RESEARCH – June 2013

Epidemiology IV. The frequency and quality of graphs in epidemiology journals

Authors: Annika Voore, Aleksei Baburin, Allan Puur, Mati Rahu

Articles PDF


OBJECTIVE. To determine the frequency and quality of graphs used for presenting study findings in epidemiology journals.
METHODS. A total of 187 original research articles were selected from 2010 issues of American Journal of Epidemiology,European Journal of Epidemiology, and Paediatric and Perinatal Epidemiology. The articles were evaluated to determine the number of graphs and of graph types. The quality of each graph, i.e., visual clarity, numeric distortion, selfexplanatoriness,chartjunk, and some other characteristics, was assessed using the methodical approach by R. J. Cooper et al (2001, 2002, 2003) and D. L. Schriger et al (2006).

Half of articles contained graphs, and there were 184 graphs in total. The five most common types of graph were a line graph (38%), vertical and horizontal bar graph (27%), one-way graph (15%), survival curve and scatter graph (each of 5%). Neither a box-and-whisker graph nor a sector graph was used. Only 10% of the graphs did not include errors. The major errors were: the graph was not self-explanatory (56%), lack of visual clarity (42%), numeric distortion (39%) and over-abundance of data (27%).
Regardless of huge developments in data processing techniques in last decades, and of established clear principles regarding the construction of graphs for research data, poorly drawn graphs are proliferating. We consider it to be highly desirable to include detailed requirements for graphs in all guidelines addressing the presentation of research results in journals, reports, master and doctoral thesis.