Who Has the Best Fantasy Football Projections? 2015 Update
135In prior posts, I demonstrated how to download projections from numerous sources, calculate custom projections for your league, and compare the accuracy of different sources of projections (2013, 2014). In the latest version of our annual series, we hold the forecasters accountable and see who had the most and least accurate fantasy football projections over the last 3 years.
The R Script
You can download the R script for comparing the projections from different sources here. You can download the historical projections here and historical performance (i.e., players’ actual points scored) here.
To compare the accuracy of the projections, I use the following metrics:
- R-squared (R2) – higher is better
- Mean absolute scaled error (MASE) – lower is better
Whose Predictions Were the Best?
| Source | 2012 | 2013 | 2014 | Average | ||||
|---|---|---|---|---|---|---|---|---|
| R2 | MASE | R2 | MASE | R2 | MASE | R2 | MASE | |
| Fantasy Football Analytics: Average | .671 | .422 | .503 | .520 | .569 | .479 | .581 | .474 |
| Fantasy Football Analytics: Robust Average | .566 | .483 | .566 | .483 | ||||
| Accuscore | .457 | .549 | .457 | .549 | ||||
| CBS: Jamey Eisenberg | .619 | .501 | .388 | .676 | .465 | .614 | .491 | .597 |
| CBS: Dave Richard | .619 | .501 | .388 | .676 | .512 | .587 | .507 | .588 |
| EDS Football | .516 | .527 | .516 | .527 | ||||
| ESPN | .528 | .577 | .393 | .684 | .483 | .591 | .468 | .617 |
| FantasyPros | .674 | .411 | .500 | .520 | .547 | .516 | .574 | .482 |
| FantasySharks | .455 | .547 | .455 | .547 | ||||
| FFtoday | .593 | .457 | .442 | .559 | .494 | .538 | .510 | .518 |
| Footballguys: David Dodds | .534 | .527 | .534 | .527 | ||||
| Footballguys: Bob Henry | .566 | .479 | .566 | .479 | ||||
| Footballguys: Maurile Tremblay | .527 | .523 | .527 | .523 | ||||
| Footballguys: Jason Wood | .549 | .495 | .549 | .495 | ||||
| NFL.com | .510 | .642 | .419 | .595 | .474 | .612 | .468 | .616 |
| numberFire | .474 | .596 | .474 | .596 | ||||
| Yahoo | .499 | .567 | .499 | .567 | ||||
- Fantasy Football Analytics: Average
- Footballguys: Bob Henry
- FantasyPros
- Fantasy Football Analytics: Robust Average
- Footballguys: Jason Wood
- FFtoday
- Footballguys: Maurile Tremblay
- EDS Football
- Footballguys: David Dodds
- FantasySharks
- Accuscore
- Yahoo
- CBS: Dave Richard
- numberFire
- CBS: Jamey Eisenberg
- FOX
- NFL.com
- ESPN
Notes: FantasyFootballNerd projections were not included because the full projections are subscription only. WalterFootball projections were not included because they do not separate rushing from receiving touchdowns. CBS estimates were averaged across Jamey Eisenberg and Dave Richard in 2012 and 2013.
Here is a scatterplot of our average projections in relation to players’ actual fantasy points scored in 2014:
Interesting Observations
- Projections that combined multiple sources of projections (FFA Average, FantasyPros) were more accurate than single projections (CBS, NFL.com, ESPN).
- The R-squared of the FFA average projection was .67 in 2012, .50 in 2013, and .57 in 2014. This suggests that players are more predictable in some years than others. It also indicates that 1/3 to 1/2 of the variance in actual points is unexplained by projections, so there is much room for improvement in terms of prediction accuracy.
- There was little consistency in performance across time among sites that used single projections (CBS, NFL.com, ESPN). In 2012, CBS was the most accurate single source of projection but they were the least accurate in 2013. Moreover, the least accurate in 2012 was NFL.com, but they were among the most accurate in 2013. This suggests that no single source reliably outperforms the others. While some sites may do better than others in any given year (because of fairly random variability–i.e., chance), it is unlikely that they will continue to outperform the other sites.
Conclusion
Fantasy Football Analytics had the most accurate projections over the last three years. Why? We average across sources. Combining sources of projections removes some of their individual judgment biases (error) and gives us a more accurate fantasy projection. No single source (CBS, NFL.com, ESPN) reliably outperformed the others or the crowd, suggesting that differences between them are likely due in large part to chance. In sum, crowd projections are more accurate than individuals’ judgments for fantasy football projections. People often like to “go with their gut” when picking players. That’s fine—fantasy football is a game. Do what is fun for you. But, crowd projections are the most reliably accurate of any source. Do with that what you will!
Appreciate what you do, this site is definitely ahead of the curve when it comes to forecasting. Look forward to future articles, they have all been helpful in furthering my programming and fantasy skills.
