Hints for Good Figures (how to do well on Design Exercises)
Here is some advice on making good “figures.” It is targeted towards design exercises, but really, they are advice on how to make good figures generally.
These ideas apply more to places where we are asking you to “tell stories” from data. This starts with DE04. If you’re putting a figure in a paper or a slide, you are telling a story as well.
The first “golden rule”: be intentional. “Task First”. A figure should have a point. There should be something it is designed to do well. What do you want the viewer to see? Your choices should be in support of the goal. If you don’t have a goal, it’s hard to make a good visualization that achieves the goal.
After that, here’s some other things, in no particular order. These are generated from the experience of grading previous years’ Design Exercises. Some of the details may not apply to all exercises, but I’ve tried to make them general enough to apply to this year’s exercises.
Many of the things that contribute to a visualization haven’t been discussed yet. For example, you will need to pick good colors before we learn about color theory. Use the intuitions that you have. Part of the idea of the early assignments is to help you appreciate what you will want to learn.
Be clear what the figure is. Have a good title, captions, and labels. (Some of the early assignments don’t ask for captions). These are important. The viewer/grader should not need to go looking for the explanation in the assignment to know what they are looking at and what they are supposed to learn.
The rationale (when we ask for it) is a crutch. Ideally, your good choices should be obvious. However, you can argue for why what you are doing are good choices (and you need to do this for some assignments).
Use your “design moves” to generate ideas. If you’re stuck, think about different data transformations, layout, and encodings (in the assignments, we typically don’t allow interaction).
Make the figure look good on a page. Each visualization is a single page (later assignments will explicitly be single page PDFs, early assignments are images). Think about how things will look at that scale. Even if it is on screen: it will be a single page on screen. Don’t make things too small (or have too low of “information density” for the space). In real scenarios, where the figure must fit into the paper, the size still must be considered. Don’t expect the viewer to zoom in (except if they are very interested and want to get the fine details).
Basic charts can be effective visualizations. But: they must be well chosen (pick the appropriate chart type), they must show the right things (which variables are being shown in the chart), and the details matter a lot. You can layer extra information into basic charts (e.g., using colors in a bar chart). Details matter a lot for non-standard charts as well, but in terms of grading, you can often make up for details through other bigger choices to create effectiveness.
The details do matter. Both to create subjective impression as well as effectiveness. When the viewer sees bad details (e.g., poor choices in numbers, illegible fonts, poorly chosen colors, …) these first impressions start things off on the wrong foot.
If you are putting together multiple charts into a single visualization, be sure to tie them together well. This can mean matching details (e.g., common color schemes, typography), layout (e.g., aligning axes), etc. Importantly: help the viewer understand how to use the pieces together. Try to make the whole be more than the sum of the parts.
Choose colors wisely. Choose colors with lots of contrast (yellow on a white background?). Note that similar colors may be difficult to distinguish (comparisons at a distance can be very difficult). Don’t expect subtle differences to be easy to see. (etc)
If you reduce the amount of data, it should be obvious - or explained. If you are binning/grouping, make sure the labels reflect that. If you are selecting a subset to show, explain. I often wondered “why am I only seeing these” - sometimes there are simple reasons (reduced data set, only showing the top-5, …).
If the key to the story is a comparison, do something to make that comparison easy for the viewer to make. Not all comparisons will be easy - but try to enhance the most important ones. For example, if you are using a stacked bar chart, between-bar comparison is easiest for the whole bars, or for the bottom piece (since it is aligned on the common baseline). The choice of which to put on the bottom matters.
Highlight elements to help the story come out. Show individuals, or point at key observations. Even if something is clear (e.g., a peak on a line graph), giving some explanation of why it is interesting may help the viewer.
The assignment asks for static visualizations (like paper or article figures). Don’t expect interaction. And don’t show the interaction (e.g., we don’t need to see the Tableau UI). If you did filtering or something that required specific settings, you might want to indicate some other way than just showing how you did it using the Tableau UI.
Deriving a new variable and showing that in a simple chart is often an effective strategy because it can focus on a more important quantity.
Stay faithful to the data (in cases where you are trying to show the data). Even if you are drawing by hand, it should be based on the actual data. Be clear what is real and what isn’t. A caption can explain.