Alyssa Columbus
Biostatistics PhD Candidate, Johns Hopkins
Vivien Thomas Scholar · NASA Datanaut
How do we make quantitative evidence trustworthy enough to act on? This question runs through everything I do, from the analytic choices behind a result, to the security of the systems that produce it, to the way evidence gets taught and turned into decisions.
Alyssa Columbus is a biostatistician drawn to a question that runs under much of modern science: how do we make quantitative evidence trustworthy enough to act on? She works on this from several directions at once, from the analytic choices behind a result, to the security of the systems that produce it, to the way findings are taught and translated into decisions. She is a Fulbright Visiting Researcher at the Ludwig Maximilian University of Munich (LMU Munich) and a Biostatistics PhD candidate at the Johns Hopkins Bloomberg School of Public Health, where she holds a Vivien Thomas Scholarship and a Center of Excellence in Regulatory Science and Innovation (CERSI) Scholarship.
As principal investigator on her Fulbright award, she leads an independent line of work on researcher degrees of freedom and trustworthy data science, spanning the security of artificial intelligence (AI) systems used in health and science and the teaching of statistics and data science. She holds a Bachelor of Science in Mathematics from the University of California, Irvine and a Master of Science in Applied and Computational Mathematics from Johns Hopkins, and she is a member of the National Aeronautics and Space Administration (NASA) Datanaut corps.
Her research is cited in scientific and policy reports from bodies including the World Health Organization, the Council of the European Union, and the Organisation for Economic Co-operation and Development, and she reviews software and grants for open-science journals and international funders. She writes and speaks for readers from policymakers to schoolchildren, builds and maintains software the field depends on, and mentors the next generation of statisticians and data scientists.
One Question, Many Threads
Researcher Degrees of Freedom
How much does a result depend on the choices an analyst makes along the way? I map that space of choices, measure how far conclusions shift across it, and study when an analysis can be called finished.
Making Uncertainty Usable
A finding is only as useful as the uncertainty reported with it. I work on clear ways to communicate this uncertainty, so it holds up in science, in regulation, and in public decisions.
Securing the Systems That Now Run the Analysis
More of the analytic pipeline now runs on AI systems. I work on how to secure autonomously acting language models and the data they draw on, from prompt injection to data poisoning.
Teaching Judgment, and Turning Evidence into Policy
Teaching people to weigh evidence is the same problem as my methods work, seen from the classroom. It matters more now that AI sits beside everyone doing the analysis.
Software People Rely On
Maintained R packages, software peer review, and seven years in industry data science and security before the doctorate, so the methods can travel beyond the paper they started in.