In the world of Major League Baseball, the 2021 San Francisco Giants season stands as a remarkable instance of overperformance, defying the expectations of analysts and fans alike. This study seeks to unravel the factors contributing to the Giants’ unexpected success by addressing the research question: what made the 2021 San Francisco Giants so much better than expected?The primary objective of this research is to analyze the discrepancies between forecasted and observed data to pinpoint areas of overperformance. To achieve this, I conducted a thorough examination of both team-level and individual-level statistics. The analysis extends beyond basic metrics to include an in-depth look at the resurgence of veteran players and the critical role of pitching. Methodologically, this research involved gathering extensive data on player performances and team outcomes. By comparing preseason projections with end-of-season statistics, I identified key variances and their potential causes. The study encompassed various performance indicators, including win-loss percentages, ERA, FIP, and WAR, both at the team and individual levels. The findings reveal that the Giants’ overperformance in 2021 can be attributed to a combination of factors. At the team level, strategic decisions and effective management played a significant role. On an individual level, several veteran players experienced career resurgences, significantly surpassing their projected statistics. Additionally, the importance of pitching cannot be understated, as key pitchers outperformed expectations, contributing to the team’s overall success. In conclusion, the 2021 season for the San Francisco Giants illustrates a complex interplay of factors leading to their overperformance. While it remains challenging to definitively classify their success as purely luck or a result of precise strategic execution, the contributions of standout players were undeniably crucial. This research underscores the multifaceted nature of baseball success and highlights the potential for unexpected outcomes even in a data-driven sport.

While Major League Baseball has long been at the forefront of sports analytics, the most commonly used metric to evaluate the quality of a pitcher's performance is called the "Earned Run Average" (ERA) which is a global figure with no consideration for the distribution of values. This paper investigates the relationship between observed win percentages among pitchers that exhibit high start-to-start variation in the quality of their starts (consistent pitchers) relative to pitchers that exhibit low start-to-start variation in the quality of their starts (inconsistent pitchers). Logistic regression is first leveraged to transform the traditional components of ERA, runs allowed and innings pitched, into a proxy for expected win probability. Then, through the use of bootstrap sampling, a simulation of hypothetical pitchers is created to compare the difference in actual and expected win totals among pitchers that exhibit different distributional characteristics. Finally, the "Probability of Superiority" between opposing pitchers in individual games is calculated in an attempt to identify optimal matchups. In summary, a statistically significant pattern of outperforming expected win totals is found among pitchers with higher start-to-start variation. However, over the time period analyzed, pitchers at the major league level are much more similar than they are different. As a result, the magnitude of the difference in actual and expected win totals between consistent and inconsistent pitchers — while statistically significant — is arguably not large enough to be considered practically significant.

In recent decades there has been tremendous growth in new statistical methods and applications for modeling random events occurring in space and time. There is a positive outlook on the demand for research in this field for the coming decades as space- and time-referenced event data will become more commonly available and high in size and resolution.

This dissertation makes methodological contributions in non-parametric and semi-parametric self-exciting point processes models and their application to infectious disease spread and quantitative criminology. For infectious diseases we demonstrate that point process models can be an effective tool for real-time descriptions and forecasts of an outbreak by comparing its performance to traditional compartmental models. We introduce a purely infection-driven, non-stationary point process model and its estimation. We propose a non-parametric implementation of the Recursive Hawkes model. In dealing with gang crime event data, we address the long standing challenge of distinguishing spatial and temporal inhomogeneity with true triggering, as well as evaluating event-based treatments that are non-randomized due to practical and ethical considerations. To this end, we propose a new method to non-parametrically incorporate spatial covariates in the background rate of crimes. We also introduce a sub-sampling procedure to evaluate non-randomized, clustered treatments in order to generate synthetic controls to improve causal interpretation. We assess this procedure with simulation studies.

