Comments Using the Player Rater Expected Points Model
Last week I highlighted Suarez and Aguero as elite attackers, and they duly delivered last week. They've gone up even further in the rankings, and these two players really do sit in a tier above all others. If you can fit both in your team then do so! Rooney also had a good week and has gone up the rankings, include him in your front three and you've got a deadly but expensive trio. Again, Lukaku, Benteke and Lambert offer cheaper options if you prefer. Sturridge, Giroud and Soldado delivered again, and they're worth holding, but if you need a new striker then do not overlook Suarez or Aguero. I also claimed RVP wasn't worth his price, and he duly delived a goal and assist to prove me wrong. He will outperform his stats consistently, but with Rooney keeping pace and Suarez and Aguero dominating, I still see little reason to consider him.
In midfield Ozil disappointed a little, Ramsey maintained his good stats, and there is little happening elsewhere. Silva looks a good buy, but isn't going to match Aguero for points. Despite a very quiet week, Hazard seems a better option than Oscar in the Chelsea midfield, I might write an article comparing the two later in the week.
I also backed Coleman who finally delivered a CS after an extended period without any points, he still does well for attacking threat. Debuchy scored over the weekend and jumped up among defenders, he has good attacking threat but I'd steer clear of Newcastle's defence for now.
New Slider - Average Time Played
You may have noticed a new slider on the graphics titled "Average Minutes", this is calculated simply as Minutes Played divided by Appearances. I received feedback saying that the model wasn't great because it was affected by those making sub appearances (such as Deulofeu ranking highest for expected goals per minute). You can now filter by those who average a certain number minutes, identifying those who do not just come of the bench at the end of games. Using this with the appearances slider should help you find those who are playing games, and playing lots of minutes in those games.
I'm considering integrating this metric into the model as well. The advantage of doing this is it allows us to better make comparisons between players. For example, Eriksen has 46% better production than Paulinho, but Paulinho averages 95 minutes a game, Eriksen only 68. Take this into account and the two players rate very similarly. Seeing as FPL is played on games rather than minutes, it makes sense. At the moment the model assumes every player will play the same number of minutes which clearly isn't true.
The problem is that I don't see a clear way to distinguish 2 players with the same average minutes. For example, say we have two players with identical stats and both average 60 minutes a game. But, one player starts every game and comes off after 60 minutes, the other plays the full 90 minutes, but is benched every 3rd game. There's a question over how to compare the stats of these 2 players, I could build in a measure of variance, but that assumes low variance (the player who plays 60 minutes every game) is desirable which may not be the case, especially with bench players who can come on.
I'd love to hear thoughts and opinions on this and what direction to take this in.
Blog Notes
Either today or tomorrow I should have a post out about which players are currently over and under performing their stats in assists and goals. The Team Rater to assess clean sheets is also in the works.
Nice work. I like the idea of using avg mins, and this is something I've struggled with too. My "solution" so far has been to just calculated everything on a per 90 minute basis and then simply leave it to the reader to ignore the anomalies. One thing I've recently added is the "risk rating" which essentially calculates the variance rating you mentioned. Perhaps you could use the "avg min" rate but then include a filter for risk, which would help readers eliminate those players who don't start every now and then. That said, I guess there's more value in a 90-90-0 player than a 60-60-60 but I'm not sure how you'd value that bit! Great work, man.
ReplyDeleteGreat to hear your input Chris, thanks for reading! I was wondering what the risk rating on your model was based on, thought it might be a subjective view but I should have known better!
DeleteIt does seem to get very tricky though, I think generally a P90 value is best but it might ostracise the FPL lay-person. If they see a Dzeko or Deulofeu ranking highly they may decide the model is a poor one and dismiss it entirely.
I'll have an indepth look tomorrow and see who jumps around the rankings when using "avg mins", but a quick look at the Top 20 would see Dzeko, Bony, Adam, Taarabt, Eto'o Vaz Te and Mutch drop out, so it looks promising.