THE BOOK--Playing The Percentages In Baseball

A few people have reported bugs already - mostly minor ones, but I’ll fix ‘em.  Just shoot me an email (linked on my name).  Thanks.

Great article and nice meeting you yesterday, Sal!

Thanks Mitchel, and it was a pleasure to meet you to.  Personally, I had envisioned you as a giant disembodied head bathed in flames.  Always nice to put a real face to the name.

How are the weights for years derived? I’ve been meaning to ask this for a while. The closest thing I’ve found explaining it is this:

http://www.tangotiger.net/archives/stud0233.shtml

I have no idea - I just used Tango’s weights from the original Marcels.  These spreadsheets are a utility, but there’s no novelty in terms of the projection algorithm.

THT now has in-season Marcels, updated weekly.

http://www.hardballtimes.com/main/blog_article/projections-now-updated-every-week-at-tht/

#6: very cool, thanks. Will pitchers’ Marcels be available at some point?

An ‘all teams’ option would also be a nice addition.

We’re working on pitchers.  Thanks for the “all teams” option, I’ll see what we can do.

Sal, good job.

Not sure if you followed the other thread, but does your PA forecast follow the Marcel model of .50/.10/200?

9: No, the PA forecast is based purely on extrapolated playing time.  That’s explained in the link.  My philosophy on playing time projection is that it’s a fool’s errand, so Marcel can project the rate stats and individuals (or communities) can adjust playing time based on pre-existing injuries, roster construction, etc.

There’s no question that if I were to poll the fans, they would give me a better playing time estimate than Marcel (or really, anything systematic I could come up with).

That said, if “Marcel” is going to be used only partly, it should be more prominently mentioned.

I’d appreciate a link at the bottom of pages like this:
http://hardballtimes.com/thtstats/other/current_marcels.php?view=team&team=MIL&stats=rest

That says:
“This page contains the rate stats based on the Marcel methodology.  The playing time estimate is based on a different methodology.”

I think it’s important that if you use the Marcel implementation by name, that you remain true to it.  Otherwise, it will cause confusion, especially if I have to start answering questions about it.

I encourage all to keep doing great work like this.

For the record, I think that a Marcel playing time estimate is silly, unless you just want to save yourself some time and you have no choice but to automate your playing time projections.  You can say the same thing about the Marcel performance projections (you, not me), but you don’t get ANY ridiculous performance projections when you do a Marcel, but you will get MANY ridiculous (to the naked eye) playing time projections when you just use Marcel.  I fact, using a Marcel playing time projection for all players can only get you in trouble.  Why would you want to use them?  Or I should say, for what would you want to use them?  If you were using a performance projection and a playing time projection to evaluate a player in a trade or for a contract, or for whatever reason, you are going to likely look at the Marcel playing time projection and “know” right away if it is clearly right, wrong, or in the ballpark.  In either of those 3 cases, you are at least as well off just going with what you “know.”

I am assuming that .50/.10/200 means 50% of last year, 10% of the year before, plus 200?

I don’t have to give all the reasons why you are much better off using “other” information for playing time, do I?  If player A got plunked in the hand last year and missed 100 games and player B does not, I don’t think (I know) that there should be a 50 game difference in their playing time projections.  If a player changes roles, obviously the Marcel playing time projections are worthless.  etc.

I am sure they work in the aggregate (just like assuming that everyone is a league average batter works in the aggregate), and it is probably the best simple model out there that reduces your overall squared error, but some things are better left to the human mind, and this is one of them.

No offense to Marcel of course. I am just glad that Sal is not using them.  You would see too many ridiculous numbers and that would defeat the purpose of the publishing them on a daily basis. 

Who wants to see that Andruw Jones is going to accumulate another 200 PA, when he is going to be lucky to accumulate another 50?  There are tons of rookie players that Marcel would screw up although I don’t know how it handles rookies who never had any playing time and are now full-time players (200 PA?).  Again, if it is 200, that may work for all rookies combined, because many of them do not pan out, but we KNOW, for example, that Longoria was and is going to play full time for the entire season, barring a terrible start for the season.  We certainly can’t give Longoria the same playing time estimate as Brian Bocock (who is gone of course) or even Votto and Bruce (we did not know if and when they would be called up or how long they would last).

I just don’t know what Sal is using.  If he is including human insight, then of course that is better, as I noted, and as you are noting.

But I went to a few of the pages, and couldn’t see exactly what he was doing.  If you are going to deviate from Marcel, then all I ask is that you highlight on any page that shows the Marcel forecast exactly where you are dumping Marcel in favor of something else.

Would it make sense for me to use Bill James’ Brock2 system and then make changes, and still call it Brock2?

I just don’t want to answer questions from people who say “Marcel said...”, when Marcel didn’t say such a thing.

That’s all I’m saying.

Assuming it’s the same as the spreadsheet, he’s doing:

( PA / Games Played In Season ) * Games Remaining

If the equation for Marcels projected PAs is:

.5/.1/200

what’s the equivelent term for Marcels projected IP?

Nevermind, found it - 60 for starters, 25 for relievers, and take a weighted average for anyone inbetween.

Took a first stab at doing Marcels for next season - already had the scripts written and the data was scraped for other reasons, so I figured why not. I’m figuring playing time as stated above, or at least best as I can figure:

http://www.editgrid.com/user/cwyers/2009_Marcels_Projections