Methods for evaluating the fit of spatial point process models using residual analysis areexplored to study fatal car accidents in Downtown Los Angeles (DTLA). Residual diag- nostics include spatial residual plots, quantile-quantile (Q-Q), and residual density plots to summarize residual distributions. Comparative analysis focuses on homogeneous and dif- ferent structures of the inhomogeneous Poisson point process models, incorporating covari- ates such as freeway proximity cub distance and environmental conditions Smoke.or.Haze. Goodness-of-fit metrics and K-function analyses assess clustering and dispersion patterns, particularly in high-traffic regions, relevant to the covariates involved. Results highlight improvements in model performance when spatial covariates are in- cluded. Residual analyses reveal that homogeneous models fail to capture local clustering, while models with covariates reduce unexplained variability and align residual distributions more closely with theoretical expectations. K-function results show that combining covariates effectively balances clustering and dispersion patterns, particularly at smaller distances. The study is only an introduction to applying locational and environmental factors to enhance the ability of point process models to explain spatial variability in fatal accidents. These findings provide a foundation for improving urban safety planning and traffic policy design. Residual diagnostics and spatial analysis indicate that future models could benefit from additional covariates, thinning techniques, and spatio-temporal extensions to capture evolving accident patterns and further improve model fits.

Employing over 10 million people, with an annual expenditure averaging over $1.2 trillion, the U.S. construction industry is massive in both scale and reach. Every day, thousands of construction builds are underway across the country, each with a unique set of costs contributing to its overall budget. For many, the cost of commodities, or materials used to build a project, is a huge variable that changes constantly and can negatively impact a project both financially and temporally. Being able to better understand and predict the way the prices these commodities change over time could lead to huge savings of time and money for anyone at the planning stages of a construction project. Statistical time series modeling is one effective way to increase this understanding and prediction abilities. This thesis models and forecasts the price indices of three construction commodities integral to the industry: lumber, iron/steel, and concrete. For each commodity, the month-to-month percentage changes of nearly 100 years of index values are modeled using Autoregressive Moving Average (ARMA) models, which are selected based on multiple model selection criteria. The models are then used to forecast an 8-month period directly following the end of the data series, which corresponds to the first 8 months of 2022. In doing so, it is found that the lumber model using ARMA methods makes forecasts indicative of what has happened in reality in the first half of 2022, where as the iron/steel and concrete models have much more difficulty in doing so.

The abundance of big data has paved a way for its use in secondary mathematics classrooms. Model eliciting activities with participatory sensing data allows us to determine how well teachers can navigate a statistical investigative process, as defined by the Data Cycle. Video coding helps to map teacher progress through the different phases of the Data Cycle, as well as periods of "stuckness'" during a model-eliciting activity.

Classroom practices in regards to technology use may have a significant impact (positive or negative) on the effectiveness of a curriculum. This paper looks at temporal frequency of technology use in the context of a high school statistics curriculum, and generates exploratory clusters of that usage with multivariate regression trees. It examines both Euclidean distance and two versions of Kullback-Leibler divergence, ultimately discovering that Euclidean clusters are more robust to outliers and have lower cross-validated error.

Major League Baseball teams play considerably more games per season than the teams in any of the other major professional sports leagues in the United States, providing baseball fans with a large selection of games and frequently resulting in large fluctuations in attendance. In addition to studying the effect of the home and visiting teams' season win percentages on attendance, this paper examines various shorter-term metrics of performance to see which recent performance metric has the most significant impact on a home team’s game-to-game attendance numbers. The number of wins in a team’s previous 10 home games is shown to be statistically significant and the best recent performance metric to estimate attendance; however, the attendance prediction benefits are not practically much better than a simpler model that ignores such recent performance. This paper also discusses how the magnitude of the variation in a team’s attendance numbers can differ from team to team and from season to season and suggests using a team-season standardized attendance response variable to eliminate this difference in variance issue between low average / low variance, high average / high variance, and high average / low variance teams. The models with the team-season standardized response variable consistently outperform models with the traditional attendance response variable.

Abstract: This paper explores binary classification and survival analysis as methods toaugment business practices in the financial sector, focusing specifically on loan repayment. By leveraging both statistical methods, I aim to deliver insights that inform a holistic framework for approaching key questions around retention

Fraudulent credit card chargebacks continue to be an ongoing issue in the live event ticket-ing industry. Using past work in the field as a guide, logistic, random forest, and k-nearest neighbor models are trained and evaluated using a Tidymodels framework. To address the imbalanced nature of the data set, upsampling, downsampling, SMOTE, ADASYN, and ROSE resampling techniques were applied to the data set. Findings suggest that past results are consistent in that unsampled random forest models perform best for predicting charge- back fraud. The potential to streamline more machine learning models using a tideymodels framework seems possible and would have potential benefit for company use. Sales Amount associated with the order stands out as an influential variable in predicting chargeback fraud.