Research summary: Nelson (2015), Scientific Basis for Polygraph Testing

This post is a detailed reading of Raymond Nelson’s 2015 review, Scientific Basis for Polygraph Testing, published in Polygraph, 44(1), 28–61. The full citation and a link to the source PDF are in Sources (entry 1).

Table of contents

1. Why the paper matters

Polygraph practice is often discussed in polarised terms: either as a settled science or as pseudoscience. Raymond Nelson’s 2015 review, published in the journal Polygraph, is deliberately neither. It is an attempt to set out, in one place, the empirical literature that supports the polygraph as a probabilistic test — a structured combination of physiological recording and statistical decision-making — while being explicit about what the polygraph does not claim to do.

Three framings run through the paper and are worth stating up front:

1. The polygraph does not measure lies as such. It records physiological responses to stimuli and uses statistical reference distributions to discriminate deception and truth-telling^[6].
2. All test results are probability statements. Like other scientific tests, they carry a quantifiable margin of error — not certainty^[7].
3. Polygraph accuracy is significantly greater than chance and significantly greater than unassisted human judgement, but is not perfect — no probabilistic test is^[8].

Everything else in the paper follows from these three framings.

2. Diagnostic vs. screening examinations

Nelson opens by distinguishing two categories of examination that behave quite differently and should be reported differently.

Diagnostic examinations

Address a single, known issue — an incident, allegation, or set of symptoms already indicating possible involvement. The purpose is a positive or negative diagnostic conclusion that may serve as a basis for action^[9].

Screening examinations

Conducted in the absence of a known incident or allegation. Their purpose is to add incremental validity to risk-management decisions — by developing information and, in some programmes, by deterrence^[10].

Nelson is explicit that screening polygraphs should not be used alone as a basis for action affecting individual rights, liberties, or health^[12]. This aligns directly with the position taken in the BPS Standards of Practice and the BPS Code of Ethics.

3. The three phases of an examination

Following the APA’s Meta-Analytic Survey, Nelson treats the examination as having three phases^[13]. Each phase carries its own evidence base.

3.1 The pretest interview

The pretest interview serves four scientific purposes:

• establishing identification, introduction, and the roles of examiner and examinee;
• obtaining informed consent, including the examinee’s right to terminate at any time;
• conducting a free-narrative, semi-structured, or structured interview to develop the examination issues and construct the test questions^[14];
• reviewing all test questions with the examinee before any physiological data are recorded, and conducting an acquaintance (or practice) test.

3.2 In-test data collection

Most field polygraph examinations use a variant of the comparison question technique (CQT), first described in print by Summers (1939), popularised by Reid (1947) and Backster (1963), and now the most exhaustively researched family of polygraph formats^[16].

All CQT formats include:

• Relevant questions (RQs) — describing the behaviour under investigation;
• Comparison questions (CQs) — providing a within-subject reference against which responses to RQs are evaluated.

Nelson addresses a long-standing criticism — that probable-lie comparison questions (PLCs) are manipulative and that the polygraph “measures lies”. He responds that neither characterisation matches the science. The polygraph records responses to stimuli and discriminates deception from truth-telling via probability models — not by measuring truth-value directly.

Directed-lie comparison questions (DLCs) are an ethically and methodologically defensible alternative. Studies show DLC and PLC techniques classify with equal efficiency and similar numerical score distributions^[18], and DLC examinations retain effectiveness across different languages and cultures^[19].

3.3 Required sensors

Polygraph instrumentation consists minimally of four sensor sets:

• two pneumograph sensors (thoracic and abdominal) for respiration;
• electrodermal activity sensors (palmar or distal);
• cardiovascular sensors for relative changes in blood pressure;
• an activity (movement) sensor — mandatory per APA since 1 January 2012^[21].

This is the same minimum sensor set specified in the BPS Standards of Practice §4.1. Test stimuli are presented not fewer than three and commonly up to five times, with responses aggregated across repetitions to improve diagnostic efficiency^[22].

4. Test data analysis — the statistical heart of the paper

Nelson devotes the longest technical section to test data analysis. His central claim: polygraph scoring is no different in kind from the scoring of other scientific tests in medicine, psychology, and forensics. It involves four basic steps:

1. identifying observable, measurable criteria (scoring features);
2. transforming those features to numerical values and reducing them to grand-total and subtotal indices;
3. comparing scores to statistical reference distributions to calculate numerical cutscores and classifiers;
4. applying structured decision policies.

