flowchart TD
M["Common<br>Visualisation<br>Mistakes"]
M --> T["Truthfulness<br>The chart distorts<br>the data"]
M --> C["Clarity<br>The chart is hard<br>to read"]
M --> S["Statistical<br>The chart misuses<br>statistics"]
M --> Ctx["Context<br>The chart omits<br>essential context"]
M --> A["Accessibility<br>The chart fails<br>some readers"]
M --> Au["Audience<br>The chart fits the<br>wrong reader"]
style M fill:#e3f2fd,stroke:#1976D2
style T fill:#fce4ec,stroke:#AD1457
style C fill:#fff3e0,stroke:#EF6C00
style S fill:#fff8e1,stroke:#F9A825
style Ctx fill:#e8f5e9,stroke:#388E3C
style A fill:#ede7f6,stroke:#4527A0
style Au fill:#f3e5f5,stroke:#6A1B9A
18 Avoiding Common Visualization Mistakes
18.1 Why Visualisation Mistakes Matter
A misleading chart is more dangerous than no chart at all, because it carries the false confidence of the visual.
A wrong number on a slide can be challenged. A misleading chart often passes through review unchallenged because it looks convincing. The reader trusts the picture, the picture distorts the data, and the decision is made on a false impression.
Two of the most influential books on this subject — How to Lie with Statistics by Darrell Huff (1954) and How Charts Lie by Alberto Cairo (2019) — were written for general audiences precisely because the techniques of visual deception are pervasive and the defences against them are not. The analyst who builds dashboards and reports has a corresponding obligation: not merely to avoid being deceived, but to avoid producing charts that deceive others, even unintentionally.
18.2 Categories of Visualisation Mistakes
Six recurring categories cover most of the mistakes a working analyst makes or encounters:
- Truthfulness mistakes — the chart’s visual proportions do not match the underlying data.
- Clarity mistakes — the chart is technically correct but hard to read.
- Statistical mistakes — the chart relies on or implies a misuse of statistics.
- Context mistakes — the chart omits the source, scale, period, or comparison the reader needs.
- Accessibility mistakes — the chart fails colour-blind readers, screen readers, or low-contrast environments.
- Audience mistakes — the chart is well-designed for one audience and shown to a different one.
18.3 Truthfulness Mistakes
These are the mistakes that make a chart actively misleading. Many of them are textbook examples that, sadly, still appear in business reports every week.
Truncated Y-Axis on a Bar Chart: Starting the value axis at a non-zero value visually exaggerates differences. A change from 100 to 102 looks dramatic on a chart whose y-axis runs from 99 to 103. Acceptable for line charts that show change; almost never for bar charts where length encodes magnitude.
Inverted or Reversed Axes: Plotting time backwards, or values upside-down, where convention dictates otherwise. The reader’s pattern recognition is wired for left-to-right, low-to-high.
Three-Dimensional Distortion: Adding three-dimensional effects to a two-dimensional chart. Perspective distorts perceived quantities — front bars look larger than back bars of the same height — and adds no information.
Misleading Area Encoding: Using the length of an icon when the eye reads its area (icons of cars, people, money). Doubling both the height and width quadruples the perceived quantity.
The Lie Factor: Tufte’s measure — the size of the effect shown in the graphic divided by the size of the effect in the data. A lie factor between 0.95 and 1.05 is honest; anything beyond that is distortion. Truncated axes and misleading area encoding are the most common sources of large lie factors in business charts.
Cherry-Picked Time Windows: Choosing the start and end of the time axis to exaggerate or hide a trend — for example, presenting a five-year stock chart that begins at the year of a market crash so the recovery looks dramatic.
Disproportionate Spacing on Time Axes: Plotting irregular time intervals with even spacing — for example, financial-year ends spaced equally even when the gaps between them are not.
18.4 Clarity Mistakes
These are the mistakes that make a chart difficult to read, even when it is not actively misleading.
Chartjunk: Heavy borders, drop shadows, gradient fills, decorative backgrounds, irrelevant clip-art. They reduce information density, distract attention, and signal that the designer cared more about appearance than insight.
Pie Chart Overuse: Pie charts encode by angle and area, both low on the perceptual hierarchy. For more than four slices, almost any alternative is clearer.
