Nicholas Horton, Smith College

This past year, I spent time working at the Bureau of Labor Statistics (BLS) as a senior research fellow. The BLS is the principal fact-finding agency for the federal government in the broad field of labor economics and statistics and responsible for the development and publication of data on employment and unemployment, prices and living conditions, compensation and working conditions, productivity measurement, economic growth, and employment projections.

My work was part of a program, coordinated by the American Statistical Association through a grant from the National Science Foundation, that facilitates collaboration between academic scholars and government researchers in survey methodology, statistics, economics, and social sciences. While in residence at the BLS in Washington, DC, fellows are provided the opportunity to address some of the complex methodological problems and analytic issues relevant to BLS programs.

For me, this was a wonderful opportunity to make connections between my biostatistical research interests in missing data methods for epidemiologic studies and real-world applications in large government statistics surveys.

I first heard about the program through an ASA announcement and checked out the website, which included a list of possible research topics. These included statistical quality control, imputation, small domain estimation, time series methods, disclosure avoidance, information dissemination, and data visualization. I then contacted Jean Fox, the BLS coordinator, who put me in touch with Daniell Toth and Polly Phipps from the Office of Survey Methods Research (OSMR). They had been investigating nonresponse bias in the Occupational Employment Statistics (OES) survey.

The OES is a large (more than 100,000 businesses sampled every six months) survey used to determine the number of jobs and hourly wages for more than 800 occupations. While this survey has a 78% response rate, Phipps and Toth found—by using data available from the sampling frame—that nonrespondents tended to pay more than respondents (details are available in their recently published paper in The Annals of Applied Statistics).

My project built on this work in an attempt to help correct for this nonresponse bias by incorporating information from the frame—specifically, information about average wages for all occupations in that establishment reported to the unemployment insurance system. Throughout the year and in consultation with the program directors, assistance of the survey programmers, and conversations and advice from others in the OSMR, we developed a new imputation scheme that accounted for the frame data while allowing for occupation-specific geographic and industry effects. The new imputed values more closely tracked the patterns among the respondents and yielded improved estimates, particularly for subgroups of industries and metropolitan statistical areas.

Being in DC afforded me many opportunities to learn more about the federal statistical system and related work in government statistics. The Federal Computer Assisted Survey Information Collection meeting, which took place at the BLS, provided me with a fast-paced introduction to modern data collection methodologies and metadata, as well as a chance to talk with survey methodologists and statisticians from other agencies. Other talks sponsored by the Washington Statistical Society allowed me to hear more about what is happening in the academic, industry, and government sectors. I also was able to check more of the Smithsonian museum system off my bucket list (but still have more to explore).

The fellowship provided salary and benefits, as well as support for housing, travel to and from Washington, DC, and presenting results from my work at the International Conference on Establishment Surveys (ICES-IV) in Montréal and Joint Statistical Meetings in San Diego. In addition, I’ve spoken about this work at meetings of the ASA’s Boston Chapter and Washington Statistical Society.

Overall, I found my time at the BLS to be a wonderful experience. My colleagues provided guidance in navigating the (sometimes intimidating) federal bureaucracy (e.g., my security clearance and access to data was in place before I arrived) and took the time to introduce me to their projects and collaborators. I left with a sense of pride in our federal statistical system and knowledge of the importance of the work done on these large and complex projects.