4.1 Primary (Kircher) features

Across several decades of research — largely funded by the US Department of Defense and conducted at the University of Utah and Johns Hopkins — a small number of physiological features have emerged as the primary scoring features (“Kircher features”)^[23]:

• Respiration — excursion length or correlated suppression patterns (reduced tidal volume, slowed rate, temporary elevation of exhalation baseline);
• Electrodermal activity (EDA) — amplitude of vertical increase in skin conductance. The single most robust contributor to the final classification^[24];
• Cardiovascular activity — amplitude of vertical increase in relative blood pressure.

4.2 Numerical transformations

Numerical scores are non-parametric integers assigned by comparing the strength of reaction to each RQ with the strength of reaction to the CQs presented in sequence with it. The discriminating characteristic is not the presence or absolute magnitude of any single response, but the relative magnitude of response to CQs versus RQs^[26].

Three manual scoring systems are in common use:

Differences among the seven-position systems are procedural and empirically inconsequential for accuracy^[27].

4.3 Decision rules and cutscores

A cutscore is an explicit tolerance for error. Nelson treats alpha (α) and cutscores as two sides of the same coin: a .05 cutscore implies a 95% confidence level for the categorical result; .10 is used to reduce inconclusives; .01 is used to reduce errors^[28].

Three decision rules are described:

Grand-Total Rule (GTR)

Compares the grand-total score to cutscores for deception and truth-telling. Used in event-specific diagnostic examinations where criterion variance is non-independent.

Two-Stage Rule (TSR)

Where the grand-total is inconclusive, subtotal scores are consulted against statistically-corrected cutscores^[29]. Bonferroni corrections prevent alpha inflation from multiple comparisons.

Subtotal-Score Rule (SSR)

Used in multi-issue screening. The test is classified deceptive if any independent question produces a significant deceptive subtotal; classified truthful only if all independent subtotals are significant for truth-telling. Šidák corrections are used where multiple independent probability events must all be simultaneously truthful.

5. Physiological basis

The recording of reliable physiological proxies for deception rests on the autonomic nervous system (ANS), which comprises sympathetic (S/ANS) and parasympathetic (PS/ANS) branches. Nelson frames the changes recorded during polygraph testing as allostatic — changes made in the service of maintaining homeostasis in response to real or perceived demand^[31].

What polygraph sensors actually record are three physiological phenomena correlated with deception and truth-telling:

• subtle, temporary respiratory suppression;
• phasic electrodermal activity indicative of S/ANS arousal;
• phasic change in the moving average of relative blood pressure.

“Polygraph instrumentation is not used to evaluate cardiovascular or respiratory health.”

— Nelson (2015), p. 38

It captures only those physiological features whose diagnostic utility in polygraph testing has been supported by replicated, peer-reviewed research.

6. Psychological basis

Nelson is more tentative here, and this is where his paper is most useful to practitioners with a scientific disposition. He directly rejects the older “psychological set” explanation — the notion that examinees focus on whichever question presents the greatest immediate threat — on four grounds:

• it cannot explain why DLCs work (since the examinee is not threatened by them);
• it cannot account for the fact that psychopathic examinees — who have been shown to have low fear conditioning^[32] — are detected as accurately as non-psychopathic examinees^[33];
• it requires the implausible assumption that polygraph sensors discriminate specific emotions — which the literature does not support^[34];
• the phrase “psychological set” itself does not appear in the scientific psychological literature in the form used by polygraph examiners^[35].

He also notes that the Cannon (1929) fight-or-flight framework has itself been significantly revised in modern stress research^[36], undermining the lay explanation further.

In its place, Nelson endorses a more parsimonious differential salience model alongside cognitive-behavioural and conditioned-learning accounts^[37]. He also flags preliminary process theory, linked to orienting theory^[38], while cautioning that more work is needed.

Nelson’s operative claim is modest: responses to polygraph stimuli probably result from some combination of cognition, emotion, and conditioned experience loaded differentially onto RQs and CQs — and any psychological theory of the polygraph must be falsifiable in Popper’s sense^[39].

