Wozzyck wrote: Well, there's always going to be a certain error associated with projections, which we won't really be able to eliminate; though I don't understand some of your comments. No way to predict the outliers? This is a truer statement for some stats more that others, but declaring that Pierre is likely to be a SB-outlier can certainly be made from a sound statistical foundation. And individual predictions do have a fair amount of use, though their confidence intervals are perhaps larger than people realize. The suggestion that it's better to have average performers in each category rather than having certain categories covered by few sources (as is often done with SB) is true (as it's just simple statistics), but it's not really because we don't know who the outliers might be. (We're just taking on more risk, in the form of larger confidence intervals, by relying on the performance of those we expect to be outliers.)

My suggestion was not that it's better to have average performers. My comment was suggesting that fantasy players focus far too much on individual player projections, when what they should focus on is the average prediction (and the variance of that prediction, though that's another issue) of the set of players they choose.

swyck wrote:If all the stats are low at more or less the same rate, then its still a useful guide to how players will do compared with each other.

Frankly I dont need a specific player to hit X HRs, I need my team of players to do better as a whole then the other teams of players. As long as I'm grabbing the better players all around I should be fine.

This is not true. Players that are outliers in the valuation curve are more valuable or less valuable based on how far they differentiate themselves from the mean. Grouping everyone closer to the mean takes away from this. Do it over 4 cumulative categories used in the standard 5x5 for hitting and you will be way off.

No, you're wrong here. If the counting stats are uniformly lowered by the same factor (say 7/8 to pick a number), a player would be the same number of standard deviations from the mean as he would have been before the scaling.

Thinking the Value remains the same is wrong. You miss the point if you think everyone is LOWERED by 7/8. They are not. Everyone regresses to the mean. So the guys at tha AVG stay at the AVG. The outliers above and below get closer.

You can ask others that do valuation, don't just take my word for it...

Examine swyck's first sentence and assumption in the post you were reacting to.

The stats are not ALL lowered. This is a huge misconception whey people see projection sets. They ALWAYS tend to compare and contrast the ABOVE AVG players. These are the players they know and love. The only problem with that is that the BELOW AVG players are not getting worse, they are regressing to the mean as well.

thinkspin wrote:The stats are not ALL lowered. This is a huge misconception whey people see projection sets. They ALWAYS tend to compare and contrast the ABOVE AVG players. These are the players they know and love. The only problem with that is that the BELOW AVG players are not getting worse, they are regressing to the mean as well.

So to think VALUATIONS remain the same is wrong.

I don't really know what you're arguing here. swyck made an If-Then statement, an implication that was more or less true. You said "Nay," and proceeded to talk about the mistakes that people make in doing VALUATIONS, complaints that had nothing to do with the implication.

Now it seems that you must object to his premise (which has nothing to do with the truth or falsity of the implication), that players' at-bats in the PECOTA projections look to be lower than actual expectations by roughly the same multiple. Now, of course, this is not going to hold perfectly (and it wouldn't be a systematic element of the model); but in perusing the projections, it's not an outlandish assumption, and it's not one that's going to get you into much trouble. What they certainly don't do (and anyone who's ever looked at PECOTA projections would probably notice this) is regress at-bats to some arbitrary fixed mean. Other than for players coming off of injuries or who spent an uncharacteristically large amount of time on the bench, a player's PECOTA at-bats are almost always fewer than that from the previous year, no matter how BELOW AVG they are, whatever that might mean.

EDIT: swyck never said "at-bats," but rather just "stats." Much of the discrepancy in the other stats though can be attributed to the pessimistic at-bats projections; even so, the above is all still valid if one replaces "at-bat" with whatever counting stat you choose.

Even though it's not germane to the argument, here's an exercise for you thinkspin: Suppose that the average OF has 100 RBI. Based on last years numbers suppose that all draftable OF have their previous years numbers regressed to the mean by 1/2 (so an OF who had 106 RBI, now is projected to have 103 RBI; an outfielder with 70 RBI is projected to have 85 RBI, and so on). If Toe Nash had an RBI total 2 standard deviations above the mean last year, how many standard deviations is he above the mean after the regression?

Wozzyck wrote:Even though it's not germane to the argument, here's an exercise for you thinkspin: Suppose that the average OF has 100 RBI. Based on last years numbers suppose that all draftable OF have their previous years numbers regressed to the mean by 1/2 (so an OF who had 106 RBI, now is projected to have 103 RBI; an outfielder with 70 RBI is projected to have 85 RBI, and so on). If Toe Nash had an RBI total 2 standard deviations above the mean last year, how many standard deviations is he above the mean after the regression?

Toe is still 2 standard deviations above the mean. But we are now way off track.

This thread seemed to go a few different ways from the start. The first post was very good and detailed, I was just trying to be funny with the Marcel the Monkey thing. I tend to think projections are useless unless they teach me something. And most of the projection sets I see are more interested with being correct or as accurate as possible as a whole than they are about telling you which players SHOULD DO BETTER and which ones SHOULD DO WORSE.

GTWMA said

My suggestion was not that it's better to have average performers. My comment was suggesting that fantasy players focus far too much on individual player projections, when what they should focus on is the average prediction (and the variance of that prediction, though that's another issue) of the set of players they choose.

That is about how I feel when it comes to projections, that is why I focus more on trying to understand how to value players stats and how others are doing the same.

thinkspin wrote:I tend to think projections are useless unless they teach me something. And most of the projection sets I see are more interested with being correct or as accurate as possible as a whole than they are about telling you which players SHOULD DO BETTER and which ones SHOULD DO WORSE.

thinkspin wrote:GTWMA said

My comment was suggesting that fantasy players focus far too much on individual player projections, when what they should focus on is the average prediction (and the variance of that prediction, though that's another issue) of the set of players they choose.

That is about how I feel when it comes to projections.

Most magazines or other fantasy sources actually have a 1 guy that does the rankings, 1 guy that does the projections and another guy who writes up the comments on guys so that is why the rankings will say one thing and the projections don't back it up or the comments on a certain guy says one thing and the rankings and/or projections don't back it up.

Now I have no idea how many mags or other sources actually split up the duties like this but I know at least a hand ful due and it can get a little silly!! You will see Player A projected at .310 35 110 and Player B projected at .300 30 100 but Player B is 3 spots ahead of player A in the rankings.

In agreement with the previous post, this will happen in the Baseball Prospectus books, where PECOTA puts out a projection and the writer doing the blurb on the player will sometimes flat-out state that they think the projection is way too low or way too high. PECOTA is their tool to get the projection but if it doesn't "feel" right, the writer will say so.

Roger Angell: I was talking with Bob Gibson and I said: 'Are you always this competitive?' He said: 'Oh, I think so. I got a three-year old daughter, and I've played about 500 games of tic-tac-toe with her and she hasn't beat me yet.'