Mixed Opinions - Fantasy Baseball Cafe 2014 Fantasy Baseball Cafe
100% Deposit Bonus for Cafe Members!

Return to Baseball Leftovers

Mixed Opinions

Moderator: Baseball Moderators

Re: Mixed Opinions

Postby noseeum » Wed Dec 31, 2008 1:41 pm

BTW, I don't think the creator of Marcel would take issue with what I'm saying. The whole point of it is to be the dumbest possible system with the highest possible accuracy, or more to the point, to show how terrible projections are for individual players.

In the end, they're all a crap shoot!
noseeum
Major League Manager
Major League Manager


Posts: 1697
Joined: 1 Mar 2005
Home Cafe: Baseball

Re: Mixed Opinions

Postby GotowarMissAgnes » Wed Dec 31, 2008 1:47 pm

noseeum wrote:
I disagree. The method to the calculation for playing time is arbitrary. Take Liriano. We know why he missed time the last two years. Is 92 innings a reasonable projection for this year? I don't think so, based on what we know of Tommy John surgery. Other projection systems take into account the injury proneness of players as well, but they also look at the individual circumstances.

Marcel's method falls flat for players that have significant injuries that keep them out for a long time.


And the problem is that those looks at individual circumstances may be just as biased as any other approach.

The advantage of Marcel is that you KNOW there is essentially no bias from the guy doing the projections.

And, I would argue that 92 IP is certainly a reasonable guess. 10-20 percent of guys returning from TJ surgey still flame out completely, so we have to factor that into the estimate. And Liriano has had other significant injuries, missing time in both 2002 and 2003. And, you never know what other factors might pop up.

Last year, for example, James predicted Liriano would pitch 152 IP. Marcel predicted 57 IP. He pitched 76 IP in MLB (200 overall), because the Twins kept him in AAA longer than expected.
"I don't want to play golf. When I hit a ball, I want someone else to chase it."
GotowarMissAgnes
Hall of Fame Hero
Hall of Fame Hero

User avatar
Fantasy Expert
Posts: 5516
Joined: 12 Dec 2003
Home Cafe: Baseball
Location: Happy Valley

Re: Mixed Opinions

Postby GotowarMissAgnes » Wed Dec 31, 2008 1:48 pm

My response would be: show me the data. I'm open to the idea that Marcel predicts younger or injured players worse, but let's see some evidence for that before making the claim.
"I don't want to play golf. When I hit a ball, I want someone else to chase it."
GotowarMissAgnes
Hall of Fame Hero
Hall of Fame Hero

User avatar
Fantasy Expert
Posts: 5516
Joined: 12 Dec 2003
Home Cafe: Baseball
Location: Happy Valley

Re: Mixed Opinions

Postby noseeum » Wed Dec 31, 2008 2:19 pm

GotowarMissAgnes wrote:My response would be: show me the data. I'm open to the idea that Marcel predicts younger or injured players worse, but let's see some evidence for that before making the claim.


I'm not sure data is needed. Marcel was designed to be a dumb monkey that all projection systems should be judged against. If you can't beat the monkey, than your system stinks. That's basically straight from the horse's mouth, the guy who created it.

Here's a pretty cool conversation between Tom and Nate Silver based on Nate's comparisons of all the projection systems:
http://www.insidethebook.com/ee/index.p ... aluations/

In it, Tango specifically says "Marcel is intentionally brainless on rookies." Marcel assigns league average performance to every rookie, so there is absolutely nothing going on in terms of a projection. It's just filling in numbers, not basing them at all on minor league performance. The value in this arbitrary assignment is to use it as a comparison to other systems.

Tango says throughout his writings that other systems spend inordinate amounts of time to get very marginal increases in accuracy. Marcel's hope was to create a baseline projection system that's open source, which other can than use to make improvements. But by having the dumb open source baseline, you can always judge the statistical benefits of any improvements.

Here are some quotes from the discussion:
[quote=tango]Nate: good stuff. For hitters, PECOTA has a 10 player advantage over Marcel. Over hitters and pitchers, that’s a 24 player advantage. I wouldn’t be surprised if almost all of that was on guys with less than 2-years experience. [/quote]

Someone actually has taken Tango's advice and created a Marcel based projection system, which Tom references in the discussion. Here's his take:
[quote=tango]As long as a forecasting uses Marcel at its core, but makes intelligent choices with the remaining known data, it leaves Marcel irrelevant as a forecasting system. [/quote]

Here's the link:
http://www.replacementlevel.com/index.p ... ctions_v01

My point is that there is no intelligence built into the projections. That's the whole point of them. At the extremes they break down, but for 80% of players in baseball they'll be pretty accurate.