7. Accuracy — the headline numbers

This is the section the paper is most often cited for. Nelson draws primarily on the APA’s 2011 Meta-Analytic Survey, which aggregated 24 peer-reviewed studies (8,975 confirmed scores) for event-specific diagnostic polygraphs and 14 studies (1,194 confirmed scores) for multi-issue screening polygraphs, alongside earlier systematic reviews by the Office of Technology Assessment (1983) and the National Research Council (2003).

7.1 Diagnostic vs. screening — side by side

7.2 Field vs. laboratory studies

A recurring objection is that laboratory results may not generalise to field practice. Nelson reviews this carefully. Field studies offer ecological validity but inconsistent case confirmation. Laboratory studies offer experimental control but unknown ecological validity. The systematic reviews — OTA (1983), NRC (2003), APA (2011), and Pollina et al. (2004) — all found no statistically significant differences in polygraph accuracy between field and laboratory studies^[42].

Nelson emphasises the distinction between ecological validity (the setting) and external validity (generalisability): the latter is often achieved from well-designed laboratory studies despite imperfect ecological validity.

8. Threats to accuracy

Nelson is explicit that polygraph accuracy is not uniform across all examinees. Several categories can reduce accuracy:

• Young juveniles — reduced at low functional maturity^[43];
• Chronic psychotic-spectrum mental-health conditions^[44];
• Below-range intellectual functioning;
• Conditions that compromise suitability more generally.

Nelson also dispatches the “friendly examiner hypothesis” — the claim that examinations conducted under attorney–client privilege are less valid than those conducted for law enforcement. Honts (1997) investigated the hypothesis through logic, case analysis, and meta-analysis, and found no evidence for it^[46].

9. Countermeasures

Countermeasures — deliberate attempts to alter the test outcome — receive a substantial section. Nelson’s summary of the literature:

• Simple information about how the polygraph works is not sufficient to alter accuracy^[47].
• Post-hypnotic suggestion was ineffective^[48].
• Benzodiazepines and stimulant medications were ineffective. An older study suggested meprobamate may have some effect, but ethics constraints have limited replication^[49].
• Trained physical and mental countermeasures can reduce accuracy^[50].
• Human polygraph examiners are not as accurate as they claim at distinguishing countermeasure use from other artefacts^[51].
• Truthful examinees who attempt countermeasures tend to increase their likelihood of being classified as deceptive. On this basis the NRC (2003) concluded that countermeasure use by truthful examinees is not advisable^[52].

Activity sensors improve countermeasure detection^[53]. On the overall question, Nelson quotes the National Research Council (2003) directly:

“Because the effective application of mental or physical countermeasures on the part of examinees would require skill in distinguishing between relevant and comparison questions, skill in regulating physiological response, and skill in concealing countermeasures from trained examiners, claims that it is easy to train examinees to ‘beat’ both the polygraph and trained examiners require scientific supporting evidence to be credible. However, we are not aware of any such research.”

— National Research Council (2003), p. 147^[54]

Nelson’s own conclusion is measured: the literature does not support the hypothesis that polygraph countermeasures are reliably effective in the field, but this should not be read as a claim that the polygraph is infallible.

10. Information gain — how much does the polygraph actually help?

A useful framing is Honts & Schweinle’s (2009) use of the Information Gain Index (IGI) — a measure of how much decision accuracy is improved by a method relative to unassisted professional judgement^[55]. Vrij (2008) showed that police officers achieve roughly 56% accuracy at unassisted lie detection — barely above chance.

Honts & Schweinle’s results^[56]:

• Diagnostic polygraphs deliver a statistically significant improvement across base-rates 0.01–0.97 for deceptive outcomes and 0.03–0.99 for truthful outcomes; peak information gain is 27× unassisted judgement.
• Screening polygraphs deliver a statistically significant improvement in detecting deception across base-rates 0.02–0.83 and in detecting truthfulness across 0.11–0.99.

Handler, Honts & Nelson (2013) extended the analysis to directed-lie screening tests and found broadly similar patterns^[57].

11. What Nelson (2015) does not claim

Nelson’s paper is occasionally misrepresented in both directions. It is worth being explicit about the claims the paper does not make:

• It does not claim the polygraph is deterministic, or that its results are certainties.
• It does not claim polygraph accuracy is uniform across all contexts or all examinees.
• It does not claim polygraph results should be used alone as the basis for decisions affecting rights or liberty.
• It does not claim the psychological basis is fully understood. Nelson is explicit that more theoretical work is needed, and any theory must be falsifiable.
• It does not claim countermeasures are impossible.

