APBRmetrics

Ben F. · Joined: 07 Mar 2005 Posts: 391

I just wanted to comment on Dan's latest post. I'll make it quick and flesh it out more later, as I don't have that much time now.

I found it very interesting, especially coming off the heels of reading Malcolm Gladwell's second book, "Blink" (he also wrote "The Tipping Point"). I love his books for the fascinating stories that he tells and the way it gives you a new perspective on everyday things around you.

In any case, I think it was quite relevant to what Dan said - one of the main points Gladwell makes in "Blink" is that often with lots of time and information, we make the WRONG decisions; that an expert who has seen something over and over again can make the RIGHT decision in the blink of an eye, and can't even explain it. This is the power of the human subconcious.

Part of the explanation is that this stems from the ability of the human subconcious to only process what information is directly relevant, whereas if we think about it more, we cannot make this separation.

More later...

mathayus · Joined: 15 Aug 2005 Posts: 193

I think Dan's post is quite strong. I absolutely agree that the nature of the game of basketball makes it much more difficult to quantify the effect of individuals than baseball.

He also brings up the basic truth about AI based computer models: Given all relevant data the model can perform perfectly, however models aren't generally given all relevant data, and so humans perform better. It is important to note however, that this is not a flaw with computer models, so much as it is a crucible for determining how well we humans understand what we are doing. If we cannot build a computer model to perform at least as well as us in data analysis, this means that we have not sufficiently analyzed all of the data and mental tools we are using. It's a bit like the idea that if you can make a lecture on a topic for college freshmen, then you don't really understand that topic yourself.

I disagree with NickS's quote that the reason to use a computer model is because we as humans are poor at analyzing probability. The reason to use a computer model is because computers can crunch data far better than human. Now, I will admit that our inability to crunch data causes our need to use a less exact approach to predicting probabilistic performance, but I think it's important to understand what the root of the problem is.

But I do agree with NickS when he says that stats are often at their best in provoking questions and analyzing specific subsets of the game.

**mtamada** · Joined: 28 Jan 2005 Posts: 376

mathayus · Joined: 15 Aug 2005 Posts: 193

**mtamada** · Joined: 28 Jan 2005 Posts: 376

THWilson · Joined: 19 Jul 2005 Posts: 164 Location: phoenix

mathayus · Joined: 15 Aug 2005 Posts: 193

gabefarkas · Joined: 31 Dec 2004 Posts: 1313 Location: Durham, NC

Ed Küpfer · Joined: 30 Dec 2004 Posts: 785 Location: Toronto

sansbacon · Joined: 04 Nov 2005 Posts: 1

Another good explanation is in Sunstein et al, "Behavioral Law and Economics", 2000.

capnhistory · Posted: Fri Nov 04, 2005 6:36 pm Post subject:

I'm not sure if this will add to this conversation or simply muddle it, but the thought is on my mind and I thought I should share it. I was just reading some article in psychology about decision processes, by Robyn Dawes. Dawes studies linear wieghts that could be applied to making any given decision. One of the interesting indications of his work was that it might not matter if the weights are calibrated accurately, so long as they apply the proper functions to each term (as in adding the good and subtracting the bad) and include as much relevant information as possible. Even in situations where the weights are inaccurate they still do a better jon of reaching optimal conclusions than unaided, untrained humans. If you look at all of the different linear weights models used in sports stats, some seem less usefull than others, yet all seem to hilight important information our minds may not identify independently. If we accept this to be true to any degree, then that certainly does raise the bar for all of us interested in studying sports through the numbers. We can't just say "hey, if you apply this formula, you'll find out some pretty cool stuff." Now we have to strive to identify what makes a given formula better than another, or what makes certain conclusions more useful than others. In other words, our task can't be merely to produce information, but to be informed.
_________________
Throw it down big man!
More expansive basketball babble at a slower pace The Captain of History