What's new
Fantasy Football - Footballguys Forums

Welcome to Our Forums. Once you've registered and logged in, you're primed to talk football, among other topics, with the sharpest and most experienced fantasy players on the internet.

Using Math to Predict WR Success or Failure (1 Viewer)

Whitney26

Footballguy
This is a follow-up that may interest a few people. I've run 816 WR's from the 1999 - 2013 drafts through some filters to see if I could identify any trends. As an amateur, it's a work in progress and will likely be tweeked many times as I continue to work on it.This particular study is primarily geared to the outside/big receiver. I've finished another one comprised of inside / slot guys but I don't have the tables formatted and entered yet. It's a painstakingly slow process. There is a lot of resistance in the draft world about this type of analysis. However, I'm convinced there are several teams that are already heavily influenced by it. I know the "eyeball scout" will always rule supreme, but I honestly feel the use of math in the process is only going to increase moving forward.The study......https://docs.google.com/document/d/1IPawJ_J3AOpHhSKolqSdri_mr7YbDKgBQxEAGK2E4Vo/pub

 
Do you factor in production on the field? Nice work!
Yes, it's called the Production Ratio, and it's the only performance based metric used. The formula is rather simple:Receiving Yards as a % of the teams total + Receiving TDs as a % of the teams total = Production RatioThanks for taking a look. I'm a terrible writer, and I struggle with the presentation if you know what I mean.
 
I like it a lot. Can't wait to see where the prospects fit in with their combine numbers included.
Thanks. I appreciate it. I did update the numbers with this years combine data included. I may be off, but I was able to add 36 WRs in my database. I'll keep updating as more roll in......pro days. The 2013 guys are listed in each table at the top.There just weren't any guys in the first table (Top Prospects). Everyone fell in the Solid Prospect and High Risk categories. Hope this helps.
 
Last edited by a moderator:
This is a follow-up that may interest a few people. I've run 816 WR's from the 1999 - 2013 drafts through some filters to see if I could identify any trends. As an amateur, it's a work in progress and will likely be tweeked many times as I continue to work on it.This particular study is primarily geared to the outside/big receiver. I've finished another one comprised of inside / slot guys but I don't have the tables formatted and entered yet. It's a painstakingly slow process. There is a lot of resistance in the draft world about this type of analysis. However, I'm convinced there are several teams that are already heavily influenced by it. I know the "eyeball scout" will always rule supreme, but I honestly feel the use of math in the process is only going to increase moving forward.The study......https://docs.google.com/document/d/1IPawJ_J3AOpHhSKolqSdri_mr7YbDKgBQxEAGK2E4Vo/pub
Wow, amazing work! :thumbup:
 
I like it a lot. Can't wait to see where the prospects fit in with their combine numbers included.
Thanks. I appreciate it. I did update the numbers with this years combine data included. I may be off, but I was able to add 36 WRs in my database. I'll keep updating as more roll in......pro days. The 2013 guys are listed in each table at the top.There just weren't any guys in the first table (Top Prospects). Everyone fell in the Solid Prospect and High Risk categories. Hope this helps.Oh, damn. Was hoping for a secret stats stud.
 
Looks pretty good. The general idea seems to make sense. One issue I have is with the production ratio. I think that approach is going to punish players who have good teammates and inflate players who play on bad teams with no competition for targets. In theory any elite NFL prospect should command a high percentage of his team's passes, but a guy like Robert Woods at USC, Reggie Wayne at Miami, or Laveranues Coles at Florida State is going to get killed there because he happens to play on the same team as a superstar. In general, I think that metric is going to punish players who play on good teams with capable teammates because that environment will result in them being force fed the ball less.

 
Looks pretty good. The general idea seems to make sense. One issue I have is with the production ratio. I think that approach is going to punish players who have good teammates and inflate players who play on bad teams with no competition for targets. In theory any elite NFL prospect should command a high percentage of his team's passes, but a guy like Robert Woods at USC, Reggie Wayne at Miami, or Laveranues Coles at Florida State is going to get killed there because he happens to play on the same team as a superstar. In general, I think that metric is going to punish players who play on good teams with capable teammates because that environment will result in them being force fed the ball less.
That's a good point EBF and something I've considered. I probably need to do another one with production entirely eliminated. I haven't figured out an easy way to present, but it'll throw some guys in there that really don't belong. For instance, Brad Smith and Joe Webb both measured out as elite but neither played the position in college. Obviously, they didn't make the cut for the PR. When I find the time, I'll take it completely out and we'll be able to see what we have. Guys that are really up on the game will know who belongs and who doesn't.I've found some some really nice ideas how to measure WR production at the NCAA level that doesn't penalize the guys on good teams......Hunter & Patterson or Woods and Wilson. It's something I'm still working on.On a final note, I'm almost finished with the Pass Rushers, and I think it's a lot more impressive than the WR's. I ran around 1800 DL & LBs through the filters including a few new metrics. I also hope to have the OL, CB and slot/inside WR's finished soon. It's my hobby and something I enjoy doing in my free-time. Probably need to pick up fishing or something.
 
