Towards A Better Way To Rank Players

It’s not a secret – I hate rankings. I’ve been writing this post once a winter since 2014. I’m pleased to finally see a few others in the industry publicly eschewing the rankings-based mentality. My quest against rankings is finally gaining some momentum. So, contrarian that I am, let’s work on making rankings actually, uh, work.

There are many reasons why traditional attempts at rankings are broken. The biggest is simply this – they’re one size fits all. Running a team is about managing categories. Rankings aggregate all that information ordinally in perhaps the least useful way conceivable. Attempts to use tiers scarcely help matters, especially since the location of those tiers often has more to do with name recognition than expected output.

A projection-based rankings system isn’t sufficiently novel, but it does at least take us a step in the right direction. And it’s what we’ll use for today’s thought experiment as we try to take a second step along the path towards relevance.

Let’s start with the “best” way to build a team – know everything. If all information is fully internalized, then every decision can be made to maximize roto-output. In reality, the closest we can get to that is “be a fantasy professional.” And that’s impractical for most of you out there in Readerland.

It strikes me that categorizing players by position by profile might offer a reasonable middle ground. We’ll still have rankings of a sort but they’ll be bucketed in a way that helps us to manage our statistical needs.

I’ve taken the 56 top-rated second basemen by Steamer and ATC projections. I’ve averaged every category which is technically incorrect for batting average and OBP categories (AB and PA projections are close enough not to fuss around).

The first step is sorting each category. Behold, the ugliest table! (Apologies, this doesn’t quite fit and isn’t sortable. Ideally, you want this information in separate tables – I have each in their own tab – but that looks even worse for our purposes. I removed plate appearances for a hint of cleanliness.)

This is where I start to get stuck. Beyond simply digesting all of this information wholesale, we need to figure out a way to bucket into profiles. For instance, we might call one bucket “Power.” Let’s imagine it’s a middle round and we have identified a need for home runs and/or RBI. We want to focus on the Muncies, Odors, and Biggioes of the world.

We shouldn’t completely discard other categories in our bucketing attempt. Perhaps we give a one-half weight to runs and average/OBP with a 0ne-tenth weight to stolen bases. But that’s a job for another day. We’ll focus on just pure power stats for now.

It sounds like I’ve unwittingly set us up to use something like weighted z-scores. If we have to, we can cross that bridge later. My inclination is always to use the least mathematically complex option. If you want complexity, talk to Alex, Jeff, and Podhorzer. And so, here’s one overly simplistic attempt at a Power Second Baseman bucket.

-Please note: these next two tables should be considered illustrative only.-

That’s home runs + RBI sorted.

Similarly, if we’re hunting speed, it’s probably a mix of stolen bases, runs, and batting average we crave. I whipped together a pseudo-speed score using just those three categories (10*AVG+3*SB+R) and came up with…

For a bit of fun, locate Mondesi (16), Villar (26), and Merrifield (24) in the power table. This is a good example of how statistical need can greatly affect player value and why vanilla rankings are built to mislead their users.

It’s at this point I’ll turn things over to you. Maybe we just need better formulae? While I prefer constructive feedback, feel free to rip this completely to shreds. Just remember that everything above is illustrative. If your grief is my use of HR+RBI as a sorting criteria – yes, I agree it’s stupid. It was, however, something I could do with a couple keystrokes. We need to get the theory tight before it’s worth spending hours on Excel work.