At some point, after slaving away, beating back the failures, and tasting the sweet success of a scientific discovery, we must ask ourselves, 'How do I present this to other people?' I'm referring to the visual representation of data.
The current trend in science seems to be that we compile our western blots, cell counts, fMRI scans, or what have you, into one bar graph and top it off with an error bar. You might be nodding your head and recalling your own data being represented in this way. So what's the big deal? Nature, Science, and Cell do not seem to care, right? I recently had a conversation with a friend who is working as a cell biologist investigating Alzheimer's disease about this topic and his opinion was that we should represent the data so that they are easy for the reader to understand. Some people go so far as to suggest that if you want to show a difference, you can zoom in on the data to create the appearance of a larger distance between groups, and to do the opposite to show no difference. Is this something that we should be doing?
What alternatives do we have to the ever present bar graph? Well the answer is that there are actually quite a few options. The bar graph is but one visual tool among many. I would like to focus on the merits of the scatter plot. Simply put, a scatter plot bares your data in its pure, naked, raw form. This can be unnerving for a scientist who opens himself to the possibility of greater scrutiny with a scatter plot compared to a neat and tidy box plot. As my friend rightly noted, if an experiment has a low sample size, a scatter plot looks silly with only measly five dots and that might not be received as well by other researchers compared to the traditional bar graph. However, a scatter plot tells the reader so much more than a bar graph.
This image below is from an article written by Dr. Christina Szalinski. What we see is that the same bar graph (seen on the left) can be made from starkly contrasting data sets. To put it simply, bar graphs make it difficult to see what is happening. Is there an outlier pulling my data away from the mean? Do I have different sample sizes between my control and experimental groups?
This image is from an article written by Dr. Christina Szalinski.
In her article, Dr. Szalinski, quotes Dr. Tracey Weissgerber, an assistant professor in the Division of Nephrology and Hypertension at the Mayo Clinic who said,
"We like bar graphs because they make data look clean and neat and pretty, but data aren’t clean and neat and pretty. Everybody’s data are messy and sometimes that messiness isn’t just noise, it’s critical information."
I still see bar graphs used commonly on scientific posters, at conferences, and in high impact journals. I, for one, will avoid using bar graphs at all costs. After we've done all the hard work and found something worth sharing with the scientific community, let's share it fully. Do not hide your data; show them in their full glory!
The title graphic was created by Leslee Lazar, PhD
Michael Paolillo is currently a doctoral student at the Interfakultäres Institut für Biochemie (IFIB) in the lab of Prof. Dr. Robert Feil.