Feeds:
Posts

## Urban Meyer: statistics and momentum in football

On the way to school late night yesterday, I randomly checked in Urban Meyer’s call in show for the first time.  I was pleasantly surprised by a question from the first caller, Adam, and was even more impressed by Urban‘s answer.

Adam’s question was asking the origin and evolution of the coach’s football philosophy and “To what extents, statistics, football statistics, play a role in your football philosophy?” Here is Urban’s answer:

Adam, that’s a. Congratulations. I appreciate your phone call. That’s well thought out and researched question. It is the essence of everything we do. You say what role statistics play in the management of a football game, a football program. In my world, it is everything.

And you will hear people say statistics are for losers. Usually losers are the one who’s saying that (made my day :)).

Statistics are very important and we. There are times, there is one thing that is not statistically analyzed. It is momentum. Momentum, to me, when you deal with young people. They are maybe inexperienced teams at some positions, it is even greater. So the higher level, think about this, the higher level football you get, momentum is not quite as much as a factor. […]

So we do play a game of statistics where we try to manage the game, try to force the team to drive the length of the field, take care of the football. Once you cross the fifty, that’s where you get more aggressive in play calling. However, there is times that we run a faked punt against Nebraska. That was not a statistically well-thoughtout play. However, we were in the quick sand and we were heading into the bad way against a good offense. So you need that momentum shift for your team. [……]

It was an interesting description of what statistics can do and cannot. Statistics and Momentum.

It reminded me a post, Winning with numbers, I wrote about Russ Rose around this time last year. Russ is the head coach of Penn State women’s volleyball. by coaching with numbers, he is the coach with the highest winning percentage among all NCAA sports. Interestingly, he also holds a master’s degree from Nebraska, where he wrote his thesis on volleyball statistics.

By the way, Ohio States are playing at Penn States this weekend. Go Buckeyes!

Now, what I really wonder is if momentum has anything to do with numbers that statistician can measure and study as well 🙂 By a quick search on the internet, I found this: (more…)

## NCAA brackets and Go Bucks!

The buckeyes are in the Final Four of the NCAA tournament and heading to New Orleans.

Getty Images

Since the Buckeyes are still alive, my bracket is doing ok as well. According to the 10-20-40-80-160-320 point system, my bracket has 820 points, with Kentucky and Ohio State alive, before the final-four games start on Saturday. Although the president’s bracket (with 860 points) is doing better than mine currently, but his national champion UNC is out of the picture.

As I see it, the bracket competition, coupled with March Madness, is such an intriguing place when endless data-related stories have been generated, told, and cared by the public. It is a place where each person, expert or not, can write down their predictions and see them getting tested. As a result, the field of Statistics is naturally advertised. It is also a place where no such statements like “I’m sure with 95% of confidence” or “The result is statistically significant at 5% level” have been mentioned. I will try to write on the prediction side of the story next time.

Counting all 2012 brackets in ESPN, close to 6.5 million ones by my estimation, the percentages of people picked final four teams to win the championship are:

• Kentucky 35.1%
• Kansas 7.1%
• Ohio State 4.8%
• Louisville  0.8%

Go bucks! Defeat the odds and win it all.

And one more thing, there will be a “Final Four” game viewing party in the statistics building at tOSU on  Saturday, March 31st at 6PM. Come to see how stat nerds react to the games and see the numbers differently!

## Should Ohio State University Privatize Its Parking Operations?

Back in 2010 when I attend the Ohio State University Winter graduation ceremony, President Gordon Gee proudly mentioned that

We are the country’s most comprehensive university. We have more of everything, academic programs, students, and even and, you’ve have to trust me on this, parking spaces (Commencement Winter 2010, 2:10 – 2:22).

In September 2011, the Ohio State University Board of Trustees authorized the university administration to privatize campus parking operations for 50 years, in return for an up-front payment of at least \$375M.

Professor Bruce W. Weide, my colleague and summer tennis partner, has this to say about the proposed privatization of parking operation with nice and detailed analysis and illustration:

More background information and documents can be found here. Most importantly, if you share the concerns, make you voice heard!

## How accurate is si(a)mple average?

I bet most of us learned how nicely sample average performs as an estimator of the “population mean” in the first few lectures of our introductory statistics class. As we go deeper into the statistical territory, we realize how good or how bad the sample average behaves in different scenarios and based on different assumptions. When assumptions are violated and situations are not ideal, what happens? Here is one of my experience with it.

On January 3rd, the first day of a ph.d. level math stat class I taught last winter, students were asked to provide an estimate for the regular season average Points Per Game (PPG) for an OSU freshman basketball player Jared Sullinger. At that moment, the college basketball was half way into the season and conference plays just started. Based on the 14 games Jared played before January 3rd, he averaged 17.643 PPG (with an sd 9.035 PPG). Here we go! We have a sample of size 14 out of a total of 31 regular season games (population), the task is estimating the population mean.

As one can imagine, we are statisticians good at building models for estimations and predictions. It might be good to throw all kinds of investigation at the data, e.g. the awkward design of data collection, dependence between the data, time series, regression, et al. Anyway, out 19 statisticians and 1 computer scientist in the class, twelve were interested in the problem enough to provide well-thought estimates. Several of the most accurate answers are list here:

• answer #1: “17.65  based on the past information.”
• answer #2: “17.6, which is computed by the sample mean. I model the points as samples from a Poisson distribution, and the maximum likelihood estimation of $\lambda$ yields the mean of samples. In addition, this estimation of $\lambda$ is unbiased. “
• answer #3: “16.96153846. I used the bootstrap method to simulate the rest of the season’s games 10,000 times. Additionally, I assumed that there was a 2% chance that he would miss any one of the remaining games. And there was a 0.5% chance that he would have a season ending injury at some point in the season. If he did, I used a random uniform to determine when it happened (hence, in about 500 seasons he ended up the exact same PPG) …”

It turned out that Jared’s PPG in the regular season was 17.290 (with sd 7.372) and he did play in every one of the 31 regular season games. As a result, our best guess is 17.6 PPG (answer #2), the rounded sample average (the Poisson model seems unnecessary). What did I learn? The sample average was fairly good but a simplified version of it won, at least in this case. The simpler, the better. Not forget to mention, answer #2 was provided by the only non-statistician in the class.

Here is the final question that puzzles me: “Do you think this whole event happens with a p-value less than 5%?”