I have to shout out my man John Paulsen, who has in back to back years been the most accurate expert in FantasyPros aggregate rankings, over 100 other experts
I’d encourage him to send us his projections, to see if he can actually reliably beat the crowd. Also, rankings are much less useful than projections:
https://fantasyfootballanalytics.net/2014/08/use-projections-not-rankings.html
John Paulsen was first in projections as well as rankings 4 out of the last 5 years
Send us his projections. Let’s see if he can beat the crowd!
[…] expert is the most accurate. So I was of course delighted to see Fantasy Football Analytics’ Who Has the Best Fantasy Football Projections? 2015 Update. And to make things even better, take a look at FFA’s customizable app (note, this is if […]
[…] expert is the most accurate. So I was of course delighted to see Fantasy Football Analytics’ Who Has the Best Fantasy Football Projections? 2015 Update. And to make things even better, take a look at FFA’s customizable app (note, this is if […]
First of all, great site! Our you going to be updating the custom projections app for 2015? Thanks
Hey Bobby, yes we’re currently working on updating them. Still waiting on more sites to release their projections. Stay tuned!
Hi Isaac. Great work! I found you on Fantasy Football Librarian’s site and I’m glad I did. Two questions: (1) Why weight the various projections based on each’s historical accuracy? My thought being that every member in the crowd being given equal weight would yield a better prediction. I know that can be done using your tools (changing each sources overall weight), just wondering the mathematical/statistical rationale of weighting based on the historical “quality” of the voice in the crowd? (2) Fantasy Pros is one of your many sources, but Fantasy Pros itself combines (this year) CBS Sports, ESPN, FFToday and Pro Football Focus. And there is no way to choose less than at least two of those sources using Fantasy Pros’ website. I’m hoping you only choose FFToday and Pro Football Focus, otherwise aren’t you double counting the CBS Sports and ESPN projections already baked into the Fantasy Pros projection figures? Thanks for any help you can provide.
Hi Tom,
1) Weighting by historical accuracy was one of the most requested features last year. Ultimately, I want to test which is better: a weighted average or an unweighted one. That’s why our app calculates both. Feel free to choose the “Average” in the tool if you want want an unweighted average. At the end of the season we’ll evaluate which was more accurate. The underlying question is whether some analysts consistently outperform others. This is an open question.
2) Yes, including FantasyPros double counts their projections. We include the accuracy weights by default for your reference, but if you don’t want to double count, feel free to set the weights of FantasyPros to zero. Note that Yahoo projections are powered by ProFootballFocus, so Yahoo’s projections are the same as PFF.
Hope that clarifies,
Isaac
Did you have any notable successes using the average of all of the voices? Did you compare your results to those with somebody who used weighted projections?
For comparisons of the accuracy of mean vs weighted average projections, see here:
https://fantasyfootballanalytics.net/2016/03/best-fantasy-football-projections-2016-update.html
Regarding ProFootballFocus (PFF), you’re right! And in addition to Yahoo’s projections, it’s possible that PFF provides ESPN with its projections. Specifically, on its website PFF claims the following: (1) “Pro Football Focus Fantasy is the exclusive provider of NFL projections to the Yahoo Fantasy Football product.”, (2) “PFF Fantasy is the exclusive provider of fantasy football player projections to ESPN.” See these quotes here: https://www.profootballfocus.com/partners-and-promotions/
My question is does it impact the quality of your Fantasy Football Analytics: Average projection to have the same PFF projection included multiple times? Specifically, would it be better if PFF was only included once? Let me know if I can help you. Obviously you could license the specific PFF projections: https://www.profootballfocus.com/blog/2014/01/23/license-our-player-projections/, but I know this is a free site without financial recources. Assuming I’m allowed to do so, I’d be happy to share PFF’s projections with you as I am already subscribed. They’re updated often so let me know how best I can help. I really like your site/project and believe that more projections included will ultimately provide the best projection consistently. Happy to help in any way that I can.
Hey Tom,
Including PFF multiple times (or giving it a higher weight) would likely improve the accuracy of projections if PFF is more accurate than others but would likely decrease the accuracy if PFF is less accurate. I think we already have access to PFF projections, but decided not to add them to the tool because they are the same as Yahoo (which are already included). Interestingly, ESPN projections appear to be different, so ESPN may be modifying the PFF projections using other information. ESPN has pretty low historical accuracy (they were ranked last in our accuracy metrics over the past 3 years, see article above), though, so I wouldn’t have much confidence in the modifications they are making. We are always looking for additional sources of projections, and for coding help on our tools (especially on the UI). Let me know if you have some ideas for how to help!
Thanks!
-Isaac
It is funny on the ESPN note in that I didn’t have data to back it up but I always felt like their projections sucked. Now I guess I have the proof.
Hi, what is average Vs. robust average??
Hey Mark,
Here’s a discussion of the average and robust average: https://fantasyfootballanalytics.net/2013/03/calculating-custom-fantasy-football.html
Hope that helps!
-Isaac
Hi Isaac,
Terrific site and I love your analytics.
Just for the heck of it, I do my own projections. Not as scientific as yours, but there are a few things that I take into account that others seemingly do not. I take the body of a players work, note his absences due to injury or bad boy stuff, and compute the likelihood of games that will be played by said player. And from there, I can calculate expectations for him. Case in point – I downloaded your data and the conclusion reached on a particular Kicker intrigued me. Name? Robbie Gould. Decent kicker and I wondered why he was rated dead last in the league amongst the projected 32 starters. Looked up his history and sure enough, two things jump out. He missed 3 or 4 games last year due to injury and his then coach, Marc Trestman very often failed to attempt a FG, rather he went for it on fourth down.
Seemingly, no one has taken this into account when doing their projections. I’m not picking nits, so please don’t take it that way, but I will guarantee you that his new coach will use him differently and that he will exceed the point totals others have projected.
In any case – love the work you do. Thanks!
Hi John,
Good points. It can be a useful exercise to calculate your own projections. The difficulty is validating them against historical projections, which I’d encourage you do to see if your changes result in greater accuracy. Unfortunately, it’s impossible to know what other sites’ projections do or don’t consider because they are black boxes. That’s why aggegating them can cancel out some of the “noise”.
-Isaac
Hey Isaac,
I wanted to say thank you for putting this site together. There is a ton of useful information here. I will definitely be back closer to my draft parties to use your customizable and downloadable csv spreadsheets. I am always looking for an edge and your website covers 100% of the variables I would consider when selecting a team. You just helped do my work for me and I am extremely appreciative!
I would like to ask for help from those who have participated in a auction draft before. The pot on the table is $1,000 and I would love to earn some more bragging rights. This is new adventure for everyone in this particular league and my buddy is hiring a professional auctioneer as well as at least one scantily clad waitress to bring us our food and beer. I know I need to ignore the waitress(es) and use them to my advantage for distracting others. Besides that are there any other rock solid advice or tips that you guys(gals) have used that just about works every auction draft?
Hey Jeff,
We’re hoping to release our auction draft tool later this month. Stay tuned! For some articles on strategy with auction drafts, see here:
https://fantasyfootballanalytics.net/2013/06/win-your-fantasy-football-auction-draft.html
https://fantasyfootballanalytics.net/2015/03/fantasy-football-is-like-stock-picking.html
https://fantasyfootballanalytics.net/2014/08/use-projections-not-rankings.html
Good luck!
-Isaac
Thanks for links Issac. I completely agree with ignoring ranks and following projected points instead. I love this website and hope no one else in my leagues will find it.
Isaac,
Just ran your FFA Apps Projections after adding my input for scoring, etc.
Ran your list and then took the top 350 and I have a question. Your list is full of rookie noise.
It lists 53 rookies with 52 rookie ADP’s in the top 240 (12 teams of 20).
That is a summary then of Fantasy folks like David Dodds, etc. and that is just plain impossible, improbable and will never happen.
In a redraft league of 240, only 8 to 12 rookies will be fantasy relevant. That happens year in and year out.
Of those 52 on the list, 24 have absolutely no business even being there and another 22 have fantasy points that are clearly too high.
Of course, if your prognasticators all rank the rookies thusly, there is no way for you to eliminate the noise.
But, noise it is.
The only rookies that I see making a significant contribution in their first year are Abdullah, Yeldon, Cooper, Agholor, Gordon, Winston, Mariota, and maybe Cobb, White and Perriman. Outside chances for a couple of others. Got any suggestions? Thanks
Hi John,
I’d be interested to know what scoring system you’re using. With our standard scoring system we have 15 rookies in top 350.
-Isaac
Hi Isaac,
Redraft, PPR, 12 teams, 20 players per team, 1 pt per reception except TE gets 1.5.
QB 1 pt per 20 pass yards, 4 pts per TD pass, -1 per INT, 1 pt per 10 rush, 6 pts TD Rush.
WR, TE, RB 1 pt per 10 pass yds and rush yds, 6 pts per TD, plus recption pts.
I also stripped out all Kickers after 13 and all Defenses after 15, because they would not be selected in a 240 player pool for 12 players. I also capped QB’s at 28, TE’s at 35, RB’s at 73 and WR’s at 76.
……and I should mention that I stripped out the players over the numbers above and then sorted by ADP and that is how, I believe I ended up with 53 rookies.
Hey John,
Average draft position is different from our rankings. We suggest drafting starters by VOR and bench players by ceiling. I suspect that you’ll get fewer rookies when sorting by VOR and ceiling. Let us know!
-Isaac
Hello Isaac,
Great work. Just a couple simple questions. Averaging across sources doe that mean that your taking the points projected (so to speak of every source) and literally taking the average? If so, how does your model change that significantly from the sources? Are you trying to make your model of previous data come close to the average? This may be too much to reply to, but I would love if you could e-mail me.
Hi Mervin,
We don’t average the projected points across sources. Rather, we average the projections across sources for the different stat categories (passing yards, rushing yards, etc.), and then calculate points based on the average projections of the stat categories.
-Isaac
You tested preseason projections (N=1)–not weekly projections (N=17)–right? If so, would you ever consider testing weekly projections? I suspect that prognostic skill will become more identifiable over the weekly time frame. I’d be pleased to see which services (e.g., PFF, Bloomberg, Numberfire) produce the most accurate game-scale forecasts and pleased to learn whether these weekly projections outperform full-season projections (Full Season PTS / 16).
Hey EW,
Yes, we tested preseason projections not weekly projections. N would be number of observations, which was 1,230 in 2014. We have plans to test the accuracy of weekly projections this season, stay tuned! My hypothesis is that preseason projections are more accurate for predicting season performance than weekly projections are for predicting weekly performance because there are more games played and more room for week-to-week variability to wash out (https://fantasyfootballanalytics.net/2015/02/stats-help-fantasy-football.html). At the same time, I suspect that weekly projections would be more accurate than preseason projections for predicting (i.e., summing to) season performance, because weekly projections take into account season changes in depth charts, injuries, performance, etc. You won’t get the current season’s weekly projections when drafting, though, so drafting depends on preseason projections!
-Isaac
Hi Isaac,
I just wanted to know what is typically more accurate: weighted, robust, or average. Thanks.
Hi Joe,
This is discussed under the “Interesting Observations” section of the following article:
https://fantasyfootballanalytics.net/2015/07/accuracy-of-fantasy-football-projections-interactive-scatterplot-in-r.html
Hope that helps!
-Isaac
Isaac,
In a word – tremendous. In many words – tremendous, awesome, superb, outstanding…..
I’d be interested in setting up a call with you. Reply back if you are interested via email.
Ted
Hi Ted,
Feel free to shoot me an email (see the About the Authors section).
-Isaac
Hey,
I was wondering what the best way to determine sleepers and busts was from looking at the projections sheet ranking all the players.
Thanks, Benji
Hi Benji,
See the article on How to Identify Sleepers (under the “How To” tab).
-Isaac
Hello Isaac,
Thanks for answering my last question. I have one more. If you take projections across sites with each site at a respected weight, do you multiply each number by that weight and then your answer is the projected statistic. for example, if you did Aaron Rodgers stats from three sources with given stats, are you weighing them and taking a sum or re-weighing the sum based on the best site? If this is too specific of a question for you to answer, I understand. Anyway you can help will be great.
Hi Mervin,
The weighted avg is calculated using a weighted arithmetic mean:
https://en.wikipedia.org/wiki/Weighted_arithmetic_mean
Here is the R function:
https://stat.ethz.ch/R-manual/R-devel/library/stats/html/weighted.mean.html
You can see an example in our code (currently line 84):
https://github.com/isaactpetersen/FantasyFootballAnalyticsR/blob/master/R%20Scripts/Calculations/Calculate%20League%20Projections.R
-Isaac
Hello Isaac,
Great work. Will you be posting the 2015 projections by source as csv’s in the data section of GitHub? Your Snake Optimizer App is great but I also like to overlay some actual projected raw data in case I want to tweak the rank order slightly. For example, I think most projections underestimate the number of receptions CJ Spiller will have so I want to be able to see that in addition to all the gret metrics you include in tha app like VOR, Ceiling, Floor, Risk…
Hopefully this makes sense.
Thanks again!
Hi Craig,
Yes, this is on our to-do list once we finish the Auction tool.
Thanks!
-Isaac
Isaac,
First, these are just amazing tools, thank you so much for them. Second, do you have any idea on time frame for when you will be releasing the raw projection data (e.g., actual projected stats)? I understand that you can’t say for sure, but any estimate you give would be appreciated.
Hi Michael,
Hope to get to actual projected stats after we finish the Auction tool. It could be a couple more weeks or so before we add the projected stats feature.
Thanks,
Isaac
Hello Isaac, our league uses a points system that awards more points for longer touchdowns. For example, a touchdown catch of 1-9 yards is worth 6 points, but a touchdown catch of 40-49 yards is worth 10 points. Do you have any plans to expand the scoring settings to incorporate this type of scoring system?
Hi Paul,
Yes, this is on our to-do list. We are currently working on finishing the Auction tool so it’s ready for draft season.
