Year: 2017
Type: editorial
Abstract: I hope you all agree that these are exciting times for science. As we pigment cell researchers will soon gather in Denver, USA, for our regular tri-annual Congress, the IPCC, we will learn about stimulating progress in pigment cell physiology and pathophysiology. And as we listen to the talks and visit posters, we will have a chance to look at hundreds if not thousands of slides and posters that all graphically distill, in often colorful ways, what words are hardly able to do as efficiently or in as short a time frame. Oh, yes, a picture is worth a thousand words! Not surprisingly, then, almost all biomedical science publications and presentations contain graphs and pictures. Among the graphs, one form stands out, the bar graph, with its black or colored columns of various widths and heights and usually adorned on top with the characteristic lines and tick marks indicating standard deviations (SDs) or standard errors (SEs), and sometimes even a number giving the actual values! Their simplicity is appealing as they seem to convey in straightforward ways what is on average higher and what is lower and what has not changed at all. But herein lies the problem: The “simple” columns are both overdetermined, as pointed out by Edward Tufte in his “Visual Display of Quantitative Information” (Graphics Press), and they may not adequately reflect the underlying raw results, as discussed by Weissgerber et al. in a series of recent papers. The typical column is overdetermined up to sixfold, namely by the height of its left border, the height of its right border, the position of its top horizontal line, the height of its filled rectangle, and, if displayed, the value's number and its position on the graph. And it may not truthfully reflect the underlying values because many different distributions of these values may give the same average or mean, and even similar SDs or SEs. Missing is the information of whether the values were symmetrically distributed or not, or bimodally or not, whether outliers were included, and whether their numbers (the “ns”) used for comparisons were equal or not. As Weissgerber et al. (PLOS Biology, 1002128, 2015) point out, there are elegant ways to circumvent these problems. For values numbering, say, <50, scatter plots offer a superb solution as they give the reader an immediate view on how the actual data were distributed. For greater numbers, scatter plots may become unwieldy, but box plots (also known as box-and-whisker diagrams) offer excellent alternatives. They make no assumption as to the underlying statistical distribution of the values and usually give minimal and maximal values as well as first and third quartile and the median, but other forms exist as well. Hence, we should now start to discourage the use of simple bar graphs and promote these alternatives and in particular educate the next generation of scientists to adopt them. In fact, some authors of recent papers in PCMR may have received decision letters containing suggestions in this direction, and we are satisfied to see that scatter plots and box plots are slowly replacing “old-style” bar graphs, not only in PCMR but also in many other journals. With this, I wish you all an excellent meeting and lots of joy with all the graphs and figures!
Source: Pigment Cell & Melanoma Research
DOI: 10.1111/pcmr.12602
Author: Heinz Arnheiter
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4234271860