Too Many Series on One Chart: Five, ten, fifteen lines colliding in the middle of a single line chart. Use small multiples instead, or highlight one series and grey the rest.
Inappropriate Chart Type: Using a line chart for unordered categorical data, a stacked bar to compare totals that should be compared individually, or a treemap for non-hierarchical data.
Cluttered Legends: Long legends that the reader must repeatedly consult to decode the chart. Direct labels on the chart itself are often clearer.
Bad Colour Choices: Rainbow scales for continuous data, red-green encoding that fails for colour-blind readers, or every series saturated at maximum so nothing stands out.
Tiny Type and Cramped Labels: Axis labels and annotations small enough that a reader at the back of a meeting room cannot read them.
Missing Direct Labels: Forcing the reader to estimate values by tracing a line back to the axis when a direct label would do the work in a glance.
18.5 Statistical Mistakes
These mistakes do not break the chart; they break the reasoning the chart supports.
Misleading Averages: Reporting a mean where the median is more representative, or hiding a bimodal distribution behind a single average. The mean of incomes in a room with one billionaire is meaningless.
Confusing Correlation with Causation: Plotting two variables that move together and implying that one causes the other. The classic example is the spurious correlation between ice-cream sales and drowning deaths, both of which are driven by summer.
Simpson’s Paradox: Aggregated data showing a trend that reverses when the data is broken down by category. A famous example: a hospital that has a higher overall mortality rate than another, but lower rates in every individual case-mix category.
Survivorship Bias: Charting only the entities that survived to the present — successful firms, completed projects, returning customers — and missing what made the failures fail.
Selection Bias: A sample that does not represent the population. A satisfaction survey answered only by satisfied customers is not measuring satisfaction.
P-Hacking and Cherry-Picked Findings: Running many statistical tests and reporting only the ones that came out significant. The reader sees one chart; the analyst tested twenty.
Spurious Precision: Reporting a forecast to two decimal places when the underlying model has wide error bars. False precision misleads the reader about how confident to be.
Confidence Without Intervals: Showing a single forecast line as if it were certain, when the model produces a wide range. Always show uncertainty.
Aggregation that Hides the Story: Anscombe’s quartet — four datasets with identical means, variances, and correlations but very different distributions — shows that summary statistics can hide the structure of the data. Always visualise.
18.6 Context Mistakes
These are mistakes of omission. The chart is technically correct but the reader has too little information to interpret it.
No Title or Vague Title: A title that does not state what the chart shows, or that uses internal jargon the reader does not share.
Missing Axis Labels and Units: A reader cannot tell whether the value is in rupees, lakhs, crores, or thousands of crores.
Missing Source and As-of Date: A reader does not know where the data came from or when it was current.
No Comparison or Target: A number with no benchmark cannot be interpreted. Sales were 92 says nothing without target was 100, last year was 87, industry average is 95.
No Per-Capita or Population Adjustment: Comparing absolute counts across populations of different sizes. Mumbai will always have more cases of anything than Aizawl.
Currency Without Adjustment: Comparing rupee values across years without inflation adjustment, or across countries without exchange-rate normalisation.
No Annotation of Anomalies: A spike in the data corresponding to a known event — a holiday, a strike, a system migration — that goes unannotated. The reader is left to guess.
18.7 Accessibility Mistakes
A chart that excludes a portion of its readers has failed those readers. The most common accessibility errors:
Red-Green Encoding: Using red and green as the only contrast for important categorical distinctions. Approximately eight per cent of male readers cannot reliably tell them apart.
Colour as the Sole Encoding: Conveying information only through colour, with no shape, label, or position fall-back. The chart fails for colour-blind readers, in greyscale printing, and on poorly-calibrated projectors.
Low Contrast: Light grey text on a white background; pale colours that fall below WCAG contrast thresholds.
Tiny Type: Labels and annotations below 9 or 10 pt that low-vision readers and those at the back of a room cannot see.
No Alt Text: Charts in digital reports without descriptive alt text or a fall-back data table for screen-reader users.
Animation Without an Accessible Static Equivalent: Important findings conveyed only through animated transitions, with no static image that captures the same insight.
18.8 Audience and Purpose Mistakes
These mistakes are not about the chart itself but about the match between the chart and the reader.
