Tunnel vision, confirmation bias and the risk of a hypothesis becoming an investigative truth
Abstract: Investigative tunnel vision represents one of the principal cognitive risks in reconstructing a criminal event, as it can progressively narrow the range of hypotheses and direct the selection and interpretation of evidence towards an explanation that has already been formulated. Closely associated with confirmation bias, the phenomenon can influence every stage of an investigation, from the choice of investigative activities to the assessment of scientific and digital evidence, generating a cumulative process that makes disconfirming information increasingly difficult to recognise. Drawing on the scientific literature and the analysis of emblematic cases, this article examines the relationship between cognitive biases, investigative methodology and miscarriages of justice, highlighting the need for procedures based on the comparison of alternative hypotheses and the deliberate search for evidence capable of falsifying the prevailing reconstruction. The reliability of an investigation therefore depends not on how quickly an explanation is identified, but on the capacity to subject it to continuous scrutiny and, when necessary, to abandon it.
Keywords: #TunnelVision #ConfirmationBias #CognitiveBiases #InvestigativeMethodology #HypothesisFalsification #MiscarriageOfJustice #ForensicEvidence #DigitalForensics #EvidenceAssessment #ScenarioReconstruction #MariaRosaSurace #EthicaSocietas #EthicaSocietasRivista #ScientificJournal #HumanSciences #SocialSciences #EthicaSocietasUPLI
When a hypothesis ceases to be a hypothesis
Formulating an initial reconstruction means temporarily reducing uncertainty. One element may suggest a direction, a witness statement may open up a line of inquiry, and a particular behaviour may make one scenario plausible. But plausible does not mean proven. Tunnel vision develops when a possibility progressively begins to acquire a different status: it is no longer treated as an explanation to be tested, but as the explanation around which subsequent information is organised.
Findley and Scott have shown how this phenomenon can affect different stages of criminal proceedings, from the investigative phase through to subsequent stages of the judicial process. Its most insidious characteristic is precisely its progressive nature: an initial interpretation can guide subsequent decisions and, through those very decisions, become increasingly difficult to question.
Confirmation bias operates at a complementary level. Once a particular view has been formed, elements consistent with it may acquire greater salience, while disconfirming evidence risks being regarded as less relevant or less reliable. A cycle is thus created: hypothesis, targeted search, new information, interpretation, reinforcement of the hypothesis.
The problem is that this process can create a sense of confirmation even when what is increasing is not necessarily the accuracy of the theory, but rather the amount of information being interpreted through it.
Elaad’s 2022 study is particularly interesting because it avoids an overly simplistic interpretation of the phenomenon. In the hypothetical crime scenarios examined, investigators were more influenced by tunnel vision than non-investigator participants and displayed greater confidence in the suspect’s guilt. However, when exculpatory information was presented, their confidence in the suspect’s guilt also decreased.
This finding suggests that investigative experience does not automatically confer immunity from cognitive bias, but neither does it mean that investigators are incapable of responding to contradictory evidence. Experience makes it possible to recognise patterns and anomalies more rapidly. Yet that same speed can encourage associations between a new case and an already familiar scenario.
The issue, therefore, is not to prevent investigators from formulating hypotheses. It is to ensure that no hypothesis becomes too costly to abandon.
The suspect can become the focus of the wrong question
Identifying a suspect inevitably changes the direction of an investigation. It becomes possible to examine the person’s movements, relationships, communications, behaviour and possible motives. But there is a fundamental difference between investigating a person and investigating in order to prove that the person is responsible.
In the first case, the question remains open. In the second, the suspect becomes the centre around which the available evidence is organised.
This distinction can even influence which investigative activities are undertaken. Research by Rassin, Eerland and Kuijpers in 2010 showed that an initial assessment of guilt can influence the selection of subsequent investigative activities. In their studies, as perceptions of guilt increased, so too did the selection of activities aimed at finding incriminating evidence. The authors nevertheless pointed out that, overall, incriminating activities were not always preferred in absolute terms.
