PER Ratings: Atlanta Dream

On Swish Appeal, I've gone ahead and posted the WNBA's player efficiency ratings for 2009.

If you're interested, here's how the various players of the Atlanta Dream did in PER:

Angel McCoughtry (23.2)
Sancho Lyttle (21.3)
Erika de Souza (20,0)
Chamique Holdsclaw (14.8)
Michelle Snow (14.6)
Ivory Latta (13.5)
Tamera Young (12.1)
Coco Miller (10.4)
Iziane Castro Marques (9.8)
Shalee Lehning (8.7)
Jennifer Lacy (5.8)
Armintie Price (4.2)

If the numbers aren't pleasing, you might want to beat up on John Hollinger, the guy who created the formula.

Expected Wins Method: Pat, Geno...and MaChelle?

A while back I wrote an article about rating WNBA coaches using a system created by John Hollinger. His idea was to find a mathematical way to predict the record of a coach before the season started and then to observe if the coach was able to achieve his or her expect win total. The coach would be graded by increase in expected wins per seasons played.

Hollinger weighted a coach's record based on

25 percent of the winning percentage two seasons before, plus
50 percent of the winning percentage one season before, plus
25 percent of a .500 winning percentage

The .500 winning percentage is a good idea: it raises the expectations for a coach performing poorly, and helps to not penalize a coach that is very successful. Both successful coaches and coaches on the hot seat should reach at least .500 every year. Coaches like Laimbeer and Dan Hughes dominated the metric, winning about 2 to 3 more games a year on the average than expected.

I then thought: if this metric worked for WNBA coaches, why wouldn't it work for women's college coaches? After all, those coaches play about 34 games a year, too. Their records are adversely impacted by injuries in the same manner as a WNBA coach. The only difference would be that winning coaches generally keep winning due to their skill at recruiting.

Furthermore, the major coaches have coached at their institutions for 20 years or more -there would be no paucity of data. I decided to choose two very successful coaches and one "test" coach who, thought generally successful, is not as bedecked in glory as my other two choices.

The first coach is MaChelle Joseph, the current coach at Georgia Tech. The two years before Joseph started, the Yellow Jackets had records of 15-14 (.517 win percentage) and 20-11 (.645 win percentage). Her predicted percentage based on the two years of her predecessor should have been (.25*.517 + .50*.645 + .25*.500 = ) .576, which translates to a 17-12 record over a 29 game season. Josephy went 14-15 instead, and had an expected win difference of -3.

We add all of the expected win differences over Joseph's eight-year career. From 2001 to the present, Joseph has gone 106-76 and has an expected win difference/season of 1.167. In general, Joseph will outpredict the formula above by 1.167 wins a season. You can think of the number - very loosely - as meaning that Joseph adds value at the rate of 1.167 wins a year.

Now, let's look at the biggest coaches of all.

Pat Summitt's started coaching in the 1974-75 season. She has eight national championships to her credit and a 1005-193 lifetime record. Her best year by Hollinger's system was the 1998-99 season where she won the NCAA Championship after coming off a 29-10 season with a 39-0 record. She had a +11 in expected wins that year.

Summitt's expected win difference/season is +2.886. That's pretty amazing, implying that three of Tennessee's wins per year are solely attributable to the coach. This year was a bad year for Summitt, who went -5 expected wins over the 2008-09 season. However, the last time she had a negative differential was way back in 1981-82, when she went 22-10. It really was the worst season for the Lady Vols in two decades.

And of course, you can't talk about great coaches without talking about Geno Auriemma. He took control of a Connecticut program that finished 9-20 in 1983-84 and 9-18 in 1984-85. His first season, he went 12-15 with the Huskies...and had never had a losing season since then.

Auriemma has had three undefeated seasons as head coach of Connecticut. He has six national championships of his own. Auriemma's expected win difference/season is an amazing +3.792. Hollinger's system says Geno is better than Pat!

We can argue with John Hollinger's methodology, of course. However, arguing between "who is better, Geno or Pat?" is like arguing "who was better, Babe Ruth or Ted Williams?" As a Tech fan, I'm hoping that someday Joseph's name will mentioned along with the greats, with the numbers to prove it.

