UPDATE: I’ll refer you to our updated (beta) projections as of January 30.
Can’t wait to think about who to draft for next year? We have just published our early projections for 2009!
We’ve covered all the 5×5 categories, save saves. Those will be added as we get closer to Spring Training and find out who will close for each team. The main changes you will see from now on are park adjustments as players move around, as well as playing time adjustments as team announce who their starters will be at each position and on their pitching staff.
Methodology
We’ve taken the 3 year statistics for each player, much like the Marcels method, but instead of looking at the counting stats, we prefer to use rate stats and batted ball data in order to arrive at our numbers.
Batting average is calculated in two ways and averaged. The first is based on historic contact rate and other rate stats, and the second looks at GB/LD/FB distribution as well as historical BABIP data. HRs are extrapolated from estimated balls struck, FB% and HR/FB%. Runs and RBI are done on a simple ratio with PA, since batters are likely to be used the same way they have been in the past. Steals are calculated based on OBP, historic percentage of steal attempts and percentage of successful steals.
For pitchers, we estimate walks, hits and strikeouts based on an estimated number of batters faced and historic rate stats. That gives us WHIP and strikeouts. HRs are estimated by HR/FB% data. ERA is simply an expected FIP based on their 3-year rate stats. For starting pitchers, we look at ERA differential from the league average to get a rough Pythagorean winning percentage, adjust for number of expected decisions, then apply that to the expected number of starts. Reliever wins are calculated simply from historical win rate data.
6 comments
Comments feed for this article
November 15, 2008 at 5:55 pm
tangotiger
Please click on my name for an invitation to a Forecasters Challenge.
Tom
December 16, 2008 at 10:46 pm
Dave
Guessing you missed the boat on the Cory Lidle projections.
December 17, 2008 at 3:56 am
redsoxtalk
Yeah, that’s pretty funny that he’s in there. I projected anyone who played in the last three years and reached threshold values for PA or IP. In future releases of my projections, I will weed out these types of players.
December 31, 2008 at 5:59 pm
Brandon
i cant open your projections can you email them to me?
bragaller@yahoo.com
January 6, 2009 at 11:33 pm
steve shane
Isnt using 3 year avg’s as the basis of your projections not fundamentally sound? Players increase their skill (and stats) as they reach their late mid-late 20’s and their skill (and stats) decrease (unless theyre a BALCO client) in their mid-late 30’s.
Just seems like if a 26 yo player, who has consistently improved each of the last 3 seasons, and the underlying stats support his improving stats, that you would avg the last 3 seasons instead of projecting a continuing improvement.
Also, some projection services use a ‘regression towards the mean’ approach which I also believe is fundamentally flawed bc players can and do increase or descrease their abilities depending on what part of their career they are.
January 7, 2009 at 2:04 am
redsoxtalk
Hello Steve, thanks for your questions. The year to year fluctuations in stats means that using a multiyear average is a lot safer than any given year’s numbers. “Consistent improvement” could be nothing but a bad year, an average year, then a good year in that order. Take Coco Crisp, who looked like he was going to be a star his first three years. A youngster showing improvement everywhere, only to have his stats crash to his “real” skillset in Boston (some of that was injury-related, but not all of it).
Sample size is always a problem, as even a few lucky base hits can really spike a player’s average. The way to get around that is to look at basic rate stats, (like K rate and BB rate) and batted ball data. These actually don’t change too much from year to year, and tell us more accurately about what a player is really capable of doing, and is not governed as much by the competition he faces, the opponent’s defense, bad official scoring or even dumb luck.
You are right in that players tend to peak in their late 20s and start declining at 30. However, the amount of true improvement is merely incremental in most cases. Also, players age differently, so this trend is not so straightforward to apply on an individual basis. Still, I am working on an aging component as part of my revised projections, based on the player type (superstar, speedster, etc).
There is nothing wrong with regression towards the mean, so long as you choose “the mean” carefully. I regress players conservatively based on their position and player type.