Each plate reproduces one of the chart forms from Visualize This (2nd ed.) — hot dogs, postage, Walmarts and all — then translates it into the question it answers on store / date / category / net-sales data and names the Power BI visual that does the job. Hover any mark for its value; every chart has a table twin.
Start with the default barplot(), then take away: borders off, one hue for the bars that carry the story, gray for the rest, a wrapped annotation for the 2001 jump.
Weekly net sales by fiscal week with record weeks (or promo weeks) highlighted. The story is "which weeks broke the pattern", not "here are 52 bars".
Clustered column + conditional formatting on Column color driven by a flag measure, e.g. Is Record Week = IF([Net Sales] > [Prior Max Net Sales], 1, 0). Annotate with a text box, not a data label on every bar.
plot(type="s"): a value that holds until the next change is a step, and interpolating between changes invents movement that never happened.
Shelf price, fuel rack price, contract cost, tax rate — anything effective-dated. Draw the step from the effective date; your line chart of monthly average price is smoothing away the actual events.
Line chart → Format → Lines → Step on (2023+ builds). Feed it the effective-dated table, not a monthly average. For older builds, use Deneb with "interpolate": "step-after".
A time series with two periodic components wants a grid, not a 288-point line. One hue, six ordered steps, white gaps between cells, and the axes labelled in human units (12am, noon, 6pm).
Transactions or fuel gallons by day-of-week × hour per store — the labor-scheduling chart. Or week-of-year × year for seasonality drift.
Matrix visual with rows = hour, columns = weekday, values = measure, and Cell elements → Background color on a single-hue gradient. Turn off values if you want a pure heatmap. Or the Deneb rect mark with a scale: {"scheme":"blues"}.
Pie charts are for ≤ 6 slices at a glance; sorted bars beat them for reading values; and when the proportions have a time axis, stack them — then reshape the long table to wide (his prepare-household-types.R) before plotting.
Category mix of net sales by fiscal period (tobacco / beverages / fuel / lottery…). The stacked-100% view answers "is the mix shifting"; the plain stacked view answers "is the total growing". Ask which question first.
100% stacked area (or stacked column for ≤ 24 periods). Sort legend by last-period share, fold anything under 3% into Other with a calculated column, and cap at 7 series — the palette stops being distinguishable past that.
The D3 barbell: a band scale for rows, a line for the gap, two circles for the ends. Sort by one end so the gaps read as a shape. Label MEN / WOMEN once, on the top row, never in a legend the eye has to travel to.
This year vs. last year per store, or plan vs. actual per category. The connecting line is the variance — far more legible than clustered bars for 15+ rows.
No core dumbbell. Options: (a) Deneb — rule mark from [Prior] to [Current] plus two point layers; (b) the "MacGyver" scatter with Error bars; (c) a table with a variance column, which is often honest enough.
Two lines plus a clipped area fill between them, colored by which is on top. The fill is the message; the crossing year is the annotation.
Net sales vs. budget by week (fill = variance, colored above/below), or inside vs. fuel gross margin dollars — the classic c-store "who's carrying the store" chart.
Core area charts can't clip by sign. Deneb: two area layers with a filter transform per sign, or one area mark with y/y2 and a condition on fill. Keep the arithmetic in DAX ([Variance to Budget]), not in the Vega spec.
Load boundaries with sf, project to Albers equal-area (never Mercator for area comparisons), join the census table on GEOID, bin the rate (not the count) into five classes, and shade with one hue.
Comp-store growth or gallons-per-store by county/region; store locations as points; new openings as small multiples by year. Shade rates and per-store metrics, never raw totals — big regions win by size alone.
Shape map (filled, needs a TopoJSON for custom regions) or Azure Maps with Filled map layer; bubble layer for store points with size = √metric via a SQRT([Gallons]) measure. Enable map visuals in tenant settings first — it's off by default.
Stack the dots, then bin them; bin width is a real choice that changes the picture (try 1, 3, 6). Compare groups with small multiples on a shared x-axis and a mean line, ordered by the thing you expect to shift.
Basket size distribution by store format; days-of-supply distribution by SKU; margin % distribution by category. A mean in a KPI card hides the two-humped distribution every time.
No core histogram. Bin the column (right-click field → New group → Bin) and use a column chart, or Deneb bin transform with a bin-width parameter. Box plots: the AppSource "Box and Whisker" visual or Deneb boxplot mark.
Built from the companion code and data for Visualize This, 2nd ed. (Nathan Yau, Wiley 2024), chapters 4–8. Chart forms follow Yau's; colors are a CVD-validated palette in fixed slot order rather than the book's per-chart hex codes; every chart carries a hover layer and a table twin. Retail translations assume a star schema at store × date × category grain.