While physical objects have been used to represent information for a long
time, physical visualizations only recently started to attract attention from
the InfoVis and HCI communities. In this article we present our early
experiments in designing physical visualizations for supporting data
analysis. Based on Amar's taxonomy of analytical tasks we show that physical
visualizations can support a number of analytical activities but that further
research is needed to support all activities. Based on our analysis, we
propose promising directions for future research.