SubscribeI updated the composite projections to include playing time votes from the Community Forecast through this morning.
Reassuringly, there has been very little movement since the early version on Friday. Of the 1081 players listed originally:
11 gained more than $2
37 gained $2
257 gained $1
561 stayed the same
181 lost $1
22 lost $2
12 lost more than $2
Most of those 23 players who changed by more than $2 are of little fantasy interest. Here are the players you might care about:
Chris Snyder -$1 (+8)
Pablo Sandoval $15 (+4)
Torii Hunter $19 (+4)
Jason Bulger -$1 (+4)
Kazuo Matsui -$3 (+3)
Russell Branyan -$11 (-3)
Howie Kendrick $9 (-4)
Carlos Santana -$6 (-4)
The Angels were low on votes last week, and so their players are some of the biggest movers over the weekend. Branyan drops a little on news that he’ll start the season on the DL.
This change also adds 68 new players to the mix, such as fantasy non-factors Livan Hernandez (-$16), Todd Wellemeyer (-$15), and Robb Quinlan (-$26). Of the new players, Jeff Weaver is the highest ranked, at -$7. Nothing to see here.
I’m feeling really good about how the numbers look now. I’ll probably keep updating every couple of days until Opening Day, but I doubt much will change. As vote totals accumulate, any movement will probably be smaller than even this update.
Related posts:
First of all, as a first-time commenter, I have to give a thousand thanks to Mays for a terrific site. The methodology is sound and well-reasoned, and I fully trust the price guide’s dollar values for all my leagues of various configurations and sizes.
I have a question concerning playing time from the community forecasts. Is there any control for mass error? Every team pitched between 1418 and 1473 innings last year — http://www.baseball-reference.com/leagues/MLB/2009-standard-pitching.shtml — but the Nationals are forecast to pitch a collective 1808 innings, more than 11 innings per game. In fact, every team but STL and CLE is forecast to pitch more than 1473.
Are the differences between team IP numbers brought back down to earth before the playing time projections are fed into the price guide, or are players on one team given a leg up for being forecast to play a 200-game season? Same question goes for PA, although the community forecasts are less egregiously wrong there.
Seems like it may be another manifestation of the selection bias I mentioned a few posts ago.
One thing I do know about forecasting (from other disciplines) is that imposing constraints on the aggregate invariably yields inferior predictions from the models on the individual components.
For instance, you’ll often see projected records for each team and the total losses will not equal the total wins.
1808 IP is nuts though. This is a case where it might force people to be a little more realistic though, but if you’re committed to the method it seems like you should just leave it, as with any other.
Indeed, that is true for most estimations. But is TangoTiger’s survey really that kind of statistical estimation? I agree that we shouldn’t go tweaking the individual rate stats from the computer-based projections because, say, the collective HR/AB is too high. But the survey isn’t an analysis of prior observations, it’s a collection of guesses, and it’s plausible that lots of MLB fans just don’t know how many IP a team has during the year.
This is outside of Mays’s control, but I think a better design for the survey would record each player’s percentage of the respondent’s total IP, and keep a running average of those percentages. The whole basis for the community forecasts being a good estimator for individual IP is that survey takers know about their teams’ *tradeoffs* at a position, not each individual’s playing time. They don’t know, in the abstract, exactly how many innings Batista will pitch, but they know about how much more often Clippard will pitch than Batista.
Between your objections and a few that I thought of after my post, I see there’s not much Mays can do to tweak the community forecasts as they’re currently reported. In fact, I think most of the error comes from the 26th-50th+ roster spots all getting too much playing time, and the fantasy-relevant players are probably all right.
But a guess at one person’s playing time in isolation is precisely meaningless. I would feel a lot better about the numbers if the survey somehow accounted for fans’ perceptions of the playing time distribution *among teammates*. Non-responses could at least be recorded as zeroes, which it seems they are not currently. Maybe I should have sent this as an e-mail to Mr. Tango rather than posting it here, but perhaps Mr. Tango or Mays has thought of this and possible solutions?
Not sure if this is the right place to post this but I was wondering about harnessing the CF system to predict trades that could affect valuations. For many leagues, the player pool is restricted to either the NL or AL only, and a player traded out of that league “disappears” from the available player pool immediately.
To pick an obvious example, Adrian Gonzalez is widely rumored to be traded sometime this season. Ideally I would want to know the community’s best guess on (A) the probability of a trade being completed; (B) the expected date of the trade; and (C) the probability that trade will be inter-league, rather than intra-league. That way I could tweak down his expected appearances in the NL appropriately. More realistically, we could just try to estimate the probability of trade, and assume for (B) two weeks before the trade deadline (one could research the average actual trade date of a starting player to get a better assumption) and for (C) just assume each team has an equal chance at him, so 14/29th chance he goes to the AL.
{A quick note as a first time commenter — in a word: awesome!)