Overview

1. Background and Motivation
2. Taxonomy of Dishonest Chart Techniques
3. Related Work

Dishonest Charts

• Visualizations can convey a misleading impression.
• Occur by accident or design.
• Data is always biased and never objective or neutral.
• Result from design choice or data manipulation.
• Effects on interpretation:
  • Patterns
  • Trends
  • Relationships

9 Dishonest Chart Techniques

Cherry Picking

• Selecting only specific data points, groups, or timeframes that support desired outcome.
• Deliberately hides broader context or opposing trends.
  • Showing short-term cooling trend to deny long-term warming.

Cherry Picking Demo

Cherry Picking Example

• Judd et al. (2024) Climate Reconstruction.¹
• Uses a massive 485-million-year geological timeline.
• Visually compresses and hides rate of modern warming.

Underbinning

• Reducing number of bins (e.g., in histograms or heatmaps).
• Obscures true underlying distribution.
• Hides:
  • Important clusters
  • Gaps
  • Outliers

Underbinning Demo

Underbinning Example

• Distribution of NBA player heights.¹
• Too few bins (overly broad intervals).
• Data grouped into coarse categories, true distribution is hidden.

Axis Truncation

• Truncating the baseline
• Chart where y-axis does not start at zero.
• Differences appear larger.

Axis Truncation Demo

Axis Truncation Example

• Fox News Tax Cuts Broadcast (2012).¹
• Truncated y-axis (starts at 34% instead of 0%).
• Exaggerates minor 4.6% tax increase, making second bar appear much larger.

Chart recreated from original in Python by the authors.

Artificial Flattening or Steepening

• Flattening chart's visual appearance:
  • Extreme aspect ratios.
  • Overly broad y-axis.

Artificial Flattening / Steepening Demo

Time Gap

• Using uneven or irregular time intervals on continuous axis but spacing them evenly.

Time Gap Demo

Oversmoothing

• Applying aggressive smoothing algorithms or over-fitted curves to raw data points.
• Hides critical short-term fluctuations, volatility, or local extremes.

Oversmoothing Demo

Manipulating the Visual Encoding

• Manipulates choice of visual encoding.
• Often by misleading colour, size and shape (encoding).

Visual Encoding Demo 1: Size

Manipulating the Visual Encoding Example 1

• "Shrinking Family Doctor" (Time Magazine, 1990).¹
• Maps a 1D percentage drop to a 2D icon's height.
• Viewers perceive area, making the 1990 doctor look drastically smaller than the data supports.

Chart recreated from original in Python by the authors.

Visual Encoding Demo 2: Color Mapping

Perspective Distortion

• Misuses 3d perspective.
• For example, tilting a pie chart to make foreground slices look larger.

Perspective Distortion Demo

Perspective Distortion Example

• Context: Apple Macworld Keynote (2008),presented by Steve Jobs.¹
• Misuse: 3D tilt pushes the 19.5% slice into the middle.
• Effect: Perspective distortion makes 19.5% appear physically larger than 21.2%.

Chart recreated from original in Python by the authors.

Concealing Uncertainty

• Not including:
  • Error bars.
  • Confidence intervals.
  • Variance indicators.
• Suggests absolute precision in data that actually estimates or forecasts where there is uncertainty.

Concealing Uncertainty Demo

Related Work

• Awesome Misleading Visualizations: Curated GitHub List.
  https://github.com/UKPLab/awesome-misleading-visualizations
• FlawViz: Collection of Misleading Data Visualization.
  https://flawviz.github.io/
• FlowingData: Defensive Chart Techniques by Nathan Yau.
  https://flowingdata.com/projects/dishonest-charts/
• How Charts Lie: Book by Alberto Cairo.
  https://wwnorton.com/books/9781324001560