My only problem with ZIPS and CHONE has been playing time forecasts. The non-rate stats are somewhat irrelevant if you don't consider who is actually going to be playing/pitching. Projecting 500 ABs for a guy who is the 5th OF because that's what his age would indicate makes no sense. Projecting 300 ABs for a guy who is now the starter b/c that's what his history has been makes just as little.

The problem with projecting playing time is that the vast majority of it is deterministic; IE, guys get as much playing time as their teams give them. Transitioning from predicting ability to predicting behavior is very difficult, and raw statistics can only do so much here.

I'm not saying that the computer modeling should predict the playing time. That has to be done manually based upon the best available information before the season.

They're not designed to project playing time, though, and I'm not terribly interested in seeing Brian Bocock's line for the Jays when it reflects the 1% chance he's called up for a week in September or the 98% chance he's not called up at all. I want to see the 1% window where Brian Bocock is used as a major league regular/utility man.

Brian Bocock ss 25 .188 .255 .245 107 400 51 75 15 1 2 39 35 123 9 8 35

This is a glorious projection and your wants would deny me it, so bully to you.

Anyhow, I doubt either Sean or Dan would strenuously object to someone making detailed depth charts and reprojecting their numbers based on that. It's just not what they're personally trying to do.

I have never even bothered looking at anything other than the rate stats on these projections. I think Teal has it right, just tell me what a guy is going to do when he plays.

My only problem with ZIPS and CHONE has been playing time forecasts.

This is also the problem that Dan and Sean have with them as well, so you are echoing what the creators of those systems are saying.

Fantisitcs apparently is a person or persons that creates playing time forecasts. They did a better job, apparently than the Crowdsource from my Community Forecast project last year.

In my Forecasters Challenge last year, ZiPS, MGL, Steamer and Marcel all used the playing time forecasts from the Community, and they all ended up in the top third of the 22 systems (essentially competing with themselves frankly).

What we also see is that the playing time forecasts are more important than the rate forecasts (for fantasy baseball anyway).

So, I would highly recommend going with Chone or ZiPS on rate, and then going to the Fangraphs playing time forecasts, or Fantistics.

This should pretty much be the best you can hope for.

I'm projecting playing time, but not necessarily major league playing time. Last year I projected 536 PA for Brandon Wood. He actually got 474, but only 46 in the majors. To project 500-750 guys who are mostly going to play in the minors, and forecast 10-50 PA for each doesn't give you much room to qualitatively show their abilities. So I prefer showing more of a full season for presentation purposes.

You want to use your own depth charts with these projections, be my guest.

I have created a spreadsheet for people who want to do their own team record projections. Check it out. I strongly encourage you to form your own team W-L projections. Spreadsheet link in the text at the bottom of that page.

Chone, thanks for the link.

Sean,

That's helpful to know about the PA projections. Looking at them in a vacuum, they almost seemed like hopes or based upon age based/career line projections. They may be the latter, but if they're regardless of level, then they make sense. The only problem becomes projecting distribution. But, that's not computer based.

Cool analysis. Last year, we at Razzball used a Marcel/CHONE/ZiPS average for rates and Baseball Prospectus for playing time. Glad to see this general strategy is best (will use just CHONE and ZiPS this year) - surprising that BP gets beat both on projecting stats and playing time. Oh how the mighty have fallen.

What about CAIRO?

I want to very much echo point #4 - both as to the general point and, specifically, on the glorious awfulness of Brian Bocock.

What about CAIRO?

Only 2 years of history, so probably too soon to include it with the systems that have been around longer.

That said, I'd trust it over Pecota any day.

#10, It'll be interesting to see how things go with Clay running the forecasts. The last time he had to step in he went very conservative. I can't swear to this, but my recollection is that he cut the standard error quite a bit but left them with fewer projections to really brag upon.

Kind of confirms what Darrell Huff wrote in one of his books. Since nobody really expected projections to be accurate, Gary Huckabay got more credit for a few surprisingly accurate projections than Clay did for a generally more accurate set.