Choosing a Chart: A Taxonomy of Visualisation Methods

1Every graph is one of four element types

This is the useful abstraction. Instead of memorising chart names, ask which element is carrying your data, and the chart follows.

Fig 2.1Point, bar, line, area, drawing the same twelve months

element carrying the data

Four charts, one dataset. The element you pick decides what the reader can do with it: compare a single month, follow a trend, judge a total, or look for a relationship. Point, bar, line, area is the list to write down.

The seven core visuals

About a dozen chart types cover roughly 90% of cases, and around 5% of work needs a genuinely custom visual. So master a dozen deeply rather than a hundred shallowly.

Fig 2.2The core set, with the job each one is for

Each of these is doing one job well rather than several jobs adequately. The table beside the chart is what a grader is looking for: the visual, the use case, and one sentence on why that visual and not a neighbouring one.

When simple text beats a chart

Fig 2.3Two numbers, drawn and then simply stated

A two-bar chart forces the reader to compare two lengths and then do the subtraction themselves. A sentence hands them the answer. When there is no need for a visual, do not build one. Gestalt still applies to text, so bold, colour and size create the emphasis inside the sentence.

Lines and slopegraphs

A line implies continuity, so the eye reads a line as a continuous variable. That is the continuity principle doing the work.
A slopegraph connects just two points in time, which lets you show change even in categorical data.
Slopegraphs answer "what changed, and how fast" between two periods, without drawing everything in between.

Fig 2.4Eight product lines over two years, as lines and as a slopegraph

The multi-series line has all the data and gives up its message to clutter. The slopegraph throws away everything between the two endpoints and gains the one thing being asked about, which is who rose and who fell. Multi-series lines work, but clutter destroys them quickly.

Waterfall

The chart for explaining a change rather than stating it. Standard business use is a revenue-to-profit bridge, or a headcount plan.

Fig 2.5Headcount moved from 100 to 116. The waterfall says why

Two endpoints alone say the total went from 100 to 116, which is a fact with no explanation. The waterfall decomposes the move into its components, so the reader sees what pushed up and what pulled down. Positive and negative bars carry direction intuitively, so growth and loss need no legend.

Design concepts: form follows function

Borrowed from product design philosophy. The best products start from the function they must deliver, then design a form that delivers it. The function of any visualization is to communicate information to an audience, and everything else is downstream of that.

Step 1

Affordances

Things you add or remove that make the message easier to take in. Highlight, declutter, simplify the axis, show only start and end. The test is processing time.

Step 2

Accessibility

It works for a wide audience, including colour-blind readers and people seeing it in print.

Step 3

Aesthetics

Pleasant enough that somebody engages with it rather than skipping past.

Step 4

Acceptance

The audience receives it in the intended form. The step most people skip, and the reason good visuals still get rejected.

Building acceptance

People resist change, so a new format meets resistance by default. Creating acceptance is the analyst's job, not the audience's: explain the reasoning behind the design, show how it improves on the earlier version, win over a few audience members first and let them advocate for you, then take feedback and iterate. Get affordances, accessibility and aesthetics right and acceptance usually follows on its own.

Visual hierarchy

Fig 2.6A cloud of points becomes named segments

A plain scatter becomes far more useful when you cut it into quadrants on both axes. Grey out the segments that do not matter and colour the one that does, so attention lands on the actionable group. Do not overdo it. Added elements that do not support the message destroy the original purpose.

Dissecting model visuals

Taking a finished, well-made visual apart to see why each choice was made. The governing idea is intentional design: every element, from colour to alignment to labels, serves a specific purpose.

Technique	What it means
Emphasize key data	Bold colour, thicker lines, clear labels on the thing that matters
De-emphasize context	Lighter, thinner lines and smaller fonts for comparison data and footnotes
Use annotations	Text placed directly on the chart to explain an event, a trend or a forecast assumption
Layout and colour	Horizontal bars for long names. One attention-grabbing colour for the critical category
Maintain clarity	Label only essential points. Remove an axis entirely if data labels already give the numbers

Fig 2.7The five techniques applied one at a time

apply

Two lines in the same colour say the two series are equally important. Making one thick and one pale says which one to read. A reference line for a target makes the gap visible without any arithmetic, and an annotation puts the finding on the chart instead of leaving the reader to locate it.

Storytelling with data

Data informs partially, a story convinces. The end goal of every visualization exercise is to move an audience from insight to action, and that happens only when the story is consistent and compelling.

Audience-centric, not data-centric

This is the line to remember. You are not visualising because you have data. You are visualising because somebody needs to consume it.

Question	Why it changes the design
Who is the audience?	Sets the vocabulary, the chart complexity and the assumed background
What do you want them to know or do?	Decides whether you are informing, persuading or inviting challenge
How will they consume it?	Live presentation against email changes everything about the visual

In a live session you can clarify in real time, so the visual can carry less on its own.
In a document or email you will not be there, so the visual has to be self-evident and stand alone.
To prove a point or support a trend, use a line chart, because continuity implies the trend continues.
To invite questions when you are unsure of a relationship, use a scatter plot, which presents evidence without asserting a conclusion.
Choosing the wrong visual can impose assumptions on your audience that you never intended.

