Article: Geeking Out: Pitch To Contact and Team Batting

[John Bonnes]

You can view the page at http://twinsdaily.com/content.php?r=1720-Geeking-Out-Pitch-To-Contact-and-Team-Batting

[Teflon]

Interesting post. Thanks, John.

 

Perhaps there needs to be context around the base runner situation in which the strikeout occurred since the value of a strikeout over a fly out or ground out is that it tallies an out with no chance of a runner advancing. Also, I would be interested to know the correlation between strikeouts and runs surrendered to see if it is similar or dissimilar to the correlation (or lack thereof) to wins.

[wagwan]

Love the Calculus formulas. My students have a Calc final next week. I will send them to Twins Daily for their revision work. Not many people in Lebanon actually read this site daily. The data analysis stuff is just what my students need to see. You actually CAN use this stuff in the real world.

[Alex]

Let me first say that I appreciate any dive of this depth into statistics. It makes for an interesting read and an interesting discussion. I just read through both. However, there are many flaws in tying the correlations found in this study to value in players.

 

First, it's evaluating multi-outcome measurements against single outcome results. The more outcomes you add to a statistic, the higher your correlation will be by default. OPS, for example, measures six possible outcomes (BB, 1B, 2B, 3B, HR and outs by omission - including strikeouts). Is it any surprise that it correlates better than any other statistic or outcome when it includes all of them?

 

Hits/9 and average includes four-five outcomes (1B, 2B, 3B, HR, and the fifth again being outs) , which is pretty good, but people "savage" it (from your other response) because it does not include a key element that players can do to help their team win and add value. It's why WHIP works so well here because it essentially is a catch-all for everything -- of course would correlate. If you're giving up a ton of hits and walks per 3 outs, you're going to be in trouble.

 

Contrast that with K, HR, and BB, which measure a single outcome each. Assigning value as you do here to a single outcome based solely on correlation is dangerous. Your placement of 3Bs proves that. Is a 3B really that worthless? Sure it doesn't correlate to a win, but to argue that it is less valuable an event than a 2B would be preposterous, right? (This is why Tom Tango's work is so valuable)

 

So, in looking at evaluating a pitcher, we should be asking what tools we have, and which are predictive. Can we predict H/9 for pitchers? K's? Do K's correlate with H/9? If you're looking to criticize people for valuing strikeouts in pitchers because they don't correlate as well as you'd like to runs p/game, what other stats, along with HR and BB should we be using?

[jay]

Nice points, Alex.

 

You're absolutely right, but I don't see using multi-outcome stats to evaluate a player as a bad thing. I don't necessarily care how the contributions came about whether it was a single, walk, double, or homer -- I only care how that player contributes to the team winning over the long run. We win on offense by scoring runs, so the stat that best tells me how a player is contributing to scoring runs is by all means how I want to compare him to others. Same argument for WHIP.

 

Both OPS and WHIP are extremely easy to understand and come from basic stats that fans are familiar with. They remove the specifics of "how" and focus on the result. They also both tend to be strongly predictive of future performance. These stats are better evaluators than any signle outcome statistic which, frankly, I don't think we should use much at all.

[Willihammer]

Again I can't help myself. Its disingenuous to use K/9 to look at any relationship involving strikeout proficiency, unless you are purposely trying to discount the strength of that relationship.In the K/9 universe, walks, and hits, improve your strikeout numbers. It is not a huge discrepency, but its there, and its skewing your correlations. Example:

 

Pitcher A has a K/9 of 7.97. Pitcher B has a K/9 of 8.57. Pitcher B is the better strikeout pitcher right? No. Pitcher B strikes out 21.7% of batters, but Pitcher A is slightly better, striking out 22.0% of batters (Pitcher A is Jake Peavy, Pitcher B is Edinson Volquez). K/9 is an obsolete metric.

[Shane Wahl]

I really like this article a lot. That's pretty helpful. I also like Willihammer's response.

[nicksaviking]

Just throwing this out there and it might be grasping at straws, but bullpen arms generally have a higher K/9 than the starters. So a team that uses its bullpen a lot may have an inflated K/9. Of course a team that uses its bullpen a lot also is probably giving up a bunch of runs early.

 

Would the K/9 to runs/9 coefficient be different if we just factored in starting pitchers?

[Alex]

Nice points, Alex.

