Incident investigation—the retrospective analysis by which organizations determine the causes of accidents and near-misses—is vulnerable to cognitive bias because investigators already know the outcome and receive information in a sequence that favors hindsight-driven causal stories. Burggraaf and Groeneweg (2016) identified cognitive biases in all nine real incident analysis reports written by certified investigators—outcome bias, hindsight bias, and the curse of knowledge, with confirmation bias treated as a further example of the dual-system mechanism—and proposed countermeasures: training investigators to recognize biases, critical peer review, and enforced step-by-step reasoning that deliberately engages effortful 'System 2' thought. The open problem is twofold: which further cognitive biases demonstrably distort retrospective incident analysis, and whether countermeasures of the proposed kinds actually reduce bias-induced error in investigation conclusions. A 2025 systematic review of occupational safety and health confirms that the development and evaluation of debiasing strategies in this domain remains a significant gap: additional biases have since been documented empirically (confirmation bias, anchoring, fundamental attribution error in investigation interviews; human-error attribution bias; contextual bias from prior safety-history information and witness opinions), and one popular structured tool, the Cause Analysis chart, was experimentally shown to reduce accuracy and increase worker-blame rather than mitigate bias. A 2026 controlled experiment testing a fact-checking intervention with professional investigators improved their separation of opinions from facts in witness evidence but left credibility assessments and causal judgments unaffected. Thus no countermeasure has yet been demonstrated to reduce bias-induced error in incident investigation conclusions, and generalizability across investigators, industries, and analysis methods is untested.

Background

Incident investigation (also called accident investigation or root-cause analysis) is the process by which organizations reconstruct what happened and why after an unwanted event, typically producing causal conclusions and recommendations intended to prevent recurrence. Its conclusions drive safety-critical decisions and, in many jurisdictions, must withstand legal scrutiny. The task is inherently retrospective: investigators analyze large amounts of information, often under time pressure, already knowing the adverse outcome. These conditions—outcome knowledge, sequential information exposure, and judgment under pressure—are classic ingredients of cognitive bias, i.e. systematic deviations from normative judgment produced by fast, automatic thinking. In the dual-process view associated with Kahneman, cognition divides into System 1 (fast, automatic, effortless impression formation) and System 2 (slow, controlled, effortful reasoning); a cognitive bias arises when System 1 delivers an answer to a question that actually requires System 2 effort, and the need for that effort goes unrecognized. Three biases are classic for retrospective causal analysis: hindsight bias, the tendency, once an outcome is known, to overestimate how predictable it was beforehand (Fischhoff, 1975); outcome bias, the tendency to judge the quality of a decision or action by its eventual outcome rather than by the information available at the time (Baron and Hershey, 1988); and the curse of knowledge, the tendency of better-informed people to misjudge how others, lacking that knowledge, would reason or act (Camerer, Loewenstein, and Weber, 1989).

All three plausibly bias an investigator's reconstruction of what the actors 'should have seen'. Burggraaf and Groeneweg (2016) evaluated nine incident analysis reports written by certified investigators of an internationally operating safety consultancy: they extracted the factual elements, re-analyzed the incidents, and compared conclusions, identifying cognitive biases in all reports—outcome bias, hindsight bias, and the curse of knowledge, with confirmation bias treated as a further example of the dual-system mechanism. They argued investigators can detect these biases themselves at an early stage if they know the 'red flags' to look for, and proposed countermeasures: training in bias recognition, critical peer review, and enforcing step-by-step reasoning that forces System 2 engagement. The same research group subsequently noted that whilst extensive training in recognizing one's own cognitive biases is possible, "the effectiveness is unclear and it could be very expensive", and called for research on the success rate of interventions as future work (Burggraaf et al., 2019).