Designing for the Designer: Producing the chart the analyst finds satisfying rather than the chart the audience needs.
Wrong Level of Detail: Showing transaction-level data to a board member, or aggregate strategic KPIs to a front-line operator.
Wrong Cadence: Daily-refresh charts in a quarterly board pack, or quarterly KPIs on an operations control screen.
Wrong Medium: A complex interactive view exported as a screenshot for a printed report.
Wrong Mental Model: A chart whose underlying definition does not match how the audience thinks about the metric. Customer churn defined by the analytics team is not always the same metric as customer churn in the operating manager’s head.
Wrong Tone: A celebratory chart for a target that the team narrowly missed; a doom-laden chart for a target the team comfortably exceeded.
18.9 A Diagnostic Checklist
flowchart LR
A["1. Does the chart<br>answer one clear<br>question?"] --> B["2. Are the visual<br>proportions honest?"]
B --> C["3. Is the chart<br>type appropriate?"]
C --> D["4. Is the context<br>complete?"]
D --> E["5. Will every<br>reader read it?"]
E --> F["6. Will it survive<br>its medium?"]
F --> G["7. What action<br>does it imply?"]
style A fill:#fce4ec,stroke:#AD1457
style B fill:#fff3e0,stroke:#EF6C00
style C fill:#fff8e1,stroke:#F9A825
style D fill:#e3f2fd,stroke:#1976D2
style E fill:#ede7f6,stroke:#4527A0
style F fill:#e8f5e9,stroke:#388E3C
style G fill:#f3e5f5,stroke:#6A1B9A
A seven-question checklist to run on any chart before publication:
- Does the chart answer one clear question? If you cannot state the question in a sentence, the chart is too vague.
- Are the visual proportions honest? Lie factor near one. No truncated bar axes. No misleading area encoding.
- Is the chart type appropriate? Bar for categorical comparison, line for trend, scatter for relationship, choropleth-of-rates for geography.
- Is the context complete? Title, axes, units, source, period, comparison or target, key annotations.
- Will every reader read it? Greyscale-safe, colour-blind-safe, large-enough type, alt-text or fall-back table for digital.
- Will it survive its medium? Print, projector, mobile, email — does it still read?
- What action does it imply? A chart that does not lead to a decision is decoration.
18.10 Common Pitfalls Catalogue
A consolidated catalogue of the recurring mistakes covered above:
- Truncated bar-chart y-axis that exaggerates small differences.
- Three-dimensional distortion that adds no data and warps perception.
- Misleading area icons where doubling height and width quadruples perceived quantity.
- Cherry-picked time windows that frame the trend the designer prefers.
- Pie chart with too many slices that the eye cannot decode.
- Too many lines on one chart that the reader cannot follow.
- Rainbow scales for continuous data that introduce bands not in the data.
- Red-green encoding that fails for the most common form of colour-vision deficiency.
- Misleading averages that hide bimodal or skewed distributions.
- Correlation framed as causation that suggests intervention will produce the observed change.
- Simpson’s paradox unaddressed by the design.
- Survivorship bias in the data the chart is built on.
- Spurious precision that misleads the reader about model uncertainty.
- Missing context — no title, no units, no source, no period.
- Missing per-capita or inflation adjustment when comparing populations or values across time.
- No alt text or fall-back data table for digital screen-reader users.
- Wrong audience match — the chart is well-designed for the wrong reader.
- Charts that imply no action for the audience.
18.11 Illustrative Cases
The following short cases illustrate how common mistakes turn into misleading charts and how a redesign restores the truth.
A Sales Chart with a Truncated Axis
A regional sales team presents a quarterly performance chart with the y-axis running from 95 to 105. The drop from 102 in Q2 to 99 in Q3 looks like a collapse. The redesign anchors the y-axis at zero; the same chart now shows a small dip in a series that has remained between 95 and 105 throughout. The redesign also adds the target as a reference line at 100 and the prior-year average at 98, so the dip can be interpreted in context.
A Choropleth Map of Disease Cases
A public-health dashboard shades districts by absolute case count. Large rural districts dominate visually because they are larger, not because they have more incidence. The redesign uses cases per ten thousand population and a perceptually uniform palette. The geographical pattern of incidence becomes visible at last.
