STPF Director Jennifer Peal and STPF fellow Justin Smith at the Census Bureau. | Ryan Gallasch/AAAS
This past fall, the U.S. Census Bureau became the newest agency to host a AAAS Science & Technology Policy Fellow (STPF) when it took on a data science researcher. The Bureau, best known for its decadal population census of the US, also produces more than 100 yearly surveys, including business, housing, manufacturers, community and city surveys. It is increasingly using data scientists to help improve its data management and analysis, but it has not always been able to find them.
“One of the things we have discovered is that data scientists are out there, but we may not be able to recruit them,” said Anup Mathur, a senior computer scientist for research computing architecture at the Bureau. “Using the AAAS Science & Technology Policy Fellowships and other programs, we’re able to bring this talent in to help us,” while these scientists get experience working on new challenges, Mathur said. “It’s a symbiotic relationship.”
Data scientists can be hard to find because the field is relatively new, so graduate students who work in this area are “sprinkled across STEM areas,” Mathur said. There are chemists, biologists, physicists and others who have a lot of training in data, and “are really doing some cutting-edge science,” but they aren’t in a traditional computer science department.
The Bureau’s inaugural fellow, Justin Smith, who has a degree in physics, is a case in point. During his doctoral studies, he also developed an interest in science policy, but knew that he still wanted to do technical work. “I realized there was a space for me to apply those skills to governmental problems with policy impact and policy relevancy by shifting to data science,” Smith said. So, he took a physics post-doctoral fellowship that allowed him to focus on machine learning, big data technologies, and becoming a better programmer.
When he found out about the opportunity to work at the newly created Center for Optimization and Data Science at Census, it seemed like a good fit. “My role is still very much a research role,” Smith said – which can be the case in some STPF fellowships. He spends his time “trying to code up solutions to problems, code up data to get it in the format we want, finding out why my algorithm isn’t working, and doing a lot of reading, learning and getting more proficient.”
One problem Smith is tackling is increasing the number of survey responses that can be categorized using only a computer. For example, when Census surveys a business, it asks what type of business it is, but the responses can only be understood by the computers about 80 percent of the time. The rest of the responses need to be read by a person, who then determines the correct category. Smith and Mathur hope to find a way to add in additional data sets to help the system make more correct classifications without human input.
“If we can automate it using machine learning, that’s going to save us a lot of time and money,” Mathur said.
Including Census, there are currently 19 federal agencies that host policy fellows. STPF staff are always in talks with other potential host agencies to help find positions suitable for fellows to connect science with policy, said Olivia Payne, project director of STPF programs.
“We’re following trends across the globe,” said Payne. “There’s an increasing desire for applications of machine learning and other technologies, as well as bolstering the monitoring and evaluation of programs using the wealth of data on hand. We are looking to increase our ability to meet the demand for more statisticians, mathematicians, computer and data scientists.”