Jessica Hullman
Northwestern University, IL·Computer Science
Any statistical analysis rests on judgment calls: which data points count as outliers, which variables to include, what shape to assume for a relationship. An analyst studying how wealth relates to age might drop implausible ages and fit a straight line, which assumes each extra year adds a fixed amount of wealth; a different set of equally defensible choices could support a different conclusion.
- Fields
Computer Science
Mathematics & Statistics
Social Science
- Topics
data visualization
multiverse analysis
uncertainty visualization
- Grants
- 1 known grant, through Sep 2027
What they work on
- Conduct formative studies with practicing analysts to characterize how they currently reason about alternative models and analysis steps, and where existing tools fail to support defining or comparing those alternatives.
- Design user interfaces and programming libraries that elicit analysts' assumptions and knowledge so that alternative analysis specifications can be expressed directly in visual analysis software and computational notebooks.
- Develop visual representations and interaction techniques that let an analyst compare results across many alternative models or analysis paths, conveying both within-analysis uncertainty and uncertainty from analysis choices.
- Build software infrastructure that makes running and evaluating multiverse analyses feasible within existing data analysis workflows, and release the tools publicly.
- Test empirically how different representations of a set of plausible analyses affect readers' understanding, and create shareable interactive multiverse reports combining code, figures and narrative.
More about the research
One honest response is a "multiverse analysis": run and report many reasonable versions of the analysis instead of one. But the software analysts actually use offers almost no help for this. It is awkward to specify alternative choices, hard to compare many models and charts side by side, and there is little guidance on how to show both the uncertainty inside one analysis and the uncertainty created by the choices themselves.
Why it matters
Conclusions drawn from data often hinge on choices that never appear in the final report, which contributes to findings that do not replicate. Tools that let analysts define, compare and publish many defensible analyses, and empirical evidence about which displays readers actually understand, could make results in science, journalism and policy analysis more transparent about how much they depend on assumptions.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Michael Gleicher
University of Wisconsin-Madison, WI·Computer Science
Determine how people interpret small subsets of data items to grasp an overview of a large collection, using empirical user studies that link subset design choices to task performance.
Alexander Endert
Georgia Institute of Technology, GA·Computer Science
Build an open-source software testbed with modular AI guidance components for chart creation and visual analysis of tabular data.
Khairi Reda
University of Illinois Chicago, IL·Computer Science·Early-career
Extend image-computable models of human texture perception to chart and map images, predicting which data properties the early-to-mid visual cortex retains for the viewer.
The school
Northwestern University: 71 labs on PI Pages, $1.3B a year on research, 536 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.