I like it a lot. Can't wait to see where the prospects fit in with their combine numbers included.
Thanks. I appreciate it. I did update the numbers with this years combine data included. I may be off, but I was able to add 36 WRs in my database. I'll keep updating as more roll in......pro days. The 2013 guys are listed in each table at the top.There just weren't any guys in the first table (Top Prospects). Everyone fell in the Solid Prospect and High Risk categories. Hope this helps.
Oh, damn. Was hoping for a secret stats stud.I'm confident there's going to be some unknown guys that's going to emerge. It's still way too early to write it off yet. Lots of running left.
 
This is a follow-up that may interest a few people. I've run 816 WR's from the 1999 - 2013 drafts through some filters to see if I could identify any trends. As an amateur, it's a work in progress and will likely be tweeked many times as I continue to work on it.This particular study is primarily geared to the outside/big receiver. I've finished another one comprised of inside / slot guys but I don't have the tables formatted and entered yet. It's a painstakingly slow process. There is a lot of resistance in the draft world about this type of analysis. However, I'm convinced there are several teams that are already heavily influenced by it. I know the "eyeball scout" will always rule supreme, but I honestly feel the use of math in the process is only going to increase moving forward.The study......https://docs.google.com/document/d/1IPawJ_J3AOpHhSKolqSdri_mr7YbDKgBQxEAGK2E4Vo/pub
:thumbup: great job man
 
I think this is a great resource, thanks for sharing. Do you have any data on RBs or are they just impossible to predict?

 
I think this is a great resource, thanks for sharing. Do you have any data on RBs or are they just impossible to predict?
They are difficult. I've tried many combinations, but I haven't found anything I'm comfortable yet.However, I thought a couple of weeks ago I'd give it a shot for guys that actually have combine data. I may be wrong, but I trust the combine data more than the pro day. It'll eliminate a lot of players, but I think I'll be able to identify some trends. For example, I currently have the data for 361 RB's that went to the combine dating back to 1999. That doesn't include 2013. I have the data for another 280 that weren't there. I think 400 or so guys are enough to do something. Anyway, I'll try to get something together soon. I'm trying to do every position, but it's time consuming and I don't have a lot of time due to work and family.
 
This is very cool.One thing I think you need to try to factor in is hands. Perhaps using catch % and drops/target? Are those stats readily available on college athletes?

 
I love seeing data like this, but I think you have to be wary of putting too much stock in it, since it features a dangerous amount of curve fitting and circular reasoning. Basically, you took a sample of players, said "what variables correlated with success in this sample", then created a formula emphasizing those variables. In that respect, it is little surprise that your variables do such an amazing job at predicting success in the sample- that's specifically what that formula and those cutoffs were designed to do, and you designed them very well, indeed. Still, just because it had high predictive power within the sample on which is was derived does not mean it will have similarly strong predictive power going forward. Curve fitting can often mistake random splits for meaningfully predictive trends. Maybe explosive power really indicates which WRs are the best, or maybe most of the best WRs in your study were explosive strictly due to random chance, and a different sample would have placed a much lower emphasis on explosiveness. Typically, an analysis like this is suggestive, but can't be considered fully reliable until it has successfully predicted the behavior of several populations other than the one used to create it. In that respect, I really look forward to seeing how its predictions fare over the next several years, because there's a good chance you're really on to something, here. And in the future, if you want to be able to test a prediction model immediately, it's best to hold some of the initial data back and not use it in creating the model. For instance, you could exclude the receivers from every 3rd draft from the study, create a prediction model using all of the remaining players, and then use those WRs you held back initially to see how well your model predicted results in seasons it had never seen before.All in all, though, I don't want you to take this as me being disparaging or unappreciative, because I think this tool is potentially very useful going forward, and I definitely appreciate the amount of work that must have gone into it. I'll certainly be keeping my eyes open for more posts from you in the future.

 
I'm doing a long series of blog posts that lays out a method for determining skill-position success in NFL players. It has some of these same ideas Whitney is using here, but is further along I think. I say that with a lot of respect. ZWK and Whitney have made a lot of progress in a short time. I've just been working on it longer (seven years) and it's a bit more refined.First post is up. Will try to do a few posts a week.

 
This is very cool.One thing I think you need to try to factor in is hands. Perhaps using catch % and drops/target? Are those stats readily available on college athletes?
I'm not sure if they are readily available. I wish there was a site for NCAA football like PFF. I know there's a lot of stuff out there, but I just haven't found an easy source yet.
 
