Adjust Batters baseline

[jcard]

The “adjust players with fewer than X at bats” function is described as using a weighted regression to an “average” baseline, whereas the “weaken players” function uses a “replacement level” baseline. In operation, however, the “adjust” process is using a sub-replacement level baseline. By using thresholds of 0, 150, 200, and 300 (the default) with the 2001 season I determined that a batting average of a little under .200 was being used. Nomad Garciaparra, for example, had a .306 player editor average with a threshold of 0, and a .213 with a threshold of 300; simulating the season with the latter threshold he finished with a season average of .185.
Sure enough, when I set the threshold to 10,000 at bats so that everyone would basically be regressed to the baseline, every batter finished with a batting average in the editor between .184 and .193 with an OBP right around .250 and SLG a little over .300.
I am not nit-picking over a specific definition of “average”, and I think regression is very useful. It is no more possible, however, for a regression to any definition of “average” with any regression weight to adjust a player downward to below-average than it is to put warm water in a refrigerator at any refrigerator (this, above freezing) temperature for any length of time and end up with a glass of ice.

[jcard]

Given that Lukas himself recently spoke to the importance of this function in a post on the “Historical Simulations” sub-forum, there should at least be some clarity here as to whether this function is operating as intended. If it is, then a more precise articulation of the baseline for regression needs to be given- as it stands, it doesn’t meet any reasonable definition of “average”. While it will (at the default settings) rein in the player who hit an unrepresentative .310 in 150 at-bats by making him a .250 hitter, it does so at the cost of turning the more common case of a typical backup who hit .250 in 200 at-bats into a .230 hitter. The result is that teams have largely unusable benches and back ends of the pitching staff and bullpen.
I would argue that the “adjust” baseline is more appropriate for replacement-level baseline that is the stated definition the “weaken” baseline. Indeed, in the previous post I gave the resultant expected slash line statistics that (with extremely little variance) were found in every player’s “player editor” when setting the at-bat threshold extremely high (thus, “adjusting” everyone to basically the “average” baseline used by the function. Compare those numbers to those found in a recent study of “replacement level” described in the top half of this article:
http://www.highheatstats.com/2020/06...level-hitting/

[Brad K]

What is your view on the production of the reduced players being distributed, via the league totals function, to better players? Wouldn't it be better to make all the adjustments within the pool of players being adjusted?

[Lukas Berger]

Originally Posted by Brad K

Wouldn't it be better to make all the adjustments within the pool of players being adjusted?

What you're suggesting isn't just some sort of player ratings distribution change, but would require ground up redesign of the engine where league totals aren't even used any more, which isn't likely to ever happen.

[Lukas Berger]

As to the main suggestion here, I can try to remember to mention this to Markus at some point in the OOTP23 dev cycle, but if we change anything here I would guess it's more likely to be the wording of the blurb than the regression itself.