The effects of knowledge of external representations and display selection upon database query performance.

This study investigated the representation selection and reasoning behaviour of participants who were offered a choice of informationally equivalent data representations (tables, bar charts, scatter plots, sector graphs, pie charts, lists - {\\\\it i.e.} matrix graphics). A prototype automatic information visualisation system (AIVE) was used to present a series of questions about the information in a database. The database contained car information (manufacturer, model, purchase price, CO2 emission, engine size, horsepower, {\\\\it etc\\\\/}). The database queries posed to participants varied in terms of task demand ({\\\\it e.g.\\\\/} {\\\\it identify\\\\/} a car with a particular attribute, {\\\\it rank\\\\/} cars in terms of some feature, {\\\\it compare\\\\/} cars on dimension(s)). Participants read a question then chose, via an array of icons, the representation that they thought would best assist them with answering the query. AIVE generated the chosen representation and instantiated it with the appropriate data. Response accuracy and timing data were recorded. Results indicated that some tasks are more `representation specific' than others. The ratio of correct to incorrect responses also varied widely across representation/task combinations. Participants were observed to make surprisingly quick representation selection decisions. Participants were divided into two groups on the basis of their prior ER knowledge level (assessed via a card-sort pre-task). The groups differed most in their use of scatterplots, bar charts and sector graphs. On the whole participants in the high ER knowledge group tended to be more graphically heterogeneous in their representational behaviour than those in the low ER knowledge group. Results are discussed in relation to work on representation schemas (\\\\cite{Novick01}) - the suggestion is that participants in the high ER knowledge group possessed more developed representation schemas, including better knowledge of representation applicability conditions than participants with less well-structured ER knowledge.