Related results frame the problem's difficulty. Debiasing—the design of interventions that reduce bias-induced error—is notoriously hard: a systematic review in medical judgment found most debiasing strategies impeded or unvalidated (Croskerry, Singhal, and Mamede, 2013), and debiasing training has only recently been validated in other professional domains (Sellier, Scopelliti, and Morewedge, 2019; Heerma van Voss et al., 2025, with national security risk analysts). Within safety science itself, experimental work has begun measuring bias in investigators: MacLean and Read (2019) tested a widely used structured checklist tool (the 'Cause Analysis chart', on which investigators select substandard acts and conditions from predefined causal options) and found it made causal judgments less accurate and more biased toward blaming workers, while neither it nor an open-ended form mitigated the contextual bias induced by prior unsafe-history information; MacLean and Dror (2023) quantified a human-error attribution bias in professional investigators markedly larger than in lay controls; MacLean and Miller (2024) showed that witness opinions and case background knowledge bias investigators' evaluations; Thallapureddy et al. (2023) documented confirmation bias, anchoring, and fundamental attribution error in simulated investigation interviews; and Baybutt (2016) catalogued analogous biases in process hazard analysis. Adjacent proposals for bias-resistant investigation practice, such as tailored incident investigation protocols (Al-Bayati, 2024), remain feasibility demonstrations without bias-reduction measurement. The closest attempt so far is a controlled experiment by MacLean and Schael (2026), in which professional investigators and students either did or did not separate opinions from facts in a witness statement before judging it: early fact-checking reduced misclassifications of opinions as facts, but left credibility assessments and causal judgments unaffected—evidence processing improved, while bias in the conclusions themselves did not. What still does not exist is empirical validation that any countermeasure of the classes proposed in the source study—bias-recognition training, critical peer review, or enforced step-by-step reasoning—reduces bias-induced error in incident investigation conclusions.

Problem Statement

The problem, as posed by Burggraaf and Groeneweg (2016) and left open by the subsequent literature, has one overarching objective—making retrospective incident investigation robust against investigator cognitive bias—and two empirically linked components. First: which cognitive biases, beyond those already documented in the literature reviewed here (the biases identified by Burggraaf and Groeneweg (2016) and the further biases documented by the subsequent studies cited in this statement), demonstrably distort the causal conclusions of incident investigation (across its stages, from information collection and interviewing to causal analysis and report writing)? Second: can countermeasures of the proposed classes—training investigators to recognize biases, critical peer review, enforced step-by-step reasoning that engages effortful System 2 thought, or equivalent structural safeguards such as blinding or linear sequential unmasking of evidence (presenting case information in a fixed order that withholds potentially biasing context information until judgments on each item are recorded)—be shown empirically to reduce bias-induced error in investigation conclusions? Both components are settled by controlled empirical evidence (experiments or controlled field studies with investigators, ideally professionals): the first by evidence that a further cognitive bias demonstrably distorts the causal conclusions of incident investigation; the second by evidence that measures bias in investigation judgments with and without the countermeasure and demonstrates reduced bias-induced error in investigation conclusions. Feasibility under realistic time pressure and generalizability across investigators, industries, and analysis methods remain recognized design considerations and known difficulties. The negative branch is part of the question: a proposed countermeasure class is disconfirmed by a well-powered demonstration that it does not reduce (or worsens) bias in investigation conclusions—as in MacLean and Read's (2019) finding that the Cause Analysis chart, one of the listed countermeasure classes (a structural safeguard), did not reduce bias in investigation conclusions—whereas improvement confined to low-level evidence processing without improvement of the conclusions themselves, as in MacLean and Schael's (2026) fact-checking test, does not show such a reduction. The second component is settled by resolving any one of the listed countermeasure classes, by positive evidence or by a well-powered negative demonstration; the first component is independent of it.

