalkeeper in question. Clearly this is a good measure of the team's defensive stability, and is not a measure of the goalkeeper's ability (although GK positioning is taken into account by xG modeling), instead principaly being one of a team's ability to stop the opposition from creating high quality (high xG) chances. 

- PSxGC-GC/90 is, according to Wheeler, a stat which 'judges the shot stopping ability of a keeper'. This stat is useful as it can be used in conjunction with other metrics to measure how many saves a keeper is likely to make each game.

- PSxG/SoT is, according to Wheeler, a stat which illustrates how threatening the shots being taken are. This stat is useful as it paints a picture of, over the course of the season (or in this case over the course of the career of the manager), how dangerous the shots being taken against the goalkeeper are. This stat goes a step further than npxGC/90, as it takes into account PSxG rather than just xG, which means that it is a stat better suited to analysing goalkeeper performance.

Knowing this, which of these metrics would be helpful in our search for a GK who achieves 13.75CS and 134.25 saves? Well npxGC is certainly helpful in looking for the keeper who's team is least likely to give up high xG chances, and therefore statistically the most likely to keep a CS. PSxGC-GC/90 is a vital metric when looking at how good a keeper is at shot-stopping, and differentiating between a good keeper and a bad one, which, as we will see in the upcoming section, is mightily important, but when comparing keepers across the league, is much less important than the others mentioned, due to the fact that most keepers are within about 0.5 PSxGC-GC/90, usually either side of 0. PSxG/SoT is also helpful, as it again highlights the defensive stability of a manager's system, which is arguably the greatest contributing factor to a goalkeeper's ability to keep clean sheets.

The Candidates

Forest's new keeper Miguel comes out on top, finishing 2nd in two categories when only looking at keepers who meet the benchmark in all 3. As Wheeler states, 'Forest have had the 4th lowest npxGC/90 in the league' since Nuno's appointment. Nuno joined Forest in December 2023, and signed Goalkeeper Sels in January, starting him in every game from February 4th onwards. Since Sels began starting in goal for forest, he has faced 17.5xGOT, and conceded 27 goals (suggesting that he is actually not that good at shot stopping). In fact, Forest conceded less than 1xG in 6 of the 16 games he played during his time as Forest's number 1, just over 1xGOT/90, meaning that his team's npxGC/90 will be very close, if not lower than 1 per 90, which is exceptional.

The addition of a competent shot stopper will mean that Forest could offer some very exciting FPL options from a defensive point view, none more than Miguel. If a goalkeeper could get their PSxG/SoT down to near the xGOT (xPSxG), rather than the very poor underperformance of sels by almost 8 goals (conceding 8 goals more than xGOT suggests), then that keeper would be down towards the benchmark set by Wheeler of 0.25 (xGOT for Sels during his time at forest of 17.48 divided by the 61 shots he faced equals just over 0.28, the same as Pickford last season). Miguel, for instance in the last 365 days has maintained an average of 0.43 PSxG-GA/90, which, if replicated, would be the highest of any keeper in the league by far, with Wolves' Jose Sa coming in behind him at 0.30 PSxG-GA/90 (over the last 365 days, not just premier league). Admittedly the data set is small, and there is limited information generally on Brasilian Serie A matches, but even so... he's clearly a good keeper.

While some may be concerned by the fact that he has only played in Serie A, it's important to note that often Goalkeepers performances are much more consistent if they transfer from one league to another, for them there is little change, they still face shots on goal, they're still expected to play it out from the back etc. Furthermore, the xG of a shot doesn't change depending on what division it's in, so therefore expected stats are a good measure of a goalkeeper's ability, particularly when looking at it per 90 minutes. For example, David Raya has maintained a PSxGG-GC/90 of between -0.04 and -0.08 (per 90) in 5 out of the last 6 seasons, 3 of which he played in the championship. Therefore it's safe to assume that Miguel will be fine, and will be able to maintain a relatively good PSxGC-GC/90, which would make him an ideal candidate for a 4.5m keeper based on Forest's defensive record towards the end of last season.

The second candidate that Wheeler puts forward is Robert Sanchez of Chelsea. Now, I'm not convinced by Sanchez as a pick, and as Wheeler mentions, his stats aren't great, finishing 5th, 4th and 6th for each metric respectively.  Furthermore, it's important to note that Chelsea were 11th in the League last season for xGA and 12th for goals conceded.