Thanks,
Isaac
Hi Isaac,
Great website. I think it is the best for projections. I plan to use it for my league long draft – but what I am dreaming about is will you be doing projections for the weekly NFL games? I would love to use your projections for playing on FanDuel.
Any chance you will be doing weekly projections for the 2015 NFL season?
Thanks.
Hey Vik,
Yes, we have weekly tools planned for this season.
Stay tuned!
-Isaac
Hi Isaac,
I really like what your doing here as I do regressions with r2 and mase in my econometrics class. This really is the future of fantasy football projections and I applaud you for this. My only question is, do you provide a comparison of players to start week to week based on matchups and average analyst opinions similar to the who to start comparison on Fantasy Pros?
Hi Greg,
We are developing tools for daily and weekly leagues, and for week matchups.
Stay tuned!
-Isaac
Awesome. I just noticed my question is the same one like the commenters before me, sorry for the repeat and thanks for the quick response.
Isaac, is there a tool that enables you to select how many of each position your league uses? I play in both 2 QB and 3 WR leagues, and I’m assuming this would just simply lower the amount of points projected for the replacement leveled players. This would be a beneficial tool to have when inputting your own scoring system.
Hi Stephen,
In the Snake tool, you can modify the VOR Baseline values. In the upcoming Auction tool, you’ll be able to modify how many players you start at each position.
Hope that helps,
Isaac
Isaac, is there a tool that enables you to select how many of each position your league uses? I play in both 2 QB and 3 WR leagues, and I’m assuming this would just simply lower the amount of points projected for the replacement leveled players. This would be a beneficial tool to have when inputting your own scoring system.
Nevermind I see where you can change that now, thanks!
Hi Isaac,
How often will you be updating projection data for use with your apps?
Thanks for such a great site. I love it.
Nevermind, just saw that you update often. Thanks again!
Thank you so much for you data compilation here. I am far from a math wizard, so I have to rely on statistical guru’s like you. However, I am having a hard time reconciling the following statistical analysis, which indicates that FFToday was much worse last year and that NumberFire was great: http://datascopeanalytics.com/blog/which-site-has-the-best-fantasy-football-projections/
Hi Alexander,
“Accuracy” depends on how you define it. It looks like that article focuses on “mean error” whereas we focus on MASE. Although mean error can be useful (especially for knowing whether a source systematically under- or over-estimates players’ performance), we believe that MASE is more useful for examining accuracy. Assume 2 sites provide projections for two players: Player 1 (100 points) and Player 2 (200 points). Assume Site A estimates that Player 1 will score 90 points and Player 2 will score 110 points, whereas Site B estimates that Player 1 will score 50 points and Player 2 will score 250 points. Site B is clearly more accurate than Site A. In this example, Site A and Site B would have the same mean error (0 points), but Site B would have a lower (better) MASE. We provide the projections for lots of sites, and alow you to examine their accuracy yourself in case you want to examine other accuracy metrics. We provide lots of accuracy metrics for you to examine here:
https://fantasyfootballanalytics.net/2015/07/accuracy-of-fantasy-football-projections-interactive-scatterplot-in-r.html
Hope that clarifies,
Isaac
Just wondering if the projections analysts will include the footballguys? I see them in the accuracy, but not the 2015. Thanks
Hi Mitch,
We decided not to include subscription projections in the aggregation in our tools.
-Isaac
Hey Isaac, do you use the average for calculations or weighted average? Just wondering which one is more accurate?
Hi Jag,
I discuss average vs weighted average in the “Interesting Observations” section here:
https://fantasyfootballanalytics.net/2015/07/accuracy-of-fantasy-football-projections-interactive-scatterplot-in-r.html
Hope that helps,
Isaac
Hi, Isaac. Amazing site; I’ve been devouring your great content.
Do you have a hypothesis for why the simple average outperformed the robust average last year? Is this a single-year phenomenon, or have you seen the same trend in previous years? Is there value in including the outlier projections, since we’re leveraging the wisdom of the crowds and any single outlier can’t dramatically influence the simple average but its presence ultimately provides more accuracy?
JP,
I love this question. You can actually examine the accuracy yourself (see here for more info: https://fantasyfootballanalytics.net/2015/07/accuracy-of-fantasy-football-projections-interactive-scatterplot-in-r.html). The accuracy of average and robust average were pretty similar in 2013. We’ll see about this coming year. The key difference is that the average gives all sources equal weight, whereas the robust average gives sources furthest from the central tendency the least weight. Theoretically, the robust average would prove more useful if there are a few outliers that reflect poor judgment (e.g., if one analyst forecasts Aaron Rodgers to score 0 points). On the other hand, outliers might not always reflect poor judgment—they might reflect the uncertainty (i.e., risk) of a given player. For instance, most analysts may forecast LeVeon Bell to score about 250 points, but one analyst may forecast him to score 200 points because of injury risk. In this case, the average might prove to be more accurate than the robust average. Ultimately, we will continue to test which is more accurate!
