Monday, December 13, 2010

Accuracy of Hollinger Ratings, Simple Ratings System and Win-Loss in Playoff Series

Totals from 2004-2005 through 2009-2010

Higher Playoff Seed: 56-19
Simple Ratings System( : 57-18
Hollinger Rankings Total: 51-24
Hollinger's Formula =  Rating= (((Sos-0.5)/0.037) *0.67)+(((Sos L10-0.5)/0.037)*.33) + 100 + (0.67*(Marg+ (((Road-Home)*3.5) / (games))))+ (.33(Margin L10+(((Road 10- Home 10)*3.5)/10))))

Why is Hollinger's Ratings Flawed?

First, let me say that these results are not statistically significant. While I have proven statistical significance that determines SRS is a better statistical model than win-loss (albeit slightly, but any formula that can outperform win-loss is impressive) by going back to the 1950 playoffs, I have only done Hollinger ratings from 2004-2005 until now, and this sample is not large enough to be statistically significant.

That said, there is a trend emerging...

Some possible explanations for this trend: Hollinger's formula places an emphasis on a team's last 10 games. This is flawed for several reasons: even if a team were to win all games en route to a championship, this would be 16 games, much more than 10 games. During the season's final 10 games, there will be a number of teams that will have secured a playoff spot (anywhere from the 1st to 8th seed), meaning that a team is likely to be resting its starters, which will affect its margin of victory.

10 games is not a large enough sample size to determine a trend.

Where do Hollinger Ratings get it right?

The emphasis on margin of victory is a much better predictive factor than win-loss.

If Hollinger were to take a longer term view on margin of victory, the results of his formula would be more accurate.

He also does a good job incorporating home-road games without making this too large of a factor because home-road games is literally not a factor when weighting 33.5% of his formula (the amount of home-road games over a season) because, obviously, a team plays as many road games as it does home games over a season.

(This is still a factor in 16.5% of his formula, the last 10 games, although this, at most, accounts for only a +1.4 or -1.4 in his ratings because teams only play 4 more or 4 less road games over the last 10 games of the season.)

Purposed Changes

Eliminating games where one team has already clinched a playoff seed or clinched a certain place in the standing would improve this formula a little. The major change that would be a great benefit to this formula is taking a longer term view of a team's recent games once the team reaches halfway through the season. (For example, 25 games instead of 10 games.)

I hope to analyze whether these changes make Hollinger ratings more successful during the summer after the season.

Why is SRS Succesful?

Unfortunately,'s explanation of SRS simply says it "takes into account average point differential and strength of schedule. The rating is denominated in points above/below average, where zero is the average."

Though I do have a problem with Hollinger in that he never tested or improved upon his formula, Hollinger is nice enough to make his formula public in order to scrutinize.

Unfortunately, we cannot be sure about why SRS is so successful other than saying point differential and strength of schedule are better detrimants of team success than win-loss (again, though, just slightly.) This formula may be available somewhere on this site, but I have searched thoroughly and not been able to find it.

I have provided you with valuable information and innovative content I took a long time to create. Can you do me the huge favor of getting free e-mail or RSS updates by subscribing?