I love seeing data like this, but I think you have to be wary of putting too much stock in it, since it features a dangerous amount of curve fitting and circular reasoning. Basically, you took a sample of players, said "what variables correlated with success in this sample", then created a formula emphasizing those variables. In that respect, it is little surprise that your variables do such an amazing job at predicting success in the sample- that's specifically what that formula and those cutoffs were designed to do, and you designed them very well, indeed. Still, just because it had high predictive power within the sample on which is was derived does not mean it will have similarly strong predictive power going forward. Curve fitting can often mistake random splits for meaningfully predictive trends. Maybe explosive power really indicates which WRs are the best, or maybe most of the best WRs in your study were explosive strictly due to random chance, and a different sample would have placed a much lower emphasis on explosiveness. Typically, an analysis like this is suggestive, but can't be considered fully reliable until it has successfully predicted the behavior of several populations other than the one used to create it. In that respect, I really look forward to seeing how its predictions fare over the next several years, because there's a good chance you're really on to something, here. And in the future, if you want to be able to test a prediction model immediately, it's best to hold some of the initial data back and not use it in creating the model. For instance, you could exclude the receivers from every 3rd draft from the study, create a prediction model using all of the remaining players, and then use those WRs you held back initially to see how well your model predicted results in seasons it had never seen before.All in all, though, I don't want you to take this as me being disparaging or unappreciative, because I think this tool is potentially very useful going forward, and I definitely appreciate the amount of work that must have gone into it. I'll certainly be keeping my eyes open for more posts from you in the future.
Thanks for the response. If I'm being completely honest, it's a little over my head. I've always made up for my lack of smarts with hard work. I'm far from a statistics guru, but I think I'm pretty decent at researching stuff I'm interesting in. I've been trading stocks for the past 20 years using TA so I assume my interest in this stuff came from that. Like I mentioned earlier, this is definitely a work in progress. The potential to try other stuff is unlimited, and I'm always open to new and fresh ideas. I think you've given some good ones.It really doensn't belong here, but I did a similar study with the Pass Rush group. It doesn't include this years players, but I thought it provided some interesting information. Here is the link:https://docs.google.com/document/d/1tSHEXhceGdZDvPCMY-5DvysP2bEcNXtKcO2KclaC-OY/pubI broke it down into 3 sections:1 - Top Speed Rushers2 - Top Power Rushers3 - High Risk RushersThe first two groups were interesting, but I was particularly intrigued by the high risk guys. Unfortunately, my presentation skills are limited, and I didn't do a good job explaining what I see. Anyway, feel free to get a good laugh. It's probably not worth the space it's taking up on my Google Drive.
 
I'm doing a long series of blog posts that lays out a method for determining skill-position success in NFL players. It has some of these same ideas Whitney is using here, but is further along I think. I say that with a lot of respect. ZWK and Whitney have made a lot of progress in a short time. I've just been working on it longer (seven years) and it's a bit more refined.First post is up. Will try to do a few posts a week.
I've always been a fan of your work, and I've reading your blog since the beginning. I look forward to hearing your thoughts this year.
 
Thanks Whitney -- appreciate that coming from you. IMO you're on the right track here. Also, since you have a measure of performance in your stuff... One thing I noticed is that a lot of the guys from small schools dominate your on-field measure. Take a look at the average performance of DI guys (including the non-BCS players) and compare it to the average performance of the DII and DIII guys.I haven't updated this in a year or two, but when I looked at the average of my measure of on-field performance for DII and DIII schools applying a multiplier of 78% or 79% brought them back to the average for the DI players. Interestingly it was almost exactly the same number for both the WR and RB position.

 
SSOG's comment is a good one, but taking a step back, is there a glossary that I missed? Where is the explanation about the formulas?

 
Very cool, and thank you for sharing. A couple players are missing that I think would be interesting to see where they fit. AJ Green and Tavon Austin. Also the fact that Justin Hunter is in the High Risk category is interesting. According to National Football Post, his vertical and broad jumps were in the top 10% of all WRs since 1999. His 40 time was in the top 30% over the same period. Yet his Explosive Power is mediocre in your results. Obviously we don't know the formulas you've used, but how much of a factor is weight? If he was 5 or 10 pounds heavier would he be a top prospect?

 
Great work, Whitney.I like Production Ratio as a measure of college production. I also like that you're looking for ways to improve on it. My approach has been to average together a bunch of different measures of college production, since each one has its flaws (and many of the flaws are different from each other). If I had the Production Ratio numbers in my data set, it would have a pretty big weight in my rating of college production.I'm going to need to spend some more time looking at what you're doing with the combine numbers, and how it compares with what I've tried. It looks like two of your main metrics are based on the combine numbers that I consider to be most important - size, speed, and jumps - which is a good sign.I'm wondering if there are ways that you could simplify things to be less at risk of the overfitting problem that SSOG warns about. For example, how much do you gain by using separate cutoffs for Speed Score and Explosive Power, instead of combining them into a single metric with a single cutoff? How much do you gain by using 3 cone times, or the twitch index? What would happen if you used somewhat different cutoffs, or did something fuzzier instead of having firm cutoffs?The more different cutoffs you have, the more opportunities the model has to be tailored around the players in the existing data set who have done well, which is what creates overfitting. SSOG had a nice example of that risk on Twitter:

Important lesson: if you manipulate enough cutoffs, you can make anyone look like a stud. For instance... Lance Moore and Colston are the only 2 WRs with 50+ grabs, 6+ TDs, 10+ ypr, and 60+% catch% in each of last 3 yrs.
 

Users who are viewing this thread

Top