# Do Stats Help in Fantasy Football?

17I’ve presented statistically-informed draft strategies for snake and auction drafts, arguments to use projections rather than rankings, evidence that the most accurate source of projections is the wisdom of the crowd, and tools for calculating projections and identifying sleepers. This all begs the question: do statistics work in fantasy football? Many people have asked how my team did in fantasy football this season. Here’s your answer:

I won my league’s regular season and championship and scored nearly 100 more points than the next highest scoring team. I’m hesitant to attribute all of my success to statistics, however. Football is a game with lots of chance. Projections are somewhat accurate but have lots of room for improvement—they explain about one-half to two-thirds of the variance in fantasy performance. That means that there’s a lot we can’t currently predict (e.g., injuries, trades, number of touches from week to week). Weekly projections are likely even less accurate than full season projections because weekly projections have a smaller signal-to-noise ratio due to a smaller sample size (fewer plays). As a result, I wouldn’t bet the farm on fantasy football, especially weekly fantasy football.

Nevertheless, statistics gives us a starting point for making more accurate decisions that are less biased than those decisions made by individuals’ judgment. According to the law of large numbers, with a large sample size (games, trials, etc.), the sampled value should approach the expected value. In other words, in a small sample of games, a player may show lots of variability and randomness from week to week in their scores. Over the course of a season, however, some of this randomness will average out, and the player’s points should more closely approximate his average/expected value (not necessarily his projected value, however). Anyone can get lucky any given week and out-predict the statistics. But, just like the stock market, the statistics tend to win out over time.

## Summary

Statistics are helpful in fantasy football up to a point—projections are not incredibly accurate because football involves lots of unpredictable chance. Betting on weekly fantasy football is exactly that: a gamble. Nevertheless, statistics are helpful because they are more accurate than individuals’ judgments over the long term.

Couple Questions: Do you let the guys in your league use this website?

I love the apps, went through my own draft and used it to see a “what if” and am embarrassed at how poorly I did on my own! Can’t wait for next season, which brings me to my next question: Will you be updating and maintaining this for the 2015 Season?

Hey Cooper,

Well I can’t exactly prohibit anyone from using the website 🙂 To my knowledge, the guys I play against don’t know about the site. I’m using that as a litmus test of sorts for knowing when we’ve reached a larger audience. I’m willing to sacrifice my “comparative advantage” by sharing my tools and giving the fantasy community an open-source platform that the community can collectively improve.

Definitely plan to maintain it for the 2015 season. We’re always looking for help, though. Our to-do list for next year is to include kickers/defense and tools for weekly leagues. Let us know if you’d be interested in contributing!

-Isaac

I just came across your site when I was searching for some sort of chart showing fantasy points scored on y-axis and number of players on x-axis. Although I didn’t find it yet, I found this site to be a nice source of fantasy information. I have thought for a while that I should get my wife to run some fantasy sports analyses since she’s in the same field as you (child psychology) and does a ton of work with statistics. Maybe one of these days she’ll help me out!

Is this what you’re looking for? See the density plot—projected points on x-axis and density (related to # of players) on y-axis:

http://fantasyfootballanalytics.net/2014/06/scraping-fantasy-football-projections.html

Hey Isaac,

Just found your site and I think it’s awesome! I’ve been using a similar analytical approach for the past 9 seasons and I’m always intrigued by the question of how to validate if the approach is actually working. I’m curious if you have been using your approach for more than 1 season and if you have a larger data set to show how well your system works?

To me, the best metric for measuring someone’s fantasy “skill” is the percentage of total points they earn in the league for any given season. I’ve come to call it Percentage of Points Earned (PPE), where in a standard 10 team league, if everyone is of equal skill, over the long run everyone’s PPE should be about 10%. When I calculated the PPE for the past 9 seasons, I was relieved to see I do in fact have the highest PPE in my league at 10.38, but was a little surprised that the spread was only 10.38 to a low of 9.07. Have you done any similar analysis or know of any data so I can gut-check our league PPE’s against other leagues?

Thanks for all the great content! Keep it up!

Cheers,

Marcus

Hi Marcus,

Yes, we have validated our projections against data from the last three seasons:

http://fantasyfootballanalytics.net/2015/02/best-fantasy-football-projections-2015.html

-Isaac

I would love how to apply stats data to my daily fantasy game. I’ve done ok with it but I want to take my game to the next level.