12. Relevance for UK practice

Nelson writes for a US professional audience. The UK statutory and regulatory contexts — the Offender Management Act 2007 (ss.28–30), the Domestic Abuse Act 2021 (s.76), HMPPS administration of mandatory testing, the Equality Act 2010, the UK GDPR — do not feature. But the underlying science is not jurisdictionally specific, and the paper’s findings inform UK practice in four ways:

1. Validated accuracy thresholds. The thresholds in the BPS Standards of Practice §2.8 (90% evidentiary, 86% paired, 80% investigative, “significantly greater than chance” for screening) derive from the APA framework Nelson surveys. Practising to those thresholds aligns UK examiners with the published evidence base.
2. Minimum instrumentation and procedure. The sensor set and procedural minima — respiration, EDA, cardiovascular, activity sensor; 90-minute minimum; in-person conduct; validated techniques with documented deviations; record retention — reflect Nelson’s review.
3. Reporting discipline. Nelson’s emphasis on inconclusives as a legitimate category, on lower-bound confidence intervals when describing individual-case accuracy, and on not overstating reliability, mirrors the BPS Code of Ethics §§4 and 8.
4. Psychological framing in client-facing materials. Over-simplifications (“detects fear”) that cannot account for DLCs, psychopathy, or cognitive-behavioural findings should be avoided.

13. Historical timeline

14. Glossary

ANS

Autonomic nervous system. The part of the peripheral nervous system that regulates involuntary physiological processes. Comprises sympathetic (S/ANS) and parasympathetic (PS/ANS) branches.

Allostasis

Physiological change in the service of maintaining homeostasis in response to real or perceived demand. Contrasts with the older, narrower concept of homeostasis.

CQT

Comparison Question Technique. The family of polygraph testing formats that uses comparison questions alongside relevant questions to produce numerical scores.

RQ

Relevant question. Describes the behaviour under investigation.

CQ

Comparison question. Provides a within-subject reference against which responses to RQs are evaluated.

PLC

Probable-Lie Comparison. A traditional comparison question the examinee is manoeuvred into probably denying untruthfully.

DLC

Directed-Lie Comparison. A transparent comparison question the examinee is instructed to answer untruthfully.

GTR

Grand-Total Rule. Decision rule using the summed score across the whole test, used in event-specific diagnostic exams.

TSR

Two-Stage Rule. Uses subtotal scores to resolve inconclusive grand-total results, with statistical corrections for multiple comparisons.

SSR

Subtotal-Score Rule. Decision rule for multi-issue screening exams, treating subtotal scores as statistically independent.

EDA

Electrodermal activity. Skin conductance changes reflecting sympathetic autonomic arousal; the single most robust contributor to final polygraph classification.

IGI

Information Gain Index. Measure of how much a method improves decision accuracy compared with unassisted professional judgement.

PPV / NPV

Positive / Negative Predictive Value. Probability that a positive or negative test result is correct, given the base-rate.

Sensitivity / Specificity

The proportion of true positives correctly identified (sensitivity) and true negatives correctly identified (specificity) — the two classical metrics of diagnostic accuracy.

PCSOT

Post-Conviction Sex Offender Testing. Polygraph testing of sex offenders as a condition of treatment or supervised release; requires specialised APA training.

Base-rate

The prevalence of deception (or whatever the test is trying to detect) in the population being tested. Predictive values shift with base-rate; sensitivity and specificity do not.

Inconclusive

A formal result category used when the test data do not meet the threshold for either a deceptive or truthful categorical conclusion.

15. Sources (numbered references)

Sources are listed in the order of first citation. Each inline footnote links to the source entry here; the entry gives the full formal citation and, where available, a link to the original.

