Choosing the Right Chart for Your Data

Open the chart menu in any spreadsheet and you will find the gallery arranged by shape: columns, lines, pies, bubbles, radars. It is a catalogue of decorations. Nothing in it tells you the only thing that matters — which picture will answer the question you are actually asking. Choose by shape and you get a slide. Choose by question and you get evidence.

Process analysis runs on four questions, and each has a chart built for it. How does this behave over time? What does the whole distribution look like? Which few categories dominate? Do these two things move together? Learn the four pairings and you will produce sharper analysis than most dashboards you have ever been sent.

Start from the question, not the gallery

The discipline is to write the question down before opening the tool. “Show the data” is not a question. “Did cycle time change after the new intake form?” is. Nearly every question in process work is a variation on one of four, and the moment the question is explicit, the chart usually chooses itself.

1. 01Behaviour over time — run chart, graduating to a control chart once limits are known
2. 02Shape and spread of one measure — histogram
3. 03Relative weight of categories — Pareto chart
4. 04Relationship between two measures — scatter plot

How does it behave over time? The run chart

If your data was collected over time — and in process work it almost always was — time order is the most informative thing about it, and most summaries destroy it. A before-and-after pair of averages cannot distinguish a genuine shift from a slow trend that was already underway, a one-week spike, or a seasonal cycle: all four produce the same two numbers. The run chart preserves the story — one dot per result, in sequence. Shifts announce themselves as sustained runs on one side of the centre; trends as steady climbs; cycles as rhythm. It is the first chart to draw for any process metric, and often the last one you need.

What does the whole picture look like? The histogram

An average is a one-number summary, and one number cannot describe a crowd. A histogram stacks results into bins and shows the shape of the whole distribution: where results cluster, how widely they spread, whether the tail runs long on one side. Shape is diagnosis. Two peaks usually mean two processes hiding under one name — two shifts, two machines, two clerks with different methods. A cliff edge at a specification limit suggests results are being inspected, rounded, or nudged at the boundary. A long tail warns that the average flatters the experience of every customer living in the tail. None of this is visible in a summary statistic, which is precisely why summaries feel so reassuring.

Which few dominate? The Pareto chart

When the data is categories — defect types, complaint reasons, error codes — the useful question is almost never “what is the average?” but “where should we aim first?” The Pareto chart answers it by sorting the bars from largest to smallest, usually with a cumulative line running above them. The picture routinely reveals what Joseph Juran called the vital few and the trivial many: a small handful of categories carrying most of the pain. Its power is political as much as analytical — it moves the argument from whose problem is loudest to which problem is largest, and it makes the choice of a first target look like arithmetic instead of favouritism.

Do these move together? The scatter plot

When you suspect one measure influences another — temperature and defect rate, queue length and error frequency, experience and handle time — plot the pairs as points. The scatter plot shows whether a relationship exists, which direction it runs, how tight it is, and whether it bends. It also enforces humility. A cloud with no pattern retires a confident theory in seconds, and even a tight pattern shows association only: two measures can move together because one drives the other, because both answer to a third factor, or by coincidence in a small sample. The chart raises the question sharply; investigation settles it.

Charts that flatter more than they inform

Some chart types appear constantly in business decks and rarely survive contact with a careful question. Treat each of these as a request to see the data another way:

• Pie charts — human eyes compare angles poorly; beyond about three slices, a sorted bar chart does the same job better
• Truncated value axes — starting the axis near the minimum turns routine wobble into apparent drama
• 3D effects — perspective distorts area, and area is the message
• Dual-axis charts — two scales in one frame let an author manufacture whatever correlation the story needs
• Averages without spread — a line of monthly means with no sense of variation invites explanations for pure noise

A chart is an argument you can check; a decoration is a claim you must take on trust.

Building the habit

Choosing charts well is a habit, and like most habits it sharpens fastest with structured practice on realistic data. Our Yellow Belt program builds the practical toolkit in 14 hours for $129, and Green Belt applies every chart in this article inside a full simulated DMAIC project across 35 hours of material, ending in a 100-question proctored exam with one free retake and lifetime access. The next time someone sends you a nine-slice pie chart, you will know both what to ask and what to draw instead.