Julia's guide to figures and tables

After years of supervising student theses and grading reports and presentations, I've noticed I give the same feedback on figures and tables over and over again. A few months ago, I posted a blog with presentation tips, including tips on presenting figures and tables. This post goes deeper and shares all the common feedback I give to students about figures and tables in their theses, papers, and presentations collected in one place.

Outline:

The process of making stuff

Most of us learn academic skills by copying others. We only see the end product, not the messy process behind it. So we fumble around, trying everything until we get it right. Learning this way is slow and at times frustrating.

Here’s a common mistake: you jump straight into making the figure, messing with colours, fonts, and legend placement for hours. Or you painstakingly copy results into a table, only to realise while writing the paper that you structured it wrong. The problem is that we start creating the end product without thinking about what we want to convey.

The better approach: before you touch any software, spend 5 minutes writing down what you want to achieve. What key point do you want your figure to convey? What should readers learn from the table? This saves you tons of time editing mediocre figures and tables.

I like to think of it as writing a paragraph-long story of what I want my figure to say. Once you know your story, decisions about what to include (or exclude) become obvious. Bonus: this often helps you rewrite sections of the paper, since figures reflect the key messages from your text.

Figures

This section is an organised list of feedback on figures I give very often. By figures, I mean plots showing your results and diagrams of your methods. When I design figures for results, I aim to have one key conclusion or message per figure.

Font

Make the font size large enough! Believe it or not, but the most common mistake I encounter is that figures are simply not readable. In papers, adjust the font size to match the paper (larger is also OK). In presentations, make it large enough so people in the back of the room can read it – it doesn’t have to be AS big as the slide text font, but not much smaller than that.
Export as PDF: plots exported to JPEG or PNG are converted to pixels, which can make the text fuzzy. Export to PDF to avoid this! You can check if a figure in a paper is exported to PDF by trying to select the text. If you can, it’s a PDF. Otherwise, it’s an image.
I recommend using sans serif fonts because they look less busy. Alternatively, you can match the font of the text (I never do this because I don’t like serif fonts in data visualisations).

Colours

Choose a colour scheme that makes it easy to spot differences between categories. If your paper may be printed in black and white, choose colours that have different values so you can still interpret the figure.
Use a tool like ColorBrewer to choose colourblind-friendly colour schemes. This tool shows maps as examples, but you can use the colour schemes for any chart type.
Bonus: Use the same colour scheme throughout your paper (when appropriate). It looks messy if you switch colour schemes for no reason. A good reason to pick different colour schemes is when you show different types of data, e.g. a categorical dataset and a sequential one.

Choosing charts

Most of us have heard that you shouldn’t use pie charts, but I think bar charts are also abused in machine learning (ML). Bar charts imply counts, so I don’t like it when people use them to display accuracy. The best way to show the performance of ML models is with box plots: they show the distribution, too.
If you do use bar charts, make sure the y-axis starts at zero! It’s fine for line charts to start somewhere else, but not starting at 0 with bar charts can distort the results.

Legend and labels

Make sure the legend doesn’t overlap with any important data points.
Write descriptive legend, axis, and tick labels! Avoid tick labels like Class 0, Class 1 etc. Name them instead (e.g., Cat, Dog). Also, capitalise all labels – it looks neater.

Tables

Tables are more precise than figures, but they’re harder to interpret. An advantage of tables is that you can copy the results, which is handy when you’re developing methods on the same dataset.

In LaTeX, use the booktabs package. Don’t use horizontal lines between the rows; only use them at the top, under column headers, and at the bottom. You don’t need vertical dividers between columns either. However, you can use additional dividers to separate groups (e.g., datasets or subsets of models)
Think about how to organise your table. It’s common to make one table per dataset. It’s fine to use the same table for multiple results as long as it doesn’t get confusing.
Design a meaningful highlighting scheme to make tables easier to read. Bold is often used to highlight significantly best results, and underline for second best. You can also use colours like blue and red.

Presenting or discussing data visualisations and tables

When you present or discuss figures and tables, you should always explain the key message. Don’t just read the figure or table out loud. Instead, tell the audience what they should take away from it.

In papers

Figures and tables should always have a caption.
Captions usually go on top of tables, but below figures.
Captions should have the following structure: a sentence on what the figure shows (what data are you showing?), an explanation of the components (describe the legend), highlighted important patterns (what do you want the reader to see?), and a conclusion (what do these patterns mean? How do they relate to the rest of your paper?).
Make sure you refer to every figure/table in the text. Ideally, figures and tables should appear in the same order you refer to them, so readers encounter Figure 1, then Figure 2, etc.

In presentations

Don’t put table/figure captions in presentations!
Scientific figures/tables tend to be complex. This is fine in papers when readers have more time to parse the figure. In presentations, audiences have less time, so highlight important information by using arrows, boxes, colours, or hiding parts irrelevant parts.
Don’t copy-paste tables and figures from your paper into your presentation – make simplified versions with larger font sizes so they’re legible for people in the back.

Think about your audience

And a note on highlighting results

I’m a bit of a data visualisation nerd. I buy books on the topic and follow people on LinkedIn. If you are too, you’ve probably seen advice like this:

The idea is to highlight key results while pushing the rest to the background. This is great data storytelling and good practice for business visualisation and presentations, but it feels inappropriate in science. I don’t think there are actual rules against this, but I would never present a figure this way in academic papers. Papers try to be as objective as possible. Your goal is to present the facts as you found them so readers can make their own judgments. In business analytics, audiences don’t want to make up their own minds – they want concrete recommendations.

Academic figures should still guide the audience to specific conclusions without sacrificing objectivity. Academic readers may feel sceptical or distrust you if you try to ‘hide’ the results that don’t support your point too much. You need to strike a balance between a clear story and respecting the reader’s intelligence. Leave room open for debate of your results.

Even within academia, there are different audiences: fields have their own expectations, and short workshop papers present results differently than extensive journal publications. Either way, the foundation remains the same: decide what story you want to tell, then build figures and tables that tell it clearly.

That’s it! Let me know in the comments below if you think any advice is missing.