 

You're absolutely right, but I don't see using multi-outcome stats to evaluate a player as a bad thing. I don't necessarily care how the contributions came about whether it was a single, walk, double, or homer -- I only care how that player contributes to the team winning over the long run. We win on offense by scoring runs, so the stat that best tells me how a player is contributing to scoring runs is by all means how I want to compare him to others. Same argument for WHIP.

 

Both OPS and WHIP are extremely easy to understand and come from basic stats that fans are familiar with. They remove the specifics of "how" and focus on the result. They also both tend to be strongly predictive of future performance. These stats are better evaluators than any signle outcome statistic which, frankly, I don't think we should use much at all.

 

 

Agreed, so maybe I wasn't clear. I was pointing out what I think is a key flaw in the study in it's conclusion to discount strikeouts as a evaluative tool. In terms of single outcomes, that's a pretty significant correlation.

[jay]

Agreed, so maybe I wasn't clear. I was pointing out what I think is a key flaw in the study in it's conclusion to discount strikeouts as a evaluative tool. In terms of single outcomes, that's a pretty significant correlation.

 

Ah, I see. When you look at a stat like FIP, it's solely based on those three you pointed out -- K, HR, BB. They also all have the strongest correlations here among the single outcome pitching stats. So, without going into multi outcome stats, I think you answered your own original question.

 

However, I think John's point here is not to discount K's entirely, but that people single out K's too often and the signficance placed on them is a good bit greater than their actual value... especially among the fans in our base who have been rather starved of such value.

[Alex]

Ah, I see. When you look at a stat like FIP, it's solely based on those three you pointed out -- K, HR, BB. They also all have the strongest correlations here among the single outcome pitching stats. So, without going into multi outcome stats, I think you answered your own original question.

 

However, I think John's point here is not to discount K's entirely, but that people single out K's too often and the signficance placed on them is a good bit greater than their actual value... especially among the fans in our base who have been rather starved of such value.

 

Fair enough, but the next logical step in seeing how effective K's are is not to compare them to other stats, but to see their correlation to other stats above. Do pitchers that get more K's give up fewer homers and hits in general?

[Alex]

Again I can't help myself. Its disingenuous to use K/9 to look at any relationship involving strikeout proficiency, unless you are purposely trying to discount the strength of that relationship.In the K/9 universe, walks, and hits, improve your strikeout numbers. It is not a huge discrepency, but its there, and its skewing your correlations. Example:

 

Pitcher A has a K/9 of 7.97. Pitcher B has a K/9 of 8.57. Pitcher B is the better strikeout pitcher right? No. Pitcher B strikes out 21.7% of batters, but Pitcher A is slightly better, striking out 22.0% of batters (Pitcher A is Jake Peavy, Pitcher B is Edinson Volquez). K/9 is an obsolete metric.

 

Interestingly, using k%, the Twins starting pitchers look even worse (though it's tough to look much worse). In k/9 they hold spots 101 (Worley), 106 (Pelfrey), and 108 (Correia). In K% they hold 104 (Worley), 106 (Correia), and 107 (Pelfrey) (108 Qualified pitchers).

 

If Diamond had the innings, he'd be 105th in K% and 109th in K/9. Pedro Hernandez would be 102nd (K/9) and 99th in K%. Wow.

[snepp]

Again I can't help myself. Its disingenuous to use K/9 to look at any relationship involving strikeout proficiency, unless you are purposely trying to discount the strength of that relationship.In the K/9 universe, walks, and hits, improve your strikeout numbers. It is not a huge discrepency, but its there, and its skewing your correlations. Example:

 

Pitcher A has a K/9 of 7.97. Pitcher B has a K/9 of 8.57. Pitcher B is the better strikeout pitcher right? No. Pitcher B strikes out 21.7% of batters, but Pitcher A is slightly better, striking out 22.0% of batters (Pitcher A is Jake Peavy, Pitcher B is Edinson Volquez). K/9 is an obsolete metric.

 

 

Your larger point is right, but your final conclusion is blowing it a bit out of proportion. K/9 and K% have a very, very, strong correlation to each other. K/9 isn't obsolete, it's just ever so slightly worse than K%.

