October 2, 2009

More on Cavs Magic Series

Filed under: Uncategorized — wwinston @ 8:08 am

We have a player rating for each player during each game. Ben Wallace finished the Magic series with a -42 points rating and -216 Impact rating. After three games he had a -34 points rating so there were indications of problems. Wally Szcezrbiak  also had a subpar series. He had a -35 points rating for the series. When Wallace and Wally were both out Cavs played 23 points better than an average NBA team. Rest of the series Cavs played 20 points worse than an average team.


  1. Not that it matters in this case but what is the size of the error on your 58 minute sample? At basketballvalue it tend to be about 17 points and makes interpreting all but the most wildly positive and negative results questionable.

    What size error do you get on a 200 minute sample? At bv it is still about 9 points. I’ll look at the smaller stuff but I have the most confidence in what there is the least of- lineups played over 200 minutes and it really strong or weak. Stil hard to say much about stuff in the broad middle.

    Multi-year analysis has challenges but may help some. If a team wanted to rely on lineup Adjusted +/- I’d try to get 200+ minute samples. Probably 400+, whether it be from multi-year or not. Can’t get this on many. Especially if you waste the regular season using hundreds of small lineups. Teams ought to be pretty selective about what they are testing / using, if they are trying to be guided by the numbers.

    Comment by Crow — October 2, 2009 @ 6:45 pm

  2. There were only 16 lineups used over 400 minutes last season in the entire league and if you expand to a 2 year framework it is probably only about 50. The error at BV is still 4-6 points.

    So on average a team might sorta know ballpark how good 2 of their lineups were. At most.

    Unless you’ve cut the error significantly.

    If teams really worked at it I’d think they could know more about hopefully at least 5-10 lineups. I’d think they should but they don’t. Coaches playing the match-up game to such a degree that season stats aren’t that helpful in most cases. At least using publicly available Adjusted +/-. But even the available data might help over not knowing and considering it alongside other stuff.

    Looking at 4 man data and adjusting it might increase sample size but perhaps 5 man lineups are too unique to use this technique?
    I’d look at and then decide. Got to work any angle you can.

    Comment by Crow — October 2, 2009 @ 7:10 pm

  3. If specific player vs player match-ups can produce wildly different results it seems likely lineup vs lineup will too. Maybe you can gather enough data against divisional rivals or regular playoff opponents for starter vs starters using a multi-year approach to have some decent basis but for most situations the data will be quite thin.

    Largely back to coaches guessing, or analysts guessing.

    Comment by Crow — October 2, 2009 @ 7:19 pm

  4. standard deviation is 12 points per 48 minutes

    Comment by wwinston — October 2, 2009 @ 8:40 pm

RSS feed for comments on this post. TrackBack URL

Leave a comment

Powered by WordPress