One of the simplest issues is related to use of space in a figure, such as too much white space, or too much chart junk.
This one about post-retraction citation is one where the figure is much taller than it needs to be. In the paper itself, it takes up almost an entire page (of A5/ half-page size), when it could be about a third of that size. Additionally, the Y-axis does not start at 0, which is confusing.
In fact, here is a redo attempt using their data:
This one, from a population genetic study on domestication of chickens, could reconsider the usage of space. I would make the 4 photos as a full column on the left so that they are visible at all, make the map larger and clearer, make the PCA plot smaller with larger font since there are only a few points, make the bars shorter for the admixture plot, since there is not very much resolution anyway, and shrink the legend, which takes up a lot of space.
This is an annoyingly common problem in the metagenomics world, since authors want to show hundreds or thousands of bacterial groups at once. After about a dozen or so colors, it becomes difficult to keep track of individual colors. The authors should probably think a bit harder about what they want to say with the figure. A better way of handling this is to make the top 10-or-so factors as colors, and the rest as gray.
Here in Klinges 2019, the pie charts are only part of the problem, since the stripes and polka dots are much worse.
This next one by Busch 2022 turns the data into a circle for no reason, since it would be clearer as a square grid. Many colors are very similar, so again, hard to interpret.
Many species in this figure by Kenny 2020 end up as black, but the labels are so small that they all look like black squares.
Two nested pies are used by Ling 2010, with far too many taxa to ever make sense, and in no particular order. The most common color is hidden among the list that it is hard to identify which taxa is actually the largest green pie slice - it is about a quarter down on the right side. These data potentially could be side-by-side bars comparing the two conditions, and anything under 1% should be merged into an "other" category.
Color schemes should be consistent within a paper, and within a figure. This figure in Lebeer et al 2023 gets most of it right, limiting it to 10 colors, the "other" color being gray/brown... except that they swap the colors for different panels in the same figure! In panel A, orange is one species on the tree (L. pasteurii), then a different species in the species distribution in part C (Prevotella), then another in the pie chart in D (Bifidobacterium).
This can also be a problem without colors. This figure from Amei 2020 seems to have other figures in color in the same article, so it is not banned by the journal, but chose a scheme that is very difficult to visually grasp. There are only 10 categories (9 species and then others), so it actually could be in color.
I do not like circular plots, Circos plots, or similar. They typically have the same problem where it is difficult to compare positions since it requires mental rotation of the plot, a task that humans are bad at. The creators of Circos argue that there are some visual advantages of those plots - beyond "not the least being the fact that it is attractive." However, attractive, but hard to interpret, does not make a good graph.
These are very common in genomics. For bacterial genomes, it at least makes some sense to represent circular genomes as a circle, but this gets widely misused and abused. In this one from Kenny 2020, no scale bars and no description in the caption, so even though the purple and green peaks are identifiable, it is hard to interpret. Also, in that paper, it was not clear that the bacterial genome was complete, so it was not necessarily a circle yet.
Same goes for the Lactobacillus genome by Van De Guchte 2006. Too many similar colors, things are in the caption that should instead be labeled on the graph.
Here I agree with Claus Wilke's advice for pie charts:
- "Clearly visualizes the data as proportions of a whole"
- "Visually emphasizes simple fractions, such as 1/2, 1/3, 1/4"
One can see why these rules are suggested in the case below, where 20 amino acids are given random colors by Excel, and also start with a pie slice that is not relevant to the message of the paper by Schmidt 2020. The pie should start either at 12 o-clock or 3 o-clock with the most relevant slice, or color the slices in some logical way, either to emphasize key amino acids, or to color them by biochemical category as is done in many studies. There is also a lot of white space.
Here are a bunch more where pie charts should have been bars. Pie charts can be effective with the above rules where there is an explicit "whole" and the target is key fractions, but for something that could have arbitrarily been sampled more (i.e. there was no limit to the number of samples, like taking the first 100 or something), it is better to use bars to just show the counts, as was done here for Keeling 2014.
Here are my replots of the same datasets:
Sometimes points are used that should be lines, or lines instead of bars, or bars instead of points. This is generally true above for nearly all pie charts, but can include other chart type.
This one from Sun 2015 uses lines that should be bars, since there is no obvious reason that the species (here, the datapoints) are connected to each other - like they are not sequential, they are categorical. There was an order to the species in question based on a phylogenetic tree, which was arbitrary (can be rotated) and is also not displayed in this figure when it should be.
These got very popular at the height of the genome paper age. They just look like a sea of numbers.
This one from the banana genome even put a banana in the middle, further obscuring the numbers.
A simpler one from the mountain pine beetle genome has only three groups. The blurred edges on the figure look like it was originally rendered as a JPG, then converted to another format for the paper submission, rather than exported as SVG or PNG in the first place.
This one from the Tribolium genome only included overlaps with humans, making it look like they forgot to include the rest of the data.
Even in 2020, they were still used.
Too many lines or points on the same plot can also make graphs hard to interpret.
This one from Officer 1984 tries to show three variables on the same plot, but the lines all look the same and are directly labeled, instead of using separate shapes or lines styles (I'm assuming that colors were not an option then). They instead could have used 3 different shapes.
This one from Bar-on 2018 on global biomass fits all the proportions into a square, giving each of them weird shapes.
Their dataset had a remake as an infographic by an artist on VISUAL CAPITALIST. Here the use of cubes is very effective at visualizing the quantities.
The same group also made a paper for the biomass of mammals, with a similar infographic on VISUAL CAPITALIST.
I can take an additional moment to critique paper titles that are overly generic, in that they say nothing about the results. Vanity journals are especially responsible for this. Maybe we could reword this as don't fear the verb, since most of the first group clearly indicate a topic but say nothing about how that does anything or relates to anything else.
- Genomic anatomy of the hippocampus. (2008, Neuron)
- Low Mid-Proterozoic atmospheric oxygen levels and the delayed rise of animals (2014, Science)
- Egalitarianism in young children. (2014, Nature)
- Mummified seal carcasses in the McMurdo Sound region (1959, Science)
- Oxygen tension of the vaginal surface during sexual stimulation in the human (1978, Fertility and Sterility)
versus some ones that make a statement in the title