This post not only generated a lot of online reactions (>200 retweets already – for our standards that’s close to breaking the internet!), but also a lot of discussion among colleagues in the lab. And indeed, while plotting might seem like something you just to in addition to the actual analysis, doing it the right way is arguably as important as analyzing the right way since, after all, the figures are the things most of us look at when we try to understand the results of an article.
The gist of Page’s post is that bar plots do not allow you to see differences that might be very important for your data, since they fail to provide distributional information (also, in how many cases does it make sense that they start at 0, or whatever the starting point of your y axis is?)
The discussion amongst our labmates generated comments such as “But bar plots CAN be useful, for instance for count data” – fair enough – or “Well bar plots are evil but I also don’t like box plots since they assume a normal distribution – let’s use violin plots or pirate plots instead” – equally fair, plus who wouldn’t prefer a violin or a pirate over a bar or a box?
The retweets included some from fame locations like Journal of Neuroscience, who mentioned that they are in the process of changing their submission rules, or from dear but far away colleagues who requested a T-Shirt with the graphic (and who also write useful posts about a similar problem in the representation of ERP graphs).
The combination of these two reactions led us to the idea of the following project: We want to send a beautiful histogram-box plot-bar plot T-Shirt to the editors of major journals to spread the word and raise awareness. In addition, if they would like to, we will ask them to tweet a selfie (#barbarplots) with said T-Shirt to support the campaign.
Like the idea? Want a T-Shirt (or at least a sticker)? Please go this way to our Kickstarter campaign!