Regression and Depression

[DOUGHBOYS]

re·gres·sion
[ri-gresh-uhn] noun
the act of going back to a previous place or state; return or reversion.

I put this definition at the top of this post because this word is starting to piss me off.
By definition,
Mike Trout cannot regress.
Matt Harvey cannot regress.
Bryce Harper cannot regress.

These players cannot regress because there is no starting point to regress to!
Do we use minor league stats?
No, that's not fair.
Do we use an estimation of where their stats should be?
No, because everybody has a different opinion.
How about we stop using the word?
There's a reason, regression rhymes with depression!

Regression is a 'smart' word. When we use it, we think we're smarter cuz we're using a three syllable word.
And every time we use it in describing young players like Trout, or Harvey, or Harper we use it incorrectly.
We use the word 'regression' by this definition in fantasy baseball-
Re.gres.sion
noun, verb, adverb
'Aw c'mon! He can't be this good! There can only be bad times ahead cuz he can't get much better, right?'
A word that describes a player as TOO good.
Or, while other players are taking viagra for 10 minutes of fun, Trout, Harvey, and Harper are the one's that have to call a doctor four hours later....And folks are jealous and want them to be as limp as others.

No players had 'regression' a few short years ago.
It was not part of the fantasy baseball lexicon.
Now that it is, we think it's another way of saying, 'There's no way he can keep this up'.
And four hours later, they are proved wrong.
Every breakout year that any player has, will now be followed and hounded by the word, regression.
Some players get to regress differently.
Let's take Jose Bautista. The first few years of his career were very Yuni Betancourtish. Then, he hits a ton of homers.
Regression wasn't a big term then, so most experts 'stuck in the middled' him. Which is to say, they took his usual numbers before his breakout year, then took the numbers from his breakout year, and made their projections in the middle.
If Bautista hits 70 homers this year, they would do the same thing. Only his base point would now not be Yuniish, but Bautistaish.
But what if Yuni Betancourt himself hits 40 homers this year?
Do experts project regression and give him previous years numbers in their projections while screaming regression?
Or do they 'stuck in the middle him' too?
In this day and age, they can do both. They'll cry regression, while projecting 25 homers for Betancourt next year.
It's human nature.

There was a post that Matt Harvey has only faced poor hitting teams and will now regress.
What does he regress to, exactly?
He has always had good numbers.
He hasn't pitched a full year in the majors.
His splits show hardly a weakness.
He virtually has no place to regress to!
The easy road is to say that Harvey cannot be THIS good.
But, he can. And it's not easy to accept.
But, I regress....oh sorry, that word comes too easily, now doesn't it?
I mean, I digress.

[ToddZ]

My take on regression. I wrote about this on www.fantasyalarm.com but here's the Cliff Note's version.

In it's truest definition, regression is revert to or approach a mean. It today's vernacular, it has become synonymous with "play worse".

When someone says Trout will regress, most of the time they are saying Trout will play worse.

THIS JUST IN -- EVERYONE WILL REGRESS!!!!!

EVERYONE!

There are elements of a player's performance that are in part random. It's cliche, but part of a player's performance has a luck aspect - some are lucky, some are unlucky -- completely independent of their talent.

Performance baselines are skills driven, but in any given time, the actual performance level can be above or below the baseline. Sometimes it is because of skills maturation or decline. Other times it is simply dumb ass luck, good or bad.

Regression is bad luck reverting to normal luck.
Regression is also good luck reverting to normal luck.

Everyone regresses.

Everyone.

That said, the problem with Trout, Harper and Harvey is we don't yet know the skills aspect of their performance baseline, so when the good luck turns to regular luck, we don't know exactly where the baseline will settle.

This is the part some are too obstinate, naive or even ignorant to accept. Some of Trout's 2012 success was due to plain old dumb ass luck.

When players with exceptional skills also are favored by Lady Luck, subjective bias infiltrates and stains objective analysis.

That's what's happening with Trout.

From both sides.

[DOUGHBOYS]

As usual, Todd, SPOT ON.

[Money]

Doesn't this also apply to the good? Does it always have to be bad? What about the guy that is under performing? The regression to the mean is based off of other peripheral numbers that indicate better things to come. Does this not suggest a regression to the upside?

[DOUGHBOYS]

Doesn't this also apply to the good? Does it always have to be bad? What about the guy that is under performing? The regression to the mean is based off of other peripheral numbers that indicate better things to come. Does this not suggest a regression to the upside?

We never hear the word 'progression'. Do we, Joe?
Unless Josh Hamilton pulls a 'Dunn2011', we know that he'll start hitting and start PROgressing to his normal numbers.
It just seems like most of the public would rather talk about a player coming down to earth rather than a guy picking himself out of the mud, the blood, and the beer.

[OaktownSteve]

Doesn't this also apply to the good? Does it always have to be bad? What about the guy that is under performing? The regression to the mean is based off of other peripheral numbers that indicate better things to come. Does this not suggest a regression to the upside?

Yes it sure does. I would add to what Todd wrote that we can regress a players performance to their own statistically established baselines, which as has been pointed out, is slightly more effective the longer the track record of that player. But there are always going to be outliers in any large data set. Players exceed or fall far below their own career averages all the time. And when they do so, it is some combination of skill and luck. In the case of Bautista, many attributed it to luck and predicted a lot of regression. Many saw it as a change in the talent level (new approach at the plate) and saw less regression.

In addition to comparing a player to their own statistical averages, we also compare them to the established historical averages for all players. This is how we know, for instance, that a player who hits .500 over the course of two weeks probably isn't going to hit .500 for a whole year. They are going to regress towards (though not necessarily all the way back to) the historical batting average for all players for that season and for all time.

I bring this up in particular to talk about the case of Trout. I believe that everybody who is in the business of making predictions should have predicted some regression for Mike Trout. Not because he might not turn out to be the greatest player in history. He may. The odds are against it, but he may. But the reason that we should have predicted regression for Trout is because of the incredible distance between his season last year and the historical averages for all players. The further one goes from that average, the more likely regression becomes. A player might hit .340 for a season. They might hit .370. Maybe even .400. But they will not hit .500. Won’t happen. I’ll say it again, the further you are from established norms the more likely regression becomes.

The issue in part here is not that people don’t understand what regression is. It’s that they don’t understand what predictions are. Baseball predictions based solely on a statistical model, that are as non-qualitative as possible, are just a range of possibilities. You are just saying, given what we know about numbers this is what is most likely and it is a range or continuum of outcomes, each slightly less likely the further you travel from the mean prediction for that player. But we are not saying, “this is for sure what’s gonna happen.” Saying Trout would regress is a statistically accurate prediction and at the same time does not necessarily tell us what Trout will do.

[ToddZ]

Doesn't this also apply to the good? Does it always have to be bad? What about the guy that is under performing? The regression to the mean is based off of other peripheral numbers that indicate better things to come. Does this not suggest a regression to the upside?

As I said...

Regression is bad luck reverting to normal luck.
Regression is also good luck reverting to normal luck.

David Price's K/9 and BB/9 are within his career norms. His BABIP and HR/FB are extremely unlucky. There's probably some bad pitching elevating those numbers, but he's also been UNLUCKY.

With normal luck going forward, Price should be just fine.

Hamilton is a little different. While there are elements of his game that are unlucky, there is also evidence of a marked skills decline (mostly in contact). Yeah, things will get better, but his performance baseline could be lower than a few years ago.