Between the Truths · 02
Christian Hardinghaus · October 2026
Visitor numbers have risen by 100 percent. A success? That depends on where the story begins.
An imaginary museum records 100 visitors in January, 80 in February and 160 in March. Its director announces: “Visitor numbers double!” A critic replies: “Interest fell by 20 percent in February.” Both calculations are correct. Each tells only part of the story.
February → March
80 → 160: up 100 percent
January → March
100 → 160: up 60 percent
Invented arithmetic example; not data from an actual museum.
The selection becomes the story
Cherry picking selects evidence from a larger body of material primarily because it supports a preferred conclusion. Card stacking assembles information and arguments into a one-sided case. The terms overlap. Our concern here is the question they share: what are we not being shown?
The museum’s claim becomes misleading when it suggests more than the selected comparison establishes. The rise from February to March demonstrates a doubling against the previous month. It establishes neither lasting success nor its cause. Was the building partly closed in February? Was admission free in March? Such context could change our judgment.
A fact-check may need the missing row
Finding a quoted figure in the original table is not always enough. We also need to examine the selection. Why that starting point? Why that group? Why that particular measure?
Research methodology addresses the selective availability of results. The Cochrane Handbook distinguishes missing studies or results from selecting a reported result from several possible analyses. Both can bias an assessment. [1] This does not make every short summary manipulative. It helps explain why transparent selection rules matter.
Not every omission is cherry picking
A report needs boundaries. An article about the past three months may reasonably focus on those months. The problem arises when that limited selection is used to make unsupported claims about years, entire populations or general causes.
A counterexample does not disprove every general statement either. An individual case may differ from an average without contradicting it. A claim about every single case, on the other hand, can be disproved by a verified exception. We need to identify the kind of claim being tested before deciding what counts against it.
Three questions to ask
- Which timeframe, group and measure have been selected?
- Does the conclusion remain plausible when I include the surrounding data?
- Are the selection rules explained, or do we see only convenient results?
Back to the museum
A fair short account might say: “After falling from 100 to 80 visitors, attendance rose to 160 in March. These three monthly figures tell us little about the causes or the long-term trend.” Less triumphant, less dramatic, but closer to what we actually know.
The next time a chart looks particularly convincing, look towards its edges. The crucial information may be the timeframe underneath it rather than the most impressive bar.
Explore the mechanism
Cherry Picking · Card Stacking
Sources
- Matthew J. Page, Julian P. T. Higgins and Jonathan A. C. Sterne: Assessing risk of bias due to missing evidence in a meta-analysis. Cochrane Handbook, Chapter 13, version 6.5 (2024), especially section 13.1. Methodological source on selective availability of research results; the museum example is an original illustration.
Examples without a source reference are original illustrations.
