Don't Use Bar Graphs to Represent Data

Many different datasets can lead to the same bar graph.
A recent criticism, published in PLoS Biology, has called into question the use of bar and line graphs to represent continuous data. The authors list 3 flaws inherent to bar and line graphs.

Bar graphs hide the underlying nuances of continuous data, summarizing the differences. Things like distribution patters and sample sizes are not represented by bar graphs. Very different datasets can look exactly the same as demonstrated on image to the right.

When used to represent paired data, bar graphs can make the datapoints seem independent from each other, despite the possibility that changes are consistent between different points.

These graphs also hide outliers, which in studies with small sample sizes can give a incomplete picture to the observer.

Researchers looked at more than 600 papers published in the year 2014 and found that continuous data was represented by bar graphs in 85.6% of the cases, the majority of them having relatively small sample sizes. The authors suggest that this type of data should be represented by scatterplots, box plots or histograms, which actually represent data distribution.

Share on Google Plus