Illustration: a dossier collage of generic campaign-era posters, loudspeaker trucks and rally crowds.

Card stacking

Illustration, not an archive object

Also called Selective omission, Cherry-picking

Select, omit, or order facts so the remaining pile can only be read one way.

The tell

Missing denominator, missing counterexample, missing dates.

Not this

A short item that flags its own limits.

History and effect

IPA 1937; the technique behind most propaganda that never states an outright falsehood, since every individual fact used can be accurate.

It leaves the audience with a factually correct but structurally misleading picture, without ever asking them to believe something false.

Lineage: IPA 1937

How it works

Card stacking works on the pile, not the individual card. Each item can be accurate and checkable; the distortion lies in which items were chosen, which were left out, and the order they are laid down in. The Institute for Propaganda Analysis listed it among its seven devices in 1937, and Alfred McClung Lee and Elizabeth Briant Lee's The Fine Art of Propaganda (1939) described it as assembling the best or worst possible case by choosing among facts, illustrations and statements. The IPA's wording allowed falsehoods into the stack; the harder form to catch is built entirely from true material.

The moving parts recur across eras. Selection: favourable cases, studies or witnesses appear and unfavourable ones are dropped. Missing context: a count without its denominator, a change without its baseline, a figure without its date range. Darrell Huff's How to Lie with Statistics (1954) catalogued the statistical forms: choosing which "average" to report, cutting off a chart axis to magnify a small change, sampling for convenience. Ordering: a strong verified item placed first lends credibility to weaker items after it. Choice of comparison: the categories in a table are picked so that one side wins.

Two variants need institutional power. In suppression, the unfavourable evidence is removed from public view, so the remaining pile is the only one available. In the structural variant, no single author need intend the result: many separate decisions about what is published or passed upward add up to a skewed record, as with unpublished drug trials or intelligence caveats filtered out before a final briefing.

Why it works

Readers judge from what is in front of them. Daniel Kahneman (Thinking, Fast and Slow, 2011) calls this "what you see is all there is": fast judgement builds the most coherent story from the evidence available and does not ask how much is missing. Consumer researchers Frank Kardes and colleagues describe the same pattern as omission neglect. Base-rate neglect (Kahneman and Tversky, 1973) explains why a vivid count without a denominator persuades.

Card stacking also defeats the usual defence. A reader who spot-checks claims finds each one true, and Timothy Levine's truth-default theory (2014) notes that people assume honesty until something triggers suspicion; accurate items give no trigger. For audiences already inclined to the conclusion, confirmation bias (reviewed by Raymond Nickerson, 1998) makes a one-sided pile feel like confirmation rather than selection.

Examples across eras and sides

  • The Bryce Report (1915, United Kingdom, government committee). The Committee on Alleged German Outrages published summaries of depositions from Belgian refugees and soldiers, with witnesses unnamed and not cross-examined, placing documented mass killings beside unverified mutilation stories so that the first lent credibility to the second. Historians later established that German troops killed about 6,500 Belgian and French civilians in 1914, while finding many of the most lurid stories unsubstantiated (Horne and Kramer, German Atrocities 1914: A History of Denial, 2001).
  • "Sind wir die Barbaren?" (1916, Germany, Imperial government). Germany answered the "barbarism" charge with a chart comparing itself with England and France on selected measures of culture and social provision, categories that favour Germany and do not engage the accusations about Belgium. See Sind wir die Barbaren? (Are We the Barbarians?) (Landesarchiv Baden-Württemberg, J 151 Nr 2243).
  • The suppressed 1937 census (USSR, Soviet government). When the census counted fewer people than the leadership's public statements implied, a shortfall consistent with famine and repression deaths, its results were declared defective and withheld, leading census officials were arrested, and a 1939 census was published instead (Catherine Merridale, "The 1937 Census and the Limits of Stalinist Rule", The Historical Journal, 1996).
  • The Frank Statement (1954, United States, tobacco industry). The companies answered new lung-cancer studies by citing only the doubters. In United States v. Philip Morris (D.D.C. 2006), Judge Gladys Kessler found it part of a long effort to maintain public doubt. See A Frank Statement to Cigarette Smokers (Allan M. Brandt, The Cigarette Century, 2007).
  • Gulf of Tonkin and Iraq (1964 and 2003, United States, executive branch). The NSA historian Robert Hanyok concluded that the 4 August 1964 attack did not occur and that signals intelligence was presented so as to deny decision makers a complete picture (President Johnson's Gulf of Tonkin 'Midnight Address'). In 2003 Colin Powell showed the UN mobile weapons labs whose main source, "Curveball", had been doubted inside U.S. and German intelligence; the Senate Intelligence Committee (2004) faulted the intelligence community's handling of the source; its report examined the intelligence agencies, not how policymakers used their product (Mobile Production Facilities for Biological Agents (UN Security Council slide)).
  • Unpublished antidepressant trials (2008 study, United States, pharmaceutical research). Comparing 74 FDA-registered trials of 12 antidepressants with the literature, Erick Turner and colleagues found about 94% of published trials positive against about 51% by the FDA's analysis, because negative trials were mostly unpublished or reported with a positive slant (Turner et al., New England Journal of Medicine, 2008).
  • The Gyirong flood toll (2026, China, public security authorities). Official coverage centred on rescue effort and a toll of 43 dead on the Tibet side, while police punished higher counts as "rumours" and platforms deleted them. The higher counts are unverified; the tag rests on what was left visible to check the toll with. See The Gyirong flood and the official death toll (the Guardian, 8 September 2026).