Rating WNBA Coaches: The Expected Wins Method

One of two challenges given to me by the author of the Rethinking Basketball blog was to rank which coaches have been the best coaches of the WNBA based on the method used by John Hollinger.

Hollinger - who is mysteriously listed as "Dave Hollinger" in the article - uses a very simple method. Figure out how many games a team should win and compare the actual number of wins to that number. If a team exceeds the expected wins, we credit those wins to the coach. If the team fails to meet the mark, we debit the coach.

The first problem becomes determining the number of expected wins for a team. Hollinger uses the following method:

expected winning percentage for a team = (0.25)*(winning percentage of season before last) + (0.5)*(winning percentage from last season) + (0.25)*(.500)

Note that 1/4 of the formula above is based on a .500 record. This serves two purposes - to give the coach of a sub-.500 team something to strive for, and to give the coach of a winning team a reward for continuing to win.

Hollinger examined all coaches that had five complete seasons of experience. However, the WNBA presents a problem with its multiple expansions. The formula does not handle the case of a coach coaching in the first or second year of a franchise's existence. It's probably a good idea that the formula fails, since it's more difficult to judge a coach given the dearth of talent in those first two years.

For example, Marynell Meadors loses three complete years - the first two years with the Charlotte Sting in 1997-98 and the first year with the Atlanta Dream in 2008. Van Chancellor's first two years with the Houston Comets can't effectively be judged as Houston was an "expansion team". Who is to say that he did a great job? Maybe the talent just fell into his lap.

Therefore, the list of coaches below:

a) have coached five complete seasons in the WNBA, where
b) any season which is one of the first two seasons of a new franchise isn't counted.

This gives us six coaches to look at who meet the criteria: the following lists the coaches name, the total number of wins above expected over the number of seasons evaluated, the number of seasons evaluated, and the wins above expected per year.

Hughes 23 7 3.29
Laimbeer 18 6 3.00
Donovan 13 6 2.17
Thibault 8 5 1.60
M. Cooper 9 6 1.50
Chancellor 0 8 0.00
Adubato -1 8 -0.13

For example, let's look at the leader, Dan Hughes. Hughes started out by finishing Marynell Meador's 1999 season with the Charlotte Sting where Meadors was fired midway through. We won't count that season. Hughes was let go by the Sting after that season anyway, replaced with T. R. Dunn.

Hughes then shows up in Cleveland in 2000 and coaches his first complete season. The franchise has now existed for at least two years, so we can use the formula.

expected wins for Cleveland Rockers in 2000 = (0.25)(1998 win percentage) + (0.5)(1999 win percentage) + (0.25)(.500 win percentage) = 0.40125.

The Rockers played 32 games in 2000, which means that the Rockers were expected to win 32 * 0.40125 = 12.84 games, or 13 games. Hughes won 17 games that year, giving him a +4 in expected wins.

Now let's look at Van Chancellor. He started coaching with the Comets in 1997, but we can't start evaluating his seasons until 1999 since the formula doesn't work for any season which is one of the first two in the franchise history.

expected wins for Houston Comets in 1999 = (0.25)(1997 win percentage) + (0.5)*(1998 win percentage) = (0.25)(.500 win percentage) = 0.73575.

The Comets played 32 games in 1999, which means that the Comets were expected to win 32 * 0.73575 = 23.544 games, or 24 games. Chancellor won 26 games that year, giving him a +2 in expected wins.

(* * *)

Do the results meet the smell test? In other words, if I had asked you to order these coaches from best or worst, would the answers be what you expected?

Seeing Hughes and Laimbeer in the 1-2 positions might satisfy some readers, but others might think that Laimbeer belongs on top. Donovan, Thibault and Michael Cooper at least have positive values in expected wins per year, lending credence to the belief that those three coaches are at least better than average. (Some Sparks fans might dispute that about Cooper, though.)

Van Chancellor coming out with zero expected wins above average per year is a bit shocking. Was the Comets dynasty more a function of its players than its coach? Chancellor doesn't get any credit in the metric for the first two years of the Comets performance. Richie Adubato is the only coach on the list with a negative value in expected wins per year, and both he and Chancellor are the only coaches on the list that are no longer coaching.