8 Graphing Best Practices

A chart is an argument. It takes a table nobody wants to read and turns it into a claim a reader can check at a glance – which means a chart can mislead as easily as it can inform, and usually without anyone noticing. This chapter is about the decisions that come before you open a chart menu: what question the chart answers, which shape suits it, and what to leave out.

Learning Objectives

By the end of this chapter, you will be able to:

Say what question a chart is supposed to answer before you build it
Choose a chart type that suits the comparison you are making
Label a chart so a reader who did not make it can read it
Spot the common ways a chart misleads, and avoid them in your own work

8.1 The Principles

Before we get into specific chart types, some general principles. These come from a long tradition of work on data visualization, especially by Edward Tufte (The Visual Display of Quantitative Information) and more recently by Cole Nussbaumer Knaflic (Storytelling with Data):

A chart should answer a question. Before you make a chart, decide what question it is supposed to answer. “Show the yield data” is not a question. “Which variety had the highest yield?” is.
Always add a descriptive title and label every axis. A reader who did not make the chart must be able to tell what it shows without asking. “Yield” is not an axis label; “Yield (bu/ac)” is. In Excel, add these from Chart Design → Add Chart Element → Chart Title and → Axis Titles. Unlabelled charts are the single most common — and most easily avoided — presentation failure.
Minimize clutter. Gridlines, 3D effects, drop shadows, patterned fills, excessive tick marks — all of these are usually just noise. Tufte calls the noise “chart junk.” Default to removing it.
Label directly when possible. A chart where each line is labeled at the end is easier to read than one with a legend that forces the eye to jump back and forth.
Be careful with axis ranges. Where a chart’s bars or areas encode magnitude, starting the axis somewhere other than zero exaggerates differences. Whatever you choose, make it obvious to the reader.
Use colour purposefully. Colour should encode information (the blue line is canola, the orange is wheat), not be decorative. Use a colour-blind-friendly palette — about 8% of men cannot distinguish red from green. Default Excel colours are mostly fine; avoid red-on-green combinations.
Be wary of pie charts. Humans judge angles and areas poorly; a bar chart almost always does the job better. If you do use one, keep it to three or four slices.

8.2 Excel vs R for Charts

I want to be honest about this: Excel charts are fine for many purposes. They are easy to make, they look acceptable with default settings, and they are immediately editable by non-technical colleagues. If you are making a one-off chart for an internal meeting, Excel may be the right choice.

R (ggplot2) wins when you need:

The chart to be reproducible from a script.
The chart to update automatically as the underlying data changes.
Many similar charts (via faceting or loops).
Fine-grained control over appearance.
Statistical features (trend lines, confidence bands, quantile ribbons).

Most working analysts use both. In this course we will favor R because it builds the reproducibility muscle, which is harder to build than the Excel-chart muscle.