This week, we want to tell a story. A story of wastefulness, pride, and a team resting on its laurels after almost winning the World Series. The setting, North Texas. Globe Life Park in Arlington, Texas is one of the largest ballparks in the country, with an attendance capacity of 49,115 fans. In 2012, the Texas Rangers had just come off of a second consecutive World Series appearance. However, they were still not achieving sell-outs of every game. In fact, in 2011, they did not have a single sold out game. This means that there was still room for improvement to try and draw more fans to their games. Unfortunately, either they assumed that back to back World Series appearances would be enough to draw fans or they did not believe in the power of giveaways.
For most teams, one of the more popular incentives to draw fans to the stadium are providing promotions, like fireworks nights or even giveaways to a certain amount of early arriving fans. In 2012, the Texas Rangers offered ball caps once, bobbleheads three times, shirts seven times, and fireworks twelve times. When taken cumulatively, we can see if these selected promotions actually had any effect on attendance.
Exploratory graphics help us find models that might work for predicting attendance. Figure 1 shows distributions of attendance across the days of the week and Figure 2 shows distributions of attendance by month for the 2012 season. By looking at these boxplots, we can make comparisons by day and by month across the distributions of attendance.
As we can see by the graphics above, attendance rises on weekends (Friday-Sunday), and decreases in August. We can explore these data further using a lattice of scatter plots. In Figure 3, we mapped the relationship between temperature and attendance, controlling for the time of game and clear or cloudy skies.
There does not seem to be a moderate or strong relationship between temperature and attendance. However, we can see that fireworks always seem to have high attendance. More telling perhaps is a strip plot of attendance by opponent or visiting team. This plot is seen in Figure 4. Teams from large metropolitan areas, such as New York, Chicago, and Houston are consistently associated with high attendance. However, there are only sixteen visiting teams in this study and only eighty-one games. Accordingly, utilizing the visiting team as a categorical predictor presents problems.
To advise management regarding promotions, it is prudent to see if promotions have a positive effect on attendance, and if they do, what that effect might be. After performing a cost-benefit analysis, we had no choice but to draw some troubling conclusions. We first created a table listing the days of the week that the Texas Rangers offered promotions.
All promotions were given out on weekends, including Fridays, when attendance was usually higher anyway. Our hypothesis was that this would reduce the effects of promotions on attendance. The average attendance on weekend days without promotions was 45,839. The average attendance on weekend days with promotions was 46,794. We then created a linear model for predicting attendance using month, day of the week, and an indicator variable for promotions. We entered these explanatory variables in a particular order so we could see if promotions increase attendance when controlling for day of the week and month. Below is the output from this model.
Here is how this model worked on train and test data:
This model only accounts for around 49% of the variance in attendance. Additionally, the ANOVA table shows that promotions are not a statistically significant predictor. Only weekend days, and August, are significant predictors. This model estimates that promotions have a slight positive effect on attendance, and that attendance is expected to increase by 1,136 on days where promotions are offered, month and day of the week accounted for. As stated earlier, on weekend days with promotions, attendance only increased by 955 on average, against weekend days without promotions, almost 200 people less than predicted by this model.
Considering costs for the forthcoming season, the unit cost of a promotions, on average, is expected to be around $2 when ordered in large quantities. We then performed a cost/volume/profit analysis to assess profit contributions. We used the mean attendance last year of 42,719 and an increase of 1,136 due to promotions, up to 43,855, to test days without and with promotions respectively. The results of this analysis is below:
As you can see, an increase of such a small number in attendance due to promotions reduces our total net income on days with promotions. While overall revenue increases, the variable costs increase too much to make it verifiable. We believe that this small increase is due to only offering promotions on days when attendance is high anyway. If the Texas Rangers want to utilize promotions effectively, they should offer them on other days of the week. Tuesdays consistently have lower attendance. Perhaps offering promotions on that day, like the Dodgers do, would increase attendance significantly. All things considered, currently the increase in attendance due to promotions is ineffective. A restructuring of this strategy should be implemented.
When does your baseball team offer promotions? Are you more likely to go to a game with a giveaway? Let us know in the comments below!
As always, our code can be found on Github.
The SaberSmart Team
Miller, T. W. (2016). Sports Analytics and Data Science: Winning the Game with Methods and Models. Old Tappan, N.J.: Pearson/FT Press. [ISBN-13 978-0-13-388643-6] Chapter 7: Promoting Brands and Products (pages 101–118, R code exhibit 7.1, and Python code exhibit 7.2) and Chapter 9: Managing Finances section on cost-volume-profit analysis (pages 134–138). Data sets and programs (R and Python) available in the GitHub repository: https://github.com/mtpa/sads