looking for projected wins vs actual wins standings

[Thurston Fluff]

I've been interested in looking into anomolies in projecting win/losses to see if it is simply random chance or if there is some small variable that got overlooked. I'm basically following the idea that many times with scientific calculations anomolies that don't fit the paradigm (as outlined in Thomas Kuhn's Theory of Scientific Revolutions) get ignored as random when in fact there may be some sort of minor variable that could be used to further enhance the calculation. Unfortunately I can't find an easily accessable simple projection vs. actual standings. I'd like to get as many years as I can so I can study the teams that didn't fit the projections. If someone has a link or something I'd greatly appreciate it. I don't really need the math or proofs but I would appreciate opinions on which projections are most reliable. Thanks in advance for any help given.

[igor67]

If you to the bottom left grouping BP has a number of different standings

www.baseballprospectus.co.../sortable/

 

The adjusted standings combine adjustments for runs scored and allowed plus a difficulty adjustment for the schedule. They also give you the real standings and just the simple unadjusted pythagorean standings

[Thurston Fluff]

Thanks Igor. This, along with stuff Russ gave me, is going to take some time to sort through. Most likely to find out nothing is going on and I wasted my time. Thank God my time isn't worth much.

[rluzinski]

I'm basically following the idea that many times with scientific calculations anomolies that don't fit the paradigm (as outlined in Thomas Kuhn's Theory of Scientific Revolutions) get ignored as random when in fact there may be some sort of minor variable that could be used to further enhance the calculation.

 

I think people KNOW that there are variables that are being ignored. The real question is, how significant are those omissions. For instance, the pthygorean equation assumes a normal distributions of runs scored per game (x amount of games scoring 0, 1, 2, etc.. runs). We know that not all types of offenses will score the same distributions but it's generally close enough to ignore. We accept a small amount of loss in precision to simplify the model but the overall accuracy of our model shouldn't theoretically suffer.

 

You seem to be suggesting that a significant variable might be being omitted, which may be causing a significant and systematic error in the projections? I won't begin to pretend that it's not possible but would assume a fairly rigorous study would be needed to find it. These projections are no longer something people throw together in their free time. They are constructed and tested against the actual results and each other over and over again. I don't think you'll find an obvious omission.

 

Also, as I mentioned in the PM, there's a lot of statistical noise in a seasons worth of games for a team, so it's difficult to find systematic problems with a model without looking at many years worth of results. Assuming a binomial distribution, a team that has a 50% chance of winning every one of their 162 games still has about a 60% chance of winning 85 or more games or 76 or less games. That means that you could construct an absolutely perfect model and still expect it to be wrong by at least 4 games more often than within 4 games!

 

The reality is that any projection system will always have a high degree of uncertainty associated with it. Of course, if that wasn't true, the sport it was modeling would be incredibly boring, since people could have a pretty good guess with how the season would play out before a single game was played.

[dadofandrew]

The one constant is that a very good team bats a lot less than others...if they win most home games, those missing innings add up. Because of that minor glitch, you'll see most of them outperform their "expected" win total.

[Thurston Fluff]

You seem to be suggesting that a significant variable might be being omitted, which may be causing a significant and systematic error in the projections?

 

Not at all. I simply want to see if there is any minor adjustment that might make it more accurate. I tend to believe in stats and hard numbers but I also question them. Mostly becuase I believe asking questions about them and analyzing them to the nth degree keeps statistical analysis updated and as accurate as can be.

 

Also, as I mentioned in the PM, there's a lot of statistical noise in a seasons worth of games for a team, so it's difficult to find systematic problems with a model without looking at many years worth of results

 

I'm finding that out.http://forum.brewerfan.net/images/smilies/wink.gif

I'm working now on taking individual stats. Things like run differential and seeing what those anomolies look like. I most likely will find out the null hypothisis is right, there is nothing goig on. But why not look? The best stats are the ones that had the most questions asked about them and survived IMO.

 

The one constant is that a very good team bats a lot less than others...if they win most home games, those missing innings add up. Because of that minor glitch, you'll see most of them outperform their "expected" win total.

 

If this is right then it should follow losing records should be closer to the predicted number correct?

 

Again thanks all. I'm not at all agianst stats and am not out to prove they are all full of it. I just believe all of them should be questioned. Just like any good skeptic.

[rluzinski]

The one constant is that a very good team bats a lot less than others...if they win most home games, those missing innings add up. Because of that minor glitch, you'll see most of them outperform their "expected" win total.

 

Wouldn't that effect be very small?

 

Even if they very good team wins the majority of their home games, it just means they are losing more than expected on the road, and causing a lot of 8 inning pitching performances (making them look slightly better in terms of RA/G). It evens out.

 

So, it basically comes down to the difference in total wins. The difference between a very good team and a mediocre team is only 20 wins in a year. The very good team loses 20 innings worth of offense or has 20 innings they have to pitch more than an average team. An average team has about 1450 innings, so 20 innings is only 1.5% of that. That might cause a team to outperform their expected win total by about 1.

 

Am I missing something?