Does Decision Quality Predict the Future?

If swing/take decision quality is a real, stable skill, it should do more than repeat itself from one season to the next. It should help predict how many runs a hitter produces the following year, beyond what you'd already expect from that hitter's raw talent and recent results.

We tested that using every hitter with at least 750 pitches seen in two consecutive seasons from 2021 to 2026: 1,354 season-to-season pairs. 2020 is excluded because the 60-game season is too short and atypical for year-over-year comparisons.

A hitter's RAA/100 in one season lines up strongly with the next (average Spearman correlation +0.60 across the five season pairs). Hitters who chose well last year tend to choose well this year.

Diff/100, the gap between what actually happened and what the hitter's choices were expected to produce, is less stable at +0.32. Diff/100 mixes luck with real bat-to-ball skill: for most pitches the expected value uses league-average contact rates, so a hitter who consistently makes better contact than the league will show a persistently positive Diff/100.

Same population and seasons as the year-over-year table on the ESV methodology page.

Next, we predicted each hitter's run production in year N+1 from what we knew in year N: their actual run production, their quality of contact (expected wOBA on balls in play, plus hard-hit rate), and their RAA/100.

Adding RAA/100 barely helped. A model using only prior production and contact quality explained 22.7% of the season-to-season variation in run production. Adding RAA/100 left that unchanged at 22.7%, and the same was true with hard-hit rate also in the model (24.4% both with and without RAA/100).

The plain reading: “this hitter's decisions are consistently good (or bad)” is a claim the data supports. “Good decisions predict future production” is not supported by this test. Plate discipline appears to be a real, stable skill, but its effect on next season's run production is small next to how hard a hitter hits the ball and how they produced last year.

Consistent with this, whether a hitter got a full season the following year tracked their actual results, not their RAA/100 or the expected value of their swing/take choices.

This is a limitation we found by testing, not a caveat we assumed. It bounds what RAA/100 should be used for: describing a hitter's approach, not forecasting output.

We set out to show that decision quality predicts future performance. It mostly doesn't: the stronger claim didn't hold up once talent was controlled for. We're publishing it deliberately. Knowing where a metric stops working is part of validating it honestly, and a front office should know exactly which claims RAA/100 can carry (a stable read on a hitter's approach) and which it can't (a forecast of next year's production).