Thursday, November 20, 2008

Evaluating a Player's Career "Range" (Part II)

My first attempt at evaluating a player's career "range" can be found in Part I. I had some reservations about the list that that initial attempt generated, so I approached it in a slightly different manner.

This is what I came up with on a second look:
So, I ran the numbers using 60% of Peak Avg as the basecamp level.

Just to recap: we're trying to measure a player's consistent "greatness". Our first try used a flat rate of 20 (or 27) Win Shares as the floor - or basecamp, in our mountain metaphor - of that greatness, and then summed up the total Win Shares earned above that 20 (or 27) in a single season. By doing that, we would be measuring both the breadth and height of a "mountain range" (a single player's career), and this would hopefully allow us to more logically quantify a player's career (to compare "consistently very-good-to-great for many years" players, like Hank Aaron, and "good-players-with-ridiculously-high-peaks" like Sandy Koufax).

That first run gave us some results that we weren't quite sure about. (talked about in a posting up the thread) The suggestion, then, was to use a varying basecamp number, based off of some percentage of the player's peak average. The thought behind this was that it would give us a measure of a player's consistency *in relation to his peak*. Players with a low peak (ie, players who never had a seasn better than, say, 18 WS) would naturally fall off the listing because they would never accrue big enough numbers to compete with the hall-of-fame caliber players.

You can see the list with the variable basecamp here:
http://wezen.net/stats/ws-seavar.txt

Did it work? Honestly, I don't think so. All of a sudden, we have Hank Aaron rated as the #1 player (up from #9 on the 20WS list); Babe Ruth drops down to #2 (from #1 on all other lists); Bonds goes from #2 to #8; Honus Wagner drops from #4 to #12; Mantle falls from #8 to #22; Walter Johnson falls from #15 to #28; Gary Sheffield climbs from #35 to #21; and Rickey Henderson and Paul Molitor break the top 30. Now, I like that Rickey shows up higher, but I don't like the way we got there.

Why do Ruth and Bonds and Wagner and Johnson and Mantle (among others) fall so hard in this ranking? Well, take a look at the following charts (forgive the lack of active links):

http://wezen.net/stats/charts/ws-ruth.jpg
http://wezen.net/stats/charts/ws-bonds.jpg
http://wezen.net/stats/charts/ws-aaron.jpg
http://wezen.net/stats/charts/ws-rickey.jpg
http://wezen.net/stats/charts/ws-molitor.jpg

For players like Bonds and Ruth, who have such high Peak Avg's, their new base camp (the orange line - based off of 60% of Peak) is much higher than the old basecamp (the cyan line - shown here at 25 WS). In short, those players are being punished for having a high peak average - the method here is ignoring their win shares in the 20-30 range, a range which would be completely counted for most other players on the list. Looking at the charts for Molitor and Rickey, you can see the that new basecamp (again, the orange line) is lower than the 25WS line and, in Molitor's case, much lower. They are getting credit in this ranking for Win Shares that are basically junk to someone like Bonds or Ruth.

Players like Aaron and Gehrig benefit the most from this ranking, it seems, and I think that's because their Peak Avg was such that their variable base camp gets calculated to right around 25 WS, placing them right in the middle of that too-high and too-low problem that Ruth and Molitor exemplify.

Yes, we are measuring their consistency and their consistency in relation to their peak. And this is valuable because it helps us determine who stayed at that level for a long time. But I don't think it works for ranking who was best across their career (I mean, is Paul Molitor really 13 spots better than Joe DiMaggio? and I'm a Milwaukee guy...).

In the end, we are trying to determine an "all star win share" value, like alljoeteam suggests, and I don't think it makes sense for us to define this value as a varying number. Somewhere, there should be a value we can agree on as "all star level" and use that.

[I did try running the numbers where the basecamp couldn't go higher than 25, so that players like Bonds and Ruth were using the 25WS line; it definitely changed the rankings, but ended up making them look too close to the 20/27WS runs I did previously].
I gave this process some further thought after originally posting this. I do think there is value in this process, but it's just not what I was originally looking for. In the end, we get a list of the most consistent quality players in baseball history. We get the Paul Molitor's and Hank Aaron's of the world, and that's a worthwhile list to have.

I did one last tweak of the list, weighting the variable basecamp values by the peak. The list that it generates gives us the best ranking of consistency, I think. It seems weird to see someone like Babe Ruth so far down the list, but it makes sense. As excellent as Ruth was, he didn't consistently reach those peak values (yes, his seasons were consistently better than, say, Paul Molitor, but they weren't all at that *same* level of excellence), so his showing up a little lower is fine.

No comments: