Read on Twitter
Sup, Friday ... one of my fav podcasts was talking a bit about analytics in football - and it’s a subject that I’ve always been super interested in. So it inspired this thread on “advanced metrics” in football...
First off, prob important to distinguish between advanced metrics in baseball vs football.
MLB NextGen measures really useful data in a purely objective manner. Exit Velocity, Spin Rate, Swinging Strike %, etc - all provide highly reliable and meaningful info.
MLB NextGen measures really useful data in a purely objective manner. Exit Velocity, Spin Rate, Swinging Strike %, etc - all provide highly reliable and meaningful info.
the thing about Football "advanced" metrics is theyre way more volatile and way less useful = bc they essentially all involve some level of interpretation. Some incl subjective “grading” of players .. Stat I heard being discussed was “CPOE (Completion Percentage Over Expected)”
Quick sidebar: I dunno who created CPOE; not trying to make thread about personalities involved - but assume it comes from some smug StatsBro World dude… article introducing CPOE headlined “The NFL is Drafting QBs All Wrong” so typical, I’m-smartest-guy-in-the-room stuff
Anyway, the point is to look at whatever the components of CPOE are to see if it could potentially be as useful as the objective NextGen stuff that we get in baseball.
So what are the components of CPOE? Are they hard data like exit velocity and spin rate?
So what are the components of CPOE? Are they hard data like exit velocity and spin rate?
apparently, CPOE merges various data points, then adjusts them, to create a new metric. Seems that Step 1 is to combine Raw Completion % + Avg Depth of Target (ADOT) to get “Expected Completion %”
... and we’re already off the rails as far as objective, hard data.
... and we’re already off the rails as far as objective, hard data.
just randomly took Completion %; mixed it with ADOT; and called it "Expected Completion %"Apparently, the hypothesis there is that combination “helps correct the deficiencies in raw completion percentage.” Shall we continue?
Once you have “Expected Completion %” from Step 1, the next operation is to adjust for “the level of competition a player faced”. The goal is to boost players in situations where it is “tougher” to complete a pass and ding players where it is “easier”
Oh my lawd…
Oh my lawd…
I didnt see anything in the intro article that explained exactly what went into the "boosting" and "dinging" but I'm sure it was very scientific 
The end result of randomly creating an "Expected Completion %" and then randomly adjusting it based on some unstated criteria is...
The end result of randomly creating an "Expected Completion %" and then randomly adjusting it based on some unstated criteria is...
CPOE!
Well, technically, create Expected Completion % in Step 1; then adjust it based on how “tough” or “easy” things are in Step 2; then take whatever % you got from Step 2 and subtract it from the QBs actual Completion %
Wheeeeee! "advanced" metrics!
Wheeeeee! "advanced" metrics!
CPOE looks like it will be easily as predictive as a coin flip; and probably far more reliable than something like the old 80s Magic 8-Ball that attempted to predict the future (with very poor results in most cases!)
