Stop Using DvP; Creating the New Standard for Opponent Matchups in NBA DFS.


Banner DFS DivesStop Using DvP

If you've ever played DFS (or any fantasy sport), then chances are you've faced a tough decision between two players, and your final decision came down to the little green (or red) number next to the players matchup. The well known defense verse position (DvP) metric has been widely accepted as the primary source to consult when considering whether a player has a good or bad matchup.

DvP

On the surface, DvP makes sense. If a team has given up more points on average to a certain position, then they are more likely to in the future, right? Unfortunately not. The standard DvP metric you see on popular sites fails to account for multiple crucial factors. For one, certain players are better than others, and have different salaries (see Salary Adjusted DvP). But second, and perhaps just as important, positions are becoming less and less relevant in the NBA each year. You have guys like Porzingis and Drummond both being listed as centers, despite playing two COMPLETELY different roles. At Analytic City we attempt to solve this problem with our proprietary player profile DvP metric.

What is a Player Profile? 

Simply put, each player in the NBA is given a classification (player profile) based on how they score their FGs, how they accumulate fantasy points, and how many minutes they play. Let's look at an example:

Player: Andre Drummond, Center (Big)

Hypothetical FG Scoring Profile: 45% from the restricted area, 35% from the paint, 18% from midrange, 2% from three.

Hypothetical Fantasy Point Accumulation: 41% from scoring, 40% from rebounding, 10% from assists, 9% from defense.

Hypothetical Minutes Per Game:  30.2 MPG

Final Classification: Paint_Big-Rebounder-Starter

In all, there are 20-25 total final classifications we use based on various thresholds we have set.

From this, we can create our Player Profile DvP metric by looking at how players with the same player profile have performed against a players opponent in past games. To read more about Player Profiles and how they are constructed/used, click here.

So, Does Player Profile DvP Have Predictive Power?

Now for the million dollar question, our player profile DvP metric sounds and looks awesome on the surface, but does it actually have any predictive power? I decided to use the second edition of our DFS dives series to take a look. 

The first thing I did was filter the data into what I believe will give the most accurate picture as to whether our metric has predictive power. I limited our sample to only the top 50 salaried players from every slate, and also removed any player who played fewer than 10 minutes. This should give us a really clean dataset to look at, as we won't have to deal with player injuries, or things like cheaper players who may have performed extremely well due to expanded roles (in which case DvP data is hardly relevant). That being said, it also means the results of this study can only be extrapolated to higher salaried players, and are less relevant to cheaper/inexpensive DFS plays. After filtering the data from the first half of the season, I was left with 1320 entries. Before running any regressions, I graphed the value created from each players Player Profile vs each players Fantasy Points Scored. Value created is defined as the difference in dollars per point between that profile against the players opponent and the league average. In other words, how much cheaper have players with the same profile been against the player in question's opponent when compared to the league average? A positive value indicates players with that profile have been cheaper for each point scored, while a negative value indicates players with that profile have been more expensive for each point scored.

XY Scatter

I was incredibly excited to see a noticeable positive relationship between the two variables. You may be thinking to yourself "That barely looks like anything", but it is important to keep in mind the HUNDREDS of other factors that go into every NBA game, so creating a statistic with predictive power is incredibly difficult (there are very few of them). And from an initial look, it seems as though we may just have done that! For reference, here are some other matchup based statistics that I have found to have some (although how much is still in question) predictive power in NBA DFS:

- Pace (for players under $7,000)

- Over/Under

- Salary Adjusted DvP

End of list. I am sure there are more, but my point is that the list of matchup based metrics you can use to help you make NBA DFS lineup building decisions is very limited. 

Is Player Profile DvP Actually Predictive?

To test whether or not there was a statistically significant correlation between these variables, I ran a linear regression.

Regression

The values with the yellow background above are what we care about. You can see in the bottom left that the slope of our line is 0.0267, which means that for each 1 point increase in the value of our player profile, you can expect a 0.0267 point increase in fantasy points scored. Bringing this to the extremes, the difference in expected fantasy points scored for a player with a value of -50 from their player profile and a value of 50 from their player profile (an extremely bad matchup vs an extremely good matchup) is a whopping 2.67 points! The next thing to look at is how confident we are in these results; our given P-value of 0.0235 indicates that our results are statistically significant at the 5% level, a commonly accepted threshold for statistical significance". A p-value of 5% or lower is often considered to be statistically significant." (investopedia.com) Simply put, a correlation exists!

How Can You Use This Information in Your DFS Lineup Building?

If you're using our projections, good news! We account for Player Profile DvP in each and every player point projection (unless the sample size is under 10). But even if you aren't, my main takeaway is this: Looking at the value gained (or lost) from each players player profile should be one of your primary tools when considering if a player is in a positive matchup in NBA DFS. 

 

Our 'DFS Dives' series aims to take an in depth and analytical look at popular DFS statistics to see if they REALLY make a difference in player performance. 

Have a stat or question you want answered about NBA DFS? Email us and we will select our favorite question to cover in an edition of our 'DFS Dives' series! Exclusive to AC Plus members!