Tuesday, September 20, 2005
Individual Stats Primer
Since there will be more discussion of individual stats on this site this season, I thought I’d throw together a post to let people know what values of each are exceptional. These are the measures I refer to on a regular basis. I’ll add to and adjust this document as events warrant.
Percentage of possible minutes played (%Min): Self explanatory, I think. La Salle’s Steven Smith (97.3%) led all D1 players in 2005. He missed just 32 of the 1,185 minutes that the Explorers played. Wyoming point guard Jay Straight (96.9%) was second, including a Ripken-esque nine consecutive games without a rest. Few players exceed 90%.
Offensive rating (ORtg): A measure of personal offensive efficiency developed by Dean Oliver. The formula is very complicated, but accurate. For a detailed explanation, buy Basketball on Paper. Anything over 110 is good, and 120 is excellent for a player that is the workhorse on his team. The best I saw from last season was Utah State’s Spencer Nelson at 133.
Percentage of possessions used (%Poss): A measure of personal possessions used while the player is on the court. Simply assigns credit or blame to a player when his actions end a possession, either by missing a shot that isn’t rebounded by the offense or committing a turnover. 20% is average, and 25% indicates a go-to guy. 15% is a player with a limited role in the offense. Higher values do not indicate a player is better, merely that he is more involved in the offense. It’s almost impossible to get to 30% in the college game, although Vermont’s Taylor Coppenrath did so in 2004 and 2005. It is difficult to combine high possession usage with high offensive rating.
Percentage of shots taken (%Shots): This is the percentage of a team’s shots taken, while the player is on the court. This is a pretty good proxy for %Poss, and significantly easier to calculate. It is PlayerFGA / (%Min * TeamFGA).
Effective field goal percentage (eFG%): Same as regular field goal percentage, except that made three-pointers are appropriately given 50% more credit. The top ten in this category with at least 300 FGA from the 2005 season, with players in bold returning in 2006:
1 Salim Stoudamire Arizona 64.8 2 Jaycee Carroll Utah St. 64.2 3 John Reimold Bowl Green 63.6 4 Dee Brown Illinois 63.6 5 Josh Almanson Bowl Green 63.2 6 Andrew Bogut Utah 63.0 7 Eric Williams Wake Forest 63.0 8 Daniel Kickert St.Mary's 62.8 9 Michael Harris Rice 61.6 10 Seamus Boxley Portland St 61.1
It’s mostly players that shoot a lot of threes and shoot them well, but there are some post players that sneak in also.
Offensive rebounding percentage (OR%): This is the percentage of possible offensive rebounds a player gets:
PlayerOR / [%Min * (Team OR + Opp. DR)]
The denominator is scaled based on the percentage of a team’s minutes played by the player. Anything over 10% is good. Sean May of North Carolina was the best I saw for the 2005 season at 16.7%.
Defensive rebounding percentage (DR%): This is the percentage of possible defensive rebounds a player gets:
PlayerDR / [%Min * (Team DR + Opp. OR)]
Anything over 20% is good. Bogut was the best I discovered for the 2005 season at 31.0%.
It is generally believed that offensive rebounds are more attributable to individual effort than defensive rebounds. Due to its relative rarity, an offensive rebound is considered more valuable than a defensive rebound.
FT Rate: Free throw rate is calculated by 100*FTA/FGA. This measures a player’s ability to get the line using the number of free throws shot per 100 field goal attempts. Players that shoot a lot of free throws tend to be efficient scorers, so a high free throw rate is a good thing unless the player is horrible from the line. Anything over 50 is good, and 70 is excellent. Dwayne Jones (109.3) of Saint Joseph’s was the only player in 2005 to exceed 100 in this category among players with at least 200 FGAs. Jones shot only 54% from the line, so all those trips to the line didn’t hurt the opposition much. The oft-injured Jason Fraser of Villanova deserves mention for posting 115 FTAs with 90 FGAs for a free throw rate of 127.8. Below are the top ten players in FTRate with at least 200 FGAs. Only two return in 2006.
1 Dwayne Jones St. Joseph's 109.3 2 Ellis Myles Louisville 98.8 3 Steven Thomas Texas Arl. 91.9 4 Jamar Howard Wichita St. 91.4 5 Jason Maxiell Cincinnati 90.4 6 Ronny Turiaf Gonzaga 83.8 7 Blake Hamilton Monmouth 83.2 8 Ike Diogu Arizona St. 78.1 9 Chad McKnight Morehead St. 77.6 10 John Bowler E. Michigan 77.6
Turnover Rate (TORate): This is the percentage of personal possessions used on turnovers. It is highly dependent on context, but anything below 15% is great. Average in 2005 was around 20%. Point guards are typically in the 20-25% range due to the nature of their position. Players that do little passing or dribbling will have an artificially deflated TO%.
Assist Rate (ARate): This is assists divided by the field goals made by the player’s teammates while he is on the court. [Changed 4/9/06]
Block Percentage (%Blocks): This is the percentage of opponents’ two-point shots that are blocked by the player while he is on the court. It is computed by Blocks/(%Min * Opponents’ two-point attempts). Anything greater than 8% is very good.
Steal Percentage (%Stls): This is the percentage of possessions that a player records a steal shile he is on the court. It is computed by Steals/(%Min * Team Possessions). Anything greater than 5% is very good.
Monday, August 01, 2005
Points Distribution
The points distribution page illustrates where a team’s (and their opponents’) offense in coming from. The numbers in each column indicate the percentage of points scored (or allowed) by each type of shot. Keep in mind that the numbers on this page do not say anything about the quality of a team’s offense or defense. The data provides another piece of the puzzle of how each team plays offense or defense.
