Clarifying Hypotheses by Sketching Data

Abstract

Discussions between data analysts and colleagues or clients with no statistical background are difficult, as the analyst often has to teach and explain their statistical and domain knowledge. We investigate work practices of data analysts who collaborate with non-experts, and report findings regarding types of analysis, collaboration and availability of data. Based on these, we have created a tool to enhance collaboration between data analysts and their clients in the initial stages of the analytical process. Sketching time series data allows analysts to discuss expectations for later analysis. We propose function composition rather than freehand sketching, in order to structure the analyst-client conversation by independently expressing expected features in the data. We evaluate the usability of our prototype through two small studies, and report on user feedback for future iterations.