Monday, August 18, 2008
What Does Xavier Nady’s Fluke Season Tell Us?
The Yankees still have a long hill to climb if they're going to be in the playoffs this season, but that's not Xavier Nady's fault. Since being acquired from Pittsburgh on July 25, Nady has been killing the ball, with a line of .312/.391/649 as a Yank, with 7 HRs in only 77 AB. Nady has an OPS+ of 145 overall this season including his time in Pittsburgh, after never breaking 107.Nady is 29, which means he's past the point where gains are typically made, but that doesn't mean that he hasn't improved. He's likely not as good as he appears to be right now, but there's a very good chance that he'll be better in 2009 than he has been in every season other than 2008
Nady's performance to this point got me thinking about fluke seasons and if they have any predictive value. So, I pulled up my Lahman database and started messing around with it to see if there was anything I could put together that would help me answer my question.
After fiddling around with a few different things, I came up with an approach that made sense to me.
1. I did very basic Marcel-like retroactive projections back through 1978. I used a 3/2/1 weight for seasons N, N-1, N-2 (where N is the current year). I did not do any projecting of the first two years of any player's career, and I did not include any MLEs or park factors. I then added in 500 league average plate appearances from the season in question to regress towards the mean and factored in aging for the component stats as well.
2. When that was done, I compared the projections for every season and player to what they actually did. If a player exceeded his projection by a certain amount, I considered that a fluke season.
3. I then calculated two separate projections going forward (N + 1) from every fluke season. One using the actual fluke season data, and one used the projection entering the fluke season instead. Those two projections were then compared to the actual performance in year N+1.
4. For the comparisons, I used wOBA (weighted on base average), which is basically the rate version of linear weights, scaled to OBP. So, a wOBA of .300 is bad, a wOBA of .330 is around average, .400 is great, etc., wOBA is easily converted to runs using the formula PA x wOBA / 1.15.
5. I set my cut off for a fluke season at a wOBA 15% better than projected. I also eliminated any seasons of fewer than 300 plate appearances. Why 15%? Because according to this post at The Book blog, the standard deviation for wOBA is SQRT(wOBA*(1.1-wOBA)/PA). So looking at an average .330 wOBA over 500 PA, we get an SD of .022, which means that 2 standard deviations better than that would be around 14%. Also, I'm only looking at positive fluke seasons for now.
6. This is not supposed to be a rigorous study. If it was, I'd want to do a lot more adjusting for context, with league and park adjustments and I'd probably include more than just the past 30 seasons. This is just a quick and dirty look to see if there's something there.
I don't know about most of you, but when the term 'fluke season' comes up, I instantly think of Brady Anderson in 1996. Anderson entered 1996 with 72 HRs in 3271 career ABs, with a career SLG of .393. At 32, odds of him improving would seem slim. So what happened? Anderson hit 50 HRs and slugged .637. It was one of the most incongruous performances in MLB history.
Now, obviously Anderson was never that good again, but it's interesting to look at what happened in 1997.
| Player | Brady Anderson |
| Year | 1996 |
| Age | 32 |
| Projected wOBA for 1996 | .346 |
| Actual wOBA for 1996 | .431 |
| wOBA Difference | 24.7% |
| Original Projected wOBA for 1997 | .342 |
| Revised Projected wOBA for 1997 | .379 |
| Actual wOBA for 1997 | .375 |
| Original Difference | 9.6% |
| Revised Difference | -1.0% |
So we see that coming into 1996, Anderson was projected to have a wOBA of .346, but he was 24.7% better than that at .431. Anderson's original projection for 1997 just used his 1996 projection instead of his actual 1996 numbers. The revised projection used the actual numbers. You can see that a subsequent comparison shows that even though his 1996 was a fluke, it did tell us something, as Anderson's revised projection was much closer to his actual 1997 performance than his original projection would have been.
One example doesn't make a case of course, so here are a few others to look at.
In my mind, the second biggest fluke season ever was Adrian Beltre in 2004. Beltre had youth on his side as well as a very good reputation as a prospect, but had hit just .262/.320/.428 for his career entering 2004, with an OPS+ of 97 and a career wOBA of .313. His 2004 retro-projection would have been for a line of .250/.303/.425 (wOBA of .310) thanks in large part to a dreadful 2003 where he hit just .240/.290/.424.
So what happened?
| Player | Adrian Beltre |
| Year | 2004 |
| Age | 25 |
| Projected wOBA for 2004 | .311 |
| Actual wOBA for 2004 | .418 |
| wOBA Difference | 34.5% |
| Original Projected wOBA for 2005 | .314 |
| Revised Projected wOBA for 2005 | .359 |
| Actual wOBA for 2005 | .305 |
| Original Difference | -2.7% |
| Revised Difference | -15.2% |
Yeah, Beltre hit .334/.388/.629 instead, good for a wOBA of .418. When compared to his projection coming into 2004, this was an even bigger fluke than Anderson's, although as I mentioned before Beltre was relatively young which made improving more realistic for him.
Unlike Anderson, Beltre gave back all of his gains and then some in 2005, as his projection for 2005 would have been closer if we used his 2004 projection instead of his 2004 actual line.
I've got one more single case to look at, which I'm hoping is predictive in the case of Nady,
Paul O'Neill came to the Yankees in a trade for Roberto Kelly in the 1992-1993 offseason. O'Neill was about to turn 30 and to that point had hit .259/.336/431 for his career in Cincinnati (wOBA of .323). He'd have projected to hit .256/.345/.418 (wOBA of .322) in 1993. Instead, he hit .311/.367/.504(wOBA of .370).
| Player | Paul O'Neill |
| Year | 1993 |
| Age | 30 |
| Projected wOBA for 1993 | .322 |
| Actual wOBA for 1993 | .370 |
| wOBA Difference | 14.9% |
| Original Projected wOBA for 1994 | .325 |
| Revised Projected wOBA for 1994 | .346 |
| Actual wOBA for 1994 | .435 |
| Original Difference | 33.6% |
| Revised Difference | 25.7% |
In O'Neill's case, he had tangibly improved and 1994 saw him hit even better.
Lastly, here's a look at the combined data for all players who exceeded their projected wOBA by 15% or more in any given season from 1978 - 2007.
| Player | All |
| Year | N |
| Projected wOBA for year N | .324 |
| Actual wOBA for year N | .389 |
| wOBA Difference | 20.1% |
| Original Projected wOBA for N + 1 | .324 |
| Revised Projected wOBA for N+1 | .352 |
| Actual wOBA for N+1 | .346 |
| Original Difference | -6.5% |
| Revised Difference | 1.6% |
The general point here is that a fluke season can often be an indicator of a change in a player's skill. While the fluke season itself overstates it, when it is rolled into the player's projection going forward, we can see that in many cases the player has tangibly improved. This is probably common sense but I always like to validate CW statistically if i can.
| Player | Xavier Nady |
| Year | 2008 |
| Age | 29 |
| Projected wOBA for 2008 | .339 |
| Actual wOBA for 2008 | .411 |
| wOBA Difference | 21.2% |
| Original Projected wOBA for 2009 | .335 |
| Revised Projected wOBA for 2009 | .353 |
| Actual wOBA for 2009 | TBD |
| Original Difference | TBD |
| Revised Difference | TBD |
Let's hope Nady's more Paulie than Adrian.
Update: More Random Pie Charts!
Page 1 of 1 pages: