Hot Spring to Hot Summer Connection

[bigh0rt]

No, not temperature. The common belief is that there is absolutely zero correlation between Spring Training performance and Regular Season performance. This is what I have thought for some time now. One of the fellas over at ACTA Sports thinks there may be at least some form of predictive nature in the Spring...

A few years ago we discovered that there is a way to use spring training stats to predict future performance. We took all spring training hitters and found that, as expected, about half of them do better than their career norms in the upcoming season, and about half of them do worse than their career norms. However, when we chose only those players doing exceptionally well in spring training, we found that about three-fourths of them performed better than their career average during the upcoming season.

Our definition of “exceptionally well” was slugging 100 points higher in spring training than their previous career slugging percentage. Here’s the list of players who are currently 200 points higher so far this spring training. These 24 players might be heading for above-average seasons.

Some of the Usual Suspects made the list (Butler, Hamilton), as well as some players who have already taken off a season ago (Granderson, Upton), but at the very least, I'll be Watch Listing and keeping an early eye on some of the bounce back candidates (Furcal, Mora, Pudge), and the guys who have dropped off the radar (Burke, Shealy, Gwynn Jr.).

ACTA Sports | FULL STORY

[Pogotheostrich]

Setting the bar at 200 career AB seems pretty low to me. Especially when all they need to do is best their career averages for the year. I like to see it based on players with at least a few MLB seasons under their belt.

[Inukchuk]

Very cool link. I'll be keeping an eye on that list to see how they pan out. It could essentially help identify breakout candidates if it's consistent...

[Amazinz]

I'd really like to see the research they use to support the claim. However, both articles on that site concerning that topic make the assertion without evidence. (If I missed the article that does, let me know.) There's a reason people say spring training stats don't matter: sample size. I'd like to see how they overcame that and how in depth they researched it, until then I'm not buying. As Pogo pointed out, I question the merit of 200 AB as criteria for eligibility. I suspect that they are observing random chance and not a real correlation.

[GotowarMissAgnes]

I'm researching this, because if you look at their archive, they've created a list like this each spring since 2005. I'll do the rest later, but here's what I found from their 2007 list.

They names 36 players on that list.
22 of the 36 improved their OPS+ in 2007 compared to 2006 (for players with less than a full season, I compared their 2007 performance to their career performance; for other players, I compared 2007 to 2006). 14 of the 36 had their OPS+ decline in 2007.

15 of the 36 had their OPS+ improve by more than 10 points; however, 7 of those 15 are questionable, because they include guys who had limited playing time. Milton Bradley's OPS+ went from 114 to 153, but he also played in only 61 games. Similarly, big OPS+ increases by Mike Rivera, Cristian Guzman, Timo Perez, Greg Dobbs, Willie Bloomquist and, possibly, Garret Anderson, had little fantasy relevance. 9 of the 36 had it decline by more than 10 points; 3 weren't really relevant.

So, overall:
8 improved by more than 10 points and had significant PT
7 improved by more than 10 points, but had limited PT
7 improved by less than 10 points
5 declined by less than 10 points
3 declined by more than 10 points, but for PT or other reasons weren't really relevant
6 declined by more than 10 points

Some amount of the predictive power also seems to come from the fact that they compare players to their career slugging; thus, players that "broke out" in 2006, but still have somewhat low career slugging, fall into the list. To some extent, the spring numbers might "confirm" the 2006 breakout, but a better analysis for that would compare a group of 2006 breakout players and see whether those who showed up on this list were more likely to maintain or improve upon their prior year breakout.

[GotowarMissAgnes]

A look at this method from their 2006 data is even less positive. In that year, they again identified 36 players. They break down this way:

13 improved their OPS+ performance over prior year/career by 10 points
4 improved their OPS+ performance over prior year/career by less than 10 points
5 had their OPS+ performance decline from prior year/career by less than 10 points
14 had their OPS+ performance decline from prior year/career by more than 10 points

So, over the 2 years we now have:

28 improved by 10+ points
11 improved by 1-9 points
10 declined by 1-9 points
23 declined by 10+ points

I'm not seeing much predictive performance here.

[Matthias]

GTWMA,

I tried emailing this question to Tom Tango; haven't heard back a real response yet. Thought you might have an opinion on it.

Beyond a simple predictive power (X out of Y did Z) there should be a level of improvement that is statistically significant. I'm not sure what the normal ST vs. prior year SLG or OPS variance is in the MLB, but there must exist some level of improvement (even if it is extraordinarily high) that would be statistically significant, even for a limited ABs. Any idea on what that might be?

[GotowarMissAgnes]

Finally, their 2005 set of 30 players:

12 had an improvement of 10+ points
3 had an improvement of 1-9 points
5 had a decline of 1-9 points
10 had a decline of 10+ points

Three year totals:

40 had a 10+ improvement
14 had a 1-9 improvement
15 had a 1-9 decline
33 had a 10+ decline

Not seeing much there.

Yes, Matthias, you would probably want to test the statistical significance of these changes, but as we see with things like clutch hitting, none of these differences in such a small subset of at bats is in any way likely to have statistical significance. You could run a standard difference in mean score test, but it would show little. In addition, there's nothing in their list that controls for things like park factors and such.

All in all, I would not put much credence in this theory.

[Matthias]

Alright. Thanks for the number crunching, GTWMA.

[Amazinz]

Interesting. Thanks for the legwork, GTWMA.