Preseason stats: Information or Noise?

"While it's best to be cautious of such players, don't throw out preseason stats altogether. The evidence indicates there is some predictive value to how players play even in seemingly meaningless games." - Kevin Pelton, former Sonics stat geek

The tiny sample size of preseason rather than the lower competition appears to be the major factor for why it's not more predictive. Nevertheless, it does a surprisingly nice job of showing who put up stats in the regular season.

about 1 year ago poster 6 comments 1 recs |

Story-email Email Printer Print

Comments

great link!

I’ve been wondering about this.

Boomshakalaka

by jksnake99 on Oct 13, 2008 11:10 AM PDT reply actions 0 recs

a correlation of

0.4 between pre season and regular season record is considered good? I imagine that low of a correlation value taken in isolation would lead you to not put much weight into it.

Off the cuff the stats seem to have a low signal to noise ratio.

Life is exhausting when you are this stupid.

by jonestr on Oct 13, 2008 12:12 PM PDT reply actions 0 recs

I think it depends on the field.

In pure math or the hard sciences, 0.4 is pretty weak. In some soft sciences like psychology, it is incredibly good. In basketball, 0.4 is about the correlation between box score stats like PER and plus/minus stats. So, my guess is 0.4 is a decent correlation for dynamic sports like basketball. However, it is obviously far from conclusive.

by poster on Oct 13, 2008 12:21 PM PDT up reply actions 0 recs

true

I was just playing around and generating some random numbers for various sets sizes and checking the correlations and it is pretty easy to get something on the order of 0.4 with smaller data sets. It seems like if you believe that something like correlation is good then you would do some more filtering to improve your estimate, as was done in this article. Agree with jksnake that it is a pretty good read.

Life is exhausting when you are this stupid.

by jonestr on Oct 13, 2008 12:36 PM PDT up reply actions 0 recs

I'm surprised they even report the correlation...

A quantity of greater interest is the average absolute value of the prediction error or the standard deviation of the prediction error. If someone is using preseason games to predict regular season stats, take preseason projection subtract the actual performance, get the absolute value and average it. Or, just find the standard deviation of the prediction error. Either of these numbers would tell you a lot more about how accurate the prediction is. Say we predict Oden to average 17 points per game, from the preseason, the standard deviation or the absolute value of the prediction error would give us an idea of the margin for error of the that prediction.

by PoliSam on Oct 13, 2008 9:44 PM PDT up reply actions 0 recs