Free Preview
The following is a free sample of a column from Baseball HQ, one of hundreds readers receive during the course of their subscription. Get a competitive advantage in your league by subscribing today!
FANALYTICS: The Most Accurate Player Projections in the Universe
Ron Shandler - February 3
Sam Walker's book, Fantasyland officially debuts on your local bookstore shelves in about a month. I received my advanced copy this week; most of you will surely find it to be a fun and fascinating read. I'll be devoting my February 24 column to a preview of the future best-seller.
One of the many underlying themes of the book is how each of the "experts" in Tout Wars has his own corner of the universe that he protects with his life. Frankly, I'm one of those; this is my livelihood. However, part of our respective empires is a certain highly-prized offering, and I can't help wondering if it is worth protecting... our player projections.
From the beginnings of this industry two decades ago, "accurate player projections" has been the promotional power play that industry touts have always lorded over each other. This was never a friendly battle, but recent skirmishes have become more brutal and the stakes are now higher. Before it all goes nuclear, I think it is time to end this war, and so I plan to do so right now with a single statement:
Baseball HQ publishes the most accurate player projections in the universe.
It's true. Last year, Ray Murphy projected that Andruw Jones would lead the National League in home runs. I projected that Derek Lee would have a breakout season and Brian Roberts was the best buy at 2B for the price. We told you, right here on this very site, that Brad Wilkerson would collapse, Mike Lowell would implode and Rafael Palmeiro would get nailed for steroid use on August 1.
We said it all. Right here. It's true because I say it is.
We can say anything we want after the fact because very few people are going to take the time to check. Of course, the statements have to give the appearance of being authoritative and sound at least marginally reasonable. The last sentence in the above paragraph crosses that line, obviously, but the first two sentences look pretty sure-footed, don't they?
Taking it a step further, if we use numbers instead of words, it's even easier to play fast and loose with the truth. Mean squared errors? Correlation coefficients? Oh, ours was 0.75 and everyone else's was no higher than 0.73 so we clearly thumped the competition. Who is going to check? Heck, how many readers are going to know what I'm even talking about?
The introduction in the 2005 Baseball Forecaster discussed how projective accuracy is an elusive concept. I expanded upon that in a long essay called The Great Myths of Projective Accuracy, which we had hyped in our ads last spring with the headline, "Accurate Projections are a Crock!"
That's what I really believe.
But readers don't want to hear that. If you scan nearly any public message board, you will surely find a thread entitled, "Who's got the most accurate projections?" It would be nice if there was a fair way to answer that question, but there isn't. Any attempt to do a comparative analysis is going to be flawed. There are two reasons why: a) it is nearly impossible to remove bias from the process, and b) these analyses are, of themselves, just single data points in the global comparison of different systems. The variability of results from one year to the next is so great, and the variances so small, that claiming dominance is just silly.
In last week's column, I wrote a line that embodies the entire reality of baseball player prognosticating:
This is not a game of precision. This is a game of human beings and tendencies.
Once we embrace the implications of this concept, only then can we truly leverage our knowledge at the draft table and focus on what's most important. It's not the numbers. It's the marketplace. In fact, with all the advanced quantitative systems being developed and placed carefully up on pedestals, the truth about predicting a player's performance is so exceeding simple as to be mind-boggling.
All projections are relentlessly driven by two powerful, unstoppable forces: Regression to the Mean and the Gravity Principle.
What that means is, if you projected every player to simply perform closer to his career average or league average, you would end up with a compilation of numbers perfectly reasonable for going into your draft. In fact, the Gravity Principle says you should simply project every player to perform worse than last year, and that would be perfectly reasonable as well.
Todd Zola of Mastersball.com published a report a few years back that held some outwardly startling conclusions. In a new light, they make more sense now. Here are a few of them, based on data from 1996-2003:
- Of all hitters, 70% hit fewer HRs from any year to any subsequent year.
- Of players who hit 35 HRs in any year, 79% fell off the following year.
- Of all players, 75% stole fewer bases from any year to any subsequent year.
- Of all hitters who earned positive roto value in any year, only 32% earned more value the following year.
It's all just regression and gravity in full force.
This is all easy to grasp with a player like Andruw Jones. Regression and gravity tell us that the odds are stacked against another 51-HR performance. His career average is in the 35-HR vicinity; anything he does between 35 and 45 will be within an acceptable range of probability. When you get to the draft table, his final price will be determined by market forces, not the expectation that he is going to hit exactly 41 HRs in 2006.
Of course, that exact number is something we need to publish in order to calculate his expected dollar value. And that exact number is what all these comparative analyses will be screaming about at this time next year. But that "41" figure is almost completely irrelevant.
In the end, the true challenge for us is to identify those players whose performance will defy the regression. That's the gold for fantasy leaguers. It's no surprise that Jones will regress; what we want to know are the odds that he might maintain, or even improve. If the leading indicators are strong in a opposing direction, we'll project a gravity defying performance. We do that for many players. But those are 30% percentage plays, according to Zola's study.
Most players -- the other 70% -- are just going to follow some type of regression. That's a percentage play strong enough to provide a good foundation for prognosticating accuracy. But here's one better...
"Marcel the Monkey" is the assertion by folks on some of the sabermetric blogs that a "chimp forecasting method" — a simplistic averaging of the last few seasons and making minor adjustments for age — is nearly as good as any other, more comprehensive system. Well... this is mostly true. If 70% accuracy is the best that we can reasonably expect, Marcel gets us about 65% of the way there. All of our "advanced" systems are fighting for occupation of that last 5%."
Hmmm... wait a minute.
If 70% of players naturally regress...
And the best accuracy we can expect is 70%...
Then it seems that it doesn't take much to achieve the high water mark of prognosticating accuracy.
And creating the most accurate player projections in the universe becomes a no-brainer.
Receive weekly Baseball HQ insights via e-mail with a FREE SUBSCRIPTION to
Baseball HQ Friday. News analyses, gaming strategies and
Ron Shandler's Master Notes, every Friday, always FREE. Subscribe today!
