The grammar of graphics — Data Viz Module 2 | LeadAfrik Public Economics Hub

In 1999, Leland Wilkinson published The Grammar of Graphics, a book that did for charts what linguistic grammar did for sentences: it gave you a system of composition instead of a closed list of templates. ggplot2 in R, Vega-Lite, Observable Plot, and Tableau's marks card are all direct descendants of his framework. Power BI's visualisation engine is the same idea, hidden behind clicks.

The seven layers

Data: the dataset — rows and columns.
Aesthetics: the mapping of data variables to visual properties (x, y, colour, size, shape, opacity).
Geometries (geoms): the shapes that represent the data — points, lines, bars, areas.
Statistics: any transformation applied — counts, means, regressions, density.
Scales: how data values map to visual values (log axis, colour palette, size range).
Coordinates: Cartesian, polar, geographic projections.
Facets: small multiples — repeating the chart for sub-groups.

Composition over selection

Spreadsheet chart pickers give you 'Bar Chart', 'Pie Chart', 'Scatter'. The grammar of graphics says: there is no 'bar chart' — there is a geom_bar with x mapped to category and y mapped to count. Subtle change; massive consequences. You compose charts from primitives instead of choosing from a menu.

Try it now — Viz Studio playground

Open the LeadAfrik Viz Studio at /learn/viz-studio to feel this. Drag fields onto x, y, colour, size shelves; pick a mark type; the chart renders live. It runs entirely in your browser, on Kenya bank-rate, pension, and M-Pesa data. The same grammar transfers directly to Tableau and Power BI.

Why this matters even if you live in Power BI

Once you see the grammar, picking up a new tool takes a day, not a quarter. Tableau's 'marks card' is just aesthetics. Power BI's 'fields well' is just aesthetics. ggplot's aes() function is literally named for it. The mental model transfers. You stop being a Tableau expert or a Power BI expert and become someone who knows how charts compose, applied through whichever tool is in front of you.

# ggplot2 — the grammar made explicit
library(ggplot2)

ggplot(data = sales,
       aes(x = quarter, y = revenue, colour = segment)) +
  geom_line(linewidth = 1) +
  geom_point(size = 3) +
  scale_colour_brewer(palette = "Set2") +
  facet_wrap(~ region) +
  labs(title = "Enterprise revenue grew 4x while SMB stayed flat")

Each '+' is a grammar layer. Same composition in Tableau requires clicking through the marks card, but the mental model is identical.

The practical payoff

Once you think in grammar, decisions get cleaner. 'Should colour encode the segment or the product line?' becomes a real question with a real answer, not a vibe. 'Do I need facets or a single chart?' is a structural decision, not a design preference. You stop arguing about 'which chart type' and start arguing about the right mapping for the claim.

Exercise

Pick a chart you've made recently. Write it out in grammar-of-graphics terms: what is the data, the aesthetics (x, y, colour, size), the geom, any statistics applied, the scales, and any facets? Are any of the mappings doing unimportant work — and if so, can the chart be simpler?