The theory and practice of effective scientific figures

CORE Workshop - Ocean Sciences

Alexa Fredston

2024-02-14

What am I looking at, exactly?

I made these slides “programmatically” (in code) using a script type called Quarto in the RStudio IDE.

You can access the slides or the code easily in the GitHub repository, or navigate to github.com, look up the username afredston, and click on the pinned repository entitled “core-figures”.

Why did you do this, Alexa?

Great question. Partly, I wanted an excuse to use Quarto. Mostly, I wanted to show you this because it’s actually extremely useful for visual communication of scientific results, especially at preliminary stages. Quarto:

Is designed to integrate coding and formatting, which allows you to clearly and heavily annotate your figures / tables / results as you generate them. I often make these files to explain complicated methods and preliminary results to myself and collaborators.

Why did you do this, Alexa?

Generates stuff (figures, tables, etc) programmatically: if you update the data / analysis / model / whatever, the stuff will all automatically update. This saves you time if you frequently re-visualize results. If you tell Quarto to generate text results from data (which is easy to do), it can also automatically update in-text stats.

Why did you do this, Alexa?

Allows you to write up results – and potentially entire papers – that are much less prone to error (and wasting time) updating and copy-pasting statistics, tables, and figures.

None of that made sense

Don’t worry about it. The main goal today is to talk about figures! I just wanted to explain how I make mine, and why the slides look the way they do.

Today’s presentation

How to get from data and results to a figure

Some color theory

When to use figures or tables

Examples are (mostly) from my own papers, but this is for two reasons: (1) I wanted to be able to critique them freely and (2) I ran out of time to find other people’s cooler figures

Is this really worth my time?

Lots of people – lots of people – only look at your title, abstract, and figures

It is extremely obvious to reviewers who spent time on their figures and who didn’t

Your papers (and talks, and meetings) will be so much easier for people to follow if there are great figures

Know your message

Figures are like papers: they should really only contain one idea

If you can’t summarize the main point of a figure in a single, relatively brief sentence, it may have too much information

Panels are OK for dividing up results but just be sure they are clearly delineated and annotated

It can take a while to hone in on this message, but you should know what it is by the time you are writing up final results

##Example: one message, one figure

Burgess et al. 2023

Use your imagination

Making and customizing beautiful and creative figures is much easier than it used to be

Don’t be constrained by the default way your personal research workflow generates plots

Example: simple but effective

Freeman et al. 2018

Plot types

Work backward from what you want to communicate

Some key questions to ask yourself:

What quantities do you want to visualize – absolute values, relative values, grouped values, etc.?
What summary statistics (e.g., means, trends) do you want to emphasize? How can you do this while still plotting the raw data?
Do you need to plot your data over space? If so, does a map or an abstraction of space work better?
Do you need to plot your data over time?
How will you visualize uncertainty or variation in your data?

Stay close to the data

Some journals require you to plot raw data

Be very hesitant about omitting outliers

Violin plots are the new box-and-whisker plots

Fredston et al. 2021

How do I match my idea to a figure?

Google is your friend, but you need to know how to describe what you’re looking for

For example, “best plot types for two continuous and one categorical variable”, or “how do I visualize relative change over time for many groups at once”

Try from data to viz

Example: “ridge” plots that show density by group

Example: “ribbon” plots for uncertainty

Exploratory plot by Alexa

Example: “tile” plots that abstract space

Exploratory plot by Alexa

Conceptual figures

I am a huge fan of the conceptual diagram, which can summarize a huge amount of what you did in a simple, appealing format

Some ways to use them:

Describing an experimental procedure
Outlining a model workflow
Summarizing a collaborative process
Characterizing a complicated dataset

Example: conceptual diagram

Maureaud et al. 2024

Example: conceptual diagram

Halpern et al. 2022

Example: conceptual diagram

Brodie et al. 2023

Annotation

Don’t hesitate to draw on and with your plots!

Remember that point size, shape, color, and shading can all communicate for you

You can plot icons from The Noun Project or biorender or silhouettes of taxa from phylopic

I often do this in a software called Inkscape (a little buggy on Macs though), which is also great for resizing figures to meet journal guidelines

Figure captions

Should be as short and straightforward as possible while still explaining everything in the figure

Try to avoid acronyms and jargon. The more intuitively you design your figure, the less you need to explain here

Do not make readers hunt through your manuscript to find explanations of things being plotted that are not in the figure caption!!

If your figure caption is getting incredibly long that is a sign your figure may need editing

Maps: additional considerations

What to plot depends on what you want the viewer to take away – could be bathymetry, country boundaries, major roads or rivers, model outputs… but don’t try to put more than one or two of these on the same map

Choose your projection to align with the areas you want to be most visible

Do not assume people know where your study region is; consider including a zoomed-out map situating it on the continent/globe

Be aware it is extremely difficult to plot uncertainty on a map. People often plot multiple maps side-by-side to get around this

Example: study areas

Fredston et al. 2023

Example: heatmaps (“chloropleths”)

New York Times

Color choices

Use color only when you need color; remember people will often print papers in black and white

You can find color palettes or ramps online and copy over their IDs (look for “hex” or “RGB” codes)

Match colors to your variable types:

Continuous variables –> continuous color gradients
Ordered variables –> discrete color gradients
Categorical variables –> discrete color palettes

Make your plots colorblind-friendly. Somewhere between 5-10% of the general population has some limitation to their color vision. This site has some great palettes for continuous and categorical data that are colorblind-friendly, and you can read a simple explainer here.

Use colorbrewer to explore color choices, especially for maps

Matching scales to variables

From the Datawrapper blog

Example: Barbie plot

Fredston et al. 2023

More on color choices

Be mindful of the interpretation of your colors. For example, red colors are often interpreted as literally hotter, and blues are often interpreted as literally colder; green colors are often used for vegetation or productivity

If you aren’t using a variable that is commonly associated with a color try to visualize it with a neutral color that won’t be confusing

Be consistent across plots and use the same colors for the same variables in all your figures

Example: color choices

Antão et al. 2020

Example: using color over two dimensions

Bivariate color ramp

Interactive plots, gifs, etc.

People increasingly make interactive plots to go with their papers; see Gavin Fay’s example

You can also generate gifs of your figures to embed in talks, which is really helpful for showing (for example) things evolving through time

You didn’t talk about tables

Tables in the main text of a paper can often be replaced by figures that show extra information, like variance in parameter values

However, sometimes tables and text boxes are important for summarizing descriptive information

I use them for reporting lots of p-values and models in supplementary information

Example: a table I begrudgingly made

Fredston et al. 2023

More resources for data visualization

Ten Simple Rules for Better Figures

Journal of Biogeography post, a reminder to always follow the journal guidelines

Betsy Mason article, full of examples of how scientists can make better visualizations

Claus Wilke’s book, a deep dive into data visualization

The Data Visualization Catalogue, for browsing plot options

Data Imaginist, on color theory