This thesis first examines the history and contemporary landscape of school mental health, offering evidence for schools as an essential component of the child and adolescent system of care. It then provides contemporary discussion around the importance of design in public administration, as well as analyzes the current design model of school-based mental health services, including key actors, normative assumptions, and underlying conceptual models to demonstrate the outdated presumptions that have led to a model that is not designed to adapt to the unique needs of students, especially after the COVID-19 pandemic. Building on contemporary theory of design in public administration, I argue that the largely fragmented, decentralized, bureaucratic, complex, and underdeveloped design of school-based mental health services mainly developed in the 1970s and 1980s has reached its limits and cannot adapt to new societal variables. Lastly, I discuss said limitations of this model to argue for a conceptual and practical re-design of the current system of school-based mental health systems in the United States.

Due to the elimination of the established instructional methods from the impact of COVID-19, the implementation of mass synchronous learning created a new strain of educational experiences for students that took a toll on social interaction. In the Spring 2022 semester, a survey was conducted of students that were previously or currently enrolled in the principal undergraduate biological sciences course, BIO 340: General Genetics, to assess both the prevalence of social interaction in the lives of the students and the potential ways this information could be molded to improve student’s educational and motivational experience. The results of this survey indicated that there was a considerable lack of social interaction and motivation among students that have taken or are taking BIO 340. Through a process of collecting qualitative data of students by 1-on-1 interviews, the majority of students requested that professors communicate with each other to learn more about ways they can incorporate social interaction as external technological applications and tools have been developed. Students brought up many external tools that professors in other biological sciences courses have been utilizing to engage student-to-student interaction and found these resources to increase their level of understanding and motivation. The driving interest behind this creative project is to understand the importance of peer-to-peer learning that may help guide professors that are new to synchronous teaching so that they may increase their level of understanding and comfortability of accessing resources that students themselves have shown to increase their educational experiences. The mixed-method design served as a means to understand what types of social interaction enhance students’ education and motivation.

Significant health inequalities exist between different castes and ethnic communities in India, and identifying the roots of these inequalities is of interest to public health research and policy. Research on caste-based health inequalities in India has historically focused on general, government-defined categories, such as “Scheduled Castes,” “Scheduled Tribes,” and “Other Backward Classes.” This method obscures the diversity of experiences, indicators of well-being, and health outcomes between castes, tribes, and other communities in the “scheduled” category. This study analyzes data on 699,686 women from 4,260 castes, tribes and communities in the 2015-2016 Demographic and Health Survey of India to: (1) examine the diversity within and overlap between general, government-defined community categories in both wealth, infant mortality, and education, and (2) analyze how infant mortality is related to community category membership and socioeconomic status (measured using highest level of education and household wealth). While there are significant differences between general, government-defined community categories (e.g., scheduled caste, backward class) in both wealth and infant mortality, the vast majority of variation between communities occurs within these categories. Moreover, when other socioeconomic factors like wealth and education are taken into account, the difference between general, government-defined categories reduces or disappears. These findings suggest that focusing on measures of education and wealth at the household level, rather than general caste categories, may more accurately target those individuals and households most at risk for poor health outcomes. Further research is needed to explain the mechanisms by which discrimination affects health in these populations, and to identify sources of resilience, which may inform more effective policies.

Machine learning is a rapidly growing field, with no doubt in part due to its countless applications to other fields, including pedagogy and the creation of computer-aided tutoring systems. To extend the functionality of FACT, an automated teaching assistant, we want to predict, using metadata produced by student activity, whether a student is capable of fixing their own mistakes. Logs were collected from previous FACT trials with middle school math teachers and students. The data was converted to time series sequences for deep learning, and ordinary features were extracted for statistical machine learning. Ultimately, deep learning models attained an accuracy of 60%, while tree-based methods attained an accuracy of 65%, showing that some correlation, although small, exists between how a student fixes their mistakes and whether their correction is correct.

Using a dataset of ASU students from the 2016-2017 cohort, we interact gender and parent education level to observe gaps in academic achievement. We see a statistically insignificant achievement gap for males across parent education level, but a statistically significant achievement gap for females across parent education level. We also observe dropout gaps among these interaction groups. We see the widest dropout gap being between males across parent education level, with the smallest dropout gap being between females across parent education level. So with males we see an insignificant achievement gap but the widest dropout gap across parent education level, and with females we see a significant achievement gap but the smallest dropout gap across parent education level. What is driving these gaps and causing more similarly performing students to drop out at wider rates? At the aggregate level, we see larger gaps in grade- associated dropout probability across parent education level for males which may be able to explain the larger difference in overall proportions of dropouts between males. However, when predicting dropout probability of the semester with the most first generation and non-first generation dropouts, we see that females have the largest differences across parent education level in grade-associated dropout probability. This suggests that our model may be best suited in using college achievement data to predict overall dropout probabilities, not next-semester dropout probabilities using current semester data. Our findings also suggest that first generation students’ dropout probability is more sensitive to the grades they receive than non-first generation students.