A Trend Chart Without Inflation Adjustment
A government economic dashboard plots a thirty-year revenue series in nominal rupees. The chart shows steep “growth” that is largely the result of inflation. The redesign plots the same series in inflation-adjusted constant rupees and adds a second axis with the nominal series in muted grey for reference. The honest growth picture is far more modest, and the comparison with the nominal trend itself becomes a useful element of the chart.
A Dashboard With No Refresh Time
A weekly operations dashboard does not display the time of last data refresh. After a pipeline failure, the dashboard continues to show last-week numbers as if they were current; managers act on the stale figures for two days before the issue is noticed. The redesign adds a fixed footer with the last-refresh timestamp and an alert indicator when any panel’s data is more than 24 hours old.
A Chart Designed for the Designer
An analyst presents a complex slope-graph with twelve overlapping lines and a custom colour palette to a board audience. The board cannot decode it in the time available. The redesign replaces it with three simple bar charts answering the three questions the board actually had: are we on plan, where is variance, what is the projection. The original chart returns to the appendix where it serves the analyst-audience for which it was originally designed.
18.12 Hands-On Exercise: Chart Redesign Workshop
Aim: Take three flawed charts representative of the categories of mistake covered in this chapter, diagnose what is wrong, and rebuild each in Power BI or Excel applying the redesign principle.
Scenario: An analyst at Yuvijen Stores Pvt Ltd has inherited a deck of charts from the previous quarter. Three of them are clearly flawed, and the new analyst’s first task is to redesign them before the next quarterly review.
Deliverable: A Power BI or Excel report with three before-and-after pages — the original chart on the left, the redesigned chart on the right, and a one-line note explaining the fix.
18.12.1 Chart A — The 3D Pie Chart
TipThe Original
| Source data: monthly revenue share by category | |
|---|---|
| Kitchen | 36 % |
| Bath | 22 % |
| Apparel | 14 % |
| Home | 11 % |
| Stationery | 7 % |
| Toys | 5 % |
| Garden | 3 % |
| Pet Care | 2 % |
The original chart is a 3D pie chart with these eight slices. The three front slices appear visually larger than the slices of equal size at the back, the small slices on the rim are illegible, and the angles cannot be compared accurately by eye.
TipDiagnosis and Redesign
| Step | What to do |
|---|---|
| Diagnose | Clarity mistake (pie with too many slices), Truthfulness mistake (3D distortion), encoding by angle and area both low on Cleveland’s hierarchy |
| Redesign principle | Replace with a sorted horizontal bar chart; encode by length on a common scale |
| Tool | Power BI |
| Steps | 1) Insert a Bar Chart (horizontal). 2) Drag category to Y-Axis and share to X-Axis. 3) Sort descending. 4) Format: remove gridlines, add direct percentage labels at bar ends. 5) Group the smallest three categories into “Other” if the audience does not need them individually. |
The redesigned chart reads in a glance, the comparison is honest, and the small categories no longer compete for attention with the headline finding.
18.12.2 Chart B — The Truncated-Axis Bar Chart
TipThe Original
| Quarterly customer-satisfaction scores | |
|---|---|
| Q1 2025 | 78 |
| Q2 2025 | 80 |
| Q3 2025 | 82 |
| Q4 2025 | 81 |
| Q1 2026 | 83 |
The original is a vertical bar chart whose y-axis runs from 75 to 85. The bar for Q1 2026 looks roughly twice as tall as the bar for Q1 2025, when in fact the underlying value has risen by only about 6 per cent.
TipDiagnosis and Redesign
| Step | What to do |
|---|---|
| Diagnose | Truthfulness mistake — truncated y-axis on a bar chart exaggerates change; the lie factor is well above 1 |
| Redesign principle | For a bar chart, the y-axis must start at zero. If small differences need to be visible, switch to a line chart with a clearly labelled non-zero axis, or annotate the percentage change directly. |
| Tool | Excel |
| Steps | 1) Highlight the data range and Insert → Recommended Charts → Clustered Column. 2) Right-click the y-axis → Format Axis → Minimum: 0. 3) Add a horizontal reference line at the long-term average using Add Chart Element → Lines. 4) Add a data label +5 between Q1 2025 and Q1 2026 showing the actual change. |
The redesigned bar chart looks visually less dramatic, which is the honest answer. If the team wants to emphasise the trend, the right move is a separate line chart with explicit percentage-change annotations — not an axis trick.
