THE BOOK--Playing The Percentages In Baseball

Thursday, September 23, 2010

Does money buy wins?

Suppose you go to Macy’s, and you find a great coat for 200$, and a crappy one for 125$.  You happen to like shopping around, and you hit the discount stores (TJ Maxx, Marshalls) and you find that identical great coat for 100$.  What did we just learn?  Well, money buys quality, but context gets you quality too.  This is how it works:

talent + noise -> wins
talent + service time + miscalculations -> salary

You have a great manufacturer, and he can lead to you a great coat.

You can also have a great manufacturer who cuts a deal with discount stores but not regular stores, and you get the appropriate cost of that coat.

Would you then run a correlation of the quality of coat to the cost of the coat?  Well, yeah, you can.  But, that’s not really what we are after, is it?

Does money buy wins?  Well, talent costs money, and talent leads to wins.  But there are other known, easily identifiable parameters that also are important to know (service time).  Without knowing that, it looks like you can never have a strong relationship between money and wins.  That’s what alot of the economists will tell you.  Well, I agree with them that if you intentionally ignore an important parameter, you will know less about the relationship. 

The other thing, as we know, is that even if god ensured that all teams had equal players, that you will get 5 teams winning more than 87 games, and 5 teams winning less than 75 games.  So, you will always get a lower correlation than a casual observer might think.  When noise makes up such a large share of the variance, and when service time is ignored, it’s easy to see why your correlation will be under .50.  It doesn’t prove anything.

That’s why when I do my quick studies, I collapse ten years of data for a team.  Service time is still a parameter, but gets mitigated (unless you have a team that always jettisons free agents).  The noise around win% gets reduced substantially.  When I do that, I get r=.70 between wins and salary.

On the other hand, if you do a correlation of talent to wins over a ten year period for each team, I have no doubt that r would be at least .90.  And if you do correlation of talent+serviceTime to payroll (at the team level), you will get r of at least .80, if not also approaching .90.

Of course, if you do that, we have no story.  The narrative of money buying or not buying wins will show that it’s a canard.  But mainstream media loves buying those ducks.