February 18th, 2012 — 7:05pm

One cool thing we can do with the rest-of-season simulation is look at the effect that the outcome of a specific game can have. As an example, take today’s headline BracketBusters game between Murray State and St. Mary’s that just finished. Entering today, the Racers had a 92.9% chance to get an at-large bid should they fail to win their conference tournament. With a loss today, that would have dropped to 88.6%, but Murray State was able to pull out the big victory at home and–at least according to the Achievement S-Curve–punch their ticket to the Big Dance.

Comment » **|** College Basketball, March Madness, Quick Slant, simulation, team evaluation

August 7th, 2011 — 10:32pm

Neil Paine over at the PFR blog wrote basically what I was going to follow up with (albeit much better than I would have). I just wanted to add in a couple other correlations with current metrics that I looked at (correlations are for all stats from 2008-2010).

STAT | r |

EPA per Play | 0.924 |

EPA | 0.900 |

WPA per Play | 0.899 |

WPA per Game | 0.892 |

WPA | 0.886 |

Passer Rating | 0.878 |

All the EPA and WPA metrics are from Advanced NFL Stats (leaderboard here, if you don’t know what they mean check out my last post). As you can tell, EPA per Play correlates best with Total QBR, and is on par with VOA according to Neil’s article. This makes sense: the way QBR handles Clutch Index–first multiplying by it, then dividing by the sum of it–essentially cancels it out, leaving us with EPA per play and the division of credit. The Clutch Index serves to reward QBs who make their best plays in relatively clutch situations, but this appears to be minimal.

Whether or not QBR turns out to be more useful than EPA per Play or VOA probably lies in how well the division of credit is handled. At one extreme, it could be the next step in advancing QB metrics, rewarding those QBs who can get the ball downfield and put the ball on the money while punishing those who don’t. On the other end of the spectrum, if not handled correctly, it could end up adding unneeded complexity and throwing out useful information. As of now, we have no way of assessing which it will be as ESPN has yet to release any details on how their division of credit is handled. Let’s hope we can get a peek inside at some point and see exactly what’s going on.

1 comment » **|** Football, player evaluation, Quick Slant, review

January 31st, 2011 — 10:53pm

This is the first of my posts that I’ll call “Quick Slants”. These posts will be short and to the point, sometimes discussing a current event, a hot topic, or just a random topic I’m interested in. In this post, I want to discuss the difference between **descriptive** and **predictive** statistics, in preparation for my next post which will compare what wins games in the regular season versus the postseason.

A descriptive stat is one that describes the past. These types of statistics tell you what happened, not what will happen. Wins and losses are a great example; they tell you exactly what happened but they are no guarantee of the future. Other examples: Win Probability Added (a concept that originated in baseball, and implemented by Brian Burke of Advanced NFL Stats), yards gained, turnovers, and tackles.

Predictive stats are better at doing just what the name implies: predicting the future. For the most part, I am going to be more interested in predictive statistics. I’m more interested in how a player will do than how he has performed. Certainly the past can be a good indicator of the future, but it is in determining which statistics hold predictive power and which are the product of luck and randomness that the real value lies. Stats that are more predictive than descriptive are yards per play, points per possession, and Expected Points (another Brian Burke creation).

Now this is an oversimplified view, but I wanted to give a summary. In reality, statistics are some combination of predictive and descriptive. It is also, obviously, extremely important exactly what you are trying to predict. Predicting which team is going to win is certainly going to rely on different numbers than trying to determine how many yards a running back will rush for. The point is that when tackling (pardon my pun) a problem, it is important to define exactly what you are trying to do. Are you trying to describe the past? Or are you trying to predict the future? That is the first step in any process to study a question, and will guide you towards the numbers that will help you best achieve your answer.

1 comment » **|** descriptive, Football, predictive, Quick Slant