Some days writing about football is awesome. I like to think of this ‘job’ as predictive analytics, mainly because it sounds much more important than writing about fantasy football but also because it masks the reality that there are other days when hours of research don’t add up to much.
This week I fell deep down the rabbit hole of our Trends Tool and Bryan Mears’ Market Share Report in search of a deeper understanding of target share data. I specifically looked at four-game market share data for wide receivers.
Intuitively, it makes a ton of sense that more recent data would be the most predictive. Things definitely change — see Marvin Jones. But is it possible that we are overemphasizing the importance of four-game target market share?
I’m going to make this short and sweet.
What’s Most Conclusive Is That The Results Are Inconclusive
I started by taking four-game target share data from Weeks 5-9 (a five-week sample). From there, I filtered out players with projections lower than five DraftKings points and also those who fell below a five-percent target share threshold. This left me with a sample of 443 data points.
The following charts display the correlations between four-game target share and both raw DK points and Plus/Minus.
Yeah, the correlation result is very weak (0.13) in regards to raw points and essentially non-existent from a salary-based perspective.
To be clear, there hasn’t been a lot research done on this, and it covers only five weeks, so don’t read too much into this, but I do have two main takeaways.
— You may want to emphasize raw target data over market share data.
— Market share data likely still has value when looking at patterns. For example, you could weigh these percentages more heavily in pass-heavy situations or when distributing the targets of a recently injured player.
Basically, what I’ve found is that I need to find more data on market share.
Going Forward
I wasn’t sure what I would find, but I do know that you can’t force the data to show you what you want to see. A five-week sample of production data is too small. Also, it’s possible that if I looked at year-to-date or 16-game market share data, the results would be different. I plan on exploring these possibilities more in the offseason.
