FanPost

How good was PERM in predicting Blazer wins and the Conference standings?


Season greetings fellow Bedgers

As some may recall, I posted a couple of articles in the preseason where I tried to predict the Blazers win total and projected conference standings. The basic premise was that I used the prior year's PER x Minutes Played (PERM) of each player on this seasons rosters and added them to give a team total. I used a regression to predict the Blazer wins based on the historical relationship of PERM for the Blazers season wins.

See them here for more background:

http://www.blazersedge.com/2013/9/23/4761078/using-perm-to-forecast-a-52-win-blazer-season

http://www.blazersedge.com/2013/9/26/4774942/perm-follow-ups-plus-conference-rankings

Firstly, once I adjusted the PERM for a maximum minutes played I got a season win prediction of 50. So the Blazers bettered this with 54. Well done Blazers!

As for the season standings, here is the comparison of predicted with actual for the Eastern conference.

Easternconferencestandings-actualvspredicted-april14_zps841184b8_medium

So PERM predicted 5 out of 8 Eastern Playoffs seeds. It overrated Cleveland based on Andrew Bynum's failure to return to his Laker form. But it got Charlotte right, which was a non-consensus pick. The three teams I picked but didn't make it still finished 9-11.

And here here's the Western Conference comparison.

Westernconferencestandings-actualvspredicted-april14_zps4d215b78_medium

For the West it did much better, predicting 7 of the top 8. It only missed Golden State (predicted Denver instead). I think Denver would have done much better but were hit by injuries to McGee, Gallinari etc. Of course it didn't predict the Suns finishing ninth!

So there you have it. I wanted to find a metric that would explain why I thought the Blazers would be better than most pundit's predictions. While not perfect, I think it was pretty good.

Will do this again next season and may consider throwing in some extra wrinkles if I can find the time.

Cheers