AP Research: Statistics that discriminate winning from losing Oxford College of Emory basketball games
Advisor: Bryan Meyerowitz
My research paper examines the game-related statistics that distinguish winning and losing teams in
Region X, a men's college basketball league in Atlanta that includes Division I, II, and III schools. While
previous research has analyzed NCAA, NBA, and youth basketball, Region X remains an underexplored
area in basketball analytics. My paper aims to fill that gap by identifying key statistical factors that
contribute to winning. Using descriptive statistics (means and standard deviations), I will summarize
performance trends. Then, through comparative analysis (independent t-tests or one-way ANOVA), I will
determine which statistics show significant differences between winning and losing teams. Finally, I will
apply discriminant analysis to identify the most influential performance indicators. If needed, a
classification and regression tree (CRT) model may serve as an alternative approach. The findings of my
research paper could benefit coaches and players in Region X and other college leagues by highlighting
which statistical factors—such as field goal percentage, rebounds, or assists—are most predictive of
success. My research contributes to the broader field of basketball analytics by offering insights into how
different competitive levels within a single league impact performance. Potential challenges include
sample size limitations and multicollinearity among variables, but the study aims to provide meaningful
takeaways for improving team strategies and player development at various collegiate levels.
Project Blog
Post 1: Third Blog
Third Blog (2/14):
METHOD
– What type of method have you selected? Why?
– Explain and justify 3-5 steps of your method.
– Identify and explain at least TWO limitations/challenges of your method.
My method consists of Descriptive Statistics Calculation, Comparative Analysis, and Discriminant Analysis. First, I will compute means and standard deviations for key game-related statistics to summarize performance trends. Then, I will use independent t-tests (or one-way ANOVA if necessary) to identify significant differences in game-related statistics between different groups, such as winning vs. losing teams or top vs. bottom performers. Additionally, I will likely incorporate Discriminant Analysis to determine which statistical variables best differentiate performance groups, potentially using the leave-one-out cross-validation method (which validates the model by training it on all but one data point, repeating this for each point, and averaging the results to assess accuracy). I chose this method not only because several similar sources used these three types of analysis but also because it made the most sense when comparing statistics between groups. Additionally, a few sources conducted their method on SPSS(some statistics software) which I plan on using as well. A key limitation of this approach is potential multicollinearity, where some game-related statistics may be highly correlated, potentially distorting results. To mitigate this, I will exclude variables with high correlation coefficients (r>0.60r > 0.60r>0.60) from the discriminant analysis. Another limitation is the sensitivity to small sample sizes, such as in leave-one-out cross-validation, which may lead to unreliable classification accuracy.
Which criteria did you use to discriminate among the perspectives in order to reach a conclusion?
Since each source is using similar methods to analyze different leagues/levels of play, I chose the three tests that were most common among them. Additionally most of them used 2-3 tests so my method is pretty similar to existing research. Some sources focused on preseason vs post season and some about starts, vs non starters so I am taking a similar spin to the research by focusing on D1 vs D2 vs D3 levels since Region 10(my gap) consists of teams from each of these divisions. I also think this is a good approach since this method will have implications for all levels of college basketball.
Post 2: Blog 4
Thomas Hoppner Blog 4
1) Provide any data that you've already collected in a readable and comprehensible form (this might be in chart or table format, as quotes assembled from various media, as some sort of word cloud or graph, etc.).
2) Explain this data to your reader. What is significant here and why? What does it mean in terms of your thesis/hypothesis/theory? Does it confirm, refute, or qualify your central claim?
3) If your data set is incomplete, you can explain what you still need to do to complete your data and why (we should already know how if your method section is any good).
4) Whether complete or incomplete, evaluate your data. What does it seem to be missing for you to make a full and comprehensive argument? What has surprised you? What value does this data have for your inquiry, for your field, or for various fields? Why does this data matter and to whom?
1)
2) I did all the statistical analysis on google sheets and I first used ANOVA (analysis of variance) tests for every basketball statistic(rebounds, steals, blocks) and I found that none of them had a pval of less than 0.05. This means that none of the values for each separate division were statistically significant in differing from each other. So, I used Cohen’s d which was used to calculate effect sizes for differences in game-related statistics (d = 0.2 for small effects, d = 0.5 for moderate effects, and d = 0.8 for large effects). I individually calculated each static’s effect size from D1-D2, D1-D3, and D2-D3. To list a few, I found that for Free throw percentage, the effect size from D1-D3 was 0.86(L) and from D2-D3 was 1.0(L), as I investigated, the mean for D3 was very low compared to the other Divisions, making free throw percentage (percent of free throws made) an important differentiating statistic for Division 3. I found offensive rebounds has an effect size of 0.47(M) from D1-D2 but low for D2-D3 and D1-D3. I also found that for turnovers, the effect size for D1-D3 and D2-D3 was 0.51 and 0.61 respectively, so D3 clearly is different from the other two divisions in this aspect; as I looked at the mean, they actually had much less turnovers than D1 and D2 which is odd since they are like a worse division which means a worse division is losing the ball(turnover) less, this interesting point does bring up strategic ideas. Specifically, maybe D1 and D2 are willing to be more risky on offense and are okay with turning the ball over more, or D3 is just super careful in not turning the ball over, or something else. There's a lot more but the whole thing does somewhat support my claim that there are differences in the amount of each statistic per division but not to the level of pval<0.05. There are a few more tests I could do(I do not know what to focus on now) that could bring more depth into my method. My main thing I'm missing is a bigger data set for Division 3 and Division 1 I feel like. When I look at the variance for these two, It is usually higher than DIvision 2 which has 10 teams compared to the 8 for D1 and 4 for D3 but there's nothing I can really do about it for this paper in my opinion. At first when I saw that I has no statistically significant values at pval less than 0.05, I thought I was done for but as I saw Conhen’s D on some sources with values greater than 0.05, I was surprised to find some actual data. That data provides interesting theories about not just Region X but also each NCAA division in general. Although the sample isnt so big, there can be implications for other region’s leagues and conferences and differing levels of NCAA play for many sports like soccer and baseball. This matters because finding out which statistics separate the better divisions from the worse once can give people who are either playing collegiate basketball now, are planning to play, or are just spectating, a clear idea as to which statistics to look at and try to improve upon to make it to the next level. SImilarly, coaches and staff at colleges can use this data to refine their game strategies and improve practices/training their players to tailor them to stronger Division’s expectations.
Think about the original curiosity that led to your inquiry. What other areas of inquiry might that same curiosity lead to? This statistical analysis stuff could point me in the direction of sports science. However, I feel like it picked this just because I enjoyed my time playing basketball in Atlanta and college basketball so I felt this was an exciting idea. I feel like more than just sports, data analytics can lead me to business or finance or anything like that.
