Causation and correlation are two concepts that are essential to understanding modern science. They are, however, difficult to interpret. For example, people may believe that the mumps, measles, and rubella (MMR) vaccine causes autism (even if these two variables increase in different years). On the other hand, the mantra that correlation does not imply causation may conceal how correlation can be a precious hint to causation (for example, the correlation between smoking and lung cancer).

This project attempts to mitigate these challenges by providing guidelines for designing interactive installations for informal learning places like museums. We hypothesized that different ways of interacting with a data visualization could affect how we make sense of causation and correlation in data. To test this hypothesis, we iteratively designed and implemented three prototypes of an interactive museum installation. In the full-body prototype, visitors interacted with a geo-referenced data visualization using hand gestures and body movements (like jumping or grabbing). In the tangible prototype, they used paper-made objects or a 3D-printed globe. In the traditional prototype, visitors used a gamepad to control the installation.

We tested our prototypes at Discovery Place, a science museum in Charlotte, NC. We placed a 65" TV in a busy area of the museum and allowed visitors to interact with the installation freely. During our initial study sessions, we also identified interactivity problems that affect data visualizations in museums. In particular, screens in museums often compete with surrounding stimuli (e.g., people, exhibits, signage, objects). This limits the number of people who notice them or who understand that the display is interactive (two phenomena known in the pervasive display literature as "display blindness" and "interaction blindness"). To address these issues, we investigated how different ways of representing the user next-to a data visualization (as a skeleton, avatar, or by using the full camera feed) impact people's engagement with the installation. For example, in a study with 731 museum visitors, we found that by representing the user as an avatar or using the full video feed from the camera, the screen could attract more people than using a skeleton.

Because of COVID-19 related closures and restrictions, we continued our studies in-lab at IUPUI in Indianapolis, IN. Consistently with our original hypothesis, our experimental findings indicate that different interaction styles alter how we make sense of causation and correlation in data. In particular, our experimental sessions showed that people report different levels of agreement with statements that portray correlation and causation across the data on display, depending on the prototype they used. Participants tended to agree less with such statements after interacting with a gamepad. Vice versa, interacting with body movements may promote the detection of causal patterns.

This grant allowed a diverse pool of undergraduate and graduate students to participate in research activities actively. It enabled the research team to produce peer-reviewed publications that were presented at highly competitive international conferences in the field of Human-Computer Interaction and that were shared with professionals at the partner museum. 

Last Modified: 06/28/2022
Modified by: Francesco Cafaro