Four Factors
The offensive and defensive summary pages are based on Dean Oliver’s four factors concept. You can read Dean’s more detailed explanation of the four factors here. Essentially, the four factors are the building blocks to the efficiency formula.
Efficiency data gives you an idea of the quality of a team’s offense or defense, but the four factors tell you why a team is good or bad when they have or don’t have the ball. Here’s a breakdown of how each statistic on this page is calculated…
Effective field goal percentage is like regular field goal percentage except that it gives 50% more credit for made three-pointers.
eFG% = (.5*3FGM + FGM) / FGA
Turnover percentage is a pace-independent measure of ball security.
TO% = TO / Possessions
Offensive rebounding percentage is a measure of the possible rebounds that are gathered by the offense.
OR% = OR / (OR + DRopp)
Keep in mind that rebounding percentage is computed from box score data which does not contain team rebounds. Therefore, the figures shown here may differ slightly from calculations made on the rebounding totals provided by a school.
Finally, free throw rate captures a team’s ability to score from the line.
FTRateoff = FTM / FGA FTRatedef = FTA / FGA
Defensive FTRate uses FTA in the numerator since the defense has little control on the percentage of free throw attempts made by the opposition.
In Dean’s piece, he mentions the relative importance of each factor. In the NBA, eFG% is easily the most important factor, followed by TO%, OR%, and FTRate. A “RoboScout”-type analysis of games from the 2005 season shows that the importance of each factor is similar in college, with free throw rate being slightly more important in the college game, but still taking a back seat to offensive rebounding. Each team is different though. For instance, Gonzaga’s free throw rate was the second most important contributor to their offensive success. For Michigan State, offensive rebounding ranked second.
College Basketball Data
The college basketball game file posted on my site is freely available for anyone to use with three requests on my part. (1) Give me some sort of acknowledgment on your site. (2) Let me know you are using the data and why you are using it. I am always curious. (3) While I do not expect you to actively quality control the data, if you do find errors or somebody reports an error to you, please pass it along to me.
Now a little bit about the format. The home team is listed last. Games not played on a home court are denoted by a letter after the last team’s score. A capital ‘N’ indicates a game played on a neutral court. For a game where the listed home team is not playing on its home court, yet still getting a home court advantage, a lower case ‘n’ is used. For my ratings system, I apply half of the stated home court advantage to the home team listed in ‘semi-neutral’ games. Because these distinctions are made solely at my discretion, I have included the site of the game in these cases, so you can use your own judgment if desired.
I do use some logic on how to classify a game. If a game is played at the home team’s home arena, then it is always classified as a home game. This seems obvious, but there are a few cases where these could be considered neutral games, mainly during post-season tournaments. Semi-neutral games are indicated where a game is not played at a team’s home arena, but is still close enough to the team’s home so that they will benefit from some home court advantage. Usually these cases are obvious, but in some cases there can be debate. Rare exceptions are made where a team plays away from its home arena, but they still are considered home games in the database. These cases occur when the home team is playing very near its home against a team traveling a considerable distance. Also, a few teams regularly play home games at more than one arena (Connecticut and DePaul are two examples).
There are some other letter codes that are used to classify a game.
T - Conference tournament game
P - Postseason game
S - A game between two conference teams that is
not a conference game
Additionally, I include how many overtime periods (if any) were played. This will be indicated by a number after the last team’s score.
All games involving at least one of the teams in my ratings are included.
Pomeroy Ratings FAQ
- Schedule Strength is computed by averaging the rating of each opponent, factoring in home court advantage as appropriate. For schedule strength purposes only, unrated opponents are given a rating of the worst rated team.
- Data in the ‘LAST 5 GAMES’ column reflects a team’s performance in its last 5 games against rated teams, based on its opponents current ratings, using the same weighting principles that are used to calculate the season ratings.
What is the purpose of your ratings system?
This system is designed to be predictive. One can get a prediction by simply taking the difference in the ratings of two teams and make appropriate adjustments for home site advantage. You can probably save some work by looking at individual team pages. There you can find predictions for future games, along with the chances of winning the game outright. Check out this site to monitor the accuracy of the major systems out there.
What information goes into the ratings?
The only information I use from each game is the margin of victory/defeat and the site of the game. The result of the game (won/lost/tied) is ignored, other than it being incidental to the margin of victory/defeat. Because the system relies on only past data, it can’t anticipate personnel changes that might affect the relative strength of two teams competing in a future game.
How are the ratings calculated?
The ratings are calculated using a least squares algorithm which develops an equation based on each game. If Team A beats Team B by 15 points, then A = B+15. Of course, some adjustment is made for home site advantage where appropriate. All of the equations are then solved to minimize the mean squared error of each game. Each game is given a weight based on two factors - its significance and when it was played. The significance increases for games involving teams of similar ratings. Significance also increases for games involving teams of disparate ratings where the result is much closer than expected.
Increasing weight is also given to more recent games. In a 30 game schedule, Game 1 will weigh about 40% as much as Game 30, assuming equal significance. For about the first month of the season, some weight is given to the preseason ratings. This is done to prevent the massive amount of daily fluctuation that would otherwise occur with so little data.
Do you cap margin of victory/defeat at all?
Yes. The limit on margin of victory is based on the distribution of margin of victory for all games in a particular season. For college basketball, this works out to something around 16 points by the end of the year.
How do you handle home site advantage?
Pretty much any system out there has shown that teams play better at home. This system applies a fixed home advantage for all teams. I don’t adjust this on a daily basis during the season, instead choosing to use a home site advantage that I have calculated from previous seasons.
Any more questions? Write to ratings@kenpom.com