I know there's a new manager so their defense may* improve, and I know that they signed a lot of players last season, making squad harmony an issue (which is never good for defensive solidity), but one which some may argue will be improved thanks to a year of playing together. Regardless, as mentioned just then, they have a manager, so there will be a period of acclimatizing to his methods, tactics and ideas. Furthermore, last season in the championship, Moresca's primary keeper had a PSxG/SoT of 0.30, which is 0.05 higher than the benchmark set out by Wheeler. Therefore, even though Leicester's defensive numbers look good from last season (in my opinion due to the fact that they had a lower half premier league defense in the championship), if Leicester had faced more shots, then they could have shipped a lot of goals due to the quality of their goalkeeping. And unfortunately Sanchez' PSxGC-GC/90 in the last 365 days is -0.07 (substantially less than Hermansen's +0.13) and according to fbref's data, he has never had a season where he has outperformed his PSxGC, so he is completely relying on the defense of the team for which he is playing to not concede lots of big chances. There's just a lot unknowns about Chelsea, which means, if I had to pick one, I'd go for another option.

I actually think Wheeler's 3rd place pick of Johnstone could be a decent option (behind Miguel), so long as he starts ahead of Henderson. In 5 of their 13 games under Glasner, Palace conceded less than 1xG, and in their final 7 games they went unbeaten only conceding 4 goals. The issue I have with Johnstone is that his shot stopping is extremely volatile. Unlike the other examples I mentioned, despite playing 3 of the last 4 seasons in the same league, his PsxGC-GC/90 has varied between -0.30 and +0.11. This, therefore, raises some concerns for me, especially considering that there are some great Palace defensive and midfield options, which could cover the CS points that palace do get, whilst also providing some attacking returns. I'd go for a Palace defender first, but if you're looking for a double up due to the good fixtures, then he'd be a decent option.

Rotation

The other options mentioned are much less appealing to me if your plan is to set and forget, so let's have a look at the 7 GKs mentioned in Wheeler's thread, and see if there are any good rotation options. (sorry about the quality of images)

Sanchez and Miguel

Pretty decent fixtures all the way up to GW10. Facing 0 of last seasons top 8, and only playing Palace and West Ham from the top 10, both of which will be played at home. Furthermore, these two teams are the only in this rotation who were in the top 10 for goals scored last season. What I like about this rotation, is that there's a number of GWs where both teams have a decent fixture, so you could make a decision based on form.

Johnstone and Miguel

3 Promoted sides in the first 10 GWs is very appealing, along with 6 of the fixtures I've selected being home fixtures, but could go up to 7 if you go for WHU in GW2 and CRY in GW8. The only real problem with this rotation is the GW7 conundrum, with Liverpool at home for Palace and a trip to stamford bridge for Miguel. It's a decent rotation, but post GW10 it becomes a bit sticky, whereas the Sanchez / Miguel version is much better in the medium term. The Palace / Chelsea rotation isn't bad either, but I think Palace / Forest is slightly better from a defensive point of view.

The other goalkeepers in my opinion don't have great fixtures to start the season, so I haven't included them in this post, which is primarily about GW1. Flekken faces Liverpool, City and Spurs (all away) in his first 5 games, Areola faces Villa, City and Chelsea (and a challenging trip to Selhurst park), Sa plays Villa, Newcastle, Chelsea and Arsenal in his first 5 and Neto plays Newcastle, Chelsea and Liverpool in the same timeframe, so it's impossible to find a rotation that works.

In my current draft I have Miguel and Sanchez, but that may change between now and the deadline as I haven't necessarily settled on rotating 2 4.5 keepers, but if I do, that will be my selection. If I had to go for one, I'd agree with Wheeler that Miguel seems like the best option, particularly for GW1.

All credit for the original thread goes to @FPL_Wheeler. The original thread can be found here. Hopefully I was able to add something to his original thread :). I was very tempted to call this article, 'criticisms of the wheeler model', but decided against it (even though I think it sounds funny), as I'm not really criticizing Wheeler, but providing context and expanding on his ideas.

Dylan