One of the things that’s helpful to ask when doing football research is: Have we seen this before? For instance, when I look at Devin Funchess‘ production at the end of 2015 it seems like it’s pretty impressive considering how bad he was at the beginning of the season. But is it actually impressive, or is that something that happens all of the time – a rookie going from horrible to kind of good in one season.
Answering that question of how often have we seen this before is fairly important. It’s the difference between a lot of what we see on Twitter, where someone will say that a player had X% of their runs for zero or no gain, and yet no context is provided to know whether that’s good or bad. And Pro Football Reference screeners can answer the question in some cases even though all of their cutoffs are hard cutoffs. If I’m looking at a player who averaged 50 yards per game and then I search on PFR for any player over 40 yards per game and under 60 yards per game, that will be helpful, but it will also leave out potentially relevant past examples where a player averaged 61 yards per game, but is similar to the subject player in every other way.
To make some headway on the problem I’m talking about above I’ve added some functionality to the RotoViz Screener that will conduct a similarity search using the variables you want to look at, for any multi-year time frame you want to look at. If you want to find other players who have had two year starts to their career similar to Mike Evans on the factors of age and targets, you can do that. If you want to find other running backs who have had three year spans of production similar to Frank Gore from 2013-2015, and where age is a factor in the similarity search, you can do that.
Let’s walk through an example. I’ll use the Mike Evans search that I mentioned above.
Let’s search for:
- Two year time periods, as specified in the “Max Seasons” slider.
- Wide receivers
- Per Game statistics
- Display the variables Age, Draft Position, Receiving Targets
- The click “Search the Database” in the middle of the screen
After you search the database you’ll get a table below in about one second or so.
Now let’s look at the output part of the app. First, it’s divided into Player Level and Team Level. If you’re looking for something like “most runs inside the 5 yard line over the last three years” to see if the Patriots run a lot inside the five, then you would use the Team Level tab. But we’ll spend our time on the Player Level tab today.
Then inside the Player Level tab there are several different data displays you can look at that run the gamut from single year to multi-year. We’re going to look at the Similarity Search tab. Once we click on the Similarity Search tab the only other thing we really need to do is type “Mike Evans” into the dropdown and his span of 2014-2015 will be selected.
Then we can inspect the results. You can see that the most similar two season spans to the one that Evans just finished (when accounting for only draft position, age, and targets per game) include Larry Fitzgerald, Justin Blackmon, Charles Rogers, and a number of other names. If we changed the display variables it’s possible that the names returned in the Similarity Search would also change.
Maybe the killer functionality of the app is that if you select a row on the table (just click it) then it will pre-populate the Compare Multi-Year tab with information. So if I’m working on a RotoViz article and want to discuss how Mike Evans’ early target numbers look a lot like Larry Fitzgerald1 then I can just click on those two table rows. Then when I click over to the Compare Multi-Year tab I have a ready made table that I can save for insertion into my article. If you write for another fantasy site you should always feel free to use these tables in your writing. Part of what we’re doing at RotoViz is trying to provide some infrastructure to fantasy writers so that they can streamline the mindless data accumulation and cleaning required to write articles.
And there’s the resulting table. There are lots of ways I could alter that query to look for more information. For instance, if I thought that Mike Evans was the beneficiary of playing in lots of blowouts then I might look at only his targets when trailing and see if there is anything remarkable about the players he is similar to in that regard.
Perhaps as a word of caution on the use of the Similarity Search tab, I find it to be extremely useful for informing my sense of a player’s accomplishments. But it is still a limited framework to assess those accomplishments and should only be combined with your other judgments of a player. For instance, yesterday I wanted to look at players similar to Jamison Crowder to see if his rookie year might indicate any future success. The names returned by the similarity search couldn’t have been any more underwhelming and included guys like Davone Bess. Does that mean I’m totally off Crowder? No, because I still think that if the right circumstances materialize in WAS he could be Jarvis Landry-lite. But it also made me realize that lots of young players have a season like Crowder’s and then go on to do nothing else. Maybe a decent rule of thumb is to take whatever you think the comparables list might be telling you about a player, and cut that conclusion in half.
Experiment with any searches you want to do in the app and play around with the Similarity Search for fun or for real league research. If you want to add a second dimension to the data – like if you wanted to split up a player’s first eight weeks versus his second eight weeks – you can do that by using the controls for Query 2. And the really cool thing about doing that is that you can still use the Similarity Search. I’m not aware of another tool on the web that would allow you to drill down and find similar seasons to Devin Funchess’ 2015 where you can look at both first half and second half splits to see what other players had such Jekyll and Hyde rookie seasons.
- mentions of Justin Blackmon and Charles Rogers are optional!!! (back)