Upload
james-mcdowell
View
21
Download
0
Embed Size (px)
Citation preview
SOCON Basketball EvaluationJamey McDowell, Ian McConnell, Eli Stutzman
Objective
Taking NCAA game data from the 2007-2012 tournament seasons, create a metric that predicts how teams will perform against one another
Identify key aspects of the game that have a significant impact on the outcome
Predict with good probability the outcomes of games using our metrics
Preliminary Steps
Look at games that only involve teams that were identified to be in comparable conferences America East, Atlantic Sun, Big Sky, Big South, Big West, Northeast,
Ohio Valley, Patriot, Southern, Southland, Western Athletic Run the analysis on games up to and including January 31
Gives ratings on 26 metrics Each rating is given on a per/100 possession scale, and so the
actual point margin of the game is also adjusted
Identifying Key Metrics
Using the difference of ratings between involved teams, run a regression analysis to identify metrics that more greatly affected the actual point margin of the game, and assign values to them Identified 9 metrics: tm_efg_pct, op_efg_pct, op_tov_pct, tm_orb_pct,
op_orb_pct, tm_ftr_pct, op_blk_pct, op_thpttnd_pct, totalopp_ft_pct Multiplying the value of each metric by the difference in rating of
the two teams involved in a game, and adding them together, we get a “Ketchup” value
“Ketchup” and “Cocktail”
From the initial analysis, each team is given a SRSRating. When comparing teams, the difference between Team 1 and Team 2’s SRSRating, is the predicted score of the game
Taking the “Ketchup” value and adding it to the difference in SRSRating, we get our “Cocktail” value for the game, which is our adjusted predicted score of the game
“Cocktail” tailors the predicted score to the individual teams playing, rather than looking at the first half of the season alone
Logistic Regression
Results
Over the 2013-2014 tournament seasons, we predicted with 66.7% accuracy the winner of games taking place after February 1st (1202 games) “Close” Games (0-5 point spread predicted): 55.1% (314 games) “Contested” Games (5-10 point spread predicted): 56.9% (276
games) “Normal” Games (10-15 point spread predicted): 67.7% (189 games) “Blowout” Games (15+ point spread predicted): 81.6% (423 games)
Furman Ratings
Year 2013 (116 Teams) 2014 (117 Teams)Tm_efg_pct 54 15Op_efg_pct 71 47Op_tov_pct 68 110Tm_orb_pct 63 73Op_orb_pct 42 97Tm_ftr_pct 67 43Op_blk_pct 25 93Op_thpttnd_pct 14 62Totalopp_ft_pct 22 49
Predicting Furman Games
2013: 41.7% “Close”: 2-2 “Contested”: 1-5 “Normal”: 0-0 “Blowout”: 2-0
2014: 60% “Close”: 1-4 “Contested”: 3-0 “Normal”: 1-0 “Blowout”: 1-0