Last week a means to investigate the reliability and variance of the different positions was introduced with a look at quarterback, defense and kicker. Today the focus will be on running backs.
By means of reminder, the expectation of the player is compared to their performance. Each DFS site’s salary serves as a proxy for expectation. The correlation between expectation and outcome is determined. A correlation of 1.00 means the performance was predicted perfectly while a correlation of zero indicates complete randomness. A negative correlation means the lesser players performed better than the highly ranked players.
In terms of DFS game theory, conventional wisdom say to pay for predictability in cash games while embracing variance in GPP tournaments. Relating this to correlation, the higher the number, the better the play in cash games.
Data from the first five weeks was investigated. So that some measure of conformity was present when comparing across the sites, the number of running backs included each week was twice the number of NFL teams playing. So in the first three weeks, 64 running backs were examined. In Week 4, six teams were on by so 52 made the study. Last week, there were only two byes so 60 were evaluated. Weekly rankings from several sites were averaged into a consensus ranking which determined the pool each week.
The first study looks at the total inventory together for each site.
| Draft Kings | FanDuel | Draft Day | |
|---|---|---|---|
| WEEK 1 | 0.47 | 0.43 | 0.44 |
| WEEK 2 | 0.40 | 0.31 | 0.33 |
| WEEK 3 | 0.49 | 0.39 | 0.44 |
| WEEK 4 | 0.48 | 0.44 | 0.47 |
| WEEK 5 | 0.38 | 0.30 | 0.31 |
Draft Kings has the reputation for sharper pricing then other sites and through five weeks, these results support that notion. Their correlations are a bit better every week which on paper means it's harder to find a bargain on Draft Kings, but they're not that much better and leave plenty of opportunity to find the underpriced players.
In a vacuum, these numbers don't tell us a whole lot; we'll need to wait until all the positions are studied. That said, across the board the correlations are a little better than quarterback which furthers the idea that taking a flier on a quarterback in a GPP tourney is wise.
The most relevant takeaway will come when wide receivers are added to the equation. Specifically, the relative reliability of running backs and wide receivers will help discern which position is better suited for flex. Citing the reliability argument again, the better correlated position is more suited for cash games while the lower correlated position could be a spot to embrace variance in a GPP. This will be the subject of next Tuesday's posting.
Similar to how quarterbacks were broken into tiers last week, the tables below will show the correlation between the upper, middle and lower tier of running backs. Thinking this process through a little more, it’s flawed in that if all the players in one tier play at a different level, but do it proportionally, the correlation would be high. That is, if the upper tier all did poorly but the top player in the upper tier was still the top performer of the tier and everyone else scaled down, the correlation would be high. Looking at the data, this isn’t the case but it is something to keep in mind. A higher correlation with that tier does not HAVE to mean the player performed as expected but empirically, that is the case. I’m pontificating on further studies to really hone in on this issue. But I am confident a higher correlation indeed means that tier performed as expected.
| TOP THIRD | Draft Kings | FanDuel | Draft Day |
|---|---|---|---|
| WEEK 1 | 0.27 | 0.15 | 0.12 |
| WEEK 2 | 0.09 | 0.31 | 0.22 |
| WEEK 3 | 0.29 | 0.09 | 0.29 |
| WEEK 4 | 0.56 | 0.32 | 0.34 |
| WEEK 5 | 0.41 | 0.44 | 0.29 |
Assuming Draft Kings does adjust pricing more than the others, Week 2 isn't as odd as it seems. On the whole they do a better job of projecting prices so when they miss, the others are likely to appear better.
| MIDDLE THIRD | Draft Kings | FanDuel | Draft Day |
|---|---|---|---|
| WEEK 1 | 0.16 | 0.10 | -0.03 |
| WEEK 2 | 0.14 | -0.32 | 0.03 |
| WEEK 3 | 0.16 | 0.13 | 0.11 |
| WEEK 4 | -0.26 | 0.02 | 0.54 |
| WEEK 5 | 0.27 | 0.47 | 0.05 |
The main takeaway here is there's more variance in the middle tier as compared to the elite (which is also the case with quarterbacks) so this could be a place to look for the difference-maker in a GPP>
| LOWER THIRD | Draft Kings | FanDuel | Draft Day |
|---|---|---|---|
| WEEK 1 | 0.17 | 0.42 | 0.17 |
| WEEK 2 | -0.14 | 0.14 | 0.24 |
| WEEK 3 | -0.15 | 0.17 | 0.23 |
| WEEK 4 | 0.02 | 0.12 | -0.27 |
| WEEK 5 | 0.22 | -0.18 | -0.19 |
The lower third is similar in predictability to the middle tier, leaving the top tier as the most predictable. As such, at least one elite running back should populate your cash game roster.
Given that my personal strategy is to use an injury-replacement running back (assuming one exists), I'll almost always choose one of the elite. We'll have to wait a week before using this data to help determine the optimal position for flex.