The Skill(s) of Winning 1-Run Baseball Games | Disciples of Uecker

Disciples of Uecker

We'd like to go to the Playoffs, that would be cool.

The Brewers came up short this weekend to a Cubs team that has finally started to play a bit closer to its true talent.  The Brewers are now 27-17, and their one win in the three-game weekend series came by a one-run margin. The concept of one-run victories is going to be our topic today.

The ability to win close games can make an enormous difference in a club’s end-of-year record.  The 2012 Baltimore Orioles were an 82-win team by run differential, but ended up with 93 wins and a postseason bid, thanks to a 29-9 record in games decided by one run. The 2013 New York Yankees were a below-.500 team by run differential, but ended up winning 85 games, thanks in large part to a 30-16 record in one-run games.

The Milwaukee Brewers are off to the best start in franchise history this year, and their record in one-run games is an important part of that success. To date, they have gone 10-4 in such games, a success rate that exceeds even their overall winning percentage this year, which is one of the best in baseball.

It is common to dismiss success in one-run games as “luck.” That is incorrect. Random variation certainly has plenty to do with it, but winning close games is something affected by skill, and it is a trait that this Brewers team fortunately seems to possess.

To prove that, we’re going to need to do some math.  (Sorry).

I selected five skills at which the Brewers tend to excel this year. I would describe them as follows: (1) Reliever Effectiveness (as measured by SIERA); (2) Team Defense, as measured by UZR; (3) Power Hitting, as measured by ISO; (4) Quality Start Percentage (three or less runs over six innings) for the rotation, and (5) Clutch Hitting, as measured by Tango’s Clutch statistic.

Clutch Hitting may seem like a curious thing to describe as a “skill,” as the concept of the “clutch hitter” has also widely been dismissed over the years as luck. To the contrary, though, Tom Tango and his colleagues found in The Book that about one in eight hitters possess the ability to be “clutch” — to hit above their normal statistics in high-leverage situations. This early in the season, I suspect that the “clutch” statistic is also useful for another reason: it can identify as “clutch” players who are under-performing at the moment, but showing flashes of their true talent. A few players in the Brewers lineup certainly could fit into that category.

In any event, we can use multiple regression analysis to determine if these attributes in fact correspond with an improved record in one-run games or not. To do that, I looked at all team seasons from 2010–2013, using the five skills I identified as predictor variables and each team’s end-of-season winning percentage in one-run games as the outcome variable.  The regression was designed to test the null hypothesis (a/k/a “the narrative”) that these skills are irrelevant and that the Brewers have nothing other than luck to thank for their success in close games.

The results were encouraging. There is in fact a moderately-strong correlation (~.5) between these attributes and a team’s success in one-run games. Although random variation still likely plays a significant role, about 25% of a team’s success in one-run games is explained by performance at the skills I laid out above. 25% may not sound like much, but given how closely most teams are stacked around .500, it matters. Only two of the attributes appear to be statistically significant in and of themselves (Reliever Effectiveness and Clutch Hitting), but even the others helped add stability to the model. Overall, the model’s findings are highly significant (p<.001). We can reject the null hypothesis, and safely conclude that it is no coincidence that the Brewers are doing well in one-run games this season.

That said, exactly how good should the Brewers be in one-run contests? The math is in the Appendix below, but the model predicts them to have a .608 winning percentage in those games, or essentially play like a 98-win team in one-run games. That’s pretty darn good, particularly considering that no one believes that the Brewers are a true-talent 98-win team (nor need they be, to get the postseason).

At the moment, with that one-run-game record of 10-4, the Brewers are outperforming even the .608 WP the model predicts. So, yes, that differential will probably come down to earth a bit. But if these other peripherals remain similar for the rest of the season, do expect the Brewers to continue to have an above-average winning percentage in close games, and for those who tell you it is nothing but luck . . . feel free to send them over here.

Statistics courtesy of Fangraphs and Baseball Reference.

Thanks to the many Twitter users who noodled ideas with me for predictor variables over the weekend.  Their social lives are obviously as exciting as mine.  They include Jeff Wiser, Kyle Ashauer, Ross Bukouricz, Corey, BerniesMustache Blog, Vineet Barot, Curt Hogg, That Enrico Palazzo, and Brad W.

Follow Jonathan on Twitter @bachlaw. 

 

Appendix

Here is the equation, per the model, for predicting winning percentage in one-run games, along with that equation solved for the Brewers on the date I ran the numbers:

1-run game WP = .0.6534 + (-0.0678 * SIERA) + (0.0001 * DEF) + (0.0015 * QS%) + (0.0783 * ISO) + (.0110 * Clutch)

Brewers predicted % = .6534 + (-0.0678 * 2.68) + (0.0001 * 31.4) + (0.0015 * 76) + (0.0783 * .140) + (.0110 * 72)

= .6534 – 0.181704 + 0.00314 + .114 + 0.010962 + 0.00792

=  0.607718, or ~.608 WP

Share Our Posts

Share this post through social bookmarks.

  • Delicious
  • Digg
  • Newsvine
  • RSS
  • StumbleUpon
  • Technorati

Comments

Tell us what do you think.

  1. Nicholas Zettel says: May 19, 2014

    Yes! Winning one-run games is a skill. This is excellent stuff.

    You know, years ago, I looked at team park factors in order to explain whether / how teams outplay (or underplay) their run differentials. I also wonder whether playing in an extreme park leads a team to win more one-run games…

  2. rk says: May 19, 2014

    this is cool stuff, but I was wondering if there isn’t another, trivial reason why the statistics you mentioned might be correlated with performance in 1-run games. Specifically, many of the situations which make up Clutch Hitting and Reliever Effectiveness are the very same situations which you are testing–1-run games. If you are going to regress them on each other, then you should remove the overlap between them, in other words, all the 1-run games.
    Maybe you did that and I missed it, in which case, comment withdrawn. But if you didn’t, then the correlations you observe may not be predictive at all (or at least less predictive than you report). They could be a result of players having done well in 1-run games, which led to them winning the 1-run games. The question, however, is whether those performances are repeatable in the next 1-run game.

    On another note, I wonder whether Clutch Hitting in some sample of plate appearances is very well correlated with Clutch Hitting in the next few hundred plate appearances.

    • Jonathan Judge says: May 20, 2014

      Thanks for reading.

      I think I understand what you are saying, and I disagree. The point of the analysis is to look for just those sorts of coincidences. If you do various combinations of other variables, you don’t see any significant correlations. If Clutch Hitting and Reliever Effectiveness correspond with one-run game success, well, that is exactly what we are trying to find out.

      I haven’t looked up the sustainability of clutch hitting for a while, but The Book does discuss the topic in some detail.

  3. studes says: May 27, 2014

    Performance in one-run games correlates with overall performance. So do the skills you’ve included in your correlation. To get anywhere, you first need to isolate this impact.

  4. Matt Swartz says: June 22, 2014

    The other comments are dead on, but effectively the question is that you need to ask is how this model works out of sample. Can you take this regression applied to the stats in one year and determine how it predicts to the following year? Or can you take this regression and apply it to the stats in the first half of the season, and see how well it predicts the results in the second season. Using in-sample data is going to bias your results. Don’t strive for R^2 of .25 or anything remotely that high. There’s no universal threshold for useful R^2 like some people think. If you could get an R^2 of .05 on data like this on out-of-sample data, you would probably have discovered something amazing.

Trackbacks

Websites mentioned my entry.

  1. Brewers Streak Heads to Minnesota | Disciples of Uecker

Add a Comment

Fill in the form and submit.