Communicating population health statistics through graphs: a randomised controlled trial of graph design interventions
- Equal contributors
1 Centre for Epidemiology and Research, New South Wales Department of Health, 73 Miller Street, North Sydney NSW 2060, Australia
2 Hunter Valley Research Foundation, 55 Downie Street, Maryville NSW 2293, Australia
BMC Medicine 2006, 4:33 doi:10.1186/1741-7015-4-33Published: 20 December 2006
Australian epidemiologists have recognised that lay readers have difficulty understanding statistical graphs in reports on population health. This study aimed to provide evidence for graph design improvements that increase comprehension by non-experts.
This was a double-blind, randomised, controlled trial of graph-design interventions, conducted as a postal survey. Control and intervention participants were randomly selected from telephone directories of health system employees. Eligible participants were on duty at the listed location during the study period. Controls received a booklet of 12 graphs from original publications, and intervention participants received a booklet of the same graphs with design modifications. A questionnaire with 39 interpretation tasks was included with the booklet. Interventions were assessed using the ratio of the prevalence of correct responses given by the intervention group to those given by the control group for each task.
The response rate from 543 eligible participants (261 intervention and 282 control) was 67%. The prevalence of correct answers in the control group ranged from 13% for a task requiring knowledge of an acronym to 97% for a task identifying the largest category in a pie chart. Interventions producing the greatest improvement in comprehension were: changing a pie chart to a bar graph (3.6-fold increase in correct point reading), changing the y axis of a graph so that the upward direction represented an increase (2.9-fold increase in correct judgement of trend direction), a footnote to explain an acronym (2.5-fold increase in knowledge of the acronym), and matching the y axis range of two adjacent graphs (two-fold increase in correct comparison of the relative difference in prevalence between two population subgroups).
Profound population health messages can be lost through use of overly technical language and unfamiliar statistical measures. In our study, most participants did not understand age standardisation and confidence intervals. Inventive approaches are required to address this problem.