How Graphs Lie
Edward Tufte, in "The Visual Display of Quantitative Information," systematized the ways data visualizations can deceive readers, whether intentionally or not. The most common technique is Y-axis truncation: starting the Y-axis at a value above zero to make small differences appear dramatic.
For example, comparing Company A's revenue of 10 billion yen with Company B's 9.5 billion: starting the Y-axis at zero makes the difference barely visible, but starting at 9 billion makes Company A's bar appear twice the height of Company B's. A 5% difference is visually amplified to look like 100%. The same manipulation occurs in ranking data visualizations.
Scale Selection Manipulates Perception
The choice between linear and logarithmic scales produces entirely different impressions from identical data. COVID-19 case counts displayed on a linear scale appear as "explosive growth," while the same data on a logarithmic scale shows "a constant growth rate." Both are factually correct, yet the communicated message is opposite.
Income distribution graphs face the same issue. On a linear scale, top earners dominate the visual space and differences among lower earners become invisible. A logarithmic scale makes the overall distribution legible but underrepresents the absolute magnitude of the gap between top and bottom. Scale selection is never neutral; it always emphasizes something while concealing something else.
Cherry-Picking - Selecting Convenient Time Windows
By arbitrarily choosing the display period for time-series data, one can "prove" either an upward or downward trend. Showing stock prices from March 2020 (the COVID crash) to December 2021 demonstrates "remarkable recovery," while December 2021 to December 2022 shows "severe decline."
The same caution applies when examining ranking changes over time. The claim "Japan's ranking dropped 5 places in 10 years" can be easily manipulated by choosing start and end points. Evaluating long-term trends requires a sufficiently long time horizon and verification that neither endpoint represents an anomalous year.
3D Graphs and Area Illusions
3D pie charts and 3D bar charts cause elements in the foreground to appear larger and those in the background smaller due to perspective. This is a purely visual illusion that impedes accurate data comparison. Tufte termed this "chartjunk" - decoration that contributes nothing to information transmission.
Bubble charts (representing data through circle area) are similarly prone to misinterpretation. Humans are poor at comparing areas: a circle with twice the radius "feels" twice as large, but its actual area is four times greater. In data visualization, encoding through position (scatter plots) or length (bar charts) is perceived far more accurately than encoding through area or angle (pie charts).
Design Ethics in Ranking Displays
Ranking tools like MyRank influence users' emotions and behavior through display choices. Whether "top 25%" is shown in red or green changes the emotional response to the same number. Progress bar length, font size of the figure, choice of comparison group - every design decision steers interpretation.
The principles of honest data visualization are to represent data accurately and provide readers with the information needed to form their own judgments. The goal is facilitating understanding rather than manipulating impressions: not hiding uncertainty, providing context, and presenting multiple perspectives. Viewers of rankings can also make more measured judgments by remaining aware of how display choices shape their reactions.