Choosing the Right Chart

Why this matters

Module 2 built a machine that can answer nearly any question about Lantern & Pine. Module 3 is about the last three feet: the distance between a correct answer and a reader understanding it. That distance is crossed by charts, and here is the discipline this module runs on: a chart is a claim. A bar chart claims "these categories compare like this." A line chart claims "this changed over time like this." Choose the wrong chart and you make the wrong claim with correct numbers — which, as this course keeps demonstrating, is the failure mode nobody warns you about.

What each chart claims

Two consequences of the claims model. First, the question chooses the chart — not taste, not variety, not "the page looked samey." Second, this course's standing advice, offered as an opinion and owned as one: when you are not sure what to use, use a bar chart. Comparison is what most business questions are, bars are the most accurately read encoding humans have, and nobody has ever been criticized for one.

Building the core four, properly

Rebuild the greatest hits from your model, now with craft:

• Bar — Revenue by Category. Sorted descending by value, never alphabetically: "Furniture leads" should be visible before a single label is read. Data labels on (five bars can afford them); axis title off when the title says it.
• Line — Revenue by Month. Month from the Dates table (typed, ordered — the alphabetical-April disease died in Lesson 2, stay vigilant). Markers on for twelve points; off for daily grain.
• Matrix — Category × Year revenue. The reconciliation workhorse: exact values, row and column totals, and your first place to check any chart that surprises you.
• Card — [Total Revenue], named measure, formatted at the source. One number, one claim.

Charts that lie

None of the following produce an error. Every one renders professionally. Each is a lie a reader cannot detect without doing your job for you — which is why detecting them is your job.

1 · The truncated axis

Take the 2025 regional revenue bars — Midwest $2.76M, Northeast $2.37M, West $2.35M, Southeast $2.20M, Southwest $1.76M — and let an axis start at $1.5M instead of zero. The Southwest's bar shrinks to a sliver a quarter the Midwest's height, screaming "collapse" where the honest ratio is about six-to-four. Bars encode value as length; a non-zero baseline breaks the encoding, always. (Lines tolerate zoomed axes — they encode by position — which is why your stock app does it. Bars never.)

2 · The meaningless aggregation

Drop DiscountPct onto a card and Power BI cheerfully sums it: 8,599.35. Eight thousand five hundred ninety-nine… what? Percent? The number is the sum of 165,042 per-line discount rates — arithmetic without meaning, rendered with total confidence. (The honest cousin — average discount, 5.21% — is one aggregation-dropdown away, and even it deserves the L05 question: weighted by what?) Every implicit aggregation is a claim someone should have chosen. This card is why the course rule about explicit measures exists.

3 · The overloaded pie

Five categories in a pie is already at the edge (quick — is Outdoor 25% or 30%?). Forty products in a pie is a kaleidoscope. Angles are the hardest common encoding for humans to compare; slices within a few percent of each other are indistinguishable without labels, at which point the labels are doing the chart's job. Share questions with more than three or four parts: use a bar with a % of Total measure — you built it in Lesson 6.

Titles are claims too

"Revenue by Category" is a filename. "Furniture and Outdoor carry two-thirds of revenue" is a finding — and it is what the chart is for. Write visual titles as the sentence the reader should leave with, and let the axes carry the bookkeeping. Two tests: if the title could sit on any chart of the same fields, it is a label, not a title; and if the title makes a claim the chart does not visibly support, the chart or the claim has to change. (This discipline also catches your own errors — a title you cannot write is usually a chart that is not saying anything.)

Color, honestly

The Iris theme's categorical colors were chosen to be distinguishable — including by the roughly one reader in twelve with color-vision deficiency — and the course rule follows from that: never encode meaning by color alone. Color supports position and length; it does not replace them. Semantic colors (the theme's teal-good, red-bad) mean those things only — a chart that uses red decoratively has spent the alarm color on nothing. One lead color for the finding, muted neighbors for context, is worth a hundred rainbow defaults.

Common mistakes

• Choosing charts for variety. "The page has too many bars" is not a problem. Five bar charts answering five comparison questions beat one of each species answering nothing.
• Alphabetical bars. Sort by value unless the axis has inherent order (months, age bands). The reader's first glance should meet the finding.
• Pies asked to do comparison work. Parts-of-whole with 3–4 parts, maybe. Rankings, never.
• Titles that label instead of claim. If it could caption any chart of those fields, it is not doing its job.
• Decorative red. Semantic colors spend their meaning every time they appear. Save the alarm for alarms.

Summary