Hi Arron,

We have apps for daily/weekly leagues on our to-do list.

Thanks,

Isaac

Hello Isaac,

Let me start by expressing my gratitude. You’re accurate and easy to understand R-blogger posts led me to this website and i’m very excited to be here. I’m starting a masters program in Business Analytics at the University of Colorado starting on July 29th and I’ve used some of your techniques to increase my engagement with prep-work material we have been assigned before the classes start.

I’ve been playing competitive fantasy football for 13+ years in a 12 team single keeper, (1 year holding period), ppr league and would love to have a simple chat regarding your expertise and possibly leveraging some interesting topics for my formal studies/class projects.

Please reach out if you are at all interested,

Alex Prodoehl

[email protected]

Keep up the good work man. I used your R apps last year as a starting point to make my own valuations of players and won my league too (not surprising if we valued players similarly in the draft, then it’s likely we had a few of the same ones). I too am a PhD student (Electrical Engineering), so I can imagine what the demands on your time must be, so thanks for taking the time to post information, and make useful, user-friendly applications.

Hi Isaac

I ran into this site today and I am impressed. I am the creator of ffauctionaid for android. I only work on the app on my spare time and wish I could work on it more but kids and a real job have to get done. You have written down many of the ideas I have had and in general done a much better job. I’d love to correspond with you more. I’d like to invite you to join our league and see how you do as well. It is one of the craziest leagues but is managed by a good commissioner. I will tell you more if you are interested.

Wow…..what a great website!

Came across your website searching analytics help and i’m glad I found it, this is my second year as a DFS player on Draftkings & Fanduel and the information on this site is incredibly helpful and have great insight. Love the apps!

Keep up the great work!

This is my first season playing draftkings for NFL (which is where my question will be centered, but this should apply with any NFL DFS league/GPP). I’m just now discovering your site, but for the first 8 weeks of the season I have used my own methods to build my teams. I have written a java program to get fantasypros projections (with dk salaries), and built trillions of teams (per week randomly) while only keeping the highest projected teams. My program can do this for average, min, and max projections with some minor tweaking. Lately I’ve been getting in the big leagues and submitting 10 teams, handpicking them based on the top teams my program spit out. For example: The 3 best teams + 1 team with more reserved for Defense, 2 teams based on max projections, 2 teams based on min projections, etc….

My goal has been to break even most weeks on average and have a big payout every so often.

My conclusion is this: Whatever I’m doing isn’t working. This week, all 10 of my teams lost, and only a couple finished in the top 50% of the league.

Needless to say, I’ve spent a ton of time. As of today, I think I’ve read all your articles, and several others referenced here. You have certainly give me lots to think about, but I have no idea how to apply these things.

My real question for you Isaac is this: How does one win at DFS? Good question, huh?

I can’t figure out how I can be losing money at this. Sure there are really good people out there and there is a lot of luck to go around. You say the best strategy is definitely not to find the ideal team. I think I am learning that because that was my intent when I started writing my program. Apparently it’s also not an ideal strategy to try to find the 10 most ideal teams. After reading your article, I don’t know if I even know what an ideal team is anymore.

Is using a program to build your team a bad idea? Should you just use the programs to create information from data, to help build your team? How many teams are best to maximize chances of winning? Should I limit how many times a player is on all my teams? Should my teams try to maximize/minimize diversity among the team and across the teams? Exactly how much should the position matrix even be paid attention too? (these questions don’t necessarily have to be answered directly)

How do you win at DFS? I don’t know, I don’t play DFS because I consider it gambling. There is so much random variability (i.e., chance) in any given week that is unpredictable. Seasonal performance is easier to predict than weekly performance because there are more trials and some of the randomness averages out. You’re asking good questions. Diversity within and across teams can help reduce risk. You might try focusing on teams with the highest sum of floor. Keep in mind, you might be competing/betting against people with inside information.

DFS is also playable for free, and is just an adaptation of old salary cap leagues (yahoo used to do them) which were always my favorite anyway.

This is an awesome idea. I always wanted to get a data set of player stats and run multiple regression to see if i could make weekly projections better than ESPN Yahoo ect, but my we scraping abilities are a work in progress. The best I ever got was explaining about 44% of the variance. Do you think it would be worth while to expand the data set and more variables to get better predictions?

I absolutely think it’s a good idea to consider other variables. The challenge is acquiring them, and figuring out how much weight to give them. What variables did you have in mind and where would you get them?