Mathematics graduate student Aditi Sen's research refines survey analysis when few people respond—with sweeping applications. 

By Jason P. Dinh

Aditi Sen (M.S. ‘25, applied mathematics & statistics, and scientific computation) loves statistics for its ubiquitous nature. 

“I love statistics because it is everywhere—quietly shaping data-driven decisions across every field,” said Sen, who is now a Ph.D. candidate in applied mathematics & statistics, and scientific computation at the University of Maryland. 

Guided by that interest, Sen’s dissertation focuses on how to make the estimates generated from real-world survey data more precise. 

Sen develops new methods to generate more precise survey estimates for small areas, where few or no people respond. Her approaches merge insights from multiple datasets—a technique called statistical data integration. This method leverages the strengths of different sources—for instance, in sample coverage or the types of questions asked—to cover for each one’s weaknesses, yielding more precise estimates with little added cost. 

Sen’s work can be applied to public health interventions and policy design. Governments, for example, rely on surveys to gather localized data to guide how they target their policies, funding and resources. 

“The problem is that when you look at granular levels, like counties or districts, you may not even have data,” Sen said. “That’s where we want to contribute.”

Born to be a statistician

Sen may have always been destined to become a statistician. She grew up in Calcutta, India—the hometown of Prasanta Chandra Mahalanobis, known as the father of Indian Statistics, who introduced a popular statistical metric called Mahalanobis distance. 

Sen earned a bachelor’s degree in statistics from Presidency University in India, followed by a master’s degree at the University of Calcutta. She went on to work as a data analyst for HSBC—Europe’s largest bank by total assets—where she applied her statistical acumen to analyze issues related to banking transactions, personal loans and credit card transactions. 

“Those were some of my very formative years,” Sen said. “But it was always in me to learn more about the subject and improve the methods.” 

So, she enrolled at UMD to pursue her Ph.D. 

From COVID-19 to presidential elections

Working with her advisor, Partha Lahiri, a professor in the Joint Program in Survey Methodology and the Department of Mathematics, Sen quickly took on projects that could benefit the world outside of academia. 

Compared with her prior coursework, “survey statistics is taught at UMD in a very different way,” Sen said. “It focuses on the applications and how it is useful to us.” 

Early in her Ph.D., Sen researched how the American public perceived masking in response to COVID-19. She analyzed data from a coronavirus survey conducted by the University of Southern California. This dataset struggled to generate reliable estimates in smaller states such as Delaware and Wyoming, where only three to four people responded. 

To better grasp public opinion in these areas, Sen integrated insights from multiple large datasets produced by the U.S. Census Bureau. Her approach gleaned valuable information from these massive surveys, even though they didn’t ask specifically about COVID-19. She built a predictive model that characterized the geographic and demographic attributes in each state using the Census data, then fed those characteristics back into the COVID-19 survey. In states with few survey respondents, the new model drew on responses from demographically similar states to improve estimates. The research was published as an editor’s invited paper in the journal Statistics in Transition New Series in 2022. 

Later, Sen used open-access Pew Research Center data to improve U.S. presidential polling data for areas with few or no respondents. Her new method, published in the Journal of Survey Statistics and Methodology in 2025, started by identifying economic and demographic characteristics that predicted presidential preference using Pew’s dataset. Then, similar to the COVID-19 study, the model pulled relevant Census Bureau data to estimate presidential preference based on demographics in districts where few to no people responded. The paper won the student paper competition for the Washington, DC chapter of the American Association for Public Opinion Research in 2024.

“Aditi’s research on COVID-19 masking has given us some really critical insights into how public health actually works on the ground,” Lahiri said. “Beyond that, I’ve been incredibly impressed with her work on election forecasting; by finding ways to combine different data sources to make projections more precise, she’s tackling one of the biggest challenges in modern polling.”

For the promise and impact of her research, Sen won the American Statistical Association’s Edward C. Bryant Scholarship for an Outstanding Graduate Student in Survey Statistics at the Joint Statistical Meetings—the largest gathering of statisticians in North America—in 2025.

“Being named the sole winner of the 2025 Edward C. Bryant Scholarship from the American Statistical Association is a huge deal,” Lahiri added. “It really highlights her technical skill and her ability to apply survey statistics to messy, real-world problems.”

Inside the statistical mind of ChatGPT

Sen’s latest project looks under the hood of transformer models, which are a type of advanced neural network behind the mass proliferation of generative artificial intelligence. Examples of transformer-based models include ChatGPT (the “T” stands for transformer), the image-generation program DALL-E and the Nobel Prize-winning protein-folding software AlphaFold. 

Transformer models can learn rules, context and meaning at scale, but the fundamental statistics that govern them are murky. Because transformer models are, at their core, predictive models like those used in survey analysis, Sen suspects that demystifying their foundations could yield profound insights for her field. 

Sen hopes to continue this work as she seeks postdoctoral and faculty positions, sharpening her skills in statistical learning theory and applying abstract concepts to benefit people outside of academia. 

“If I can make a difference in public health applications, that would be something I would feel good about,” Sen explained. “That's what I love and find really useful about survey statistics—that this work can be applied to benefit any person in any field.”