Lecture video · L07 Choosing the Right Chart

[Slide 1] Module Three begins tonight, and it begins with a confession about everything we have built so far: the machine is invisible. Nobody in a boardroom will ever see your star schema or your measure family. They will see charts — and the distance between a correct answer and a reader understanding it is crossed entirely by the charts you choose. So here is the discipline this whole module runs on, in four words: a chart is a claim. Tonight you learn what each chart claims, how to build the core ones with real craft, and how to spot the three most professional-looking lies in business graphics — none of which, in this course's oldest tradition, produce a single error message.

[Slide 2] Every chart species makes a claim, and here is the whole taxonomy in five rows — let me read it. A bar or column chart claims: these things compare like this. A line claims: this changed over a sequence like this. A card claims: this one number is the answer. A table or matrix claims: you will want to look up the exact values. And a pie — used sparingly — claims: these parts make a whole. The consequence of thinking in claims: the question chooses the chart. Not taste, not variety, not the page looking samey. And when you genuinely are not sure? My standing advice, offered as an opinion and owned as one: use a bar chart. Most business questions are comparisons, length is the encoding humans read most accurately, and nobody has ever been criticized for a bar chart.

[Slide 3] The core four, now with craft. Bars sort by value, descending — Furniture's lead should be visible before anyone reads a label; alphabetical order is for dictionaries. Lines take their months from the Dates table, typed and ordered, because the alphabetical-April disease from Lesson Two stays dead only through vigilance. The matrix is your reconciliation workhorse: exact values, totals both ways, and the first place you check any chart that surprises you — tonight our matrix's grand total matched the revenue card to the cent, which is exactly the kind of free audit you should collect. Cards show named, formatted measures, always — no implicit sums ship, per the standing rule. And data labels follow one principle: on when the reader will use the individual values, off when they would bury the shape.

[Slide 4] Prediction beat. A dashboard ships with a card reading eight thousand five hundred ninety-nine point three five, labeled discount percent. One minute: where did that number come from? Here is the trace: someone dropped the discount-rate column onto a card, and Power BI's implicit aggregation summed it — one hundred sixty-five thousand per-line rates, added together. Eighty-six hundred percent of discount. The arithmetic is flawless and the number means nothing, because rates do not sum. The honest cousin — average discount, about five point two percent — was one dropdown away. This card is why the explicit-measure rule exists: every aggregation is a choice, and a number nobody chose is a number nobody can defend.

[Slide 5] The first great lie: the truncated axis. Take the twenty twenty-five regional bars — Midwest at two point seven six million, Southwest at one point seven six — a real ratio of about one point six to one. Now floor the axis at one and a half million, quote, to show the detail. The Southwest's bar becomes a sliver, and the visual ratio balloons to roughly five to one. Every label on the chart remains perfectly correct; the lengths are doing the lying. The rule is geometry, not etiquette: bars encode value as length from the baseline, so the baseline is zero, always. Lines encode by position, which is why your stock app zooms its axis legitimately. Bars never. If small differences matter, data labels carry them honestly.

[Slide 6] Lie number three is gentler: the overloaded pie. Five slices is already the edge — quickly now, is Outdoor twenty-five percent or thirty? Forty slices is a kaleidoscope. Angles are the encoding humans compare worst, and once every slice needs a label to be read, the labels are doing the chart's job. Share questions with more than a few parts get a sorted bar of the percent-of-total measure you built in Lesson Six. And while we are near the subject of saying things: titles. Revenue by Category is a filename. Furniture and Outdoor carry two-thirds of revenue — that is a title, because it is a claim. The test is simple: if your title could sit on any chart of the same fields, it is a label. Write the sentence the reader should leave with, and write it before you build — a title you cannot write is usually a chart that is not saying anything.

[Slide 7] Last craft point: color. About one reader in twelve has some color-vision deficiency, which is why the course theme's categorical colors were chosen distinguishable — and why the rule is: never encode meaning by color alone. Color supports position and length; it does not replace them. And semantic colors are words: the theme's teal means good, its red means bad, and they mean those things only if they are never spent on decoration. A chart that colors every third bar red for rhythm has taught its readers that red means nothing — and the day something is genuinely on fire, the alarm is already worn out.

[Slide 8] Tonight in five lines. A chart is a claim, and the question chooses the species — bars compare, lines move, cards headline, matrices reconcile. When unsure, a sorted, zero-based, labeled bar chart, and own the choice. The three professional lies all render without errors: truncated axes exaggerate real differences, implicit aggregations manufacture confident nonsense like our eighty-six-hundred-percent discount, and crowded pies hide rankings in angles. Titles state findings — write the sentence first. And color supports encodings for every reader, including the one in twelve the rainbow default forgets. This week's Hot practice puts you in the chief-of-staff's chair reviewing five charts — and fair warning: one of them is genuinely good, and clearing it is part of the grade. Next week: the charts beyond the core — maps and their pitfalls, scatterplots, small multiples — and conditional formatting used like an adult.