Key Takeaways — Chapter 5: Choosing the Right Chart

1. The Question Comes First, Always

You do not choose a chart based on what the data looks like, what tool you know best, or what looks visually appealing. You choose a chart based on what question you are answering. The same dataset requires completely different chart types depending on whether you are asking about comparison, distribution, relationship, composition, change over time, or spatial pattern. Before you select a chart type, articulate the question in a single sentence. If you cannot state the question clearly, you are not ready to make a chart.

2. Two Inputs Drive the Decision

Every chart selection decision has exactly two inputs: the type of data you have and the type of question you are answering. Data types are: categorical, continuous/quantitative, temporal, spatial, and network/relational. Question types are: comparison, distribution, relationship, composition, change over time, and spatial pattern. Classify both inputs, and the chart selection matrix gives you a ranked list of candidates. This two-input model replaces guessing, scrolling through chart galleries, and defaulting to whatever chart type you made last.

3. The Chart Selection Matrix Is Your Reference

The data type x question type matrix is the core artifact of this chapter. For every combination of data type and question type, it identifies the most effective chart types ranked by perceptual accuracy and communication clarity. Memorize the most common cells — bar chart for comparison, line chart for change over time, scatter plot for relationship, histogram for distribution — and consult the matrix for less common combinations. Print it. Keep it next to your monitor.

4. The Decision Tree Narrows the Field

The matrix gives you candidates. The decision tree narrows them. Walk through the steps: identify question type, count your variables and categories, check your dataset size, then run the context check (audience, medium, purpose). The tree is an algorithm — follow it mechanically and you will arrive at a defensible chart type every time.

5. Six Common Mistakes and Their Fixes

The most common chart selection errors are: (1) pie charts with too many slices — fix with bar charts; (2) dual-axis charts that imply false correlations — fix with small multiples or normalization; (3) 3D bar charts that distort comparisons — fix by removing the third dimension; (4) spaghetti line charts with too many overlapping series — fix with small multiples or the highlight strategy; (5) stacked bar charts used for segment comparison — fix with grouped bars or small multiples; (6) chart types that do not match the question — fix by going back to the question classification. Learn to diagnose these on sight.

6. Context Settles the Final Choice

The matrix and decision tree get you to two or three candidates. Three contextual factors settle the choice: audience (executive audiences need familiar, simple chart types; technical audiences can handle complexity), medium (print demands simplicity and colorblind safety; interactive dashboards allow progressive disclosure), and purpose (exploratory analysis values speed; explanatory visualization values clarity; confirmatory reporting values precision). A violin plot may be the most informative chart for a distribution question, but if your audience has never seen a violin plot, a box plot or histogram will communicate more effectively.

7. Every Real Dataset Is a Mix of Data Types

Do not classify your entire dataset as one data type. Classify each variable. The Meridian Corp dataset contains categorical data (product line), continuous data (revenue), temporal data (order date), and spatial data (region). The chart you choose depends on which variables you are plotting for a specific question, not on the "type" of the dataset as a whole.

8. Rules Can Be Broken — But Only With Reason

The decision framework is a guide, not a commandment. Data journalism, artistic visualization, and novel chart types are legitimate departures from the matrix recommendations. But every departure should serve a purpose: it should match the visual encoding to the data structure and the communication goal better than the standard recommendation does. Break the rules knowingly, not accidentally.