Thread by @AdamHarstad

Volume is not more important than efficiency. Most of the time, volume *is* efficiency.

https://twitter.com/ir_dannyt/status/1368084874305929217?s=21

See also: Harstad’s Razor. (I’m never going to stop trying to make this happen.)

https://twitter.com/adamharstad/status/1233428466353221634?s=21

Toy model. Imagine NBA players see 5 “amazing looks” per game, shots that the average player shoots 60% on.

And they see 5 “good looks” (50%) and 5 “okay looks” (40%) and they see as many bad looks (30%) as they want to take.

Imagine 2 players. One shoots 58% on 4.5 shots per game and the other shoots 52% on 19.3 shots per game.

The former is “more efficient”, right? Nah, he’s cherry-picking amazing looks. *SOMEONE* has to take some okay shots sometime, and the latter is doing it (and doing it well).

The context of this thread: late-drafted receivers average about as many yards per route run as early-drafted receivers. So drafting a WR early doesn’t get you a more efficient player, right? Wrong! Early WRs are running way more routes.

Toy model: 1st round WRs run 3x as many routes as 6th round WRs. Both average 1.7 YPRR.

That 1.7 YPRR represents the average of the top X% of 6th rounders (say 15%) but the top 3X% of 1st rounders (say the top 45%).

If we looked at the top 45% of 6th rounders, they’d have way less than 1.7 YPRR because the 16th-45th percentile can’t even get on the field.

If we only looked at the top 15% of 1st round WRs, they’d have way more than 1.7 YPRR. (I’ll leave the proof to the reader.)

So in this case, the higher volume of routes among 1st round WRs means they’re more efficient, on average, *even if they aren’t “more efficient” on average* (because selection effects prevent a large part of the later sample from dragging things down).

Volume here is efficiency.

Another example: when Kyler came in I saw people comparing career YPA of sub-6’ QBs to 6’+ QBs and finding they were identical or the short guys even had an edge.

Proof that height doesn’t matter? Nah, because short guys were wildly underrepresented. There were barely any.

The short QBs in the sample were predominantly Drew Brees, Russ Wilson, Doug Flutie— good quarterbacks! The taller guys included Andy Daltons and Brock Osweilers.

Height can be overcome, but if it didn’t matter why were there so few mediocre short QBs?

Systemic bias is a plausible answer here: a similar analysis in 1985 could use the shortage of black QBs as evidence of the inferiority of black QBs despite their good efficiency stats. Just saying “the average is the same!” isn’t prima facie proof that something doesn’t matter.

(Similarly, part of the explanation for early WRs getting more routes *has to be* systemic bias, as teams give them more opportunities. Though giving more opportunities to bad receivers should actually drag the average down, so this is another point in their favor.)

Anyway, that’s the mechanism underlying my claim that volume is often efficiency. Given two groups of unequal size but equal “efficiency”, if one group is much smaller because it systematically excludes the lower-efficiency members, that group’s apparent efficiency is a mirage.

Similarly, if you looked at the YPA average of QBs by a certain characteristic (draft round, height, games started, number of vowels), but only included guys with 7.0+ YPA in the first place, I’d imagine you’d find little difference between the groups.

The key factor here is the percentage of QBs with a certain characteristic that makes it past the filter, not the average performance once they do.

(This is also the mechanism underlying why yards per target is not an efficiency metric. Because volume in this context is efficiency!)

Also, I hesitate to call these in advance rather than just letting time tell, but I suspect “volume is efficiency” might be making it into heavy rotation alongside “everything is selection bias” and “people underestimate the pace of modern medicine”.

My schtick is expanding.

@threadreaderapp unroll pls