1. Nelson, R. (2015). Scientific Basis for Polygraph Testing. Polygraph, 44(1), 28–61. PDF on polygraph.org.
2. American Polygraph Association (2011). Meta-Analytic Survey of Criterion Accuracy of Validated Polygraph Techniques. Polygraph, 40(4), 196–305.
3. American Polygraph Association (2011a). By-laws, effective 1 January 2012. polygraph.org.
4. Vrij, A. (2008). Detecting Lies and Deceit: Pitfalls and Opportunities (2nd ed.). West Sussex: Wiley.
5. Senter, S., Weatherman, D., Krapohl, D. & Horvath, F. (2010). Psychological set or differential salience: a proposal for reconciling theory and terminology in polygraph testing. Polygraph, 39(2), 109–117.
6. Rovner, L. I. (1979). The Effects of Information and Practice on the Accuracy of Physiological Detection of Deception. Doctoral dissertation, University of Utah. See also Rovner (1986), Polygraph, 15(1), 1–39.
7. Wilson, J. & Jungner, G. (1968). Principles and Practice of Screening for Disease. WHO Chronicle, 22(11), 473. Geneva: World Health Organization.
8. Summers, W. G. (1939). Science can get the confession. Fordham Law Review, 8, 334–354.
9. Kvale, S. (1996). Interviews: An Introduction to Qualitative Research Interviewing. Sage. See also Kahn & Cannell (1957); GAO (1991); Campion, Campion & Hudson (1994).
10. Honts, C. R. & Alloway, W. R. (2007). Information does not affect the validity of a comparison question test. Legal and Criminological Psychology, 12, 311–320. See also Honts & Reavy (2009); Nelson, Handler, Blalock & Hernandez (2012), Polygraph 41(3).
11. Amsel, T. T. (1999). Exclusive or nonexclusive comparison questions: A comparative field study. Polygraph, 28, 273–283. See also Horvath (1988); Horvath & Palmatier (2008); Palmatier (1991).
12. Nelson, R., Handler, M. & Morgan, C. (2012). Criterion validity of the directed lie screening test and the Empirical Scoring System with inexperienced examiners and non-naive examinees in a laboratory setting. Polygraph, 41(3), 176–185.
13. Honts, C., Amato, S. & Gordon, A. (2004). Effects of outside issues on the comparison question test. Journal of General Psychology, 131(1), 53–74. See also Krapohl & Ryan (2001); Abrams (1984); Hilliard (1979); Podlesny, Raskin & Barland (1976).
14. Bell, B. G., Raskin, D. C., Honts, C. R. & Kircher, J. C. (1999). The Utah numerical scoring system. Polygraph, 28(1), 1–9.
15. Kircher, J. C. & Raskin, D. C. (1988). Human versus computerized evaluations of polygraph data in a laboratory setting. Journal of Applied Psychology, 73, 291–302.
16. Ansley, N. & Krapohl, D. J. (2000). The frequency of appearance of evaluative criteria in field polygraph charts. Polygraph, 29, 169–176. See also Ansley (1999); Offe & Offe (2007). On EDA: Boucsein (2012), Electrodermal Activity, Plenum Press.
17. Raskin, D. C. & Hare, R. D. (1978). Psychopathy and detection of deception in a prison population. Psychophysiology, 15, 126–136. See also Kircher, Kristjiansson, Gardner & Webb (2005).
18. Senter, S. M. & Dollins, A. B. (2002). New decision rule development: Exploration of a two-stage approach. DoDPI00-R-0001. See also Senter (2003); Senter & Dollins (2008), Polygraph, 37(2), 112–124.
19. Barland, G. H., Honts, C. R. & Barger, S. D. (1989). Studies of the accuracy of security screening polygraph examinations. DTIC AD Number A304654. Department of Defense Polygraph Institute. See also Podlesny & Truslow (1993); Raskin, Honts & Kircher (2014).
20. Sterling, P. & Eyer, J. (1988). Allostasis: a new paradigm to explain arousal pathology. In Fisher & Reason (Eds.), Handbook of Life Stress, Cognition and Health. Wiley. See also Berntson & Cacioppo (2007) in Handbook of Psychophysiology (3rd ed.), Cambridge University Press; Handler, Rovner & Nelson (2008), Polygraph, 38, 228–233.
21. Birbaumer, N., Veit, R., Lotze, M., Erb, M., Hermann, C., Grodd, W. & Flor, H. (2005). Deficient fear conditioning in psychopathy: a functional magnetic resonance imaging study. Archives of General Psychiatry, 62, 799–805.
