THE BOOK--Playing The Percentages In Baseball

When comparing the correlations of different statistics, isn’t relative frequency an important consideration?

If a player’s Contact% in a year is based on 1,000+ swings, while BUH% is out of on only 10 or 20 attempts, a much higher y-y correlation for Contact% might mean that it is a more predictive stat, but it may not say anything about the ability being measured (i.e. bunting skill could be totally static, but it would still have low correlation due to the small sample size).

Maybe this is part of the reason that so many of the top performers are plate discipline numbers?

My fantasy pts. league has a .95 correlation between OPS and Pts/PA. I’m perpelexed as to what stats to focus on for next season. I was thinking something like w/OBAfip. However, when I look at 2010 to this season I don’t necessarily see very good results. I see OPS y-t-y has a .60 correlation to OPS the next year in the link. Why not just use OPS? What am I missing? Any ideas how to approach next year’s draft?

Eggplant, I think you have a great point.  However, I think the question the author was asking was: Given stats you see in one season, how likely would you to see the same stats the next season from the same player?

So, yes, most players will have more Contact% opportunities in a given year than BUH% opportunities, but that’s part of the story of comparing one year to another year.  If you want to compare attempt to attempt numbers (i.e.  “Given 1000 Contact% attempts versus 1000 BUH% attempts, which correlates better?"), then that’s a different question entirely.

Thanks, Tango. Much appreciated.

Geri, I agree completely with your answer to the question you say the author’s asking. I’m just warning against drawing broader conclusions than the data supports.

For example, the author comments that “the majority of the plate discipline statistics show fantastic correlation year-to-year. It would appear that the degree to which a hitter is patient, a free swinger, shows good selection, etc, really doesn’t vary all that much.”

I’m arguing against the implication that these skills are more consistent than those measured by stats with lower correlations. Probably in some cases they are and in others not, but I don’t think you can tell from the correlations alone.

Derek Carty’s post from a few weeks back addresses the issue of “opportunities”.  Think of those two articles as two sides of the same coin.

Can someone explain what the values mean going across, and what they mean going down in the spreadsheet. Thanks.

"What’s puzzling to me is that while LD% is highly variable year-to-year for hitters, GB% and FB% are not. One possibility is that his reflects a coding error in the batted ball data, but if that was the case I would assume the other types would show similar variability. But they don’t.”

Typical results are GB 40, LD 20, FB 40

LD may be confused with GB or FB, but GB and FB are almost never confused with each other.

So an 3 point error in determining LF could look like this

GB 39, LD 23, FB 38

GB and FB look like the correlate Y2Y, LD does not.

From the spreadsheet, it looks like OSwing% is a really good metric for plate discipline.

For BB%, there is a good correlation to next years Isolated Power (.40) and Zone% (.37). This seems to imply that either power hitters are more disciplined, or that power hitters are thrown fewer strikes.  It’s probably some of each, with maybe the latter being a stronger effect.

ZSwing% has a noticeable correlation to IP (.21). So either power hitters hit for more power because they swing at more strikes (not wasting good pitches to hit and getting behind in the count), or they swing at more strikes because they have power and want to leverage it as much as possible. Either way, there is something else at play here besides just strike zone judgment.

But OSwing% has no correlation to next season’s IP (-.01), or BA (0.00). Even though batters with a low OSwing% tend to also take more strikes and therefore get more behind in the count, they apparently make up for it by putting fewer OOZ pitches into play, and swinging at better pitches to hit of those in the strike zone.

But the league OSwing% league average has varied quite a bit from year to year since 2002. Since the overall Swing% has stayed the same while the overall Zone% has also varied, I can only assume that the absolute OSwing numbers are unreliable. Better to look at OSwing relative to the avg for that season.