I am a recent Master’s in Statistics Graduate from UMass Amherst seeking a full time role in data analysis for social good. I have a strong background in statistical analysis and programming using R and Python, data visualization, and communication of data to non-technical audiences. Along with my graduate studies, I was a teaching assistant for Elementary Statistics. I led weekly discussion sections for the students to review and practice the material, and I have developed creative ways to convey fundamental ideas of statistics to people who have never seen it before. Prior to my Master’s in Statistics, I completed my bachelor’s degree in Statistical and Data Sciences from Smith College with a concentration in Community Engagement and Social Change. I am interested in using statistics to better the wellbeing of people and communities, so many of my internships and projects are centered around this intention. I have worked with several non-profit organizations on work which has tangibly helped to move their mission forward. Through my experiences in my internships and academic projects, I have seen the positive impacts of this work in various contexts, and I am excited to further expand the scope of my impact as I move into a full-time role.

During this past year, my Master’s program has been an amazing opportunity to deepen and expand my understanding of statistical methods through coursework, projects, and teaching.

Through the coursework, I gained understanding of a variety of statistical methods including regression modeling, methods for categorical data analysis, Bayesian methods, Causal Inference methods, and Machine Learning. In learning these methods, I found it especially interesting to conduct simulation studies to gain more insight into how the methods work and how their performance changes in different contexts.

The projects helped me to develop an added layer of critical thinking through the experience of applying the methods I learned in class to real-world data, which rarely fits as neatly into the structure of the methods as I would imagine when first learning. Some of my most valuable project experience has been through a Statistical Consulting course where each week, a new client who was seeking support with data analysis would come in, and we would offer the clients guidance as a class. Some clients only needed a small amount of guidance which could be discussed in one class period, and others sought more long-term help. For clients seeking long-term help, a small group of us would volunteer to continue working on the project outside of class.

A central idea I have learned from this experience is the value of taking my time to fully understand the context of the data and questions before jumping to an analysis. Many times the research involves a lot of terminology and domain-specific context which I am completely unfamiliar with, and it can take patience to fully understand the data. Often, one of the most challenging pieces of the work is understanding the research questions enough to translate them to statistical questions and methods, and this piece is essential. In the statistics curriculum, the focus is mainly on mastering and implementing various methods, but I have learned that the true work is in matching the question to an appropriate approach, and often trying multiple approaches.

Building off of this, I have learned the value in the most simple statistical methods. Most of the time, real research data does not fit neatly into all of the assumptions of simple statistical modeling, but I have been surprised at the frequency with which the results of simple approaches actually are consistent with more complex approaches. For example, in a recent project, the goal was to identify unusual points in the context of a question which could be most simply modeled by simple linear regression. However, there were violations of the linear regression assumptions including non-constant variance and the range of response values being restricted between 0 and 1. Further, there was concern that the linear regression fit can be heavily influenced by outliers, and the goal is to find outliers, so we may want an approach which is less affected by outliers. Based on this, we fit a simple linear regression, a beta regression (for restricted response range), weighted linear regression (for non-constant variance), and a quantile regression (for robustness to outliers). We found that the results were mostly consistent between all methods. This approach allows us to center the simple approach in communicating to the client but also ensure the statistical robustness of our results. Before this course, I would have generally thought that if an assumption is violated and there is a more complex approach to address it, it would be best to use the complex one to be sure of the statistical rigor. However, there is extreme value in having the ability to convey a method to someone with no background in statistics in a way that they can fully understand and interpret. Further, in some fields, there are standard statistical methods that are commonly used, and reviewers of publications may even be skeptical if methods they have not heard of are used instead of the standard. Although statistical rigor is important, the results cannot have an impact if they cannot be communicated to a broad audience.

In addition to my coursework and project experiences, I enjoyed working as a teaching assistant for an Elementary Statistics course. In this role, I led weekly discussion classes to help the students review and practice the material they had learned in lecture and held weekly office hours to answer further questions. The challenge of communicating fundamental statistical concepts to students who had never heard of them before helped me to understand and appreciate the concepts more deeply myself. I also found joy in developing creative activities to make abstract concepts more concrete. For example, to convey what a 95 percent confidence interval means, I gathered about 1000 small objects where 30 percent were marbles and 70 percent were plastic pearls and created many random samples of 50 objects such that each student had 50 objects in a small bag. Then, each student created a confidence interval for the “population” proportion of marbles based on the proportion of marbles in their sample. At the end, we had about 20 different confidence intervals, and we noticed that about 19 out of 20 of them, or 95 percent, contained the true population proportion of 30 percent. This helped the students to more easily conceptualize what 95 percent confidence means.

In my free time, I enjoy taking walks outside, spending time with friends, cooking, knitting, and teaching and practicing yoga. I am a certified 500 hour yoga instructor, and I currently only teach one informal class per week, but I would love to start teaching more often soon! I started to learn knitting over the past few months and just finished my first project, which is a scarf!