Data visualization for reports and papers
A figure that takes a reader ten seconds to understand does more work than three paragraphs explaining the same result. But a badly made figure can do the opposite — confuse a reader, misrepresent your data, or simply get ignored. The gap between a good figure and a bad one is usually a handful of deliberate choices, not raw design talent.
Choosing the right chart type
Start from what you're actually trying to show, not from whichever chart type happens to be familiar or looks impressive:
- Comparing categories — bar charts.
- Showing change over time — line charts.
- Showing the relationship between two numeric variables — scatter plots.
- Showing the distribution or spread of a variable — histograms or box plots.
- Showing composition/parts of a whole — used sparingly, and often better replaced by a simple bar chart, which is generally easier for readers to compare accurately than a pie chart.
A useful habit borrowed from data science: distinguish between exploratory visualisation (quick charts you make for yourself, to understand your own data) and explanatory visualisation (polished figures made to communicate a specific finding to someone else). The bar for polish and clarity is much higher for the second kind — that's what actually goes in your report.
Principles that make a figure genuinely readable
- Label everything. Every axis needs a label with units; every figure needs a clear, specific title or caption that could stand on its own if separated from the surrounding text.
- Don't distort the data. Starting a bar chart's y-axis somewhere other than zero can make small differences look dramatic and misleading — a genuine ethical issue in data presentation, not just a style preference.
- Cut the clutter. Unnecessary gridlines, 3D effects, decorative backgrounds, and excessive colours (sometimes called "chartjunk") make a figure harder to read, not more impressive. Every element should earn its place.
- Use colour intentionally, and check it's colourblind-friendly. Avoid relying on red-green distinctions alone; tools like ColorBrewer provide palettes designed to remain distinguishable for colourblind readers.
- One clear message per figure. If a figure is trying to show five different things at once, readers will likely walk away with none of them. Split it into multiple simpler figures if needed.
- Check what your target format requires — a report or thesis, a printed document, or a specific journal template may have resolution, font-size, or colour requirements worth checking before you finalise figures, not after.
Why is this important?
A reader (or marker) skims a document's figures before deciding how carefully to read the surrounding text — a strong set of figures can carry a lot of the persuasive weight of your results section on its own. Poor figures, on the other hand, can make even a genuinely strong result look weaker or more confusing than it actually is.
Tips
- Write the caption as if the figure might be seen entirely on its own, disconnected from the surrounding paragraph — a good caption explains what's being shown and what the reader should take away from it.
- Show individual data points where feasible (jittered scatter over a bar chart, for example), particularly with small sample sizes — it's more honest about the underlying variation than a single averaged bar.
- Get a second opinion on your key figures before finalising a report — someone unfamiliar with the data is a good test of whether the chart actually communicates what you think it does.
- For code-generated figures, tools like Python's Matplotlib/Seaborn or R's ggplot2 give you fine control over all of the above and are worth learning if you'll be making figures regularly.
References
Rougier, N. P., Droettboom, M., & Bourne, P. E. (2014). Ten Simple Rules for Better Figures. PLOS Computational Biology — a short, widely cited, practical guide to making clearer scientific figures.