Hope that helps,
Isaac
Hi Isaac-
Kudos on the awesome tools you’re providing.
How do the Fantasy Pros accuracy ratings (rankings) for their experts compare with your accuracy ratings using R?
I’d also be interested in whether the accuracy ratings improve significantly during the week as we approach gameday.
Thanx
-Bob Weil
Hi Bob,
I’d imagine that the accuracy of projections increases as the forecasters have better information to make their forecasts (e.g., injuries, suspensions, etc.). We have plans to examine the accuracy of rankings vs. projections. We’ve argued that projections are better:
https://fantasyfootballanalytics.net/2014/08/use-projections-not-rankings.html
Hope that helps,
Isaac
Wouldn’t league rosters affect the auction values in the custom projections app? If its a two qb league or you must start 3 wr, obviously QBs and wr will be more valuable than RB.
Hi Drew,
I answered your question here:
https://fantasyfootballanalytics.net/2014/06/custom-rankings-and-projections-for-your-league.html#comment-34139
-Isaac
Isaac,
I am interested in looking at individualized rankings for weekly fantasy football. This is in reference to weekly projections of players. I was wondering if you had any insight on this, or if there is a section of the website that I am missing. I was looking for something that forecasts how skill positions will perform each week as well as how they fare, statistically against certain DB’s.
This website provides analytics that I used as reference last year:
http://www.fantasyomatic.com/?page_id=9887
I also noticed that they were not evaluated in your rankings above, so maybe they are just too new? Please let me know your thoughts.
Thanks,
Chad
Hi Chad,
We have plans to release weekly rankings/projections for this coming season. Stay tuned!
-Isaac
Hey Isaac,
Have you considered evaluating each website’s ability to correctly predict sleepers? The probability of predicting alone is quite low, but do some websites do better than listing at least one?
Best,
J.T.
Hi J.T.,
I don’t think there are standard ways websites present sleeper predictions (which would be necessary for comparing their predictions). You still could examine each analysts’s projection accuracy, however. You can do this yourself here: https://fantasyfootballanalytics.net/2015/07/accuracy-of-fantasy-football-projections-interactive-scatterplot-in-r.html. We have plans to allow you to look at accuracy at lower position ranks (i.e., sleepers).
Hope that helps,
Isaac
Isaac,
That is exciting news. Are these in the works to be released prior to Week 1 of the regular season? Also, how would their accuracy compare to the ‘industry standard’ and sites such as ‘fantasyomatic’?
Thanks,
Chad
Hi Chad,
Yes, we will release projections before Week 1. We will examine their accuracy throughout the season.
-Isaac
Isaac,
I do love the website, thanks! How can you tell when the other sites have been updated in your system? For Example Jordy is still ranked on the average, I know he can be can excluded, but just want to make sure I have updated data. Also CBS has Forte Number one for PPR? That seems high and does not jive with their website as of yesterday. I have draft on Thursday night and would like to have an update. Thanks Again!
Andrew
Hi Andrew,
We have “Last Updated” for each source on our to-do list. We update the data very regularly. Where do you see that CBS has Forte #1? We don’t provide CBS rankings but rather CBS ADP (and I’m not seeing that ADP is 1 for CBS).
Thanks,
Isaac
I noticed that in your 2015 apps you no longer use any of the FOUR Football Guys or FantasyPros, which were by far the best evaluators other than your own aggregate projection approach. Given that, should i be concerned about this year’s projections being much weaker since you’re not using the best projection systems you’ve monitored while continuing to use the weakest ones (espn, nfl, fox).
Hi Jeff,
We include all the sources that FantasyPros includes (and more)—we want to avoid “double counting” their sources, and we’ve shown that the average is more accurate than all of the individual analysts, so we expect the projections to be strong.
Thanks,
Isaac
“Combining sources of projections removes some of their individual judgment biases (error) and gives us a more accurate fantasy projection. No single source (CBS, NFL.com, ESPN) reliably outperformed the others or the crowd, suggesting that differences between them are likely due in large part to chance. In sum, crowd projections are more accurate than individuals’ judgments for fantasy football projections. ”
Using your programs, you can find value plays where the VOR looks like a relative bargain for a certain draft position. My question is this, if the “crowd” really represents the most accurate info, has it ever been reviewed at the end of the season to see which data point was more accurate?
hypothetical example VOR = 100 in given league would give a draft position of 25
But assume this player is routinely available at 35. Looking at multiple examples of this, what data point was more correct, the VOR (or projected points based on the crowd), or did the ADP do a better job reflecting the final performance?
