[Business Week] Extreme Moneyball: The Houston Astros Go All In on Data Analysis

Yep, right now he has no out pitch. That is what caused him to give up those 2 runs in his last 27 2/3 innings, because right now, he simply has no out pitch

 

Yep, it also caused him to give up 25 runs in the 26 innings before that.

See how easy it is to play with sample sizes.

 

No, I don't. I say that the data as a whole is suggesting that for this particular pitcher, based on all the inputs they have, it's expected that he would regress. There are no absolutes in baseball - ever - but I'm suggesting that it's the best prediction based on all the data they have - inputting everything from age to stuff to competitiveness to results to mechanics and on and on. All of that gets analyzed and the computer spits out its best projection of what a player is likely to do going forward. So from there, you have to pick between trusting the data or your eyes. Which do you choose? The answer to that question - when the data and eyes disagree - is whether you support the system or not.

The analytics take into account of these things, though. It's not like the machine doesn't think 24 year olds can improve. But it looks at other factors to see the likelihood of those improvements for a *particular* 24 yr old. Has he shown improvement in the past? Do the mechanics lend themselves to small tweaks? What's the underlying cause of the walks - bad control, trying to do too much, inability to get swinging strikes, etc? All of this is part of the analytical process.

All of these can be inputs into the computer - really if the system is any good, it absolutely should and must be taken into account. Without them, I would argue any analytical system is pretty pointless.

That's certainly true. But common sense can be used to understand the Astros' approach. We don't have any idea about the specific inputs or outputs of the black box, but the general approach is public information.

Again, you can keep saying that - but in reality, the Astros data told them they should make this trade. We know that or they wouldn't have made it. So if you're advocating against the trade, you're saying you're picking what you see over the data that the Astros are using. It's not a bad thing and you may be absolutely right. But it's like me saying the Sun is Blue. By definition, I'm arguing that the Sun is not Yellow, even if science says otherwise. I can't say I think the Sun is Blue but I support the idea the science that says the Sun is Yellow. It's one or the other. I'm sure there will come trades that I disagree with too - but when I do, I will be arguing that I think their system is wrong or needs to be changed or whatever. Like you, I may be right or wrong in my analysis, but I will still be arguing against trusting their data and thus against their system.

Absolutely. Not sure the relevance, but I don't think anyone in the front office would disagree with that.

 

In a data driven system, or even in a scouting-based system, nothing will ever be 100%. Hell, this trade may not have even had a 65% approval... it may be closer to 55-45% chance of it succeeding vs. chance of it blowing up in their face. Obviously, even the slightest possible "advantage" was enough for them to pull the trigger.

Its still humans making the decision. Many more variables/factors likely go into play that the computer has no chance of taking into account... #1.) selection bias - Cosart wasn't a Luhnow guy. #2.) Likability - perhaps the front office didn't admire the way he handled himself, went to the media, or his ability to offer suggestion. #3.) Moran may have been the guy Luhnow really wanted in the draft, but more of his staff wanted Appel and the computer may have favored Appel by one or two percentage points.

Again, all variables that one has to take into account as to why the deal went down. You will never eliminate 100% of the human component... and with that, you will have success/failures that cannot be explained by analytics alone.

So my final take on this topic is yes, the computers/data can help guide decisions... but they're ultimately executed/carried out by humans that have thousands of inner variables that the computer could never have accounted for. I LOVE data... we all do... millions of people do fantasy football drafts every year based on nothing but analytic/past-performance/projection-based data that has nothing to do with the eye-test. I am a fan of analytics and everything the Astros are attempting to do (and hell, the way things are going right now, its the ONLY thing they're really doing that is newsworthy)... but it doesn't mean I have to accept every move they make as a for-sure "can't miss" deal (especially when data may only suggest a mild improvement, if at all, after a deal such as this)

With a possible 50-50 chance of this trade blowing up in their face, I'll still offer simplicity and take the young MLB pitcher over the prospect.