• Home
• Teaching
  Philosophy & Courses Online Education R Shiny Teaching Apps Teaching Mathematical Statistics Blog
• Open Ed
  Basics of SAS Course Basics of Python Course Basics of R Course Teaching with R R4Reproducibility Basics of R for Data Science and Statistics Improving R Programs R for Automating Workflow and Sharing Work
• Other Activities
  Outreach & Diversity Undergraduate Research Statistical Learning Group Sports Statistics Articles, Workshops, Quantitative Literacy, & More
• CV

Sports Statistics

From 2014 to 2018 I had the pleasure of being a co-faculty adviser (with Jason Osborne) of the Sports Analytics Club (SAC) here at NC State.

Jason and I connected over our shared interest in the Pittsburgh Steelers. We started out doing a presentation about umpire effectiveness. This was before PitchFx data was widely available and way before they put the strike zones on the telecasts! We presented a poster at the 2013 NESSIS conference titled Modelling Umpire Misclassification of Balls and Strikes using PitchFX Data. The full poster is available here.

We helped form a sports stats club (later renamed Sports Analytics Club) headed by a truly fantastic group of students (they really drove the success of the club). Some have gone on to work for professional teams. Below is a list of the most influential students from my time with the group (the first five were the founding members) and sports related roles that I can remember:

• Nick Kapur - Has worked for the Yankees and Dodgers and currently works at a sports consulting firm
• James Gilman - Director of Football Analytics with the Eagles
• Michael Dickey
• Caitlyn (Phelps) Kapur - Worked with Richard Childress Racing
• Marschall Furman
• Will Burton
• Jason Thompson
• Rebekah Colonnese, Megan Tabor, Cole Adams - Worked extensively with NC State Baseball
• Sam Gasell - Data Scientist with the Reds
• Bradey Baker - Senior Data Analyst with the Pacers

I gave an invited presentation about the success of the club at increasing student involvement and promoting undergraduate research at JSM in 2015. You can find slides for the talk: Increasing Undergraduate Student Knowledge and Interest Using a Sports Stats Club here.

Below are a few of the projects I worked on with club members!

Using NFL Draft Metrics to Predict Player Success - James Gilman & Nick Kapur

We did a speed session on this topic at the 2014 Joint Statistical Meetings.

The NFL draft has become a 3-day spectacle that can change the fortunes of a team. There are many draft gurus that attempt to rank players and project their futures. In an effort to make draft grades more objective and data driven, we attempt to use NFL scouting combine data and college affiliation to predict NFL impact. Many players do not complete every drill during the combine, resulting in a lot of missing data. Therefore, we use multiple imputation methods and linear regression models in order to predict player success. We look at which drills relate to having the most success at the different positions, which teams draft well based off our model, and which collegiate conferences produce the most NFL ready players.

Effect of Shot Location Trends on Offensive Efficiency in the NBA - Michael Dickey

If you’ve watched the NBA over the last 20 years you’ve seen a big shift in how the game is played. We did a speed session on this topic at the 2014 Joint Statistical Meetings.

After observing a substantial increase in three point shot attempts in recent NBA seasons, one may wonder what effect this has on the amount of other shots taken and how efficient offenses have become as a result. Using NBA shooting data from the 2000-2013 seasons we investigate trends in distance of shots taken from 2 and 3-point ranges. Specifically, we use multiple logistic regression to determine the relationships between number of shot attempts at differing distances and points per possession from across seasons.  We attempt to identify types of shots resulting in the most points-per-possession and which NBA teams are adapting this strategy. Points-per-game of teams found to follow this trend are compared with teams that have not through the use of pairwise comparison tests. The results will find the most efficient shots in basketball that coaches could use to design their offenses.

NFL Play Predictions - Will Burton and Michael Dickey

We did a speed session on this topic at the 2015 Joint Statistical Meetings. It actually got a good bit of attention and Will got to go to the superbowl!

A valuable skill for NFL coaches is to be able to anticipate whether the opposing team will call a pass or run play. If the offensive play type can be predicted, the defensive team can call for example, a blitz or coverage play to gain an advantage. Based on various factors we develop a classification model, trained using play by play data for the 2012 and 2013 NFL seasons, to predict future play types. The factors used for initial consideration are: average yards per pass play, average yards per run play, previous 3 plays, distance to first down, distance to goal line, down, point differential,quarter, and time remaining. Factors used in the final model are determined using variable selection techniques. We compare the accuracy of multiple classification methods including logistic regression and decision trees to determine the optimal model. Once a final model is determined, an interactive visualization is created using the R Shiny package. This visualization is an intuitive and easily interpretable way to make real time decisions given the current game conditions.

Scheduling Effects in the NBA and NHL - Jason A. Osborne

Jason presented this poster at the 2017 New England Symposium on Statistics in Sports. The main idea was to consider scheduling effects for these two sports and how that could be used in betting (unfortunately we aren’t rich now).

Teams in the NBA are giving increasing consideration to the issue of fatigue, recognizing that without rest, performance might be compromised. Using generalized linear mixed models we estimate effects of schedule conditions, whether home/road teams were idle, home or away the day prior to a regular season contest in professional hockey and basketball. We attempt to control for team strengths with factorial effects for home and visiting teams. On the scale of win probabilities, we find the magnitude of the estimated effects of one day of rest for the home team and the away team to be highly significant, additive and comparable across the two sports. The relative change in estimated win probability from the least to the most favorable schedule conditions is as large as 17%(NHL) or 18%(NBA).

We also find evidence that bettors are taking schedules into consideration. Published point spreads and over/under (o/u) lines in the NBA differ significantly across schedule conditions, achieving an ordering that is consistent with the conventional wisdom that having to play the day before a contest and then travel has a negative effect on performance. Interestingly, the observed over/under lines do not differ significantly across scheduling conditions, nor reflect this conventional wisdom. Lastly, inclusion of an overtime effect can “debias” factorial team effect estimates and improve predictions made prospectively (without knowledge of whether wagered-upon games went into overtime), identifying a possible market inefficiency.

Hockey Scoring Rates - Sam Gassell

This project gathers the average shift and scoring data during the 2019-2020 NHL regular season, and analyzes their relationship with Poisson regression. We didn’t end up presenting this or publishing this anywhere.

NFL Momentum - Trevor Hughes

We worked on a project to quantify momentum in the NFL. We didn’t end up presenting this or publishing this anywhere.