I spent three months of my PhD chasing a result that I thought was revolutionary, only to realize during my defense that the committee hadn’t actually seen my data—they had just seen a cluttered, over-engineered mess of a plot. I had fallen into the trap of thinking that more color, more labels, and more “sophisticated” styling would somehow mask a weak signal. It’s a common mistake in research: we treat designing readable figures as a decorative afterthought, a way to make our work look “professional,” when in reality, a figure is a functional component of a distributed system. If the mechanism of your data isn’t immediately visible, the entire argument collapses under the weight of its own visual noise.

I’m not here to teach you how to use Photoshop or which trendy color palettes are currently popular on social media. Instead, I want to talk about the actual mechanics of information transfer. I’m going to share what I’ve learned from years of seeing brilliant research die in the margins of poorly constructed charts. We will focus on how to strip away the cognitive load so that your reader can focus on the actual logic of your system, rather than squinting at a legend to figure out what a dashed line represents.

Improving Visual Hierarchy in Plots via Cognitive Load

When I look at a cluttered plot, my brain immediately tries to figure out what the “hero” of the image is. If everything—the axes, the gridlines, the data points, and the legend—is competing for the same level of visual prominence, you haven’t actually designed a figure; you’ve just dumped data onto a canvas. This is where most people fail at improving visual hierarchy in plots. To fix this, you have to treat your ink like a finite resource. The data should be the most visually “heavy” element, while the structural scaffolding, like tick marks and frame lines, should recede into the background. I often find that thinning out the weight of my axes or even removing the outer box entirely helps the reader focus on the actual signal rather than the container.

The goal is to manage the reader’s cognitive load so they don’t burn out before they reach your conclusion. This involves more than just aesthetics; it’s about effective chart design principles that respect how humans process information. If you use a bright, saturated color for a secondary trend line, you are lying to the reader’s eyes by suggesting it is more important than your primary finding. I prefer using muted tones for context and reserving high-contrast colors only for the specific mechanism I am trying to demonstrate.

Effective Chart Design Principles Beyond Surface Aesthetics

When we talk about effective chart design principles, we often get distracted by the “look” of a plot—the font choice or the hex codes—while ignoring the actual signal-to-noise ratio. A common mistake I see in recent preprints is the inclusion of excessive “chartjunk,” like heavy gridlines or 3D effects, which serve no purpose other than to compete with the data for the reader’s attention. If a line or a bar doesn’t represent a specific measurement, it shouldn’t be there. You want to minimize the non-data ink so that the reader’s brain can skip the decoration and go straight to the trend.

This extends directly into scientific figure accessibility. It isn’t just about making things “pretty” for a conference presentation; it’s about ensuring the mechanism of the data is legible under sub-optimal conditions. For instance, if you rely solely on color to distinguish between three different experimental groups, your figure is fundamentally broken for anyone viewing it in grayscale or with color vision deficiency. I always suggest layering redundant encodings—using different line styles (dashed, dotted) or marker shapes alongside color—to ensure the information survives the medium.

Five Practical Constraints for High-Fidelity Figures

• Stop using color as your only way to distinguish data series. If you’re plotting three different experimental runs, use different line styles—dashes, dots, or solid lines—in addition to color. If someone prints your paper on a monochrome laser printer, a color-only legend is just a pile of indistinguishable gray lines.
• Label your axes directly instead of relying on a distant legend. My eyes shouldn’t have to travel back and forth between a line in the middle of a plot and a tiny box in the corner to figure out what a single curve represents. When you place the label near the data, you reduce the cognitive overhead of decoding the mechanism.
• Respect the scale of your error bars. It is tempting to use tiny error bars to make your results look “clean,” but if your error bars are smaller than the stroke width of the line itself, you aren’t communicating uncertainty; you’re hiding it. If the variance is that small, consider a different way to represent the distribution, like a shaded confidence interval.
• Audit your gridlines for “chart junk.” Gridlines should be a scaffold, not the architecture. If your gridlines are as dark as your data lines, they compete for the viewer’s attention. I usually set them to a very light gray or remove them entirely if the axis ticks provide enough context. The data must always be the highest-contrast element on the page.
• Avoid the “magic” of non-linear scaling without explicit justification. If you use a log scale to compress a massive range of values, you must make sure the viewer realizes the visual distance between 10 and 100 is the same as 100 to 1000. If the physical distance on the paper is misleading, you aren’t visualizing data; you’re performing a sleight of hand.

Summary: Mechanisms over Aesthetics

Stop treating figure design as a layer of “polish” applied at the end; clarity is a functional requirement of the data representation itself, much like how a distributed system requires low latency to be useful.

Always design for the failure modes of human perception—if your figure relies on subtle color shifts or tiny labels to convey its primary mechanism, it will fail the moment it is printed in grayscale or viewed on a low-resolution screen.

Prioritize the signal-to-noise ratio by aggressively removing non-data ink, but do so with the understanding that every removed element must be replaced by a clearer structural hierarchy, not just left as a void.

The Mechanism of Clarity

At the end of the day, designing a figure isn’t about making it “look professional”—it’s about reducing the friction between your data and the reader’s brain. We’ve discussed how managing cognitive load through visual hierarchy and choosing meaningful design principles are more than just aesthetic choices; they are technical decisions. If you fail to establish a clear path for the eye to follow, or if your color choices make the data unreadable in grayscale, you haven’t just made a “bad graph.” You have actively obscured the mechanism you spent months researching. A figure should never be a puzzle that the reader has to solve; it should be a direct conduit to the insight.

As you move back to your plotting libraries and LaTeX editors, I want you to resist the urge to rely on default settings. Defaults are designed for the average case, but research is rarely average. Take the time to interrogate every axis label, every legend, and every line weight. When you treat your visualizations with the same rigorous scrutiny that you apply to your proofs or your code, you stop being a mere reporter of results and start being a translator of truth. Build figures that don’t just show what happened, but explain why it matters.

About Dr. Ingrid Falk-Weller

I write for the person who wants to understand the mechanism, not memorise the conclusion. If a claim has a caveat, the caveat goes in the paragraph, not a footnote.