Question about outliers

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • steve227
    SBR Hustler
    • 08-20-09
    • 61

    #1
    Question about outliers
    when working your own set of values for lets say a data set of values in any half time score probability do you guys crop out outliers and to what degree.if you did ,would data set be more accurate or not
  • roasthawg
    SBR MVP
    • 11-09-07
    • 2990

    #2
    I leave them in but it's something I go back and forth on.
    Comment
    • Pokerjoe
      SBR Wise Guy
      • 04-17-09
      • 704

      #3
      There are times a median is more useful than an average. Play around with that. If you're using excel, look into TRIMMEAN. But it is important not to let your numbers get warped by extreme results.
      Comment
      • steve227
        SBR Hustler
        • 08-20-09
        • 61

        #4
        thanks for the info guys bol this year
        Comment
        • uva3021
          SBR Wise Guy
          • 03-01-07
          • 537

          #5
          well it depends on sample size, which is obvious and banal, but i wouldn't ignore outliers, just find another way to optimize your central tendency (as mentioned maybe median, mode, etc...)

          don't remove apparent anomalies because it works mathematically, for those anomalies are part of the system itself, and variation within the system as a whole just like every other number is variation within that same system
          Comment
          • Wrecktangle
            SBR MVP
            • 03-01-09
            • 1524

            #6
            Outliers can be a very serious problem especially if you are using regression. You must examine each to see if it is an erroneous point as it will "punch way over its weight" in the regression. Typically I regress with all points in and then with suspected outliers removed to see how much change results in the regression equation. Almost always I remove them.
            Comment
            • Maverick22
              SBR Wise Guy
              • 04-10-10
              • 807

              #7
              Do like they do in scoring gymnastics.(or is it figure skating)?

              Throw out the worst. Throw out the best. And use the rest.
              Comment
              • Pokerjoe
                SBR Wise Guy
                • 04-17-09
                • 704

                #8
                I don't think it's that simple, Maverick, though TRIMMEAN does exactly that.

                Consider a team that (however you do it) rates 90, 94, 76, and 89 in it's last 4 games.

                I wouldn't throw out the high and low scores. They're all fair info on the team's ability. I'd average them, or median/average them (that being a half median, half average).

                But there are other data sets where the extreme results should be discounted or ignored, sure.
                Comment
                SBR Contests
                Collapse
                Top-Rated US Sportsbooks
                Collapse
                Working...