Reader Poll, Fangraphs Community Projections

by Jeff Sullivan on Dec 2, 2009 6:50 PM PST in Miscellaneous

I'm curious as to how people arrive at their forecasts, and hopefully you guys are some approximation of a representative sample. As far as I can tell, there are two ways:

(A) Project what you consider to be the most likely season

(B) Project what you consider to be the weighted average of all possible seasons

Poll

With which do you more closely align?

A

356 votes

B

87 votes

443 votes | Poll has closed

Tweet Comment 37 comments | 0 recs |

Do you like this story?

Comments

Without enough mathematics knowledge

Nor, really, the interest in spending large amounts of time on my projections, I pick what seems to me to be the most likely season (which may really be my internal weighting of several possibilities). I sort of treat the whole exercise as “reach into a hat with X number of seasons (with the more likely seasons being represented more often) on it, and pull out one,” then submit what I “pull out (of my ass, obviously)”

I'd rather know a little about a lot than a lot about a little

by Sportszilla on Dec 2, 2009 6:58 PM PST reply actions

I really wonder what Tony Blengino's predictions look like.

I just guessed. I really didn’t put a great deal of thought into mine, because I am really bad at math.

by mark sobba on Dec 2, 2009 7:25 PM PST reply actions

I tend to do some averaging, with emphasis on seasons that just occured.

I usually find myself to be a little harsh on players, usually not projecting growth. I’d rather have them exceed my expectations. Probably negates my math a bit.

by MT Olson on Dec 2, 2009 7:31 PM PST reply actions

I'm pretty sure these predictions

better reflect the “median” production than the average production. Most of the error is probably associated with not factoring injuries and the possibility of massive collapse which cause people to overpredict playing time which inflates player value. Otherwise I bet they are probably pretty reasonable. I’d hope the rate stats are accurate at least.

by Edgar for Pres on Dec 2, 2009 7:39 PM PST reply actions

I wouldn't classify that so much as error as I would viewing the purpose of the projections in different ways.

by Aaron Campeau on Dec 2, 2009 8:44 PM PST up reply actions

Maybe I am misunderstanding this but I don't see how you can do A without B

by Poochie on Dec 2, 2009 8:17 PM PST reply actions

A gives no consideration to unlikely events like catastrophic injury or rapid decline

by Jeff Sullivan on Dec 2, 2009 8:18 PM PST up reply actions

Unless it's Bedard

by Jeff Sullivan on Dec 2, 2009 8:18 PM PST up reply actions

It is not prudent to project unlikely things

by Poochie on Dec 2, 2009 8:23 PM PST up reply actions

I project what I think the ~50th percentile will look like, which I would see as the most likely outcome.

That is also, in a way, the weighted average no?

by abender20 on Dec 2, 2009 8:24 PM PST up reply actions

Unless your prediction curve is asymmetric, I suppose.

by abender20 on Dec 2, 2009 8:26 PM PST up reply actions

I think pretty much all prediction curves ought to be asymmetric

It’s a lot easier to be more bad than more good.

by Jeff Sullivan on Dec 2, 2009 8:39 PM PST up reply actions

Yes

None of these should be bell curves

by Graham MacAree on Dec 2, 2009 8:49 PM PST up reply actions

Unless you're replacement level

by vivaelpujols on Dec 2, 2009 9:20 PM PST up reply actions

Disagree

Curve is going to be skewed in favour of positive WAR for replacement level players, I think. Just because if they do badly they’re going to get pulled fast.

by Graham MacAree on Dec 2, 2009 9:35 PM PST up reply actions

You're right

I was more responding to Jeff’s statement above.

by vivaelpujols on Dec 2, 2009 9:38 PM PST up reply actions

I don't have a good feel for that, I guess.

For Franklin, I’d put his 90th percentile wOBA at something like .375 and his 10th at .295 or something and so I’d tell you ~.335 as a most likely. Not a bell curve, but probably still something like a 5/4/3 hovering around .335 and the rest can go to hell. To me, the asymmetry lies in the tails and it’s enough of an abstraction anyway that I don’t worry about it.

by abender20 on Dec 2, 2009 8:51 PM PST up reply actions

I'm not very math savvy so I try to stay conservative and input numbers that don't look ridiculous (one way or the other)

by Poochie on Dec 2, 2009 8:52 PM PST up reply actions

I think about the range and likelihood of outcomes, but that basically just does what you do.

by abender20 on Dec 2, 2009 8:54 PM PST up reply actions

I don't do these projections because I routinely predict a player's 0.000001%th percentile and assume they'll spontaneously combust in the batter's box

by Graham MacAree on Dec 2, 2009 8:58 PM PST up reply actions

Only Delmon Young has managed to do this.

by abender20 on Dec 2, 2009 9:05 PM PST up reply actions

I love how Utley is being projected by the fans to have the highest WAR

I think if you are looking for fan bias, there it sits

by Poochie on Dec 2, 2009 8:25 PM PST reply actions

I'm confused

You don’t think Utley will have the highest, or close to the highest WAR, or are you being sarcastic.

by vivaelpujols on Dec 2, 2009 9:22 PM PST up reply actions

No I think it will be unlikely that he will be better than Pujols or a few others

I just think he’s “the guy” for people of the Fangraphs and it is leaking into the projections

by Poochie on Dec 2, 2009 9:44 PM PST up reply actions

It's probably compensation for the jackass BBWAA

You know, like the reason everyone hates Jeter.

by vivaelpujols on Dec 2, 2009 9:51 PM PST up reply actions

Pujols WAR is going to take a small hit

Due to league factors when the Cards trade him to the AL next year as the team is going down in flames.

by CMC_Stags on Dec 2, 2009 9:58 PM PST up reply actions

Pujols'

by CMC_Stags on Dec 2, 2009 9:58 PM PST up reply actions

I don't know why you would project that

by Poochie on Dec 2, 2009 10:02 PM PST up reply actions

I think that doing a weighted average would be the best way to go

But I wouldn’t want to go through all of that work to get something marginally better than the median projection.

Anyway, I love that FanGraphs displays the WAR projections at the bottom of the page – makes things a lot easier.

by vivaelpujols on Dec 2, 2009 9:40 PM PST reply actions

Maybe it would be easier

to suggest people predict most likely outcomes. Then have a few questions to build an idea of how to correct for the range of distributions. A simple system might be good enough such as:

How likely is the player to be on the 15 day DL?
How likely is the player to be on the 60 day DL?
How likely is the player to be at least 25% worse than what you project?
How likely is the player to be at least 25% better than what you project?

Just off the top of my head, it might be easier for people to project this way. It would also make sure we can account for everything correctly.

by Edgar for Pres on Dec 2, 2009 9:58 PM PST reply actions