18.12.3 Chart C — The Choropleth of Counts
TipThe Original
| Region | Customers (count) | Adult Population (lakh) |
|---|---|---|
| West | 145,000 | 850 |
| South | 132,000 | 720 |
| North | 121,000 | 980 |
| East | 88,000 | 540 |
| Central | 54,000 | 340 |
| North-East | 19,000 | 120 |
The original is a choropleth map of India shaded by absolute customer counts. The northern and western regions appear darkest because they are the most populous, not because the firm has the strongest customer presence relative to opportunity.
TipDiagnosis and Redesign
| Step | What to do |
|---|---|
| Diagnose | Statistical and Context mistake — comparing absolute counts across populations of different sizes; the chart shows demographics, not business performance |
| Redesign principle | Convert to a rate — customers per 100,000 adult population — and use a perceptually uniform sequential palette |
| Tool | Power BI |
| Steps | 1) Add a calculated column Penetration = Customers / AdultPopLakh / 1000. 2) Insert a Filled Map visual. 3) Bind Region to Location and Penetration to Color saturation. 4) Pick a sequential palette such as viridis (built-in to Power BI under Format → Data colours → Diverging). 5) Add direct labels for the top three and bottom three regions. |
The redesigned choropleth answers the right question — where is our market penetration strongest and weakest? — rather than the wrong one — where do most Indians live? — that the original silently asked.
18.12.4 Compose the Before-and-After Report
Build a single Power BI report (or Excel workbook) with three pages, one per chart. Each page is split into two halves:
- Left: the original chart, faithfully reproduced.
- Right: the redesigned chart.
- Footer: a one-line note stating the diagnosis and the redesign principle (e.g. Diagnosis: 3D pie with too many slices. Fix: sorted horizontal bar.)
The before-and-after format makes the lesson teachable. It can be reused as a training artefact for the rest of the analytics team and for new joiners.
18.12.5 Connecting the Workshop to the Visualisation Programme
A useful institutional discipline is to maintain a redesign portfolio — a small collection of before-and-after pairs from the firm’s own past charts — and to add to it whenever a new redesign is done. Over time, the portfolio becomes:
- A training resource for new analysts.
- A lookbook for visualisation review meetings.
- A safeguard against the same mistakes recurring across teams.
Power BI’s Bookmarks feature and Tableau’s Story Points are well suited to building this kind of progressive narrative as a single shareable artefact.
TipFiles and Screen Recordings
Power BI report (yuvijen-chart-redesign.pbix), Excel workbook (yuvijen-chart-redesign.xlsx), the original flawed charts as images, and screen recordings of each redesign will be embedded here.
Summary
| Concept | Description |
|---|---|
| Foundations | |
| Why Visualisation Mistakes Matter | A misleading chart is more dangerous than no chart, because it carries the false confidence of the visual |
| Six Categories of Mistakes | |
| Truthfulness Category | Mistakes where the chart's visual proportions do not match the underlying data |
| Clarity Category | Mistakes where the chart is technically correct but hard to read |
| Statistical Category | Mistakes where the chart relies on or implies a misuse of statistics |
| Context Category | Mistakes where the chart omits source, scale, period, or comparison the reader needs |
| Accessibility Category | Mistakes where the chart fails colour-blind, screen-reader, or low-contrast readers |
| Audience Category | Mistakes where the chart fits the wrong reader for its purpose |
| Truthfulness Mistakes | |
| Truncated Y-Axis | Pitfall of starting a bar y-axis at a non-zero value and exaggerating differences |
| Inverted Axes | Pitfall of plotting time backwards or values upside-down against convention |
| Three-Dimensional Distortion | Pitfall of adding 3D effects that warp perceived quantities and add no information |
| Misleading Area Encoding | Pitfall of doubling both height and width of an icon and quadrupling perceived quantity |
| Lie Factor | Tufte's measure: effect shown in the graphic divided by effect in the data; near one is honest |
| Cherry-Picked Time Windows | Pitfall of choosing axis start and end to frame the trend the designer prefers |
| Disproportionate Time Spacing | Pitfall of plotting irregular time intervals with even spacing |
| Clarity Mistakes | |
| Chartjunk | Decorative material that carries no data and slows the reader |
| Pie Chart Overuse | Pitfall of pie charts with more than four slices that the eye cannot decode |