Hi Dan,
VOR is based on projections, whereas ADP is similar to rankings. We have reason to believe that projections are more accurate than rankings:
https://plus.google.com/hangouts/_/gsbtsgifr7y73brc7z4lmvlylia
You can use ADP to determine when players are likely to be drafted, and use VOR and ADP Diff to determine good bargains.
Hope that helps,
Isaac
Just curious if you have updated projections since the Tom Brady ruling?
Just checked the projections for each individual source. It looks like FFToday, FOX Sports, RTSports.com and WalterFootball have not updated. The other ones seem to have updated.
FFA Team:
I use a few different programs for FF analysis — is there anywhere on your site (or github) that we can download the projection data that underlies the Shiny app with fantasy projections (i.e. the individual player projections used to calculated projected points) or is that not stored as a standalone dataset?
Hi Rob,
You can download the projected points from the Projections tool. Downloading projected stats is on our to-do list.
Thanks,
Isaac
When you calculate the R^2 and MASE, are all players weighted equally? For example, are you penalized more for being inaccurate about the #1 player than for being inaccurate about the #100 player?
Hey Isaac–
First off, thank you so much for all of the useful information on this site!!
A couple of quick questions: In this calculation, were you including weekly projections, or just the projections at the start of the season? Also, if I use your instructions for scraping, am I able to scrape weekly projection data? Thanks so much!!
Hi Jess,
These calculations were just using season projections, but we do have plans to calculate the accuracy of weekly projections. You can download weekly projections here:
http://apps.fantasyfootballanalytics.net/projections
Hope that helps!
-Isaac
And also, do you plan on providing weekly projections on this site? Thanks!
My question is similar to Jess’s.
I am interested in studying the weekly projection accuracy by position group to understand where it might be best to invest budget dollars on weekly fantasy contests.
I see in the ‘accuracy’ section you can look at past years but not choose weeks.
Thanks for all the great info!
Hi Barney,
Once we add weekly performance moving forward, you’ll be able to examine weekly projection accuracy. Hope to add Week 1 performance soon.
Thanks,
Isaac
Isaac,
Thought you would like to know that I compared your Week 1 projections against three others and you came out slightly #1 in two leagues.
Congrats. WIll be following you over the course of this year.
John
Thanks for letting us know!
Isaac,
Oops, I should mention that I am following NOT all NFL players, rather my 40 players on two different teams. Still, you have the early lead………
Isaac,
This is amazing. Thanks for all your hard work. I was wondering if your Projections App had an option for total rankings and points projected for the remainder of the season as the season progressed. Would this be too difficult to add? Once again, I know this took a ton of effort, and I’m so appreciative.
Hi Carson,
Let us know if you find sources of rest-of-season projections (apart from numberFire).
Thanks,
Isaac
Fantasy Pros has consensus rankings for the remainder of the year http://www.fantasypros.com/nfl/rankings/ros-qb.php Not sure if you can combine these since they don’t give a point value
FantasyOmatic:http://www.fantasyomatic.com/?p=4800
Those were the only free ones I could find
Unfortunately, we’d need projections for stat categories.
Makes sense. I was also wondering how you take into account players who are questionable to play. For instance, T.Y. Hilton and Alshon Jeffrey this week as an example. Jeffrey seems to have a pretty high ranking this week, considering the feeling is that he won’t play this week, while T.Y.’s ranking is pretty low, and he is expected to play this week now.
https://fantasyfootballanalytics.net/2015/09/running-back-and-wide-receiver-gold-mining-week-2.html#comment-35576
Yahoo has a remaining-games projection
Isaac, I was curious if you have tried to use Watson Analytics to do the work for you. I have checked some of my regressions with it and it has come up with some interesting results. May be able to eek out a few more R^2
if you’re the most accurate, where can I find your weekly projections? Do you personally make the projections or are your projections based on all of the other sites you listed?
Projections (including weekly projections) are here:
http://apps.fantasyfootballanalytics.net/projections
More info is here:
https://fantasyfootballanalytics.net/2014/06/custom-rankings-and-projections-for-your-league.html
I’m looking for projected pass yards/rush yards etc. I don’t see that in your table. Where can I find that?
Click “Download”, then “Raw Projections”.
1. Are the weekly projections automatically updated everyday? If so, how often?
2. Are there any restrictions around using the data for personal or commercial use?
1) We update the data regularly. Date updated is listed in the tools.
2) What did you have in mind? We certainly ask for links to our site where credit is due.
Thanks Isaac…you seem to be a man of the people. Take care.
First of all, I’d like to thank you Isaac for making such a useful tool. Keep up the excellent work and I will consult this site daily for tips especially on Gameday.
I’m going to send you an analysis of what I come up with for the different scenarios that I am testing out directly to your email around the middle of next week.
Please post those on the site (rather than sending by email), so others can see and weigh in too!
Isaac, you are the man. Will the week 7 projections be up soon?
