Spotlight: Beatrix Haddock, Recipient of the Best Poster Award of the CSDE Winter 2020 Trainees’ Lightning Talks
Posted: 3/21/2020 (CSDE Research and CSDE Seminar Series)
The US Census Bureau is required by the privacy law in the US Code to not reveal information about individuals, households, or businesses directly or indirectly through statistics. However, advancements in technology have made traditional methods to protect the identification of individuals and database reconstruction insufficient. Solving a huge system of linear equations and inequalities to reconstruct data is theoretically feasible with today’s computing power. The Census Bureau plans to use a new Differentially Private (DP) algorithm called TopDown to protect privacy in the Decennial Census and has asked for feedback from researchers, policy makers, and communities who use or affected by this data. Beatrix Haddock, data specialist at the Institute for Health Metrics and Evaluation (IHME) and part-time student at UW, presented research that addressed this call at the CSDE Winter 2020 Trainees’ Lightning Talks and Poster session. Her poster “Differential Privacy in the 2020 Census: Considering Acceptable and Unacceptable Biases” was one of two posters to receive the best poster award.
An algorithm is DP if, run over two databases that differ on one individual, analysis of the two databases would result in the same conclusions. The Census bureau is still conducting research to decide on various tuning parameters of the DP algorithm, how much noise to introduce with TopDown, and which counts to hold accurate. The effects of different parameter choices are not well understood; however, differential accuracy and biases in the counts across territorial divisions and units of local governments have high stakes consequences for the allotment of congressional seats and vast sums of funding for infrastructure and social programs. The counts are also the basis for denominators in vital statistics data and measures used in demographic, health and social research.
Beatrix and her collaborator, CSDE Affiliate Abraham Flaxman, Associate Professor Health Metrics Sciences and of Global Health, are examining the biases as a foundation for research into methods to mitigate them. Their work follows from a finding by Professor Randall Akee of UCLA that there is a systematic downward bias in counts of American Indian and Alaska Native (AIAN) populations living on reservations. Beatrix’s poster demonstrated the same level of undercount as Akee using example data released by the Census bureau to demonstrate TopDown. The poster also found systematic upward bias in counts of the Asian alone population living in rural blocks at the county level, and a downward bias for the Asian alone population urban blocks. The poster used simulation to examine a potential mechanism driving these biases, hypothesizing that they are due to (1) the particular distribution of Asian alone (or AIAN alone) counts across different counties (or reservations), and (2) the rural vs urban and on- vs off-reservation geographies not being part of the standard geographical hierarchy recognized by TopDown.
Beatrix is a member of the Simulation Science team at IHME and will also work with Professor Flaxman on COVID-19, using census data to map nursing homes across the US. She has coauthored a paper, “Hyperbolicity of links in thickened surfaces” that was recently published in Typology and its Applications. She plans to apply to PhD programs in Applied Math this fall.
CSDE congratulates Beatrix on her accomplishments and wishes her well on her next steps!