Principles of Data Visualisation

Why Visualisation Matters

Data visualisation is not decoration. It is the primary mechanism through which analysts communicate insights to decision-makers. A well-designed chart can convey in seconds what a table of numbers might never communicate. A poorly designed chart can mislead, confuse, or completely obscure the insight you are trying to convey.

As a data analyst, your ability to visualise data effectively directly determines how much impact your analytical work has on business decisions. Analysis that stays in a spreadsheet changes nothing.

---

The Fundamental Purpose of a Chart

Before creating any visualisation, be clear about its purpose. Every chart should answer exactly one question clearly. Ask yourself:

• What decision or question does this chart support?
• Who is the audience (analyst, manager, executive, general public)?
• What is the single most important insight this chart should convey?
• What action should the viewer take after seeing it?

If you cannot answer these questions, you are not ready to choose a chart type.

---

Choosing the Right Chart Type

The most important decision in visualisation is choosing the right chart for the data relationship you want to show:

Showing change over time

Line chart: The default for continuous trends. Use when you have many time points or when the rate of change matters. Bar chart (vertical/column): Better when you have discrete periods (monthly, quarterly) and want to compare individual values. Area chart: Useful for showing volume over time, especially for stacked comparisons.

Comparing categories

Bar chart (horizontal): Best for comparing many categories, especially when labels are long. Column chart (vertical): Better for fewer categories or when the x-axis represents time. Dot plot: Good for comparing a single metric across many categories with less visual clutter than bars.

Showing composition (part of whole)

Pie/donut chart: For 5 or fewer segments. Donut preferred (easier to read the arc than area). Stacked bar/column: For more segments, or when showing composition change over time. Treemap: For hierarchical composition (category > subcategory).

Showing distribution

Histogram: Frequency distribution of a continuous variable. Box plot: Shows median, quartiles, and outliers. Excellent for comparing distributions across groups.

Showing relationships

Scatter plot: The default for two continuous variables. Shows correlation, clusters, and outliers. Bubble chart: Scatter plot with a third variable encoded as bubble size. Heatmap: For showing a matrix of values (correlation, cross-tab).

---

The Principles of Effective Data Visualisation

1. Signal-to-Noise Ratio

Maximise signal (the data and the insight) while minimising noise (visual elements that do not add information). Everything on a chart should earn its place. Remove:

• Unnecessary gridlines (or reduce them)
• Chart borders and background fills
• 3D effects (always wrong -- distorts perception)
• Decorative icons and images
• Excessive tick marks and labels

2. Data-Ink Ratio (Edward Tufte)

Data-ink is the ink used to convey data. Every non-data-ink element should be eliminated or reduced. A high data-ink ratio means more of the visual is conveying information.

3. Pre-attentive Attributes

Certain visual properties are processed by the human brain before conscious attention:

• Colour: Use sparingly and meaningfully. One accent colour for emphasis.
• Position: The most accurate encoding -- use it for the primary comparison.
• Length: Second most accurate. Used in bar charts.
• Size: Less accurate, good for secondary information.
• Shape: Use to distinguish categories only.

4. Gestalt Principles

Our brains see patterns and groupings:

• Proximity: Elements close together are perceived as a group
• Similarity: Similar colours/shapes are perceived as a group
• Continuity: We follow lines and curves naturally
• Enclosure: A border or background groups elements

Use these to guide the viewer's eye to the most important information.

---

Common Visualisation Mistakes

Using the wrong chart type: Pie charts with 10 segments. Bar charts for time series when a line would be clearer.

Truncating the y-axis: Starting a bar chart y-axis at something other than zero exaggerates differences. (Line charts can start at non-zero, bars should not.)

Too many colours: More than 5-6 distinct colours in a chart is difficult to interpret. Use one accent colour and grey for everything else.

Missing context: A chart without a clear title, axis labels, and units cannot be understood without explanation.

3D charts: Never. They distort values and add visual complexity with zero benefit.

Dual-axis charts: Two y-axes are confusing in almost all cases. Use small multiples instead.

---

The Hierarchy of Chart Elements

Design your chart in this priority order:

1. The data: The numbers and relationships you are trying to show
2. The story: The single insight the chart is designed to communicate
3. The title: Should state the insight, not just describe the data ("North Region Leads in Q3 Revenue" not "Revenue by Region Q3")
4. Labels and annotations: Guide the viewer to the key insight
5. Axes and scales: Only as much detail as needed
6. Everything else: Minimal

---

Colour in Data Visualisation

Colour is the most misused element in data visualisation.

Sequential palettes: For continuous data from low to high (light to dark). Examples: blues, greens.

Diverging palettes: For data that varies around a midpoint (e.g., positive/negative, above/below average). Red-white-green or blue-white-red.

Categorical palettes: For distinct categories with no order relationship. Use maximum 8 colours; prefer muted, accessible palettes.

Accessibility: Approximately 8% of men have red-green colour blindness. Test your charts with a colour blindness simulator. Always provide a secondary encoding (pattern, label, position) alongside colour.

Key Takeaways

• Every chart should answer exactly one question clearly for a specific audience making a specific decision.
• Match chart type to data relationship: line for trends, bar for comparisons, scatter for relationships, box plot for distributions.
• Maximise signal-to-noise ratio: remove all visual elements that do not add analytical value.
• Never truncate bar chart y-axes at non-zero values -- this distorts the magnitude of differences.
• Colour is often misused: use one accent colour with grey for background elements, and choose accessible palettes.

Practice Exercise

Find three charts in business news, annual reports, or dashboards you use at work. For each one:

1. Identify what question it is trying to answer
2. Assess whether the chart type is appropriate
3. List all non-data-ink elements that could be removed
4. Identify any misleading design choices (truncated axes, 3D effects, too many colours)
5. Sketch a redesigned version that applies the principles in this lesson