When Sample Sizes Become Reliable

This is pretty much a mush have bookmark when dealing with small sample sizes. The comments section is very good also.

When certain statistics stabilize for individual hitters:

50 PA: Swing %
100 PA: Contact Rate
150 PA: Strikeout Rate, Line Drive Rate, Pitches/PA
200 PA: Walk Rate, Groundball Rate, GB/FB
250 PA: Flyball Rate
300 PA: Home Run Rate, HR/FB
500 PA: OBP, SLG, OPS, 1B Rate, Popup Rate
550 PA: ISO

over 2 years ago Jeff Zimmerman 1 comment 0 recs |

Story-email Email Printer Print

Comments

There's an article by Pizza about pitcher rates, too.

Googling it doesn’t work, but this THT article quotes the numbers:

K/PA: 150 BF
GB%: 150 BF
LD%: 150 BF
FB%: 200 BF
GB/FB: 200 BF
K/BB: 500 BF
IF FB%: 500 BF
BB/PA: 550 BF
BABIP: Doesn’t reach a 0.50 r-squared at 650 or below.
HR/FB: Doesn’t reach a 0.50 r-squared at 650 or below.

(FYI, r-squared of .5 is the same as an r of .70.)

Just to be clear, we’re not saying that stats all of a sudden become reliable at those BF totals. Pizza picked a somewhat arbitrary r-squared cutoff and found how many BFs were necessary to achieve that level of correlation. What this list is really good for is to show the relative significance of different stats and how soon the data stabilizes, relatively. We’re always regressing, even with five seasons worth of data, it’s just that you regress less the more data you have, and you regress less for different stats given the same amount of data. /soap box

Beyond the Boxscore // Calling BJ Upton lazy is lazy.

by Sky Kalkman on May 24, 2009 1:49 PM EDT reply actions