Graphing small data sets: Should we bother?

• Published in: Behaviour & information technology Vol. 16; no. 2; pp. 61 - 71
• Main Authors: Carswell, C. Melody, Ramzy, Catherine
• Format: Journal Article
• Language: English
• Published: London Taylor & Francis Group 01.01.1997; Taylor & Francis
• Subjects: Content analysis; Exact sciences and technology; Information and communication sciences; Information and document structure and analysis; Information processing and retrieval; Information retrieval systems. Information and document management system; Information science. Documentation; Interfaces. Software; Sciences and techniques of general use; Graphics; Histogram; Line graph; Informational content; Information layout; Comparative study; Information interpretation
• ISSN: 0144-929X; 1362-3001
• DOI: 10.1080/014492997119905

While display designers tend to agree that the communication of large amounts of quantitative information calls for the use of graphs, there is less consensus about whether graphs should be used for small, summarized data sets. In the present study, three groups of 16 subjects viewed 11 sets of time series data presented as tables, bar charts, or line graphs. Data sets varied in size (4, 7, or 13 values) and complexity (number and type of departures from linearity). Subjects provided written interpretations of each of the data sets, and these interpretations were scored for (1) overall number of propositions pertaining to the data set as a whole (global content), (2) number of propositions describing relations within a subset of the data (local content), and (3) number of references to specific data values (numeric content). For the larger (7- and 13-point) data sets, interpretations based on bar charts included the greatest overall global content, but line graph interpretations proved to be most sensitive to the actual information content (complexity) of the data sets. The greater sensitivity of the line graphs was still obtained with four-point data sets; however, this advantage was greater for men than for women. For data sets of all sizes, but especially for the smallest sets, gender differences in interpretation content were obtained. These differences are discussed within the context of more general individual differences presumed to exist in graphreading strategies.