Simpson's Paradox--so what are you supposed to _do_?

Couldn’t decide on the right forum for this. It may just be GQ. It may be debateable, but it doesn’t seem like the stuff of Great Debates.

By Simpson’s Paradox, we know that it’s possible for this to occur:

Batter A is better than Batter B against left-handed pitchers.
Batter A is better than Batter B against right-handed pitchers.
Batter B is better than Batter A against pitchers in general.

So my question is–what are you supposed to do with this information when deciding what batter to use? If it’s a left-handed pitcher, you might think you should send in batter A since A’s better against lefties. But since a left-handed pitcher is also a “pitcher in general,” you might think you should send in batter B, since B is better against pitchers in general.

So is the answer that it’s just indeterminate? Or is there something more that can be said here?

Only in strange circumstances. Specifically, if they face different numbers of left-handed and right-handed pitchers.

Imagine the following:

A bats .200 against lefties and .400 against righties. He’s faced lefties 100 times and righties 100 times. Total batting average: .300

B bats .150 against lefties and .390 against righties. He’s faced lefties 20 times and righties 180 times. Total batting average: .365

But if A and B face exactly the same pitchers, then the scenario you described is not possible.

No pitcher is a “pitcher in general”. If you know the handedness of the pitcher they are going to face (as you seem to be stipulating) the decision is straightforward.

But also, what ITR champion said. Either there is something relevant that you are not telling us about how “good” the batters are, or your scenario is not possible.

What ITR champion said. Batter A is better than Batter B. The only reason B looks better is because he’s faced fewer left-handed pitchers. If they had faced the same ratio, B’s batting average would be .270.

That is a plainly false statement. Every pitcher is a “pitcher in general,” since the phrase “pitchers in general” refers to the class of all pitchers.

Did you read the Wikipedia article on Simpson’s paradox?

ETA: I thought I linked it. Here it is. The situation I described is definitely possible, because actual.

Makes sense. I think the answer to the question of the OP is, “it depends on the specific numbers.”

Scratch that last bit.

I think another right thing to say here is that, in a scenario where Simpson’s Paradox applies, it will be illicit to make generalizations like the ones found in the OP. It may be true that Batter A has outperformed Batter B for a certain sample set etc but that particular sample set will turn out not to support a generalization that Batter A is “better than” Batter B against X-type pitchers.

Here’s an 8-page paper written by Judea Pearl about how Simpson’s Paradox occurs, and how to resolve it.

Pitching over to Game Room, from IMHO.

That’s exactly what I was looking for, thanks!

I swear I thought this thread was going to be about the Simpsons episode Homer at the Bat when Burns pulls Darryl Strawberry for Homer against a lefty.

Ha!

Even stranger, I thought it was going to be about this goofy game on my iPad I’ve started to okay called Tapped Out.

I am embarrassed. D’oh! :smack:

I’ve read the whole thread and haven’t seen any mention of Julius Hibbert or Nick Riviera.

Burns: You, Strawberry, hit a home run.
Strawberry: Okay, skip.
(hits a home run)
Burns: Ha-ha! I told him to do that.
Smithers: Brilliant strategy sir.

When Burns pinch hits Homer for Strawberry because of the LH/RH pitcher match up at the end of the game.

Burns: “It’s called playing the percentages.”

I actually was reading about the Paradox earlier today before I saw the thread.

Unfortunately, my brain’s a bit fried by symbolic logic. Could you relate the answer given in that paper to your specific example?

BobLibDem presents a real world example of this paradox in another thread.

When the Civil Rights Act of 1964 was put up to a vote, these were the results:

House:
Democrats: 152-96 (61-39%)
Republicans: 138-34 (80-20%)

Southern Democrats: 7–87 (7–93%)
Southern Republicans: 0–10 (0–100%)
Northern Democrats: 145–9 (94–6%)
Northern Republicans: 138–24 (85–15%)

Senate:
Democrats: 46-21 (69-31%)
Republicans: 27-6 (82-18%)

Southern Democrats: 1–20 (5–95%)
Southern Republicans: 0–1 (0–100%)
Northern Democrats: 45–1 (98–2%)
Northern Republicans: 27–5 (84–16%)

So a higher percentage of Southern Democrats (7% and 5%) voted in favor of the law over Southern Republicans (0% and 0%). And a higher percentage of Northern Democrats (94% and 98%) voted in favor of the law over Northern Republicans (85% and 84%). But overall, a higher percentage of Republicans (80% and 82%) voted in favor of the law over Democrats (61% and 69%).

The best known example is the Berkeley gender bias case.

This blog post was recommended by a stats prof in a course I took. Heavy on equations.

SIMPSON’S PARADOX EXPLAINED