Chart selection is a decision about what comparison to make visible. A bar chart answers a different question than a line chart; a scatter plot reveals patterns invisible in a table. This guide covers which chart types to use for which analytical purposes, common visualisation mistakes that distort data, and design principles for charts that communicate accurately.
Visualisation as Communication
A data visualisation is not a decoration for data — it is a communication of a specific insight. Every chart choice, every colour decision, every axis scale is a design decision that either helps the reader understand faster or slows them down.
The most common visualisation failure is chart selection by familiarity rather than by analytical purpose. A bar chart is used because it looks professional. A 3D pie chart is used because it fills space. The result is charts that show data but do not communicate insight.
Effective visualisation starts with the question: what comparison am I trying to make visible? The answer determines the chart type.
Chart Selection by Analytical Purpose
**Comparing values across categories**: Bar chart or column chart. Bars are positioned along a common axis, enabling accurate visual comparison of relative size. Use horizontal bars when category labels are long. Order by value (descending) to make rankings immediately readable — do not default to alphabetical order.
**Trends over time**: Line chart. Time on the horizontal axis; the connected line makes trend direction immediately visible. Multiple lines compare trends across series. Avoid lines with fewer than 4-5 data points — use a bar chart instead; a line implies continuity between points that may not exist with sparse data.
**Distribution of values**: Histogram (frequency distribution of a continuous variable), box plot (quartile distribution with outliers), violin plot (distribution shape including bimodality). Average-per-category bar charts hide distribution — a region with average revenue of $50k could have tightly clustered performance or extreme variance; a bar chart shows neither.
**Part-to-whole composition**: Stacked bar chart for comparing both total and component breakdown simultaneously. 100% stacked bar for comparing composition percentage only. Pie charts only when there are 2-4 slices and proportion is the primary message — the angular comparison of pie slices is less accurate than length comparison in bars.
**Relationship between two variables**: Scatter plot. Each point represents one observation with X and Y coordinates from two variables. Reveals correlation, clusters, and outliers invisible in any aggregate view. Add a trend line (linear regression) to make correlation direction and strength explicit.
**Single numeric value**: KPI card — large number, trend indicator (up/down arrow), comparison to target or prior period. Do not use a chart when the answer is a single number.
**Geographic distribution**: Choropleth map (regions coloured by value intensity). Use only when geography is genuinely relevant — if the insight is "West region is underperforming," a bar chart is clearer. Maps look impressive but require the reader to know the geography and decode a colour scale simultaneously.
Common Visualisation Mistakes
**Truncated Y-axis**: Starting the Y-axis at a value greater than zero exaggerates differences. A bar chart with a Y-axis starting at $980,000 makes a $10,000 difference look enormous. Always start bar charts at zero. For line charts, starting above zero to show trend variance is acceptable — but label the axis clearly.
**3D charts**: Three-dimensional effects distort perception. The back bars in a 3D bar chart appear shorter than identically-sized front bars due to perspective foreshortening. There is no analytical purpose that 3D serves — remove it.
**Dual Y-axes**: Two Y-axes on the same chart allow the creator to manipulate the apparent relationship between two series by adjusting scale. Any correlation can be made to appear strong or weak by scaling the axes differently. Use dual axes only when the relationship between the two series is the specific insight, and label both axes clearly.
**Too many colours**: Colour should encode one dimension of the data — series membership, performance vs target, positive vs negative. Using 10 different colours for 10 categories requires constant reference to the legend. Limit to 4-6 meaningful colours; make non-focal series grey.
**Pie charts with many slices**: Angles are difficult to compare accurately. When a pie chart has more than 4-5 slices, the small slices are indistinguishable. Use a bar chart ordered by size to make rankings legible.
**Sorting alphabetically instead of by value**: Alphabetical sorting of bar charts forces the reader to mentally reorder to find rankings. Sort by value (descending) unless there is a meaningful natural order (months, age bands, geographic hierarchy).
Colour in Data Visualisation
Colour is one of the most misused design elements in data visualisation:
**Sequential palettes**: For data with a natural progression from low to high. Light colour = low value; dark colour = high value. Example: revenue density, temperature, population count. Use a single-hue progression — light blue to dark blue — or a perceptually uniform multi-hue scale.
**Diverging palettes**: For data centred around a meaningful midpoint (zero, target, average). One hue for positive/above, another for negative/below, neutral for the midpoint. Example: variance from target (green = above, red = below), year-over-year growth (positive/negative). Do not use diverging palettes for data without a meaningful midpoint.
**Categorical palettes**: For nominal categories with no inherent order. Use distinct hues that are distinguishable. Never use colour-coded categories without a legend.
**Accessibility**: 8% of men are red-green colour-blind. Using red/green as the only distinction between good/bad performance excludes a significant portion of your audience. Add text labels, patterns, or shape encoding alongside colour.
**Brand colours**: Using brand colours for BI purposes often creates perceptual problems — a brand's primary red may imply danger in a BI context. Separate brand colour from analytical colour encoding.
Titles and Labels
**Chart titles**: State what the chart shows. Not "Revenue Chart" but "Monthly Revenue by Region, Q1 2024." Better: state the conclusion — "West Revenue Declined 8% While East Grew 15%."
**Axis labels**: Both axes need labels. The Y-axis label "Revenue (USD thousands)" is more useful than just the numbers. Include units.
**Data labels**: Label specific data points when precision matters. Avoid labelling every bar in a 20-bar chart — it creates clutter. Label the highest, lowest, and any notable outliers.
**Annotation**: Use call-out annotations to explain anomalies — "Holiday promotion" above a revenue spike, "System outage" below a traffic drop. Contextual annotations prevent misinterpretation of unusual data points.
The Preattentive Attribute Principle
Certain visual properties are processed preattentively — before conscious attention, in under 250ms:
- Length (bar length, line position) — most effective for quantitative comparison
- Colour hue — effective for categorical distinction
- Colour intensity — effective for sequential data
- Size/area — effective for rough magnitude, inaccurate for precise comparison
- Position along a common scale — most accurate quantitative comparison
Design charts to use preattentive attributes for the most important comparisons. The primary message should be visible in a preattentive glance; secondary detail can require closer reading.
Our BI strategy practice designs analytics and visualisation standards for enterprise BI environments — contact us to discuss your dashboard and reporting requirements.
A former Microsoft data architect audits your data foundation, identifies your top priorities, and sends you a written plan. Free. No pitch.
Book a Call →