Daily Fantasy NFL: The Issues With Defense vs Position (DVP)
Note: This is an article from last year, but I think it’s an important read.
If you have played DFS, you have probably seen and used DVP statistics. For those unfamiliar, DVP, or Defense vs Position, is a statistic that shows how many fantasy points teams have allowed by position in a given season. For NFL, these positions are QB, RB, WR and TE. My favorite website for DVP statistics is NFL.com, as they supply some other advanced statistics like targets and red zone touches. These stats seem useful in daily fantasy NFL, as they look like they can give you a good idea of which teams are the best matchups for different positions. But they are also flawed, misleading metrics, mostly because of sample size issues, the large differences in strength of schedule from team to team, and injury impact.
Despite all of the issues above, my guess coming into this piece was that DVP statistics were predictive. For example, I thought at the very least if a team was ranked dead last in DVP against RBs over several games, then I should be targeting RBs against that team, just as most of us already do. I decided to look at historical data and find out, with the intention of showing fantasy points allowed has some predictive power for future games, but may not be perfect.
To do this research, I decided not to break down fantasy points allowed by position, but instead look at team fantasy points allowed rushing and receiving. The biggest reason I did this was because it was easiest with the data I have. These are also incredibly similar statistics to the break downs by position. Fantasy points allowed rushing is mostly driven by fantasy points allowed to RBs and fantasy points allowed receiving is equivalent to fantasy points allowed to QBs. DVP and these stats are basically one and the same.
I wrote a script in R that allowed me to look at several different sample sizes of games in a season to see how well they predicted rest of season statistics. For example, the first thing I did was look at the first 12 games of the season from the past several years and see how well fantasy points allowed in that sample related to fantasy points allowed in the last 4 games of the season. I could also do the same for 8 games and 8 games, or 4 games and 12 games, or any 16 game season combination.
It turns out there is a very weak relationship between fantasy points allowed rushing and receiving for the first 12 games of the season and fantasy points allowed in the last 4 games. It gets worse as your sample gets smaller. In other words, fantasy points allowed is kind of worthless.
Last season fantasy points allowed was a completely dismal signal. The correlation coefficient for a 12 game sample was 0.18 for fantasy points allowed rushing and 0.052 in fantasy points allowed receiving. As a rule of thumb, correlation coefficients from 0 – 0.3 indicate a weak or very weak relationship. Basically, the amount of fantasy points teams gave up in the first 12 games in the season had almost no bearing on the amount of fantasy points a team gave up in the last 4 games.
As we increase the data we are looking at from 1 season to the last 4 seasons, the correlations increase, but not by much. Rushing has the stronger relationship, with a 0.33 correlation coefficient, while receiving is still poor with a correlation coefficient of 0.2. You can look at the data from 2011-2014 below. The red line is the least squares regression line.
You can see how much stronger the relationship between the 12 and 4 game samples are in the rushing data than the receiving data, especially at the tails. In the rushing data, the three of the worst teams against the rush in the first 12 games were just as bad in the next 4 games. The 2011 Tampa Bay Buccaneers allowed an average of 22.67 fantasy points rushing in their first 12 games, and proceeded to allow a remarkable 33.43 fantasy points per game in their next 4. The 2011 49ers were incredibly stout against the run in their first 12 games and stayed that way, giving up only 7.18 fantasy points per game and 11.3 for the rest of the season.
In the receiving data, it’s not clear that the first 12 games are remotely predictive of the next 4 games. The 2011 Lions were above average against receiving fantasy points for the first 12 games of the season, but were obliterated for an average of 80.72 receiving fantasy points per game in the next 4. If you just looked at DVP data that season, you would not have taken advantage of Detroit’s terrible defense for the rest of that season.
Even if you look at the absolute best fantasy points allowed scenario, look at the first 14 games of a season, and see how well fantasy points allowed in those games predicted fantasy points allowed in a teams 15th game, the relationship is even weaker than look at the 12 and 4 split. This doesn’t necessarily prove fantasy points allowed is not predictive, but it strongly suggests there is very little relationship between fantasy points allowed at this point in the season and fantasy points allowed the next week.
So why do fantasy points allowed statistics fail to be predictive in Daily Fantasy NFL?
Let’s look at the sample size issues first. We are currently 13 weeks into the NFL season, and since every team has had their bye week, we have 12 games of data for each team, which is 75% of the entire season. This is 1/5th the amount of games we would have three quarters of the way through the NBA season, and 1/10th the amount of games we would have three quarters of the way through the MLB season. That is not a lot of data. There is a large luck factor that will impact DVP statistics in such small samples, so considering this data in small stretches of early in the season seems crazy to me, especially since we have shown that even 14 games of data isn’t predictive.
Strength of schedule issues are just as big. Because NFL teams play vastly different schedules, teams play very different qualities of offense and qualities of players at various positions. A team may give up a lot of fantasy points simply because they have played a lot of great offenses. Teams that are actually very poor defenses may face a lot of bad offenses and only give up an average amount of fantasy points. Atlanta is a great example of a team that has faced a very easy schedule this season. They’ve looked like an average defense, but in all likelihood they are very poor. They have played the worst four offenses according to yards per play, the 49ers, Colts, Texans, and Eagles. They have also played 7 out of their 13 games against bottom ten offenses. On the other hand, the Rams have faced an incredibly hard schedule. They’ve had to face the top scoring offense, the Arizona Cardinals, twice already this season. They have also had 4 games against the top 5 offenses in the league according to yards per play. They have been awful the past two weeks facing the Bengals and Cardinals, but they are probably not as bad of a defense as they have looked.
Injury impact can make DVP misleading as well. Last week against Seattle, the Minnesota Vikings were without their three best defensive players, Linval Joseph, Anthony Barr, and Harrison Smith. Their defense has been good all season, above average in fantasy points allowed to every position. But with three key players out, would they really be good on Sunday? Of course not . The fact that they have been good with their three best defensive players healthy isn’t relevant. Another example is the Jets. The Jets pass defense has been good this season, but without Darrelle Revis playing the past two weeks they have been burned by Odell Beckham Jr. and Jarvis Landry. DVP only captures what has happened in the past. Injuries happen so often teams can go to average defenses to awful defenses in the span of a week.
Fantasy points allowed statistics in daily fantasy NFL seem appealing. They show you how teams have been vulnerable to different positions throughout the season, which looks like a good indication of which teams will be vulnerable for the rest of the season. However, the data shows very little indication that DVP statistics in NFL are useful for predicting good matchups.
Here are three good places to look to help evaluate matchup. You can use FootballOutsiders defense ratings to get a teams defensive proficiency adjusted by strength of schedule. I tend to use ProFootballFocus to look at where individual teams are vulnerable by rating at each defensive position. Rotoworld’s news feed gives injury news on nearly every defensive player, so make sure you give it a look every day to see which teams are banged up week to week.View all posts by Daniel Steinberg