The depth charts (and the player forecast manager, PFM) got updated this weekend. The intention was, and still is, do revise them roughly weekly from here on out; that this one stretched out to a month was entirely my fault, generated by a bout of ill health and frantic catching up across the board.
But we are somewtat better (and at least able to function) now, and I’ve tried to catch up on the charts. One of the toughest decisions to make - and one which turned out to delay the process - was to rewrite them so that they only reflected what was left of the season, not what has gone before. A player who just came back now after missing all of April could be listed as a 90-95% player (no one, ever, lists at 100%) if he’s expected to play every day. The PFM should give $$ values for the player’s remaining value only. I am pretty sure that this is the right way to do it, but it did mean a lot of effort had to go into debugging the spreadsheet that I use to tie all of the information together - the concept that all estimates are for 162 games was hard-coded all over the place, a process that by itself took the better part of a week to sort through.
At this moment, the program is still using the base PECOTA projections from the pre-season, unmodified except for playing time. That is going to change, although I’m not sure exactly how - probably by using some combination of PECOTA modified by current performance, at least at first, and I hope eventually by a new PECOTA run that explicitly incorporates the current season. On the depth chart page itself, I think I need to add some indication of the team’s record-to-date, not just their expected future record, which is what you’ll see right now.
One thing I have done is to take the team projections from the depth chart and start using it in the PECOTA-based postseason odds report, instead of the pre-season PECOTA numbers. So as of this morning, the PECOTA-based postseason odds report does know that Manny Ramirez was suspended for 50 games, and does knock some points off the Dodgers’ success chances (though they’re still strong favorites). I’ll continue to update that, as injuries and trades change team strengths.
We haven’t tried to run these throughout an entire season before, so I’m sure there’s bound to be some more hiccups, and I’ll try to keep breathing evenly through them.
I understand there are alot of things to do, many more pressing than this. But getting the batting orders reasonably right needs to be on the list at some place. The San Diego batting order is laughable, after which I didn't bother checking any more.
It's almost exactly what they used last night against the Cubs... what's laughable? 7 of the 8 batters were the same (plus Henry "Peavy's Caddy" Blanco in for Hundley). Granted some swapped spots in the order (4 and 6, 7 and 8) but you're gonna get that in just about any line up on a given night.
I'm happy we're getting it updated at all. I hadn't expected that.
But in terms of usefulness for fantasy purposes, yes correct batting order is huge. Kouzmanoff hasn't batted 4th all year, I don't think. At 6th, he'll get a couple less ABs per week. He'll have fewer RBIs from batting behind guys on base a bit less often, he'll score many fewer Rs due to the stiffs now batting behind him.
I rostered Kouz on my fantasy squads, with his projected cleanup position being the tiebreaker which persuaded me to settle for him rather than the more expensive Longoria. (owie) No complaints on missing on alot of those prior to the season. But if you're gonna do them now, you should correct the obvious mistakes. Once time allows.
Tend to agree with you about Chris Young; I've drafted him high on all of my fantasy teams for two years in a row now. I was disappointed last year, and this year isn't starting any better. The guy is a hackopotamus, and I hope he improves some day with his control and knowledge of the strike zone.
But if he pops up really high on my projections AGAIN next year I will just ignore him (hopefully I don't miss his career year) ... but every algorithm is going to have it's quirks, and PECOTA (and the PFM ... and Depth Charts for evaluating pitching staffs) has always given me a considerable edge on draft day, Chris Young excepted (so far).
I am really interested in how you're going to modify PECOTA with current performance. PECOTA is already providing a range of possible outcomes...unless the performance is significantly outside of the 95%/5% ranges, shouldn't PECOTA still be essentially correct? Would being at your 75% range for one month actually establish you as having a different baseline? The only areas I could see this being useful for are prospects new to a level with very wide PECOTA error bars.
drmboat, I would imagine Clay, et al doing some sort of Bayesian process to update their estimates. That is to say, the prior distribution they had for performance (PECOTA projections) still carries some weight in the analysis, but also the data we have observed needs to be weighted -- small sample sizes are small sample sizes, but it's still information that we can incorporate. The interesting thing to me would be to observe how Clay decides to assign a degree of confidence to the PECOTA projections; I could imagine it being either a function of the Betas in the projection, or running the same Bayesian process on past seasons and maximizing the predictive ability of the process with respect to relative weighting.
I've been doing this myself by fitting a Beta distribution (different than the PECOTA "Beta") to the PECOTA distribution. Once that's done it's very easy to do a bayesian update for BA, OBA, SLG or EqA. For example, Andruw Jones's PECOTAEqA fits a Beta distribution with the parameters alpha=44.9 and beta=127.3. This year he has had 77 PA witha .349 EqA. To get the bayesian update you add 77*.349 to alpha and 77*(1-.349) to beta. The PECOTA goes from
10% 25% 40% 50% 60% 75% 90%
.153 .233 .253 .262 .273 .280 .296 TO
.252 .269 .280 .288 .295 .307 .325
for his performance going forward
Is there any chance that the equations for player ranking will be released? I rolled my own fantasy value system (see below) and I'm fairly in line with the PFM, except that the PFM values closers far above my rankings.
Below is my basic formula, any comments welcome.
For all scoring stats I calculated the average 5 year high and low values for my league. It's a pretty standard rotisserie format, so I don't expect these to be idiosyncratic. Then, I used the following formula, assuming a 15 team league.
PS = Player's Stat
ALV = Average Low Value over 5 years
AHV - Average High Value over 5 years
ISV = Stat Value
*Note: For non-averaged stats like W, SB & HR I divided the ALV and AHV number by the number of positions that could accrue that stat.
ISV = (PS-ALV)*(15/(AHV-ALV))
The total Fantasy Value of a player is then the sum of all his ISVs.
The hope of this formula is to help one figure out a situation where, say one player has fewer HRs and RBI's but a higher average, and the other has more HRs and RBIs but a lower average.
I think you're reading me wrong, which might be my fault. I might not have been clear.
The averages aren't player specific, they're Fantasy Team specific. So, say in 2007 the Fantasy Team that was 1st in ERA had a 3.00 ERA and the team that was 15th in ERA had a 5.00 ERA. Then in 2008 the team that was 1st ERA had a 3.5 ERA and the team that was 15th in ERA had a 4.5 ERA. Then the 2 year average would be 3.25 for being 1st in ERA and 4.75 for being 15th ERA. The idea is to figure out the baseline of each scoring category and then how much each point of ERA is worth in fantasy points.
Under this system a pitcher whose PECOTAERA projection is 4.75 receives an ERA ISV of 0 and a pitcher whose PECOTA projection is 3.25 receives an ERA ISV of 15. I recognize that I've turned what is actually a 15 point scale (1-15) into a 16 point scale (0-16) but since all values are relative that shouldn't be a problem for comparing players.
15 comments have been left for this post.
I understand there are alot of things to do, many more pressing than this. But getting the batting orders reasonably right needs to be on the list at some place. The San Diego batting order is laughable, after which I didn't bother checking any more.