Winning the Major League Baseball World Series

Introduction

In this research project, I will study how certain variables may be associated with a Major League Baseball team’s shot at winning the World Series. Variables that will be studied include attendance numbers, opening day payroll, and the number of all-star players. Simple linear regression in Python will be utilized to analyze these potential associations. I chose this project because of personal interest in baseball. It is fascinating how each game plays out so differently from the next due to many different variables that affect how a game turns out. Analyzers have traditionally focused on variables such as the number of wins and who the starting pitcher is to determine the shot at which a team has at getting to the World Series. I will branch out a bit and look at some unconventional variables that are arguably just as important.

Data Analysis

I found all my datasets through a quick Google search. It took me to a large database on Kaggle with lots of information that I was looking for. I selected datasets for three variables: starting salaries, number of all-stars per MLB team, and home game attendance. I chose starting salaries because how much a team decides to spend in a given season depends on how well they predict they will do; ‘better’ teams tend to spend more on luxury players while ‘inferior’ teams decide to not splurge on players. The All-Star Game that takes place in the middle of the season features players from all teams. These players are the top players in their respective positions. Fans vote on who is selected based off of performance. I chose home game attendance because more popular teams tend to have more fans and home game turnout. This may be due to how good a team is playing in a given season. Before any data analysis was conducted, I analyzed all the datasets and did some data cleaning. I had to select for only a particular year’s data (2016), as the datasets went back to 1985. I did not have to do any further data cleaning after that. Kaggle made things easier for me. Finding win percentages for each MLB team in 2016 was a bit difficult. I could not find a public dataset on Kaggle to use, so I had to go to a credible website and manually input those values into a dataframe. After I cleaned the data, I created bar graphs to visualize the spread of data that I was working with. I then performed linear regression and analyzed the associations. It was a bit difficult writing code to clean the datasets and conduct linear regression. I had to learn how to use the library as I progressed through the project. It was definitely a very insightful process.

Visualizations

I generated bar graphs to visualize the spread of data for each of the variables. These visualizations enabled me to gain a better sense of the data that I was working with. I then performed simple linear regression to analyze potential associations between win percentages and each of the respective variables.

This bar graph displays the probability of winning the World Series (pre-season predictions). There is a tail of data on the right.

This bar graph displays the starting salaries for MLB teams. There is a slow decline in amount spent as one looks from left to right.

This is a bar graph of the number of all-star players for each MLB team. It is more common for teams to send less representatives to the All-Star game than to send multiple. In this given year, the Chicago Cubs was the exception, sending 5 players to the mid-summer classic.

This is a bar graph of home game attendance for MLB teams. There is a gradual decline in attendance numbers among the teams.

The correlation coefficient is approximately 0.48. This means that there is a positive association between win percentages and salaries. However, this association is not terribly strong. About 23% of variability of win percentages is explained by the team salaries at the beginning of the season. Based off these observations, utilizing a different method to analyze associations may work out better and be more accurate.

The correlation coefficient is approximately 0.57. This indicates a positive association between win percentages and the number of All-Stars on a MLB team. This association is not very strong. Approximately 32% of variability of win percentages can be explained by the number of All-Stars on each MLB team. This variability is higher than that for win percentages & salaries.

The correlation coefficient for this model is 0.54. This is higher than the correlation coefficient for salaries & win percentages, but lower than the correlation coefficient for number of all-stars & win percentages. Approximately 29% of variability in win percentages can be explained by the attendance numbers for each MLB team. This is a fair variability value.

Conclusion

All the linear regression models that I generated and analyzed pointed towards a positive association between the variable and the probability of a team winning the MLB World Series. However, none of these models had high correlation coefficients, so utilizing another model to analyze the data would probably be more effective. The models only slightly convinced me that attendance numbers, opening day payroll/salary, and the number of all-star players influence the win percentage of MLB teams. Despite not being able to come to any strong conclusions, this research project was very rewarding. I have gained greater insight into baseball and its unpredictable nature.

References

Barry, Daniel, and J. A. Hartigan. “Choice Models for Predicting Divisional Winners in Major League Baseball.” Journal of the American Statistical Association, vol. 88, no. 423, 1993, pp. 766-774.

Boice, Jay, and Nate Silver. “2016 MLB Predictions.” FiveThiryEight, ESPN Internet Ventures, https://projects.fivethirtyeight.com/2016-mlb-predictions/.

Lahman, Sean. “The Lahman Baseball Database.” SeanLahman, SeanLahman.com, 31 March 2018, http://www.seanlahman.com/files/database/readme2017.txt.