SCAN Statement Analysis: Claims, Methods, and the Scientific Critique

SCAN analysis starts by having the subject write a statement in their own words, without guidance or prompting, describing the event in question. The analyst then reads the statement looking for approximately twelve proposed indicators. Sapir's reasoning is broadly cognitive: a person telling the truth produces a statement organised around what they actually remember. A person constructing a false account must manage what they claim to remember, what they claim not to remember, and how to make a fabricated account feel genuine, all at once. This extra cognitive load, Sapir argues, leaves traces in the text.

The pronoun shift indicator is central to the method. SCAN predicts that deceptive writers avoid first-person singular pronouns ('I') when describing the events they are lying about, and may shift to 'we' or drop the subject altogether ('Then went to the car'). The reasoning is that using 'I' claims personal agency and first-person experience: a liar supposedly avoids it at the precise moment of fabrication. Similarly, 'lack of conviction' language (hedges, memory disclaimers) is treated as suspicious because an honest person claiming a clear memory would not hedge.

Aldert Vrij, a professor of applied social psychology at the University of Portsmouth, led the most sustained empirical programme testing SCAN. In a series of studies in the 1990s and 2000s, Vrij and colleagues had participants produce truthful and deceptive accounts, then gave the statements to SCAN-trained analysts, untrained raters, and sometimes to automated analysis. Results across studies converged: SCAN analysts performed at chance (50 percent accuracy in a two-option true/false task) and sometimes below it.

A 2016 study by Bogaard, Meijer, Vrij, and Merckelbach published in Frontiers in Psychology directly tested whether the SCAN indicators appeared more often in deceptive statements than truthful ones. Most did not. Pronoun shifts and hedging language appeared with similar frequency in truthful accounts. The indicator that came closest to significance was 'missing time', but even this showed only a weak and unstable correlation. A 2020 systematic review by ten Brinke and colleagues found no peer-reviewed study demonstrating above-chance SCAN accuracy.

The theoretical foundations are equally weak. SCAN assumes that deceptive and truthful writers consistently differ in their use of specific features. But linguistic variation is driven by many factors: education, first language, topic familiarity, emotional state, cultural norms around assertion and hedging. A writer who habitually hedges for politeness, or who genuinely does not remember a peripheral detail, will produce 'lack of conviction' language that SCAN flags as deceptive. The signal SCAN is looking for, if it exists at all, is completely obscured by the noise of ordinary linguistic variation.

SCAN's continued use in investigative practice, particularly in US law enforcement and in Dutch and Israeli police contexts, represents a genuine puzzle given the strength of the empirical evidence against it. Several factors explain the persistence.

• Proprietary structure: the SCAN course is taught and licenced by Sapir's organisation. Academic critics do not have access to the full claimed methodology. This insulates the technique from peer review.
• Selection bias in practitioner experience: investigators remember and recount cases where a SCAN flag preceded a confession or breakthrough. Cases where SCAN flagged nothing significant, or flagged the wrong person, are rarely documented or shared.
• Confirmation bias in application: once an analyst has identified a SCAN 'hit', investigators may pursue the flagged person more intensively, increasing the likelihood of eventually obtaining corroborating evidence or a confession, which retrospectively validates the original flag.
• Absence of a feedback mechanism: in most investigative contexts there is no systematic comparison of SCAN outputs against eventual trial outcomes, so the failure rate is invisible to the practitioners using it.

The critical conceptual divide between forensic linguists and SCAN practitioners is the distinction between linguistic description and validated deception detection. A linguist can describe a statement: this text uses fewer first-person pronouns than a typical account; this passage contains a higher-than-average density of hedging language; this section is organised differently from the surrounding narrative. Those observations can be accurate and useful.

What a linguist cannot do, on the basis of those observations alone, is conclude that the writer is lying. Moving from description to detection requires a validated empirical relationship between the described feature and ground-truth deception, and that relationship has not been established for any of SCAN's indicators. This is not a theoretical objection peculiar to linguistics. It is the same standard applied to fingerprint comparisons, DNA mixture interpretation, and bite-mark analysis: show your error rate, and demonstrate that your conclusions are grounded in a reproducible, testable method.