Almost everything you’re going to see in this series was not an organic idea of mine. Every ingredient placed into the work i’ve spent months on is built off the shoulders of others. If you missed Part 1, you can read it here.
IMPORTANT NOTE: Due to some incorrect records in the Database, this article is no longer relevant. Part 3 contains the actual predictive qualities. See, this is why i stick to the blog. I can make mistakes and it matters less.
We’ve established two things about the top 24 IDP Linebacker:
- They are good in college from a young age.
- They are fast in a straight line.
But what we’ve found is a correlation, not something predictive. Correlation finds the relationship between variables but it doesn’t decode which are more important and valuable. For all we know, some of the values are distractions.
Is there a way to decode which values aren’t distractions? I wouldn’t be writing this article if I hadn’t already done that, duh.
I’m not showing the steps used in the process above, but all you have to know is I threw in all the data from the previous article, and then removed what isn’t predictive. And according to this model, the most predictive elements are:
- Percentage of Team Solo Tackles in College (Positive relationship)
- Lower Body Explosiveness, aka Burst (Negative Relationship)
- Height-Adjusted Speed Score (Positive Relationship)
- Age on Draft Day (Negative Relationship)
That should make sense. Solo Tackle Percentage should matter and as it increases so does your potential to be good. A good player is better at producing than a bad player. Don’t forget, the players that succeed in the NFL are the top 1-percent of all football players, so they have to stand out amongst any group that aren’t the top 1-percent. Jumping ahead, we get to Height-Adjusted Speed Score. Again, makes sense. Going sideline to sideline is an important skill. As your height adjusted speed-score increases, so does your potential to be good. And it’s not just important to be fast, but being fast for your size is also an huge versatility bonus. Next, Age on draft day has a negative correlation. So as age decreases, potential increases. A younger player is more likely to be a prodigy, and a prodigy is more likely to be good.
But we’re told that Burst is bad. As burst increases, potential decreases. This was also (slightly) visible in the correlation graphs in part 1, but making sense of it in football terms is difficult. I can try asking the model, but it’s unable to communicate because it’s software in a machine. So we’re forced to accept it. My guess is that players with high burst are more likely to be used ineffectively on the field, possibly as blitzing linebackers. That’s not the conclusion people wanted to hear, but it’s the only one that I can make.
However, that doesn’t mean we’re done.
This is only the current version of the model. With more numbers and ideas that are getting added to the project, the elements the model finds predictive could change. Even with data science, it’s not a given that what we’ve created is the end-all-be-all. Sometimes science is more art than science. A lot of people don’t get that.
There will be more parts in this long-running project, as I keep searching for predictive elements in off-ball linebackers. Stay tuned.