Sea stack plots: replacing bar charts with histograms

Published: 29-10-2020| Version 1 | DOI: 10.17632/yf439yp445.1
Alice Stuart,
Maja Ilić,
Benno Simmons,
William Sutherland


Data accompanying Ilić, Stuart, Simmons and Sutherland. In prep. Sea stack plots: replacing bar charts with histograms. To assess the transparency of data visualisation in scientific journal articles, we analysed the plot types and summary statistics shown in figures from the most recent complete issues of Animal Behaviour (Volume 168, October 2020; 22 papers) and Conservation Biology (Volume 34, Issue 5, October 2020; 16 papers). We found that figure types differed significantly between the two journals (𝜒2 = 105.06, df = 16, p < 0.001) and the most commonly used graph overall was the bar graph. For plots showing one-dimensional data from multiple groups for the purpose of comparison, almost all plots showed a group average; most showed error bars; raw data was shown in 17.1% of plots in Animal Behaviour and 26.2% in Conservation Biology; and none showed the distribution of the data as a histogram or density curve. This data shows that the majority of plots in these two journals are not transparent in their representation of the data, and that there is significant room for improvement in this respect.


Steps to reproduce

1. Download PDFs for all articles in Animal Behaviour (Volume 168, October 2020) and Conservation Biology (Volume 34, Issue 5, October 2020) 2. Delete any that are not a journal article (e.g. essays and commentaries) 3. Treating each figure panel as an individual plot, code: a) Plot type out of 'scatter', 'line', 'boxplot', 'sonogram', 'parallel graph' (called slope_graph in coding), 'experimental setup', 'dodged bar', 'dot plot', 'bar', 'accumulation', 'model explanation', 'stacked bar', 'geographic heatmap', 'distribution map', 'notched boxplot', 'venn', 'histogram', 'robustness check' and 'percentage plot'. b) For plots with sample size n ≧ 2 and type bar graph, boxplot, slope graph, or dot plot), record whether the following features were present: average, distribution, raw data, error or range bar. Split error bars were further into the following types: standard error or deviation, confidence interval, undescribed error bar, and quantiles. 3. This coding should reproduce Table S2, the references for journal articles included should reproduce Table S3