Posted by aejones on 4/15/2012 12:16:00 AM (view original):
simply saying that while sample size is a real problem, stats are unlikely to reflect "the exact opposite of reality." there's some variance but we're likely to get a reasonable convergence on something like shooting %s, for instance. 

If you really watch stuff, that's not true. I see it every single season, all the time.

Example: I see stats indicate that one player on a team is significantly better than another, when in fact he's significantly worse. (And yes, obviously I understand to adjust for other compensating factors.) This is the fundamental problem with our small sample size -- follow this indication and you'd actually be hurting your team, not helping it.

Posted by girt25 on 4/15/2012 8:06:00 AM (view original):

Posted by aejones on 4/15/2012 12:16:00 AM (view original):
simply saying that while sample size is a real problem, stats are unlikely to reflect "the exact opposite of reality." there's some variance but we're likely to get a reasonable convergence on something like shooting %s, for instance. 

If you really watch stuff, that's not true. I see it every single season, all the time.

Example: I see stats indicate that one player on a team is significantly better than another, when in fact he's significantly worse. (And yes, obviously I understand to adjust for other compensating factors.) This is the fundamental problem with our small sample size -- follow this indication and you'd actually be hurting your team, not helping it.

If you are using them by rote as an absolute rule for setting things up, then yes its a problem>  If you are just using them as guideposts to point out things for you to think about, then that is not really a problem.   Many times I'm looking at things over several years, with similar players and the like to get ideas.  And several things I've looked at and mined statistically have changed my recruiting patterns over the years.

Just randomly poking around espn.com this morning, clicked on an article about Evan Longoria. This was the first paragraph:

"In almost every circumstance, April performance means little or nothing. Last year's April stars included Placido Polanco, Brett Wallace and Sam Fuld, each of whom would see their numbers come crashing back to reality as the season wore on. Over the course of 100 plate appearances, nearly any player can produce good or bad results."

And that's basically what we're dealing with here.

(That said, I do understand and appreciate your point re: looking for overall trends, and that can be culled out of the data -- provided that you actually have enough of it. But looking at a player or handful of players when the sample size is only X number of games or even a whole season and thinking you can draw meaningful conclusions is pure folly.)

As OR put it more succintly earlier in the thread: "Trying to figure out trends with 20-35 games, would be statistically next to impossible."

So jones, yeah, they often reflect the exact opposite of reality. It would be akin to taking the stats through 25 guys of the guys mentioned above in the ESPN article (Polanco, Wallace and Fuld) and think that they represent something meaningful, when in fact they were all totally misleading and followed shortly thereafter by a regression to the mean.

I guess we're just arguing over diction here, do you think it's more likely Placido Polanco or Albert Pujols has a higher vorp/war in the month of april?

Pujols, of course. But what does that matter?

The point is that Polanco (or practically anyone else) can have phenomenal (or terrible) numbers for a defined period of time, and it doesn't really mean anything. So with that being the case, trying to determine who is good/better/whatever from these small sample sizes is equally fraught with inaccuracies and misleading info.

I don't see the harm in having the statistics.  If you think they're stupid just don't pay attention to them. 

So if i have two players who I think have similar ratings but are constructed in a similar way, and I'm not sure which one would be a better offensive option so I give them the same distro, and one is performing significantly better than the other in advanced stats over 20 games and I'm deciding whether I should adjust the distro for the post season, would it be fair to say I'm probably making a good decision for my club's chances at success if I up the distro of the guy who is performing better?

It's hard to compare real world statistics to a simulation that is built around a formula of numbers. There are many reasons why the 3 guys you named would have great Aprils. Could be they are playing against lesser talent, or it could be that people aren't really on their A games yet as it is with MLB period early in the season, players really pick it up when it warms up, or maybe they have some personal issues you'd never know later on in the year. You never know. I do know that in WIS all things being equal, that i can count on a guy shooting 60% to deliver when given the distro. I can bet on a guy avg. 5 blocks a game to come close to that number if I put him on a player taking a lion's share of shots. Nothing is completely definitive but it just sounds like you are saying it's all for naught and means nothing in the grand scheme of things.

Adding +/- and stats per minute or stats per 40 minutes would be a great start.

Posted by aejones on 4/15/2012 11:53:00 AM (view original):
So if i have two players who I think have similar ratings but are constructed in a similar way, and I'm not sure which one would be a better offensive option so I give them the same distro, and one is performing significantly better than the other in advanced stats over 20 games and I'm deciding whether I should adjust the distro for the post season, would it be fair to say I'm probably making a good decision for my club's chances at success if I up the distro of the guy who is performing better?

No it would not. That's my point. The 20-game sample can very easily be misleading.

Posted by therewas47 on 4/15/2012 11:47:00 AM (view original):
I don't see the harm in having the statistics.  If you think they're stupid just don't pay attention to them. 

Oh, I wasn't arguing against adding anything. Add away (although I'd prefer they address actual issues/problems first ...) My only point was just to be careful with them because with the small sample size, they're often going to mislead you.