How could you possibly believe that just assuming every rookie will provide league average performance will help you determine an individual player's performance? It's completely arbitrary. And that's the point. "Here's what the monkey says. Now go beat him, PECOTA. Go beat him, Bill James."
noseeum
Major League Manager
Major League Manager


Posts: 1697
Joined: 1 Mar 2005
Home Cafe: Baseball

Re: Mixed Opinions

Postby GotowarMissAgnes » Wed Dec 31, 2008 3:02 pm

Well, the question is whether just assuming average performance leads to better predictions than trying to project MLB performance based on minor league performance.

I don't think we have any disagreement at all that Marcel is a baseline or that most projection systems do only marginally better than it.

And, I would certainly not be surprised if some of that marginal difference was young/injured players. But, you don't convince me with statements from any interview. You convince me with data that shows the evidence.
"I don't want to play golf. When I hit a ball, I want someone else to chase it."
GotowarMissAgnes
Hall of Fame Hero
Hall of Fame Hero

User avatar
Fantasy Expert
Posts: 5516
Joined: 12 Dec 2003
Home Cafe: Baseball
Location: Happy Valley

Re: Mixed Opinions

Postby noseeum » Wed Dec 31, 2008 3:52 pm

GotowarMissAgnes wrote:Well, the question is whether just assuming average performance leads to better predictions than trying to project MLB performance based on minor league performance.

I don't think we have any disagreement at all that Marcel is a baseline or that most projection systems do only marginally better than it.

And, I would certainly not be surprised if some of that marginal difference was young/injured players. But, you don't convince me with statements from any interview. You convince me with data that shows the evidence.


If you read the discussion, specific players are in there. I'm not going to copy and past it for you.

I don't think I need to prove that using actual data and actual analysis of a player's actual skills in the minors can do better than just assuming for no specific reason that a rookie will perform exactly at league average in all categories. Show me a picture of a rookie the looks like Cecil Fielder, and I'll bet you I beat Marcel at that player's stolen base projections.

It's inherent in the system that it breaks down for players that have not played full time for the last three years. You're projecting some credence to these numbers that even the creator would not put in them.

I don't understand how you can know how these numbers are created and that you would still believe you can trust them for players without significant playing time in the majors for the last 3 years. Why do you think there are no major differences between James, PECOTA, CHONE, ZIP, etc. for major leaguers that DO have significant playing time over the last three years?

Marcel doesn't make park factor adjustments or playing time adjustments. It simply plugs in the last three years data, and for those players who were not in the majors, it assumes league average for those missing years. You'll need to explain to ME how you expect it to hold up for rookies, and or players with significant injuries in the past two years. He could have just as easily said that each rookie will get his last season's worth of data projected out to 300 ABs or 85 innings for a pitcher. The system would be just as arbitrary. NO ONE is league average.

The beauty of Marcel is that it KNOWS it will be wrong, it finds that perfectly acceptable, and it lets you know it instead of trying to obscure it. I don't get what you're debating. Tango would I'm sure say the same thing. But he would add, "Yes, but the funny thing is, I'm not much worse than any other projection system out there and I'm better than most."
noseeum
Major League Manager
Major League Manager


Posts: 1697
Joined: 1 Mar 2005
Home Cafe: Baseball

Re: Mixed Opinions

Postby GotowarMissAgnes » Wed Dec 31, 2008 5:14 pm

Tango would I'm sure say the same thing. But he would add, "Yes, but the funny thing is, I'm not much worse than any other projection system out there and I'm better than most."


And that's the key point. Despite the fact that it ignores a lot of relevant factors, it STILL does better than many.

So, even when looking at injured/young players, there's little good reason to prefer other systems over Marcel. Why should you trust them when they use averages? Well, what happens with injured/young players? The ones that do poorly or get re-injured get sent down; the ones that do well, get extra playing time. And, the result is that the average prediction is not far off, just as for other players.

