2009 Marcel Projections

[TooLiveBrew]

Sorry for starting another thread, but I know folks will be interested in this one, at least. FanGraphs has a nice sortable list of the '09 Marcels. The default sorting filter is by wOBA.

 

Brewers of note:

 

.298/.352/.562/.914 -- Braun

.282/.377/.530/.907 -- Fielder

.280/.329/.483/.812 -- Hart

.278/.336/.459/.795 -- Hardy

.249/.357/.414/.772 -- Weeks

.272/.343/.439/.782 -- Salome

.243/.328/.436/.765 -- Cameron

 

 

The Salome projection kinda surprised me, and I wonder how much Cam is still 'punished' by Marcel for the pitchers parks he played in

[trwi7]

You didn't post Gamel's projection?

 

.279/.350/.436/.786

 

And Escobar's.

 

.280/.353/.450/.803

 

Don't know where they got that Escobar projection from, but they have him at a 9.1 BB%, something he's never come close to accomplishing.

[logan82]

Wow, that's a really high projection for Escobar. Pretty optimistic considering he has never had a SLG or OPS that high and only topped that OBP once. Trade JJ Hardy.

[rluzinski]

Marcel ignores minor league stats completely.

[trwi7]

Marcel ignores minor league stats completely.

So what do they use exactly? His two major league at bats?

[logan82]

Marcel ignores minor league stats completely.

In that case the projections for Escobar and Gamel seem a little low. I mean Gamel did put up a line of .500/.500/1.000/1.500 line for the Brewers and Escobar was .500/.500/.500/1.000.

[endaround]

Thats' 4 ABs!

[sbrylski]

Marcel's is generally one of the poorer projections systems, due to its simplicity.

 

In Gamel and Escobar's case, their projections are largely just the projected league average, plus the few AB's they got in the bigs last year. Just those stats, very heavily regressed to the mean. Essentially worthless.

[Brawndo the Thirst Mutilat]

Right -- that's why the Gamel and Escobar projections are virtually identical -- that's the league average hitting line (plus a boost due to their success in those 4 AB). It's important to understand what Marcel can and can't do. With established major league players, a Marcel projection gets you roughly 90 - 95% of the way toward the very best projection systems. For players with a limited MLB track record, you need to look at one of the other projection systems (ZiPs, pecota, chone, and miner all use MLEs for such players, if I'm not mistaken).

 

The advantage a Marcel projection offers is simplicity. Since they are just a weighted average of a player's recent performance (more recent = more weight), adjusted for age and regressed toward the mean of a league average player, you can easily run them mid-season (for example), whereas the better projection systems only come out once a year. For a player who is having a year that is unexpected (by the projection systems at least, e.g. Ray Durham's resurgence last year), you are probably better off including the current season info in a simple projection than omitting it in the fanciest projection you can muster. Mid-season projections are, in my opinion, the best use for Marcel. You can even do them yourself, with just a bit of copying and pasting, thanks to the Hardball Times: http://www.hardballtimes....cle/is-this-guy-for-real/

[TheCrew07]

I've always liked PECOTA personally, probably because it was one of the few projection systems that liked Corey Hart as much as I did. I haven't have a BP membership since I picked up BA though, but I'd be curious what the projections for our top 3 prospects would be for next year. If memory serves the PECOTA projections come out latish, around March?, so it'll be awhile.

[rluzinski]

Yeh, minor leaguers are simply regressed 100% towards league average. And it's not that Marcels is a poor projection, since it's not setting out to be the best projection system to begin with. The creator says that it represents the minimum that ANY projection system should do. The funny thing is that it crushes many non-statistical based projection systems (ones that don't understand regression to the mean) and isn't really much worse than any of the more advanced systems for established players.

 

If you them from tangotiger.net, he includes a "reliability" column. It's inversely proportional to the amount the projection is regressed.

[JohnBriggs12]

Marcel's is generally one of the poorer projections systems, due to its simplicity.

 

In Gamel and Escobar's case, their projections are largely just the projected league average, plus the few AB's they got in the bigs last year. Just those stats, very heavily regressed to the mean. Essentially worthless.

 

Like any projection isn't essentially worthless. I think they're all meant to be conversation starters essentially.

[homer]

Like any projection isn't essentially worthless. I think they're all meant to be conversation starters essentially.

 

Anyone know how accurate theses systems can be? I've never looked it up, but I've heard from some die hard pectoa people that they are surprisingly accurate.

[TooLiveBrew]

Like any projection isn't essentially worthless. I think they're all meant to be conversation starters essentially.

 

Nope, they're intended to provide an estimated range of production. Not intended to pinpoint a precise line imo, or if one expects that, it's setting yourself up for disappointment. The projections for Cam in 2008 iirc were quite good, as were a number of others.

[Ennder]

Anyone know how accurate theses systems can be? I've never looked it up, but I've heard from some die hard pectoa people that they are surprisingly accurate.

I've never seen a good study on it. Usually they just pick a stat and figure out the correlation to that stat but I think that is the wrong way to go about it. Obviously no projection system is going to be able to predict things like injuries so I'd find the correlation on healthy players only and see how it did and understand that you just can't predict injuries accurately. I'd also want to see correlation by multiple stats and not just one stat. A prediction of .400 OBP, .400 SLG for .800 OPS is not very accurate if the player actually puts up a .300/.500/.800 year.

 

However the correlation of all hitters is usually something like .7 and pitchers is closer to .45 or so. Another problem is they use ERA as the pitching correlation and ERA is just a terrible stat to use. Something like OPS against would be much better since ERA is highly random over even 200 IP much less pitchers who only have 80 or 90.

[rluzinski]

What matters is run production, so projections should be measured by that. I know that tangotiger (insidethebook.com and tangotiger.net) has used wOBA, which is really just linear weights. Time and time again, he's found that many of the projection systems (PECOTA, ZiPS, Marcel, CHONE, etc...) are all very close to each other. Marcel is as simple as it gets and it gets you 95% of the way.

 

Keep in mind, binomial uncertainty sets the upper limit in projection accuracy. What I mean is, even if you know what the exact odd of a base hit is for a player on every AB, you are still not going to be able to perfectly predict his BA over 600 AB. Over 10,000, you could get pretty close, though.

[trwi7]

Brewers ZiPS projections

 

They're pretty pessimistic on Parra for some reason. I think their Villanueva projection is too pessimistic as well if he's coming out of the bullpen.