Spots 
While browsing pitch-type stats on Baseball Savant, I started wondering: "Does a pitch that got a lot of whiffs one year still get them the next year?" and "Does a pitch that did well one year still do well the next?"
Was that year's number good because the pitch is good, or because it happened to be a good year? That probably depends on the metric. So I used public MLB data to look at this:
For a given pitcher and pitch type, how strongly is this season's number related to next season's (same pitcher, same pitch)? How much does that differ between metrics (whiff rate, run value, and so on)?
For a given pitcher and pitch type, how strongly is this season's number related to next season's (same pitcher, same pitch)? How much does that differ between metrics (whiff rate, run value, and so on)? I have no professional baseball experience; this is just a look at public data. Baseball Savant publishes these numbers for each pitcher and pitch type. I mainly looked at five of them.
Run value condenses a pitch's results into one number, so it is a number you see often. The second half of this post digs into it a little.
Take the same pitcher and pitch type in two consecutive seasons as one pair, e.g. "pitcher A's slider in 2023" and "pitcher A's slider in 2024". Collect many pairs and compute the correlation between the first and second season. Near 1: a pitch that was good this year is about as good next year Near 0: this year's number has little to do with next year's Subtract the pitch-type average first
Levels differ a lot by pitch type: sliders get many whiffs, sinkers few. Correlating raw values would come out high just because sliders get more whiffs no matter who throws them.
What I wanted to see is whether "this pitcher's slider is good among sliders" persists, so each value is taken relative to that season's average for that pitch type (e.g. "how many points above the average slider whiff rate").
With few pitches, numbers are noisy, so pairs are split into four bands by the smaller of the two seasons' pitch counts. Roughly, 100–199 pitches is a reliever's third pitch and 800+ is a starter's main pitch.
Baseball Savant pitch arsenal stats, 2017–2025, from a dataset I publish (Hugging Face yasumorishima/mlb-stats, marts/mart_scouting_reliability) Only pairs with at least 100 pitches in both seasons: 8,022 pairs The in-progress 2026 season is excluded Except for usage, every metric correlates more with more pitches. Fewer pitches means more noise, so this was expected
While browsing pitch-type stats on Baseball Savant, I started wondering: "Does a pitch that got a lot of whiffs one year still get them the next year?" and "Does a pitch that did well one year still…