Rink Stats

Does Brazil have a home field advantage?

Tue, 09 Aug 2016

Check out my new article at FiveThirtyEight about the Olympics. My coauthor Danyel Reiche and I discuss our recent journal article in which we analyze whether Olympic host countries get a home-field advantage. We find that they do tend to win more medals in a hosting year, compared to four years prior. But when you take into account the fact that they also have a huge increase in athlete participation, the home advantage disappears.

Conference about the Olympics at Harvard

Tue, 05 Jul 2016

This Friday (July 8) I’m helping to organize a conference about the Olympic Games. It’s called Who will win in Rio? Understanding political, economic, and athletic success at the 2016 Olympic Games.

The event will feature four academics who will present their research about the Olympics. Included are Danyel Reiche, who helped organize the conference and will present his new book about the politics of the Olympics, Andrew Zimbalist, who will discuss his extensive research on the economics of mega-sporting events, Mark Glickman, who will discuss his work on evaluating player performance, and I will discuss my forthcoming article on home advantage in the Olympics.

The flyer and additional information is provided below. The event is free and open to the public, so if you’re in the Boston area I’d love to see you there!

Series Probabilities, Just in Time for the Playoffs

Wed, 13 Apr 2016

With the playoffs starting this week, I’ve worked to update the playoff series win probability app. A few of the changes I’ve made is that the header of the graph now shows the current probabilities, instead of the current number of games won. I’ve also rewritten a lot of the code and made my data management more efficient. These changes should make it run a lot faster, particularly for past series in previous seasons. Check it out and let me know if you find any bugs.

Win Probability App Updated!

Thu, 17 Mar 2016

This afternoon I posted a major update to my in-game win probability app. I had been hosting it on a server that has been having issues, so I spent some time this weekend updating much of the underlying infrastructure of the app so that it runs better. An added perk is that the app will run faster because I’ve pre-calculated the second-by-second probabilities for every game from the 2007-2008 season until today. For future games it’ll take a couple seconds to load because the app scrapes the play-by-play from the NHL’s website, cleans the data, and calculates the probabilities. Going forward I plan to periodically update it with additional pre-calculated probabilities.

On a side note, as is very clear from the lack of posts over the last year, I’ve been working hard on trying to advance my political science research so that I can finish my PhD and hopefully get a job as a professor. The result has been that a lot of my sports research has fallen to the wayside. But fear not! I have endless sports-related projects (both big and small) that I plan to work on, so stay turned. If you’re interested in checking out my political science research, you should visit the research page on my personal website.

There's So Much More Hitting in the NHL Playoffs

Wed, 15 Apr 2015

You hear every year that playoff hockey is a whole different game from regular season hockey. If playoff hockey is so much more intense than regular season hockey, then we should probably be able to see evidence of in the play-by-play. I looked at hitting numbers for an article for FiveThirtyEight today. I found that the average playoff game has 15 more hits (36%) logged than the average regular season game.

I dug a little bit further and found that this difference is remarkably robust. When you only compare playoff games to regular season games between two eventual-playoff teams, the difference is 14. Still pretty substantial. The result also holds up if you only compare playoff hockey to games in the last few weeks of the regular season.

I had some concern that this might not be a result of changes in how the game is played, but rather in how the play-by-play are recorded. The main issue is that if the NHL added an extra person or two to the group recording the play-by-play during the playoffs, then this could drive up the number of “hit” events that get recorded. I spoke with executives at two different NHL teams and they both reassured me that the people collecting the play-by-play stay the same from regular season to the playoffs. Of course, it’s possible that some of the 36% increase is attributable to confirmation bias among the data collectors. They “know” that playoff hockey is more intense, so they’re subconsciously motivated to record extra hits. While this seems totally reasonable, a 36% jump is way too massive for it to be totally attributable to confirmation bias. Maybe a 3-5% jump, but not 36%.

Below is my original version of the graph that was included in the FiveThirtyEight article. This version actually shows the confidence intervals for each team’s difference of means between regular season hits and post season hits.

Watch My 2015 Sloan Analytics Conference Presentations Here

Sun, 29 Mar 2015

A few weeks back, I was lucky enough to be invited to present a research paper at the ESPN/Sloan Sports Analytics Conference. If you weren’t able to make the conference, you can check out the videos from my two presentation on Youtube. The first video is a 20 minute talk that I gave on the first day of the conference. The second video is a 10 minute version which I presented in front of a panel of industry judges. Unfortunately, they didn’t do a great job of syncing the slides with the video feed, so if you want to follow along with the slides, they’re posted here. Here’s the full text of the paper.

When March Madness Upsets Become More Likely Than Not

Thu, 19 Mar 2015

The men’s NCAA tournament starts this afternoon, and I’ve written an article about upsets for FiveThirtyEight. Imagine that you’re watching a game and the underdog is winning. At what point does the fact that they have the lead trump the fact that they’re the underdog? That’s the question I answer in the article.

If find that for most of the second half, the score trumps the seedings. That’s also the case earlier in the first half, but only for smaller potential upsets.

More NHL overtime changes

Wed, 18 Mar 2015

Another year, another set of NHL overtime rule changes. Right now about 60% of overtime games end in a shootout. The NHL GMs want to see that number get closer to 40%. They’re considering a couple possible changes to how overtime is played, with the hope that the changes will drive up OT goal scoring. I wrote an article for Deadspin today looking at the effect these changes might have. It looks like either of the two options they’re considering will have a much more dramatic effect than any of the rules changes they’ve implemented over the past few years.

Bracket Simulator for the Wall Street Journal

Mon, 16 Mar 2015

The Wall Street Journal came out with a nifty NCAA tournament bracket simulation tool today. I gave them a bit of help on the statistics for one piece of it, so it’s pretty cool to see how it turned out. I especially like how they allow you to give your alma mater a boost in the simulations. Mine’s going to need as much of a boost as they can get, although as a good Dawg I’ll certainly have at least one bracket where they win the national championship.

Slides for my ESPN/Sloan conference presentation

Thu, 26 Feb 2015

If you’re going to be at the ESPN/Sloan Sports Analytics Conference this weekend be sure to check out my presentation. It’ll be tomorrow (Friday) at 10:20 in room 258. [Update: If you want to see the video from my presentations, check out this post.]

If you’re not going to make the conference, you can check out my presentation slides below. Also, be sure to check out my poster presentation here. And the full paper here.

Advanced player data from NHL.com available here!

Mon, 23 Feb 2015

As you may have seen, the NHL rolled out an update to their player statistics pages over the weekend. I have a couple ideas of things that I’d like to use the data for, so I spent the afternoon putting together a cleaned up version of the data. I’ve decided to make the data available to anybody who wants it. If you use the data for anything you write, please give me a shoutout of some sort and mention my Twitter handle, @rink_stats.

Here’s the data in a zip file. Each year is in a CSV file, and you can merge across years using the ‘playerid’ variable. I only just collected the data this afternoon and haven’t started to use it, so if you find any errors please let me know. Enjoy!

Do the stats look different in outdoor NHL games?

Thu, 19 Feb 2015

I wrote an article today for Deadspin where I asked the question, do the stats in outdoor and indoor NHL games look different from each other. When you compare all games, there’s some differences in the number of shots, faceoffs, and penalties. But once you account for the teams playing in the outdoor games, the only differences that are left are in the number of faceoffs and number of shots on goal.

My poster presentation for the ESPN/Sloan Sports Analytics Conference

Mon, 16 Feb 2015

As I mentioned a few weeks ago, I’ve been invited to present some of my research at the ESPN/MIT Sloan Sports Analytics Conference. I’ll be giving a 20 minute presentation during the conference, as well as having my work displayed as a poster. I’m still working on the presentation, but I’ve finished the poster. Check it out below or see the full screen version here.

Update: If you want to see the video from my presentations, check out this post.

Overachieving and underachieving college football programs throughout the FBS

Wed, 04 Feb 2015

Earlier today I published an article today for FiveThirtyEight about the college football programs which over and underachieved given their recruiting success. In the article I only looked at schools in the Power-5 conferences. I had a handful of people on Twitter ask what the results look like when you include all FBS schools. Here’s your answer.

This first graph shows the relationship between recruiting succes (x-axis) and on-field success (y-axis). As you can see, there’s a clear relationship between the two. The more recruiting points a team earned between 2011 and 2014, the lower their Massey rating (i.e. the better their team).

The lower right quadrant are teams that recruited well and performed well on the field. Included here are semifinals teams Alabama, Ohio State, and Florida State. In the upper left are teams with poor recruiting and poor on-field performance. Most of these schools are not in the power 5.

The lower left are overachieving teams; those with relatively weaker recruiting classes but strong on-field performance. Among these are Georgia Tech, Kansas State, Utah State, and Memphis. The closer a school is to the top right corner, the more of an underachiever they are. The culprits here are Michigan, Texas Tech, and Vanderbilt.

The next graph is similar to the one in the FiveThirtyEight article, except this one includes all the FBS schools. It’s probably easier to see if you click the image to zoom in. Non-Power 5 schools who overachieved in 2014 were Utah State, Memphis, and Navy. Vanderbilt remains toward the bottom of the graph, joined by the underachieving programs at South Florida, SMU, and Southern Mississippi.

The last graph shows the average over and underachievement of all FBS schools since 2005. When you include TCU’s non-Power5 seasons (which I did not do for the FiveThirtyEight article), they jump ahead of Wisconsin in the overachiever ranking. As you may expect, Boise State dominates the metric, finishing almost 70 spots higher in the Massey ratings than their recruiting would predict.

Toward the bottom of the distribution are mostly non-Power 5 schools. New Mexico State, Akron, Eastern Michigan, and Idaho have been among the least successful teams at turning recruitment into success.

One thing I should note about these results is that they’re similar to those from the original article, but they won’t be exactly the same (even for Power 5 schools). This is because the regression that I used to generate these results came from a dataset with every FBS school, not just the Power-5.

If you’re interested in seeing the actual numeric values that fed into these graphs, you can download the data here. You’ll notice that the 2014 coefficient for three schools is missing. This is because there was some weirdness in the Rivals data, so I wasn’t able to include them in that part of the analysis.

Which college football programs do the most with the high school talent they recruit?

Wed, 04 Feb 2015

I wrote an article today for FiveThirtyEight about college football and National Signing Day. I compared the four-year average recruiting class to how well each school did on the field (measured using Massey ratings). Here’s a snippet from the article.

“I looked at how well schools’ recruiting classes translate into wins on the field and created a rating of how much each program underperforms or outperforms.”

“Anchoring the top of the overachievement scale is Wisconsin, who on average finishes 32 spots higher than predicted. Also at the top of the scale is Oregon, which turns top-30 recruitment years into top-10 finishes, and Missouri, which has turned top-40 recruits into two top-5 seasons. The chart also shows that Georgia Tech’s overachievement in 2014 was not an anomaly. Despite having had just one top-40 recruiting class since 2002, the Yellow Jackets have had multiple strong seasons over the past decade.”