| Too Many Series | Pitfall of many series on one chart that the reader cannot follow |
| Inappropriate Chart Type | Pitfall of choosing a chart type that does not match the question |
| Cluttered Legends | Pitfall of long legends the reader must repeatedly consult to decode the chart |
| Bad Colour Choices | Rainbow scales, red-green encoding, or every series at maximum saturation |
| Tiny Type | Pitfall of axis labels and annotations small enough to fail readers at the back of a room |
| Missing Direct Labels | Pitfall of forcing readers to trace lines back to the axis when direct labels would suffice |
| Statistical Mistakes | |
| Misleading Averages | Pitfall of reporting means where medians are more representative or hiding bimodality |
| Correlation vs Causation | Pitfall of plotting two variables that move together and implying causation |
| Simpson's Paradox | Aggregated data showing a trend that reverses when broken down by category |
| Survivorship Bias | Drawing lessons only from successful entities while failures remain invisible |
| Selection Bias | Pitfall of a sample that does not represent the population it claims to describe |
| P-Hacking | Pitfall of running many tests and reporting only those that came out significant |
| Spurious Precision | Pitfall of reporting forecasts to false precision when the underlying error is wide |
| Confidence Without Intervals | Pitfall of showing a single forecast line as if it were certain when the model has range |
| Aggregation Hiding the Story | Anscombe-style problem: identical summary statistics can conceal very different distributions |
| Context Mistakes | |
| No Title or Vague Title | Pitfall of titles that do not state what the chart shows or use internal jargon |
| Missing Axis Labels and Units | Pitfall of leaving the reader unable to tell whether values are rupees, lakhs, or crores |
| Missing Source and Date | Pitfall of leaving the reader without source attribution or as-of date |
| No Comparison or Target | Pitfall of a number with no target, no benchmark, and no implied baseline |
| No Per-Capita Adjustment | Pitfall of comparing absolute counts across populations of different sizes |
| Currency Without Adjustment | Pitfall of comparing rupee values across years without inflation adjustment |
| No Annotation of Anomalies | Pitfall of leaving spikes from known events unannotated and unexplained |
| Accessibility Mistakes | |
| Red-Green Encoding | Pitfall of red-green pairs that fail the most common form of colour-vision deficiency |
| Colour as Sole Encoding | Pitfall of conveying information only through colour with no shape, label, or position cue |
| Low Contrast | Pitfall of light grey on white or pale colours that fail WCAG thresholds |
| Tiny Type Accessibility | Pitfall of labels below 9 or 10 pt that low-vision readers cannot see |
| No Alt Text | Pitfall of digital charts without descriptive alt text or fall-back data table |
| Inaccessible Animation | Pitfall of important findings conveyed only through animated transitions |
| Audience and Purpose Mistakes | |
| Designing for the Designer | Pitfall of producing the chart the analyst finds satisfying rather than what the audience needs |
| Wrong Level of Detail | Pitfall of transaction-level data shown to a board or strategic KPIs shown to operators |
| Wrong Cadence | Pitfall of daily charts in a quarterly board pack or quarterly KPIs on an operations screen |
| Wrong Medium | Pitfall of complex interactive views exported as screenshots for printed reports |
| Wrong Mental Model | Pitfall of a chart whose definition does not match how the audience thinks about the metric |
| Wrong Tone | Pitfall of celebratory tone for a missed target or doom-laden tone for an exceeded one |
| Diagnostic Checklist | |
| One Clear Question Check | If you cannot state the chart's question in a sentence, the chart is too vague |
| Honest Proportions Check | Lie factor near one; no truncated bar axes; no misleading area encoding |
| Appropriate Chart Type Check | Bar for categorical comparison, line for trend, scatter for relationship, choropleth-of-rates for geography |
| Complete Context Check | Title, axes, units, source, period, comparison or target, key annotations all present |
| Every-Reader Check | Greyscale-safe, colour-blind-safe, large-enough type, alt text for digital |
| Survives-Medium Check | Print, projector, mobile, email; does the chart still read on the medium it will travel through |
| Implied Action Check | A chart that does not lead to a decision is decoration; design implies action |