References

  1. Burggraaf, J., & Groeneweg, J. (2016). Managing the Human Factor in the Incident Investigation Process. SPE-179207-MS, SPE International Conference and Exhibition on Health, Safety, Security, Environment, and Social Responsibility, Stavanger. DOI: 10.2118/179207-ms. https://doi.org/10.2118/179207-ms
  2. Fischhoff, B. (1975). Hindsight is not equal to foresight: The effect of outcome knowledge on judgment under uncertainty. Journal of Experimental Psychology: Human Perception and Performance, 1(3), 288–299. DOI: 10.1037/0096-1523.1.3.288. https://doi.org/10.1037/0096-1523.1.3.288
  3. Baron, J., & Hershey, J. C. (1988). Outcome bias in decision evaluation. Journal of Personality and Social Psychology, 54(4), 569–579. DOI: 10.1037/0022-3514.54.4.569. https://doi.org/10.1037/0022-3514.54.4.569
  4. Camerer, C., Loewenstein, G., & Weber, M. (1989). The curse of knowledge in economic settings: An experimental analysis. Journal of Political Economy, 97(5), 1232–1254. DOI: 10.1086/261651. https://doi.org/10.1086/261651
  5. Burggraaf, J., Groeneweg, J., Sillem, S., & van Gelder, P. (2019). How Cognitive Biases Influence the Data Verification of Safety Indicators: A Case Study in Rail. Safety, 5(4), 69. DOI: 10.3390/safety5040069. https://doi.org/10.3390/safety5040069
  6. MacLean, C. L., & Read, J. D. (2019). An illusion of objectivity in workplace investigation: The cause analysis chart and consistency, accuracy, and bias in judgments. Journal of Safety Research, 68, 139–148. DOI: 10.1016/j.jsr.2018.12.008. https://doi.org/10.1016/j.jsr.2018.12.008
  7. Baybutt, P. (2016). Cognitive biases in process hazard analysis. Journal of Loss Prevention in the Process Industries, 43, 372–377. DOI: 10.1016/j.jlp.2016.06.014. https://doi.org/10.1016/j.jlp.2016.06.014
  8. Croskerry, P., Singhal, G., & Mamede, S. (2013). Cognitive debiasing 2: impediments to and strategies for change. BMJ Quality & Safety, 22(Suppl 2), ii65–ii72. DOI: 10.1136/bmjqs-2012-001713. https://doi.org/10.1136/bmjqs-2012-001713
  9. MacLean, C. L., & Dror, I. E. (2023). Measuring base-rate bias error in workplace safety investigators. Journal of Safety Research, 84, 108–116. DOI: 10.1016/j.jsr.2022.10.012. https://doi.org/10.1016/j.jsr.2022.10.012
  10. MacLean, C. L., & Miller, G. S. (2024). Trust but verify: The biasing effects of witness opinions and background knowledge in workplace investigations. Journal of Safety Research, 89, 33–40. DOI: 10.1016/j.jsr.2024.01.009. https://doi.org/10.1016/j.jsr.2024.01.009
  11. Thallapureddy, S., Sherratt, F., Bhandari, S., Hallowell, M. R., & Hansen, H. (2023). Exploring bias in incident investigations: An empirical examination using construction case studies. Journal of Safety Research, 86, 336–345. DOI: 10.1016/j.jsr.2023.07.012. https://doi.org/10.1016/j.jsr.2023.07.012
  12. Sellier, A.-L., Scopelliti, I., & Morewedge, C. K. (2019). Debiasing Training Improves Decision Making in the Field. Psychological Science, 30(9), 1371–1379. DOI: 10.1177/0956797619861429. https://doi.org/10.1177/0956797619861429
  13. Al-Bayati, A. J. (2024). Tailored Incident Investigation Protocols: A Critically Needed Practice. Safety, 10(2), 37. DOI: 10.3390/safety10020037. https://doi.org/10.3390/safety10020037
  14. Heerma van Voss, B., Yoon, H., Scopelliti, I., Zweet, R., Helsloot, I., & Morewedge, C. K. (2025). Debiasing training reduces confirmation bias in national risk analysts. Scientific Reports, 15, 42529. DOI: 10.1038/s41598-025-28794-w. https://doi.org/10.1038/s41598-025-28794-w
  15. MacLean, C. L., & Schael, B. (2026). Investigating Workplace Investigators: Bias Mitigation With Fact-Checking. Applied Cognitive Psychology, 40. DOI: 10.1002/acp.70258. https://doi.org/10.1002/acp.70258
  16. Bacci di Capaci, A., & Sanderson, M. (2025). Cognitive Biases in Occupational Safety and Health: A Systematic Review of Prevalence and the Evidence for Effective Debiasing Strategies. Open Journal of Safety Science and Technology, 15(3), 242–256. DOI: 10.4236/ojsst.2025.153013. https://doi.org/10.4236/ojsst.2025.153013