Again, the test is in the data, not some hypothetical discussion. All projection systems have some arbitrary elements, and the test is whether the sysem works, not the process of getting to the projection. Marcel will be right on some, wrong on others, just as they were with Liriano. I would resist throwing a system out the window until you have proof that it is no good. I often keep Marcel in to temper the supposed wisdom of the projection systems like James that often seem to over-project young guys and returning injured guys. as tango says, "..better than most".
"I don't want to play golf. When I hit a ball, I want someone else to chase it."
GotowarMissAgnes
Hall of Fame Hero
Hall of Fame Hero

User avatar
Fantasy Expert
Posts: 5516
Joined: 12 Dec 2003
Home Cafe: Baseball
Location: Happy Valley

Re: Mixed Opinions

Postby J35J » Wed Dec 31, 2008 5:15 pm

Blah, spend less time trying to figure out other peoples rankings and ways of doing things and just do it all yourself!!! ;-7 ;-D
J35J
Hall of Fame Hero
Hall of Fame Hero

User avatar
CafeholicFantasy ExpertCafe WriterCafe RankerMock(ing) DrafterPick 3 Weekly Winner
Posts: 10467
(Past Year: 393)
Joined: 26 Jan 2004
Home Cafe: Baseball

Re: Mixed Opinions

Postby Fade2White12 » Wed Dec 31, 2008 5:17 pm

J35J wrote:Blah, spend less time trying to figure out other peoples rankings and ways of doing things and just do it all yourself!!! ;-7 ;-D


+1 ;-D
Image
Fade2White12
General Manager
General Manager

User avatar
CafeholicCafe RankerMock(ing) DrafterEagle EyePick 3 Weekly WinnerLucky Ladders Weekly Winner
Posts: 3852
(Past Year: 41)
Joined: 7 Mar 2006
Home Cafe: Baseball
Location: Michigan

Re: Mixed Opinions

Postby noseeum » Wed Dec 31, 2008 5:52 pm

GotowarMissAgnes wrote:
Tango would I'm sure say the same thing. But he would add, "Yes, but the funny thing is, I'm not much worse than any other projection system out there and I'm better than most."


And that's the key point. Despite the fact that it ignores a lot of relevant factors, it STILL does better than many.

So, even when looking at injured/young players, there's little good reason to prefer other systems over Marcel. Why should you trust them when they use averages? Well, what happens with injured/young players? The ones that do poorly or get re-injured get sent down; the ones that do well, get extra playing time. And, the result is that the average prediction is not far off, just as for other players.

Again, the test is in the data, not some hypothetical discussion. All projection systems have some arbitrary elements, and the test is whether the sysem works, not the process of getting to the projection. Marcel will be right on some, wrong on others, just as they were with Liriano. I would resist throwing a system out the window until you have proof that it is no good. I often keep Marcel in to temper the supposed wisdom of the projection systems like James that often seem to over-project young guys and returning injured guys. as tango says, "..better than most".


You have this backwards. Just because there's little good reason to prefer other systems over Marcel doesn't mean that's a reason to prefer Marcel. It just shows that for these outlier type players, systematic prediction systems fail, including Marcel. Be my guest and use Marcel's predictions for Liriano types and rookies. You will be wrong. That said, PECOTA and James outperform Marcel with rookies. And I will not even both to check this because it's been hashed over for years. This is all based on MLEs.

There is significant evidence to show that minor league stats are just as good as major league stats for predicting future performance. MLEs can show that Jay Bruce and Evan Longoria should have very successful major league careers and should be above average right out of the gate. Marcel doesn't use MLEs, so it misses this. It's inherent in the system.

If someone put out a projection system that integrated MLEs with Marcel, it would be better than Marcel, and it would prove to be so statistically, because MLEs have been proven to correlate highly with major league performance. James first introduced this in his 1985 abstract:
http://baseballanalysts.com/archives/20 ... rom_20.php

You can accept this or reject this, but it's still true no matter how you feel about it.
noseeum
Major League Manager
Major League Manager


Posts: 1697
Joined: 1 Mar 2005
Home Cafe: Baseball

PreviousNext

Return to Baseball Leftovers

Who is online

Users browsing this forum: Brisbane, threelions and 11 guests

Forums Articles & Tips Sleepers Rankings Leagues


Today's Games
Monday, Sep. 1
(All times are EST, weather icons show forecast for game time)

Philadelphia at Atlanta
(1:10 pm)
NY Mets at Miami
(1:10 pm)
indoors
Boston at Tampa Bay
(1:10 pm)
indoors
Minnesota at Baltimore
(1:35 pm)
Pittsburgh at St. Louis
(2:15 pm)
Milwaukee at Chi Cubs
(2:20 pm)
Seattle at Oakland
(4:05 pm)
Detroit at Cleveland
(4:05 pm)
San Francisco at Colorado
(4:10 pm)
Arizona at San Diego
(4:10 pm)
Texas at Kansas City
(8:10 pm)
Washington at LA Dodgers
(8:10 pm)

  • Fantasy Baseball
  • Article Submissions
  • Privacy Statement
  • Site Survey 
  • Contact