How to spot it

  • What is the denominator: the population, period and rate the count belongs to?
  • What is the strongest opposing evidence, and is it mentioned at all?
  • Who chose the comparison, time window and baseline, and would another reasonable choice reverse the picture?
  • Are verified and unverified items mixed, and does the message say which is which?
  • Is missing evidence merely absent here, or removed from public view (deleted posts, withheld data, unpublished trials)?
  • Can the full set be checked: a trial registry, the full dataset, the original report?

Where it ends: edge cases and legitimate persuasion

  • Openly partisan advocacy. A lawyer's brief or campaign advertisement presents one side's best case in a genre everyone recognises. It becomes card stacking when a one-sided selection is presented as neutral or complete.
  • Summary is not stacking. Every short message omits. The test is materiality: would the omitted fact reverse or substantially weaken the conclusion? U.S. securities law draws a similar line in SEC Rule 10b-5, which forbids omitting a material fact needed to keep statements from misleading.
  • Neighbours. Partial disclosure offered as full is Limited hangout; a disclosure that does not travel with the message is Disclosure theater; Bothsidesism stacks the other way, presenting lopsided evidence as an even split; Whataboutism answers an accusation with a selected counter-comparison.
  • Falsified numbers. Once figures are altered rather than selected, the technique is plain falsehood.
  • Honest selection states its selection rule, gives denominators and date ranges, links the full data and names the strongest counter-evidence. Clinical-trial pre-registration, required by the International Committee of Medical Journal Editors from 2005, builds this into an institution.

Sources

  • Brandt, Allan M. The Cigarette Century. Basic Books, 2007.
  • Committee on Alleged German Outrages (Bryce Committee). Report of the Committee on Alleged German Outrages. HMSO, 1915.
  • Hanyok, Robert J. "Skunks, Bogies, Silent Hounds, and the Flying Fish: The Gulf of Tonkin Mystery, 2-4 August 1964." Cryptologic Quarterly, National Security Agency (declassified 2005).
  • Horne, John, and Alan Kramer. German Atrocities 1914: A History of Denial. Yale University Press, 2001.
  • Huff, Darrell. How to Lie with Statistics. W. W. Norton, 1954.
  • Institute for Propaganda Analysis. "How to Detect Propaganda." Propaganda Analysis, 1937.
  • Kahneman, Daniel. Thinking, Fast and Slow. Farrar, Straus and Giroux, 2011.
  • Kahneman, Daniel, and Amos Tversky. "On the Psychology of Prediction." Psychological Review, 1973.
  • Lee, Alfred McClung, and Elizabeth Briant Lee. The Fine Art of Propaganda. Harcourt, Brace, 1939.
  • Levine, Timothy R. "Truth-Default Theory (TDT)." Journal of Language and Social Psychology, 2014.
  • Merridale, Catherine. "The 1937 Census and the Limits of Stalinist Rule." The Historical Journal, 1996.
  • Nickerson, Raymond S. "Confirmation Bias: A Ubiquitous Phenomenon in Many Guises." Review of General Psychology, 1998.
  • Turner, Erick H., et al. "Selective Publication of Antidepressant Trials and Its Influence on Apparent Efficacy." New England Journal of Medicine, 2008.
  • U.S. Senate Select Committee on Intelligence. Report on the U.S. Intelligence Community's Prewar Intelligence Assessments on Iraq. 2004.
  • United States v. Philip Morris USA, Inc., 449 F. Supp. 2d 1 (D.D.C. 2006).
  • The Guardian, reporting on the Clean Internet 2026 campaign and the Gyirong flood, 8 September 2026.

Case studies

Then and now

In the watch briefs

All 135 briefs

In campaigns

Related techniques