Decision theory informs us more about what to show than how to show (or how people interpret vis in practice), so I think theory building efforts need to synthesize it with theories of perception, cognition, and social/cultural dynamics.

My suggestion about using decision theory as a deductive framework during design studies is one path toward codifying what we know about decision context in ways that may be useful for making testable predictions about when a particular vis design is likely to be effective.

Thanks Carlos, I didn't realize Gordon was on here.

Your work on AVD was a huge inspiration for this paper. I'd be curious to hear your thoughts on the connects we draw between disclosure and AVD vulnerabilities.

Ultimately, this paper is about the epistemology of data communication. I close with reflections on how to develop a more robust logic of generalization about decision support, and how social norms demanding that research yield guidelines that transcend context are an obstacle.

This work demonstrates the conceptual value of decision theory in visualization research and practice, and provides a primer for unfamiliar readers. My hope is that this will increase the adoption of ideas from decision theory in the visualization community.

I draw on decision theory to analyze the heterogeneity in decision problems that are studied by visualization research. By comparing examples, I call attention to the dimensions on which decision problems vary and when we can(not) expect results to transfer to different settings.

The dominant logic of generalization in visualization research (often implicitly) holds a strong assumption that efficacy of decoding is relatively context invariant, however, much modern research problematizes this assumption. Where does this leave us?

In visualization research and practice, we care a lot about supporting decision-making. However, it's not always clear how findings from empirical research generalizes to various decision contexts.

This ambitious conceptual/theoretical work is Krisha's first paper and my first PhD-student led paper at UChicago. It's a big milestone for both of us, and I think she's done a wonderful job! Check out the paper and the talk at #ieeevis for more. arxiv.org/abs/2508.08383

I also use that data set, but I ask students to encode all the variables in the table, which rules out a CDF because it doesn’t show the bacteria names. Unless they add the names with annotations. That could be cool. Avoiding occlusion and crowding would be a challenge.

Such a disappointment. It’s illustrative of a deep political problem on the left, a sort of allergy to disagreement or information that goes against one’s priorities. This just isn’t a workable approach in a democracy, especially not for a “big tent” party.

My take is that the “pause” rhetoric is bullshit. They want to defund the universities, full stop. Idk why people are in denial that this is the game plan, to eviscerate centers of power on the left starting with educational institutions and government agencies staffed by lefty technocrats.

This is what I thought you were saying, that true marginal density estimation requires different samples, but I wasn’t completely sure. Thanks!

I think I’m guilty of misusing Monte Carlo integration in this way. Not 100% sure what to do differently, but I appreciate your explanation of the issue.

I would like to know more about this too because it seems like this is how most people use posterior samples in practice.