This is a crucial point. Bias does not necessarily intervene only when evidence is interpreted. It can intervene earlier, helping to determine which evidence will be sought in the first place.
The same mechanism can emerge in statements and interrogations. An evasive answer may be interpreted as an attempt to conceal something. An extremely detailed answer may appear rehearsed. An inconsistency may be regarded as an indication of deception. The risk is that different behaviours are made to support the same conclusion.
Indeed, once a theory has become sufficiently established, almost any response can be used to confirm it, regardless of its specific content.
The same problem applies to confessions. A confession may appear to be the element that definitively completes a reconstruction. Yet its evidential value cannot be separated from the rest of the evidence. It must be compared with the chronology, physical evidence, the information actually available to the individual and independently established facts. If a confession conflicts with objective evidence, reality should not necessarily be adapted to fit the confession. It may instead be the confession that needs to be reconsidered.
The same applies to witness testimony. Memory is reconstructive and can be influenced by the passage of time, information acquired subsequently and the manner in which questions are asked. A statement should therefore not be assessed solely according to how strongly it supports a particular hypothesis, but also by considering how, when and under what conditions it was produced.
When bias enters the evidence
Scientific evidence is often regarded as the point at which subjectivity should cease to influence an investigation. A biological trace, fingerprint, physical exhibit or technical datum may appear to speak for itself. But evidence does not reach the decision-maker directly. It must first be identified, collected, preserved, analysed and interpreted, and it is precisely during interpretation that context can play a significant role.
Forensic confirmation bias refers to the risk that expectations and contextual information may influence the assessment of forensic evidence. Kassin, Dror and Kukucka have highlighted how psychological and forensic research demonstrates the vulnerability of technical assessments to contextual effects, emphasising the need for procedures capable of limiting their influence.
An emblematic example is the case of Brandon Mayfield. In 2004, during the investigation into the Madrid bombings, the FBI mistakenly identified Mayfield, a US citizen living in Oregon, as matching a fingerprint recovered from evidence connected to the attacks. Mayfield was arrested as a material witness and detained for approximately two weeks. The identification was subsequently withdrawn after Spanish police attributed the fingerprint to an Algerian national.
The report by the US Department of Justice’s Office of the Inspector General identified several factors underlying the error, including shortcomings in the examination procedure and an inadequate reassessment of the identification after contradictory information emerged.
The significance of the case does not lie in demonstrating that fingerprint analysis is intrinsically unreliable. Rather, it reveals something more complex: even highly specialised analysis can be influenced by the way information is observed, compared and interpreted.
It is within this context that forensic literature has developed strategies for managing contextual information, including sequential unmasking: the principle of making contextual information available according to a structured sequence, thereby minimising, as far as possible, the risk that information unnecessary to the technical assessment may influence the analysis.
The same issue has acquired an additional dimension with digital evidence. A device may contain thousands of items: messages, photographs, searches, location data, histories, access records, contacts and metadata. Yet a greater quantity of data does not automatically produce greater certainty.
Digital evidence may be authentic and, at the same time, ambiguous. An online search does not necessarily demonstrate criminal intent. Location data alone do not establish participation in an event. A message may lose its meaning when separated from its context. A photograph may acquire a different meaning depending on its temporal reconstruction.
The investigative question, therefore, cannot be limited to: “What does the device contain?” It must also include: “What can we actually conclude from what it contains?”
This distinction is becoming increasingly important in investigations in which the availability of data risks being confused with the availability of answers.
How an error can become a structure
The most serious consequence of tunnel vision arises when bias is not confined to a single decision but is transmitted throughout the proceedings. A hypothesis directs the search; the search produces certain information; that information contributes to the construction of the prosecution theory; and the theory then influences the interpretation of subsequent evidence.
At that point, error can become cumulative.
It is not necessary for every person involved to make an obvious mistake. It is sufficient for each person to work with information that has already been selected or interpreted at an earlier stage. A decision may therefore appear reasonable when considered in isolation while, together with other decisions, contributing to an erroneous conclusion.