22. Patrick, C. J. & Iacono, W. G. (1989). Psychopathy, threat and polygraph test accuracy. Journal of Applied Psychology, 74, 347–355. See also Barland & Raskin (1975); Balloun & Holmes (1979).
23. Kahn, J., Nelson, R. & Handler, M. (2009). An exploration of emotion and cognition during polygraph testing. Polygraph, 38, 184–197.
24. Handler, M. & Nelson, R. (2007). Polygraph terms for the 21st Century. Polygraph, 36, 157–164.
25. Taylor, S. E., Klein, L. C., Lewis, B. P., Gruenewald, T. L., Gurung, R. A. & Updegraff, J. A. (2000). Biobehavioral responses to stress in females: Tend-and-befriend, not fight-or-flight. Psychological Review, 107, 411–429. See also Bracha, Ralston, Matsukawa, Williams & Bracha (2004), Psychosomatics, 45, 448–449.
26. Palmatier, J. & Rovner, L. (2014). Credibility assessment: Preliminary process theory, the polygraph process, and construct validity. International Journal of Psychophysiology. See also Barry (1996).
27. Popper, K. R. (1959). The Logic of Scientific Discovery. Routledge.
28. Pollina, D. A., Dollins, A. B., Senter, S. M., Krapohl, D. J. & Ryan, A. H. (2004). Comparison of polygraph data obtained from individuals involved in mock crimes and actual criminal investigations. Journal of Applied Psychology, 89, 1099–1105. See also OTA (1983); NRC (2003); APA (2011).
29. Abrams, S. (1975). A response to Lykken on the polygraph. American Psychologist, 30, 709–711.
30. Abrams, S. & Weinstein, E. (1974). The validity of the polygraph with retardates. Journal of Police Science and Administration, 2. See also Abrams (1974), Polygraph, 3, 328–337.
31. National Research Council (2003). The Polygraph and Lie Detection. Washington, DC: National Academies Press. See also Office of Technology Assessment (1983), The Validity of Polygraph Testing: A Research Review and Evaluation.
32. Honts, C. R. & Peterson, C. F. (1997). Brief of the Committee of Concerned Social Scientists as Amicus Curiae, United States v. Scheffer. See also Honts (1997).
33. Timm, H. W. (1991). Effect of post-hypnotic suggestions on the accuracy of pre-employment polygraph testing. Journal of Forensic Sciences, 36, 1521–1535.
34. Iacono, W. G., Cerri, A. M., Patrick, C. J. & Fleming, J. A. (1992). Use of anti-anxiety drugs as countermeasures in the detection of guilty knowledge. Journal of Applied Psychology, 77, 60–64. See also Iacono, Boisvenu & Fleming (1984); Waid, Orne & Orne (1981).
35. Honts, C. R., Raskin, D. C. & Kircher, J. C. (1994). Mental and physical countermeasures reduce the accuracy of polygraph tests. Journal of Applied Psychology, 79, 252–259. See also Raskin & Kircher (1990); Honts, Winbush & Devitt (1994).
36. Honts, C. R. (1987). Interpreting research on polygraph countermeasures. Journal of Police Science and Administration, 15, 204–209. Related works: Honts & Hodes (1983); Honts, Hodes & Raskin (1985); Honts, Raskin & Kircher (1987); Honts, Raskin, Kircher & Hodes (1988).
37. Ogilvie, J. & Dutton, D. (2008). Improving the detection of physical countermeasures with chair sensors. Polygraph, 37(2), 136–148. See also Stephenson & Barry (1986).
38. Wells, G. L. & Olson, E. A. (2002). Eyewitness identification: Information gain from incriminating and exonerating behaviors. Journal of Experimental Psychology: Applied, 8, 155–167.
39. Honts, C. R. & Schweinle, W. (2009). Information gain analysis of polygraph accuracy. As cited in Nelson (2015), p. 48.
40. Handler, M., Honts, C. & Nelson, R. (2013). Information gain of the Directed Lie Screening Test. Polygraph, 42, 192–202.

This summary was prepared for the British Polygraph Society. It is a reading of the source paper; readers making professional, legal, or regulatory use of the material should consult the original. The inline fact-check popovers are intended to make evidence transparent, not to substitute for the source paper. Any errors in this summary are the Society’s, not the author’s.

↑ Top