PhD Dempster Award Profile: Ambarish Chattopadhyay

June 8, 2022
Ambarish Chattopadhyay Dempster Award

“I’m from a place called, it’s a very fancy name, Diamond Harbour, in India.  There are different theories about the origin of the name. I don’t know which one is correct (maybe all of them are wrong!), but one story is that, as the European traders approached the harbor, they saw the shore glittered like diamonds in the rays of the blazing sun and named the place Diamond Harbour.”

Graduating PhD student Ambarish Chattopadhyay offers an inspirational image of his hometown.  When he arrived at the Harvard Statistics Department in 2017, he found new inspiration through his learning about research, teaching, and collaboration.  Having recently received the Dempster Prize (named in honor of Professor Arthur P. Dempster) at the May 26th Commencement Celebration for his paper “On the implied weights of linear regression for causal inference” with advisor Jose Zubizarreta, Ambarish spoke with us about what motivated his dissertation work and about other “gems” of his experience in the Statistics Department.  

Stats: Tell us about your dissertation project: How did you select your topic and what questions were you asking?

Ambarish: “For my PhD, I wanted to work on something that bridged the gap between theory and practice.  Very broadly, I was interested in working on causal inference because of how fundamental it is (and relevant to practical applications).  People make causal statements all the time: for example, my headache or my blood pressure was reduced because I took this medicine. In the Department, there has been significant work in the field of causal inference, which is one of the reasons why I wanted to join the program.  In general, I also like how causal inference is very interdisciplinary; many disciplines (e.g. in the medical field and social sciences) are exploring causal relationships.  From a statistical point of view, there were also many interesting connections to statistical topics that I was familiar with, like sample surveys and missing data.”

“Broadly speaking, in my dissertation, I explored how to efficiently learn about causal effects.  There are two main ways to study causal effects: by conducting an experiment or by observing groups.  For example, if you wanted to learn about the effect of medicine on reducing blood pressure, you could conduct an experiment with control and treatment groups, or you could observe groups of people who did or didn’t take the medicine.  In both types of studies, I wanted to know: how do I infer about the causal effects?  A fundamental idea that drove the work was the idea of balance – the idea of creating an apples-to-apples comparison. For example, if I wanted to learn the effect of medicine on Person A’s blood pressure, ideally, I would have a clone of person A, so that one person could be in the treatment group and one person in the control group.  If the responses of Person A and the clone to the medicine are different, then there is a causal effect.  Another key idea was that of representativeness – the groups should not only be similar to each other, but also similar to the population that you want to learn about. Governed by these principles of balance and representativeness, I wanted to develop and analyze methodologies to set up similar groups relative to the population of interest and then compare the results.” 

Stats: What were some of the highlights as a PhD student in the program?  Are there certain people, courses or parts of the program that really shaped your experience? 

Ambarish: “The best part was meeting with so many brilliant people working in several different fields in the Department and learning the importance of collaboration was eye-opening.  I learned so much from conversations with my committee members Carl Morris, Kosuke Imai, and of course, my advisor Jose Zubizarreta; they were instrumental in shaping my research experience. When I joined here, I had a very vague idea about causal inference, but Jose introduced me to the formal research, helping me to understand what is of essence in a research problem.  Most importantly, Jose taught me about how to communicate through writing, redirecting me from writing papers that were similar to lecture notes.” 

Ambarish elaborated on one of his formative experiences in the Department: “We have an annual Departmental retreat where faculty give lightning research talks about what they are interested in and what they are doing.  From watching Jose’s presentation, I knew I was interested in working with him on causal inference.  It’s the same experience with many other students; the retreat, along with other departmental seminars (such as Stat 300), is an opportunity to learn about research in the Department, and it is an environment that encourages you to speak to faculty members.  This is a place where everyone – students and faculty – come together.”

“I must also mention the importance of teaching during my time at Harvard and for my future career.  I was highly impressed with how the Harvard Statistics Department emphasizes teaching and, more generally, emphasizes communicating statistics. Teaching is one of the best ways to communicate ideas. Explaining concepts to a class also helps me to gain understanding; whenever I teach, I learn something new.  In my first year, I took the required course Stat 303, The Art and Practice of Teaching Statistics.  For the course, we had to record several teaching presentations of different durations (5 minutes, 15 minutes, an hour) in front of faculty members and students, which we then reviewed and discussed with the teaching fellows to determine what went right or wrong in the class.  Hearing objective opinions about my teaching was very helpful.”

Stats:  What are your goals moving forward?

Ambarish: “My goal with taking a Postdoc position at Stanford in data science is to continue to work on causal inference problems in a collaborative way.  Causal inference is not just the property of statisticians; other fields like computer science and economics have made significant contributions in answering causal questions.  Because data science is interdisciplinary, this Postdoc position will be a great way to collaborate on developing and applying methodologies to different problems across disciplines.”

Stats: What advice would you give to future or 1st year PhD students?

Ambarish: “Honestly, I don’t think I’m in a position to give advice!  Even though it’s cliché, it’s true that you should do what you love and love what you do – having fun (both inside and outside of research) is important!  Having a few close friends (you don’t need 20 or 50!) to talk to will also make a big difference in your experience.”

“I’ve learned that communicating well about your research is just as important as the research itself.   While you are here over the next 5 or 6 years, I would recommend taking opportunities, like presenting in the Stat 300 Seminar series, to improve your communication.  As I mentioned before, when I write things down, it requires a lot of thinking and makes me understand my work better.”

“Lastly, grad school is hard; setbacks happen to everyone, sometimes progress can be stalled for days or weeks, but don’t get discouraged, and seek help when you need it.  In the past, I’ve been a lone wolf, but whenever I did ask for help, it worked.  There are also a lot of resources in the department and at the university, so if you are experiencing a problem, like a mental health issue, please reach out for help.”

Despite his protest, we think that these are some sage words of advice from Ambarish and have served him well during his time at Harvard.  We wish Ambarish well during his next stage in his career at Stanford – please stay in touch, Ambarish!