NO BAR CHARTS! (Weissgerber et al. 2015)

I came across this article in PLoS Biology (Weissgerber et al. 2015 doi: 10.1371/journal.pbio.1002128) and it very clearly laid out the issues of using barcharts to display continuous data, especially with low sample sizes. Bottom line: barcharts are misleading summaries of the data, and it’s better to show the data themselves with univariate scatter plots or boxplots. Figure 1 from their paper dramatically shows how barcharts can mask very different distributions of the data:

Figure 1:  Weissgerber et al. (2015).

Figure 1: Weissgerber et al. (2015).

Figure 3 from this paper is also very striking. Here we can easily see the fallacy of using standard error of the mean (SEM), or standard deviation (SD) for non-normally distributed, low sample size data:

Figure 3: Weissgerber et al. (2015)

Figure 3: Weissgerber et al. (2015)

Some people in my lab probably remember me complaining about barcharts for qRT-PCR experiments where n=3 or n=6 on one of our recent papers. I prevailed, and we had scatter plots instead of barcharts. Maybe now my motivations are clearer? 🙂

Figure 7D from our recent paper (Coruh et al. 2015 Plant Cell doi: 10.1105/tpc.15.00228) .. note avoidance of barchart!

Figure 7D from our recent paper (Coruh et al. 2015 Plant Cell doi: 10.1105/tpc.15.00228) .. note avoidance of barchart!

Leave a Reply

Your email address will not be published. Required fields are marked *