What to avoid, and what to use

3D effects distort the encoding and add nothing. Cluttered axes, excessive detail and unnecessary gridlines are chartjunk: if a design element does not add value, it subtracts. On the other side, pre-attentive attributes (colour, size, shape and position) are processed before conscious attention, which is why key data appears to pop instantly. Placing the most critical information at the top-left uses the same effect, since that is where the eye lands first.

The five communication purposes

This is the taxonomy the module is named after, and the backbone of the week.

Fig 2.8Pick a purpose, see the charts that serve it

The five purposes in order: comparing categories, hierarchies and part-to-whole, changes over time, connections and relationships, and mapping geo-spatial data. Each sits over a family of charts, and the sections below take the ones most likely to come up one at a time.

Charts for comparing categories

Gantt chart, the floating bar

Fig 2.9Why a bar anchored at zero fails for uneven ranges

Anchor every bar at zero and all five look similar, because the bar length is dominated by the distance from zero rather than by the span that matters. Let the bars float and you read relative position and absolute span at the same time, in one graphic. One category sitting between 40 and 80 and another between 0 and 40 cannot be compared on anchored bars.

Sankey diagram

Fig 2.10Flow, with band width proportional to quantity

Shows quantities moving across stages, and the width of each connecting band is proportional to the quantity flowing. Best when elements transform, divide or combine between stages. It reads two categorical variables and their quantitative link at the same time, which a stacked bar cannot do. Typical uses are energy flows, trade between countries, and movement of resources inside a company.

Small multiples

Fig 2.11Six series crammed together, then given a panel each

Works because humans scan a grid of similar shapes very fast and spot the odd one out. It lets you compare within a group and across groups in the same space, and it replaces one cluttered multi-series chart with several readable small ones.

Charts for hierarchies and part-to-whole

Fig 2.12Fourteen categories, four ways

Treemap is the one most likely to be tested. It handles many categories and several hierarchy levels at once, which is exactly where a pie chart collapses into unreadable slivers. Power BI and Tableau both provide it, and Chapter 13 uses treemaps as dashboard filters.

Method	What it does
Pie chart	The familiar part-to-whole. Fails with many categories
Stacked bar	Composition within and across categories
Square pie (waffle)	A 10×10 grid of squares filled in proportion. Easier to read than a pie
Treemap	Nested rectangles sized by value. The standard choice for hierarchical data, and it supports drill-down
Circle packing	Constituent circles inside one overall circle. Suits a large number of parts
Bubble hierarchy	A central bubble with offshoots, each subdivided again. Shows organisation, not just proportion
Tree hierarchy	A classic branching tree, for structure rather than proportion

Charts for changes over time

Fig 2.13Stacked area against stream graph

In a stacked area chart the categories must stay constant. A stream graph handles a composition that is genuinely dynamic, where new categories appear and old ones fade, because the bands flow around a moving baseline instead of stacking on a fixed one. Width at any point shows that category's intensity, so a narrowing stream means declining share.

Flow map

Plots how a quantity moves through geography and time at once. The canonical example is Minard's chart of Napoleon's march on Russia: band width shows the army shrinking along the route, a separate band shows the retreat, and a temperature scale runs underneath. Worth knowing as the standard reference for how much a single well-designed graphic can carry.

Charts for connections and relationships

These are exploratory tools. You reach for them when you have variables but no idea how they relate.

Fig 2.14Scatter plot matrix against radial network

Why a radial network exists. A scatter plot matrix locks you into pairs, because x and y are fixed. A radial layout removes that restriction, so one variable's relationship with all others is visible at once. It costs some precision and gains dimensionality.

Mapping geo-spatial data

People live in space, so a phenomenon shown on a map connects immediately in a way a bar chart of the same numbers does not. It also uses closure, since a familiar outline reads as a whole country without being labelled.

Fig 2.15The choropleth problem: colour follows area, not people

Saturation may not represent the population underneath it, and a large sparsely populated region takes visual weight out of proportion to the people in it. Normalise against population or area where this distorts the message. An isarithmic map corrects for the underlying geographic area and population, so the saturation reflects the actual phenomenon.

Key points

The seven core visuals, and that a dozen charts cover roughly 90% of cases.
Four graph element types: point, bar, line, area.
When simple text beats a chart, and why tables speak to the verbal system.
Slopegraph against line chart, and why a slopegraph can handle categorical data.
Waterfall chart: what it decomposes and why two endpoints are not enough.
Form follows function, and the chain affordances → accessibility → aesthetics → acceptance.
What an affordance is, with two examples.
How to build acceptance when an audience resists a new visual.
Intentional design, and the five dissecting techniques.
Audience-centric not data-centric, and the three context questions.
Why live and textual delivery need different visuals.
Chartjunk, and pre-attentive attributes: colour, size, shape, position.
The five communication purposes, in order, with charts for each.
Gantt chart: why floating bars beat anchored bars for uneven ranges.
Sankey diagram: flow, stages, band width proportional to quantity.
Small multiples: why grid scanning beats one cluttered chart.
Circle packing against bubble hierarchy.
Stacked area against stream graph, and why fixed categories matter.
Scatter plot matrix against radial network, and the dimensionality trade.
The choropleth population problem, and that an isarithmic map corrects it.