[old nurse]

Again I can't help myself. Its disingenuous to use K/9 to look at any relationship involving strikeout proficiency, unless you are purposely trying to discount the strength of that relationship.In the K/9 universe, walks, and hits, improve your strikeout numbers. It is not a huge discrepency, but its there, and its skewing your correlations. Example:

 

Pitcher A has a K/9 of 7.97. Pitcher B has a K/9 of 8.57. Pitcher B is the better strikeout pitcher right? No. Pitcher B strikes out 21.7% of batters, but Pitcher A is slightly better, striking out 22.0% of batters (Pitcher A is Jake Peavy, Pitcher B is Edinson Volquez). K/9 is an obsolete metric.

You may be correct on K/9 vs K% in terms of a better strikeout pitcher but not necessarily who is the better overall pitcher. If one pitcher has a better K/9 while the other has a better K% by a large variance then I would ask myself why isn't the pitcher with the high K% pitching more innings. Hence looking at the variation between stats could help you determine strengths and weakness.

[Willihammer]

Your larger point is right, but your final conclusion is blowing it a bit out of proportion. K/9 and K% have a very, very, strong correlation to each other. K/9 isn't obsolete, it's just ever so slightly worse than K%.

The variance between K/9 and K% is small but magnified in a situation like this.

 

The number John should have arrived at is -.65. That is a difference of .09 from the -.56 coefficient he arrived at. In other words, about 10% of the variance between the two figures is explained solely by the imprecise nature of the K/9 stat in measuring what it supposes to measure.

 

There is simply no reason to use it instead of K% unless you have a bias towards minimizes the importance of strikeouts.

[snepp]

You're trying to convince me of something I already said I agreed with.

[jay]

Do pitchers that get more K's give up fewer homers and hits in general?

 

I'm not sure. We can see a strong relationship between K% and ERA here, but you can make the same case for GB% and presumably BB%, XBH%, HR%, etc as well. So, in regards to your earlier question, maybe those are some other stats we can look at.

 

I do agree with John's premise that K (/9, %, whichever) gets valued disproportionally higher in many cases. At the end of the day, you can find many ways to skin the cat. Right or wrong, the Twins seem to prefer GB% and BB% over the others.

 

However, while the Twins have found pitchers good at GB% and BB%, they are really bad at K% and arguably not good at XBH% (although more defense dependent). It seems logical that you have to be at least marginally capable in all of those areas to be better than marginally effective.

 

It's far beyond my skill level, but is it possible to find some sort of exponential equation to value a pitcher based on being good at some or several of these that also de-values being really bad at some? Park and defense factors would be a challenge, but ie -

(GB% + or - from league average)^exponent + (K% + or - from league average)^exponent + ...

[Shane Wahl]

Great geekery here. Keep it up!

[Snortwood]

Here's the problem - and if we can get the problem statement correct we have 90% of the solution worked out - so here it is: Strikeouts can be context critical. There are innings and situations where a strikeout is the only non-dangerous event (passed balls and such aside). That is, we may be looking at this from too wide a lens if we try to evaluate Ks against 9 (argued above to be obsolete) or even in the context of a game. What matters is the moment.

 

When batters come to the plate with RISP and X out, we know pretty much exactly how well they've done. We don't get anything like that when pitchers are out there huffing and puffing and men are all over the bases. WHIP is a wide angle lens. It gives you an idea. But it isn't as interesting as, say ERA for starters in an inning after 2 out and nobody on. And why the guys on the mound for the Twins lead the league in that stat. Or trail, depending on how you look at it. Stats like that.

 

It's got to be context specific. Until we drill down that tightly and focus in that closely, we just won't know why some pitchers with fewer strikeouts are more successful than those who, based on numbers overall, might be thought to be superior.

[tobynotjason]

What other people said re: K%. Every time I see someone cite K/9 or BB/9 I cry inside.

 

My rant: an R squared of .05 (for BABIP) FTW? Really? (Not to mention, McCracken actually found an R of .153 for BABIP in the original study; hits aren't a DIPS stat: they include Home Runs.)

 

And as has been widely understood for almost decade, the year-to-year correlation in BABIP was pretty much totally explained by (1) year-to-year consistency in the team's defense/home park (players who switch teams show close to zero correlation: as in an r of .04); (2) DIPS themselves, which predict BABIP better than BABIP; and (3) pitcher-type tendencies, including the knuckleballer/junkballer effect and gb vs. fb pitchers, with a heavy emphasis on (1).

 

So yeah: strikeout rate and walk rate - ACTUAL rate, not per 27 outs - are THE thing.

[Carneal&Gordon]

Hawk says TWTW