Scientific Content Analysis, sold as SCAN, was developed by Avinoam Sapir, a former Israeli police polygraph examiner. It is a method for analyzing a written statement — you ask the subject to write an account of the relevant period in their own hand, and then you analyze the text.
The criteria are numerous and specific. Changes in pronoun use, particularly dropping I. Shifts in verb tense, especially into the present when narrating the past. Missing time — gaps in the chronology. Unnecessary connectives such as later that day or after that, held to mark omitted material. Spontaneous corrections. Changes in the way people are referred to, from my wife to Nancy to she. Extraneous information. The proportion of the statement devoted to the period before, during and after the event.
It is taught internationally to police, insurance investigators, immigration officers and military interrogators, and it is used to make real decisions about real people.
Glynis Bogaard, Ewout Meijer, Aldert Vrij and Harald Merckelbach tested it properly in 2016. They collected true and fabricated statements under controlled conditions and coded them against the SCAN criteria, examining whether the criteria appeared at different rates.
They did not. Most criteria appeared with statistically indistinguishable frequency in truthful and deceptive statements. In a second study, SCAN-trained analysts did no better than untrained ones. Earlier work by Nahari, Vrij and Fisher comparing SCAN with the criteria-based approach in the next chapter found the same: no discriminant validity.
The interesting question is again why practitioners are convinced.
Two mechanisms. The first is illusory correlation, demonstrated by Loren and Jean Chapman in the 1960s with clinicians interpreting projective tests: people perceive relationships between variables where none exist, particularly when the perceived relationship fits a prior expectation. Analysts remember the cases where a pronoun shift preceded a confession, and do not tally the cases where it did not, because they never learn the outcomes of most of them. This is the feedback vacuum of chapter 52 in written form.
The second is that the criteria are numerous, vague, and available in both directions. Any statement of reasonable length contains tense shifts, pronoun changes, gaps and connectives, because that is what natural writing looks like. With enough criteria and interpretive latitude, an analyst who suspects a subject will find support, and an analyst who does not will not.
Which is where the harm lies, and it is worse than mere inaccuracy. SCAN produces a document. A hunch, run through the method, comes out as a written analysis with technical vocabulary and enumerated findings, and that document then travels — into a case file, a claim decision, an asylum determination — carrying an authority the underlying judgment never had.
The method does not detect deception. It launders suspicion.
Cross-ref: verifiability approach chapter