Last summer, three graduate students traveled to the National Institute of Statistical Sciences (NISS) in Research Triangle Park, North Carolina, to team up with postdoctoral fellows, researchers from the National Agricultural Statistics Services (NASS), and university professors. They each worked on a different, complex research problem in agricultural statistics as part of an ongoing NISS-NASS partnership. Next summer, they will return to NISS to follow up on the research they conducted. Here, each student tells of his or her internship experience.

Team 1: Multivariate Imputation of Agricultural Resource Management Survey Data for Implementation in Agricultural Resource Management Survey – Phase III
Joshua Habiger, University of South Carolina

Walking up to the NISS-SAMSI building with only a pen, note pad, and Statistical Analysis with Missing Data by Donald Rubin and Roderick Little in hand, I was struck by an ever-so-familiar emotion. It was a mixture of nervousness and excitement fueled by the uncertainty of the upcoming weeks. It was the feeling I felt just seconds before barking the first words of the semester as an instructor. But this time, the uncertainty had a full tank of gas. I had just committed myself to work on a problem called Multivariate Imputation Methods for NASS’ Agricultural Resource Management Survey—a problem I knew little about.

What is NASS’ current imputation method? What kind of software do they use? What would my team members be like? Would I be able to get a publication out of the summer’s work? I was fairly certain I would be more of an office assistant than a problemsolver. After all, I was the lowly graduate student on a team consisting of a postdoctoral fellow, statistics professor, economics professor, mathematical statistician, and economist.

It turns out my role was far from what I expected. It would be Michael Robbins (the postdoc) and I who would tackle the problem full time over the next 10 weeks. Other members of the team would spend several days or weeks at a time at NISS, providing guidance, suggestions, and a plan of action for the upcoming days, but would soon be drawn back to their busy careers. I never felt as if we were put on the back burner, however. Whenever we had questions or concerns, everyone in the group was quick to respond via email, phone, or teleconference. Even NASS employees who were not assigned to the project spent a great deal of time tracking down bits of information and sorting through SAS code to help us out.

Other teams at NISS with their graduate students and postdocs also turned out to be a valuable resource. Postdocs and graduate students from teams 1, 2, and 3 spent many days in the “super secret data room,” mulling through massive NASS data sets and creating graphs and summaries. We shared the challenge of deciphering programming problems, counter-intuitive figures, and other anomalies in the data.

After a quick 10 weeks, the problem was started—not solved. I can’t help but think, “Wow! I can’t believe how much I learned this summer!” I find myself ready to make a dent in an additional field (my dissertation topic is multiple testing, not imputation) and looking forward to next summer. While next summer will mark the end of the project for me, I believe this internship prompted the beginning of several relationships with future collaborators: Michael Robbins (NISS-NASS), Sujit Ghosh (North Carolina State University), Barry Goodwin (North Carolina State University), Darcy Miller (NASS), and Kirk White (ERS).

Team 2: Design and Estimation Methodologies for Estimating the Number of Small Farms from NASS Sampling Frames
Kenneth Lopiano, University of Florida

The opportunity to work as a graduate student with NISS on a project with the USDA and NASS was priceless. As I only recently passed my qualifying exam, this was the first time I researched a problem without a known solution. While working alongside statisticians from NASS, professors from the University of Florida and North Carolina State University, and a postdoctoral fellow at NISS, I started to learn what questions to ask when conducting research.

When the project began, we expected to implement a capture-recapture methodology to reconcile disparate estimates based on two frames. However, once we received the data, it was clear that was not going to work. We had to go back to the drawing board and ask ourselves what alternative approaches we could take.

After our brainstorming sessions, I found myself thinking about ways to improve upon each approach, what the limitations were, and how these limitations could be addressed. Rather than being discouraged by an idea not panning out, Linda Young—a senior faculty adviser—emphasized that research is usually two steps forward and one step back. For me, the take-home message was to always remember to build upon each step forward.

Once the advisers and other NASS team members left me and postdoctoral fellow Patricia Gunning to the problem at hand, I found myself coming up with ideas that took me two steps forward, but then reaching an obstacle that took me one step back. With Young’s advice in mind, I was able to recognize the net gain and persevere. Working as a team, we were able to come up with viable methods to implement.

This summer’s work gave us direction, but there is more to be done. I look forward to thinking about our problem when I return to the University of Florida and coming back to NISS next summer with more ideas.

Team 3: Statistical Multi-Source Predictive Models and Error Estimation in Support of Crop Production Forecasts and Estimates
Criselda Toto, Worcester Polytechnic Institute

I am at the point in my graduate student life at which I am evaluating possible career paths, and my experience as a student intern for NISS-NASS provided me with much insight into what is in store for me. It made me realize that research will be a major part of my career, whatever path I decide to take. This realization gave me the enthusiasm to learn valuable research skills, while the program set-up provided me with the opportunity to develop them.

Our team’s project focused on building a statistical predictive model that mimics the forecasting process for crop production conducted by NASS. This forecasting process involves multiple sources of information, including the opinions of agricultural experts. Our goal was to incorporate the subjectivity in the process into the statistical model, making the process more objective.

As the junior members of the team, postdoctoral colleague Jay Wang and I were given the major task of working on the theoretical and computational aspects of the statistical modeling. Our team mentors—both from academia and government—guided us, but our perseverance and patience were put to test, sometimes leading to frustration and misunderstanding. These trials trained us to think more critically, however, and to be more sensitive to the pros and cons of each step we took. I believe these are valuable lessons, not just in research, but in life.

Learning how statistics is used in forecasting agricultural production also brought a new area of interest to my list. Having been in academia for nearly my whole life, the experience I’ve had doing research on real-life problems is limited. This experience opened a door for possible application in my dissertation. I also made valuable contacts and built relationships with those involved in the project, which is something I deeply appreciate. Overall, being an intern at NISS was a wonderful learning experience. I am grateful to have been part of this research collaboration.