There they are nevermind. Quick note, James Jones and Willie Snead do not work for the “compare players” portion. Thank you so much for all your work!
James Jones is on bye, so he won’t have projections. Looking into Willie Snead. He’s in the table, but just doesn’t show up when using “Compare Players”. Looking into it. Thanks!
Hi Isaac,
I just discovered your website this morning and I’m seeing some fantastic stuff here. For the last several years (and with great success) I’ve been using my own home-brewed aggregate-projection valuation system, and it’s been fun to compare notes.
One of the issues with consensus projections that I struggled with this last pre-season is variation in prognosticators’ playing time assumptions for injured/suspended players. For example, Tom Brady had a 4-game suspension on our draft day, though most people believed it would be shortened to some degree. With the projections that included “games played” I could easily normalize based on points/game, but the majority of projections left me to guess whether their predicted stats would be spread over 12, 14, or 16 games. Maybe the standard deviation of the predictions is decent way to quantify the uncertainty surrounding the suspension, but I felt it would be more valuable to be able to estimate Brady’s per-game stats and therefore be able to calculate best- and worse-case scenarios after factoring in replacement value.
Any thoughts? How do you handle normalization of projection data?
In our the “Change Data Settings” of our Projections tool (http://apps.fantasyfootballanalytics.net/projections), we give users the option to impute replacement-level points for players projected to play fewer than 16 games (based on how many games players are projected to play from ProFootballFocus). Replacement-level points are based on the number of projected points of the players selected by the user-defined VOR baselines.
I get that, and I think understanding replacement value is key to handling players missing time. My question is if a player is projected to miss 4-6 weeks, how do you put together a meaningful average of expert projections when some of the rest-of-season totals assume 4 games and some assume 6? From what I’ve seen, most RoS projections don’t disclose games played assumptions.
First, we don’t provide rest-of-season projections—only seasonal and weekly projections, which have firmer projections for missed games (e.g., suspensions). Second, projections from most websites are basically black boxes—we don’t know what goes into their calculations and what they’re considering. So we don’t know how many games each source is projecting them to play. We use the number of projected games from ProFootballFocus to impute replacement-level points. It’s not perfect, but it’s better than ignoring missed games altogether. Let us know if you have additional suggestions!
Great site. Really appreciate all the work you guys put into it. I have three questions:
1. What do you think about using the Solver equation in Excel to determine the optimum lineup based on the weighted projections you provide from this site? Would the output be consistent with your app?
2. What is (at a high-level) a good DFS strategy when it comes to deciding which contests to “invest” in (i.e., what percentage should be in Grand Prize Pools versus 50/50s, etc.).
3. How do you find multiple “optimized” lineups and higher ceiling lineups using your app?
Thanks!
Robert
1. If Solver is equally accurate, the results should be the same.
2. Not sure, I don’t play DFS.
3. The app gives 3 lineups: Highest points, highest ceiling, and highest floor. If you want to generate a new set of lineups, select players you want to exclude, and it will generate new lineups.
The resources on this site are fantastic. I have always been interested in seeing pre-season projections compared to the actual projections. I see that you do make weekly projections as well, do you also link to the weekly projection data that you are using to aggregate into weekly projections?
You can see the sources included in our Projections Tool under “Change Data Settings”:
http://apps.fantasyfootballanalytics.net/projections
When do you post week 10 numbers? When is it updated? Thanks.
Projections and salaries are up for this week. Here’s our general schedule:
https://fantasyfootballanalytics.net/2015/05/2015-fantasy-football-projections.html#comment-3365
Thanks for sharing this Issac. I found it very useful.
Isaac, when will you publish the 2015 season Best Fantasy Football Projections article. Last year you did it about this time???
Hi Brian,
Yes, this is on our to-do list. Hope to do it soon.
Thanks,
Isaac
Thanks Isaac!I will look forward to it
Just posted:
https://fantasyfootballanalytics.net/2016/03/best-fantasy-football-projections-2016-update.html
I love your site Isaac. Truly great work. Using this site as a great tool I was able to make the playoffs in all my leagues.
When will projections app for 2016 start showing up? Veteran auction should start soon for us and it would be awesome to have. Thanks!
Hey Joshua,
Happy to add projections for 2016 when other sites upload their 2016 projections for us to aggregate. Do you know of any websites with 2016 projections available?
Thanks,
Isaac
Is there a way for me to look at last year’s VOR results? It’s easy enough to get spreadsheets of last year’s PPR point totals (various websites), but that doesn’t let me compare directly with the VOR projections I generated using your awesome app. I’m hoping to make a before-after spreadsheet of VOR projections and outcomes for all the players. Thanks!
Will put on our to-do list. Thanks!
Do you have a link to download the app?
You can access the app here (it’s not available for download):
http://apps.fantasyfootballanalytics.net/