In October 2004, Ben Murphy & Jared Weiss wrote an excellent article on BP's website — titled something like "Predicting Future Salaries."
http://baseballprospectus.com/article.php?articleid=3547
Murphy & Weiss used Value Over Replacement Player (VORP) and some regression math to try to predict future salary growth, estimate Beltran's yearly salary, etc etc.
They came up with these two graphs:
What was odd about this finding? … After they used VORP to determine players' actual values …. ie, John Olerud worth twice what Randy Winn is, and half what Jim Thome is, etc etc …
BP found that GMs' actual decisionmaking had already turned out to be where computer analysis said it should be!
Murphy & Weiss are obviously flabbergasted:
When we regressed the left side of the equation to the right side, we were amazed to find an r-squared of .93, and yes, the coefficients are quite statistically significant from zero. The .93 means that our model, using previous year's salary and VORP, explains 93% of the change in salary.It may seem odd that VORP is the other factor in the model, especially in light of recent findings. It is doubtful that more than a handful of GMs even consider using VORP in determining player salaries, so why does it work so well in our model?
It is our opinion that VORP, as a successful comprehensive metric for individuals, comprises a successful measure of the opinions of GMs in the aggregate. That is to say, while an individual GM may not look at VORP, the group of general managers combine to look at factors which approximate VORP.
In other words, using only their own intuition and judgment …. GM's brains were tracking with VORP with 93%-plus precision, despite never looking at VORP.
In scientists' language, Murphy & Weiss are blown away by this phenomenon…
They then go on to credit themselves for finding a model that "works so well to predict GM's."
;- ) They weren't trying to predict GM's, of course; they were trying to establish inherent market value for players.
=======================
What is amazing here, is the GMs' intuition. As we've mentioned before, the human mind processes a lot of things that we are not aware of, and processes things that machines couldn't possibly. When a human being smiles at you, you are aware of whether he is sincere, or whether he is affecting a smile. A scientist might later discover that the crinkles at his eyes are a major "tell," but you didn't need to be aware of that before you acted on it.
GM's, also, look at player value and, after 20, 30, 40 years' worth of watching players change baseball seasons, have a "feel" for what matters. They have intuition.
VORP is the culmination of 20 years' worth of sabermetrics, and here you lay a regression analysis on it …. you finally pull yourself over the last mountain peak so that you know about what players are really worth …
And you find that GMs had already gotten there. Just using human judgment.
Isn't that cool?,
jemanji

