Reconfiguration Strategies with Composite Data Physicalizations

Design Futures

Collaborators Lancaster University, University of Bath


There is a lack of understanding of people's strategies and behaviors when directly manipulating physical objects in data physicalisations. In this paper, we systematically characterize different observed reconfiguration strategies informs the design of future manual and dynamic physicalizations.


We asked 20 participants to reorganize exemplars physicalizations with two levels of restriction: changing a single data object versus changing multiple data objects.


We demonstrate how participants use reconfiguration to reorganize physical data objects within an (technologically unconstraint) physicalization. In particular, reconfigurations that affect cluster cohesion and separation could be used to further communicate patterns, facilitate transformations, and inform data interaction with physical data systems.

Composite data physicalizations allow for the physical reconfiguration of data points, creating new opportunities for interaction and engagement. However, there is a lack of understanding of people's strategies and behaviors when directly manipulating physical dataobjects. In this paper, we systematically characterize different reconfiguration strategies using six exemplar physicalizations.

We asked 20 participants to reorganize these exemplars with two levels of restriction: changing a single data object versus changing multiple data objects.

Our findings show that there were two main reconfiguration strategies used: changes in proximity and changes in atomic orientation. We further characterize these using concrete examples of participant actions in relation to the structure of the physicalizations. We contribute an overview of reconfiguration strategies,which informs the design of future manually reconfigurable and dynamic composite physicalizations

In this paper, we investigated people’s spontaneous reconfiguration strategies for the reorganization of exemplar physicalizations (based on the physical bar chart archetype) to make clusters more ‘distinct’. Our findings show that proximity change was generally the most used strategy to reorganize clusters, primarily resulting in increased cohesion and separation of the clusters.

Differences between reconfiguration strategies with or without restrictions

When participants were restricted to changing one object, they would sometimes compromise the visual consistency of the initial cluster structure, to increase cohesion and/or separation, whereas with no restrictions, they choose to maintain the cluster structure. Our study showed that regardless of more or less restrictions, participants were able to reorganize the physicalizations.

Proximity and atomic orientation as novel encodings in 3D space

Our findings are in line with related work from shape-changing interfaces and constructive visualization that showed that proximity was used to differentiate between clusters of data.

These findings enable new opportunities for designing systems whose focus is not solely on linear manipulation. The rotation of data objects could allow for novel interactions with data representa-tions. For example, rotating objects to put more or less emphasis on data points, make distinct dimensions or categories in the data or allow for easier comparisons regardless of perspective. To conclude, future work could further explore how the atomic orientation of data points provides new ways of encoding data in 3D space.

Reconfiguration for data presentation and organization in physical space

Whereas we observed a variety of reconfigurations among participants, the overarching goals were to improve visual consistency and/or reduce or eliminate ambiguity in the physicalization. Expanding our findings beyond data analysis and presentation, the physical reconfiguration of information objects poses an interesting design strategy.

Perhaps, tangible user interfaces (TUIs) could translate input into preferences and actionable results (e.g. smart home control) or be used as a management tool (e.g. with axes being priority, employees, urgency, etc.).

Another use case is allowing cluster algorithms to be tweaked in an exploratory manner, based on ad-hoc and on the spot insights from the physicalization. If a researcher argues a data point is more strongly associated with a particular cluster due to their tacit knowledge — more than the visualization indicates — pushing that data point into the appropriate cluster then provides feedback for that algorithm. Allowing for reconfiguration hereby allows the researcher to add weights to the data points based on (interaction with) the physicalization itself. A third example use case would be to use cohesion and separation changes as transformations to the dataset — informing the actuation of the dynamic composite physicalization itself. Pushing objects closer together could result in the physicalization to adapt its scale (e.g. from linear to logarithmic) to accommodate the changed data point and thereby offer a quick and intuitive adaptation of the visualization without altering the underlying data value.

The reconfiguration strategies can inform the composition of interactive elements in a TUI and provide design ideas for intuitive ways of interacting with these elements.

The future design of static, constructive and dynamic composite data physicalizations

For dynamic composite physicalizations, it could be beneficial to reconsider the current technology implementations and see how actuation plays a role in proximity and atomic orientation changes. In 3D space, data values are not solely communicated in height, but also, for example, by their location, orientation, size, shape, behavior and all this in relation to other data objects and the user(s).

Research in the field of physicalization should better understand what type of interactions and what relation between human intervention and system actuation allow the user to most effectively perform data organizations.


Reconfiguration Strategies with Composite Data Physicalizations

  1. Sauvé K
  2. Verweij D
  3. Alexander J
  4. Houben S

2021CHI Conference on Human Factors in Computing Systems