The issue of multiple testing in polygraph screening – implications for internal security

Background & Context

Pre-employment polygraph screening has become a cornerstone of personnel security in sensitive government positions worldwide, yet a critical statistical problem remains underexamined: what happens when candidates are retested after initial failure? Marcin Gołaszewski's 2023 paper addresses this gap by exploring how multiple testing scenarios affect the integrity of screening programs used by law enforcement, intelligence agencies, and special services.

Screening polygraph tests are most often conducted before a candidate is hired, and Bayesian statistics help to better understand the real meaning of test results obtained. The issue is particularly acute in high-security contexts where both false negatives (hiring deceptive individuals) and false positives (rejecting honest candidates) carry significant consequences. This research fills a crucial gap between polygraph science's technical capabilities and real-world policy decisions that directly impact national security.

The situation is particularly concerning in institutions sensitive to national security, where the author questions whether retesting practices should continue or whether doubts should be verified with other tools. The paper applies Bayesian probability frameworks to polygraph screening, providing security administrators with quantitative tools to evaluate the risks inherent in common retesting practices.

Research Design & Methodology

Gołaszewski employs a theoretical Bayesian statistical modeling approach rather than empirical data collection. The methodology centers on calculating posterior probabilities under various screening scenarios using established test accuracy parameters from polygraph validation research. The author specifically uses the Directed Lie Screening Test (DLST) analyzed with the Empirical Scoring System (ESS), which has a documented sensitivity of 0.809 and a false positive rate of 0.1468.

The research design examines multiple retesting scenarios with different prior probability assumptions (base rates of deception in applicant pools). Key methodological elements include:

• Application of Bayes' theorem to update probability estimates after first and second tests
• Analysis of "successive hurdles" approach common in screening programs
• Examination of test independence issues when the same examiner conducts retests
• Comparison with research on single versus repeated polygraph examinations

The paper models polygraph screening's typical use of a 'successive hurdles' approach, where negative test results (NSR, NDI) close the case and provide a pass for the candidate to proceed, while positive results require additional testing or verification activities. The author draws heavily on Nelson and Turner's 2017 framework for Bayesian probabilities in repeated polygraph examinations.

Results & Key Findings

When only the average accuracy of the DLST test of 85.8% is considered, the statistical picture changes substantially when multiple testing is introduced. The core finding reveals that retesting practices create complex probability distributions that differ markedly from single-test scenarios.

Key statistical outcomes include:

• Posterior probabilities shift dramatically based on whether initial and retest results agree or conflict
• Base rate (prior probability) of deception in the applicant pool significantly affects interpretation of both initial and retest results
• With DLST/ESS parameters (sensitivity 0.809, FP rate 0.1468), multiple test scenarios produce varying confidence levels
• Tests performed by the same investigator in the same activity are not fully independent of each other, even if the type of test is changed, because some variables that do not affect results co-occur

The analysis demonstrates that allowing unlimited retesting after initial failures creates statistical problems that undermine screening program integrity. Allowing thorough questioning of candidates who successfully passed the test, or relying on examiner intuition, would reduce polygraph tests to a pseudoscience, according to Nelson and Turner. The paper quantifies how successive retesting opportunities increase the probability that deceptive candidates eventually pass through statistical chance alone.

Discussion & Significance

Gołaszewski's work challenges current field practices in personnel security screening by demonstrating that common retesting policies may be statistically untenable. The findings have direct implications for how intelligence agencies, law enforcement, and special services structure their vetting procedures. The Bayesian framework reveals that decision-makers often misinterpret test results by failing to account for base rates and the cumulative effects of multiple testing opportunities.

The research connects to broader debates about polygraph validity by showing that even validated testing techniques can be compromised by flawed implementation policies. Retesting happens most often when a candidate fails on the first attempt, and the author considers various scenarios related to retesting and their consequences for internal security of institutions and countries. This is particularly significant given the widespread use of polygraph screening in sensitive positions across multiple nations.

The paper contributes methodologically by demonstrating how Bayesian inference can inform policy decisions in applied security contexts. Rather than simply reporting test accuracy statistics, Gołaszewski shows how probability theory reveals the compound risks created by procedural choices. Bayesian analysis provides intuitive and practical interpretation of results, with mathematical examples in 2x2 tables for single and repeated polygraph exams.

Limitations & Considerations

The study is purely theoretical, relying on published accuracy parameters rather than analyzing actual screening data from operational programs. Real-world base rates of deception in applicant pools are often unknown or estimated imprecisely, which affects the practical application of the Bayesian calculations presented. The author notes that approximate numbers can sometimes be drawn from various sources including historical data, research studies, surveys, and police or court statistics.

The analysis acknowledges that tests performed by the same examiner are not fully independent, representing a simplification in the modeling. This lack of true independence in retesting scenarios may make the actual risks even greater than the theoretical models suggest. Additionally, the paper focuses specifically on DLST/ESS parameters, and findings may differ for other test formats or scoring systems used in different jurisdictions.

The paper does not address alternative verification methods that might be combined with polygraph screening, nor does it provide empirical validation of the theoretical models against real hiring outcomes in security agencies.

Practical Applications

Security administrators should use these findings to develop clear, statistically defensible retesting policies. Gołaszewski suggests that institutions must either find ways to address the multiple testing problem or put an end to 'corrective' retests and verify doubts with other tools if they exist and are effective enough. Recommended practices include establishing strict limits on retest opportunities, requiring independent examiners for retests, and incorporating base rate information into decision thresholds.

For polygraph consumers and job applicants, this research reveals that passing on a second or third attempt does not carry the same statistical weight as passing on the first test. The author seeks optimal solutions so that testing procedures promote tightness in security systems while remaining efficient and useful. Agencies should consider whether current retesting practices align with their actual risk tolerance and security requirements, potentially requiring policy reforms to maintain program integrity.