The end of RBIs?

Keep in mind that the body of evidence for clutch pitching is pretty strong. Be it the ability to respond to pressure or just an extra gear they can't break out regularly, some pitchers certainly have the ability to pitch better in key situations.

 

Did I say it was all bullshit? No, I did not. For a guy so fascinated with facts, you really don't care much about getting things right, do you?

 

How does anybody explain Jose Bautista's 2010 season?

 

The human factor.

 

Forgive me for falling into the role of statistics professor -- hey, it's my day job, OK? -- but let's at least start from a common place.
Let’s take a player, say Derek Jeter. He either is, or is not, better “in the clutch.” There are no other possibilities, right?
Suppose we look at the body of evidence and make a determination with regard to his “clutch-ness”. There are four possible outcomes. They are:
Our conclusion
|_______ Better _____________ Not Better _______
Jeter actually is |
Better | OK Error
|
|
Not Better | Error OK

If Jeter really is better and we conclude that he is better, all is well. If Jeter really isn’t better and we conclude thusly, again, no problem. If Jeter’s better, and we conclude that he isn’t, we’ve made a mistake. If Jeter’s not better, and we conclude that he is, we’ve also made a mistake.
Anybody disagree with anything I’ve written so far?

 

The lede to the latest LeBatard column dovetails nicely with this discussion.

http://www.miamiherald.com/2011/03/08/v-fullstory/2102109/miami-heats-recent-losses-troubling.html

 

I think it would be fun to compare the tone of gamers from last-second shots that went in and ones that didn't. I guarantee you it affects the way the writers frame everything that happened to that point.

Shot in: Focus on all the positive things that happened.

Shot out: Focus on all the negative things that happened.

We write it like the result of the shot was preordained by everything that came before it.

 

And athletes have an exasperating tendency to "explain" their performance in such simple terms.

"I just decided nobody was going to stop me on that last shot."

"I just decided to take over."

Oh, so you mean on that last-second shot you bricked last week that you didn't "just decide" to be unstoppable? Why not?

 

Damn, there are a lot of gamers and columns and sidebars I'd love to have back.

 

I disagree. The question is not whether he has performed better or worse in the clutch, because we can easily show the answer to that with the stats. The question is whether those results represent a true skill, or if they are just random chance.

If I correctly call three coin flips in a row and another guy incorrectly calls three in a row, am I better at calling coin flips than him?

 

I think you misread my post BBB, because you're agreeing with my post (as I intended it to be read). Jeter may indeed be better in the clutch, and therefore the results we've seen are consistent with that skill. Alternatively, Jeter may not be better in the clutch (he might even, in fact, be worse), but what we've seen is simply random error masquerading as "clutch" performance.

 

The statistician, being true to the scientific method, tends to be skeptical with regard to claims such as "Jeter's a great clutch hitter." Thus, he/she sets up the test so that the burden of proof is on the other side of such a claim. He/she formulates what is known as a null hypothesis. In this case, the null hypothesis would be that "Jeter isn't actually better in the clutch." The null is then put to the test; if the evidence is sufficiently against it, the null is rejected (and, by default, the alternate -- i.e., "Jeter actually is better in the clutch") -- is the only hypothesis left standing).

Now, what constitutes sufficient evidence? The statistician knows there's a chance he/she will be thrown off by sampling error and therefore reject a true null hypothesis. Such an error (in the lower left-hand corner of my diagram) is known as a Type I error. The probability of making a Type I error (known as alpha) can be decided upon prior to conducting the experiment/analysis. Alternatively, a test can be conducted and its "significance" level evaluated; the significance level is the likelihood that, if you reject the null based on this evidence, you'll be making a Type I error.