Friday, January 18, 2008

Can you predict a clutch performance?

The clutch hitter debate is one of the longest-running debates in the sabermetric community. There's no question that there are clutch hits, and that you can look at players' performances and see how they've performed in the clutch in the past.

But can you predict how hitters will perform in the future?

The best work I've seen addressing this issue is in The Book by Tom Tango et al., who found a small but significant clutch effect...just one that is probably too small to make for reliable predictions. Nevertheless, given how many comments you hear about clutch players every season, it's clear that the majority of baseball fans and personnel don't buy into those conclusions.

Two really fun projects have been launched for the 2008 season that will serve as a great opportunity for those who advocate a predictable clutch skill to put their money where their mouth is.

First, Tom Tango is putting together a clutch project that will ask fans to vote for the guy they think is the #1 clutch hitter on their respective teams. He will then identify, before the season starts, the player projected to have the best overall performance next season. We get to follow those players all season long, and at the end, he'll take the 30 clutch players (one per team) and the 30 unclutch players and pull each player's 50 most crucial plate appearances (probably measured by pLI).

If fans can predict clutch performances, the prediction is that the clutch players will perform better--or at least comparably--in their 50x30=1500 most crucial plate appearances than the unclutch players perform in their 1500 most crucial plate appearances. If clutch skill can't be predicted, then the unclutch players should perform better.

I posted about this over at RedsZone, and it looks as though Reds fans will pick either Edwin Encarnacion or Ken Griffey Jr. as their clutch hitter. Dunn will almost certainly be picked as the unclutch hitter. It'll be interesting to watch next season!


Second, Phil Birnbaum is willing to bet any reasonable sum of real money (and maybe some unreasonable sums as well) that you cannot pick a player, set of players, etc, that will improve more in the clutch than any other player, set of players, etc.

So, if you're at least 67% sure that EDE will improve his performance in the clutch more than Dunn will in 2008 (and you get to define how "clutch" will be determined--RISP, pLI, close & late, whatever), Phil will take that bet. If you just want to bet that Dunn is terrible in the clutch, Phil will take that bet too--you can compare Dunn's performance in the clutch to that of the rest of the league in the clutch.

The prediction, if clutch is as difficult to predict as previous work indicates that it will be, is that the bets should should come out in Phil's favor 50% of the time. Given that he's asking 2:1 odds, that would result in a payout to him of $2 for every $1 he loses. The break-even point is 67% accuracy in clutch (or choke) player predictions.

I hope someone takes Phil up on it--should be fun to watch!!

5 comments:

  1. What about clutch pitching? I have not seen any information on how clutch pitching effects wins. I would be winning to bet money that a team's wins is directly correlated to how well their pitchers perform in high leverage situations.

    ReplyDelete
  2. I lost $5 on the 2006 election, and the sting still resides - no more foolish betting. And my impressions are probably not supported by hard data, but I did feel last year that one major weakness in the Reds lineup was the failure of Griffey and Dunn to deliver when needed. It was our 1-2 whiff. With Dunn especially it seemed that when a pitcher had to bear down, he was beardownonable.

    ReplyDelete
  3. First, as a brief announcement, I'm aware of the bizarro issue with my site's background, and I'm working on it. No idea what is going on, as I haven't changed anything in my template.

    On to the comments!

    I would be winning to bet money that a team's wins is directly correlated to how well their pitchers perform in high leverage situations.

    There's no question that this is true. If your closer (for example) was shaky in 1-run games and awesome in 3-run games, you'll have been worse off than if things went the other way around.

    The question is whether you can predict clutch performances for particular pitchers in the future based on past performances. ... This was addressed in The Book, but I honestly can't remember how large the effect was. :(

    I did feel last year that one major weakness in the Reds lineup was the failure of Griffey and Dunn to deliver when needed. It was our 1-2 whiff. With Dunn especially it seemed that when a pitcher had to bear down, he was beardownonable.

    One way that we can measure historical clutch performance is with the "clutchiness" stat over at FanGraphs. Basically, it measures how someone's WPA compares to what their WPA "should have been" based on their overall numbers. So, +1 clutchiness means a player's impact on games was +1 win more than their overall batting line would predict.

    Here are some select Reds hitters' clutchinesses from 2007:
    Dunn: -1.10
    Encarnacion: -0.01
    Gonzalez: -1.12
    Griffey: -1.39
    Hatteberg: +1.10
    Hopper: +0.41
    Keppinger: -0.12
    Phillips: -0.04
    Ross: -0.91
    Valentin: 0.26

    So, yes, in 2007, Griffey, Dunn, Gonzalez, and Ross earned the Reds ~1 win less than you'd expect them to based on their overall numbers--simply due to how they performed in particular situations. Gonzalez's "hot streak" performances early in the season, often came in games that were already out of reach, so they didn't help his WPA very much.

    The issue is whether those '07 performances--or even their career clutchiness performances--mean anything for their 2008 performances. Past studies indicate that it means very little, and that most (though not all) of this variation is just due to randomness. Dunn, for example, had a clutchiness rating of +0.42 from 2002-2006, and posted a +1.96 clutchiness in '04-'05. And yet, he struggled in big situations last year. This is very typical when you look across a given player's seasons.

    ...it's also worth nothing that despite the fact that Dunn's clutchiness was -1.0 last year, he still was second on the team in WPA at 2.66. He's just an extremely valuable hitter.
    -j

    ReplyDelete
  4. And clutch fielding!

    That's the excuse people give for Jeter - "well, he's money in the 9th inning" (as a fielder)

    Wouldn't that be easy to check - whether high-leverage situations or simply compare the last 2-3 innings?

    ReplyDelete
  5. I imagine that someone could do that using an approach like Sean Smith's, since they'd have it all tied to their retrosheet databases. Beyond my skillz though, at least for now. -j

    ReplyDelete