Three levels can be distinguished. The first is investigative error: a line of inquiry that is not adequately tested, a suspect identified prematurely, or information that is overlooked. The second concerns the interpretation of evidence: technical data, a confession or witness testimony is interpreted within a particular reconstruction without adequately considering alternatives. The third is a miscarriage of justice, when the proceedings result in the conviction of an innocent person.
The Italian legal system contains safeguards requiring consideration of evidence favourable to the person under investigation. Article 358 of the Italian Code of Criminal Procedure requires the public prosecutor also to investigate facts and circumstances favourable to the suspect. Article 326 identifies the purpose of investigative activity as conducting the investigations necessary for decisions concerning the exercise of criminal prosecution. Article 192 also establishes the criteria for evaluating evidence and requires circumstantial evidence to be serious, precise and consistent.
But there is a question that precedes the final assessment: what information reached the decision-maker, and what information was never sought?
Rigorous evaluation can correct an erroneous interpretation. It is far more difficult to correct something that never entered the investigative process. For this reason, tunnel vision does not concern only the risk of misinterpreting evidence. It also concerns the risk of never searching for the evidence capable of contradicting the prevailing theory.
The real safeguard is not infallibility, but the ability to change one’s mind
If cognitive biases are part of normal decision-making processes, simply asking investigators to “be objective” cannot be sufficient. Objectivity must be supported by investigative methodology.
The first step is to formulate multiple scenarios compatible with the available evidence. The question should not merely be: “Who is responsible?” but rather: “Which reconstructions of the event are compatible with what we know?” This should be followed by: “What evidence should be present if this reconstruction were correct?” And, above all: “What evidence could demonstrate that it is wrong?”
This is where falsification becomes essential. Verification means searching for evidence consistent with a hypothesis. Falsification means actively searching for evidence that could contradict it.
A 2026 study published in PLOS ONE by Sarah Lenz, Tara Zohrevand, Eric Rassin and Bruno Verschuere compared a suspect-centred investigative approach with one centred on evidence and scenario reconstruction, combining these approaches with verification or falsification strategies. The study involved 293 serving and prospective German police officers and used a real case of wrongful conviction. It found no direct effect of the experimental manipulations on assessments of guilt. However, it showed that an evidence-oriented approach increased the likelihood of proposing subsequent investigative activities focused more strongly on evidence, while a falsification-oriented approach encouraged activities designed to challenge previous assumptions.
This is an important finding precisely because it does not promise a simple solution. It does not demonstrate that merely telling an investigator to “try to falsify” is enough to eliminate bias. It suggests something more concrete: methodology can alter the way subsequent stages of an investigation are designed.
The same study refers to the Scenario Reconstruction Method, developed within the context of Dutch policing following the Schiedam Park murder case. In that case, Kees B. was convicted despite evidence incompatible with the prosecution’s reconstruction; years later, the actual perpetrator confessed and genetic evidence confirmed his responsibility. The case contributed to the development of mechanisms for testing and comparing hypotheses in Dutch investigative practice.
The underlying logic is simple, yet far from trivial: shifting the starting point from “Who did it?” to “What happened?”
From this perspective, several practical safeguards can be identified:
- explicitly formulate alternative hypotheses;
- identify in advance the evidence that could disprove them;
- document why a particular line of inquiry is being prioritised;
- deliberately search for disconfirming information;
- provide for independent review whenever possible;
- limit forensic analysts’ exposure to unnecessary contextual information;
- document the criteria and methods used to select and interpret digital data.
None of these measures can completely eliminate bias. Their function is different: to prevent a single hypothesis from controlling the entire investigative process.
When the investigation stops searching
The most delicate moment in an investigation may not be when an answer is missing. It may be when an answer finally appears sufficient.
That is when a theory risks ceasing to function as a question and becoming a certainty. A trace appears to connect the evidence. A witness statement seems to confirm the reconstruction. A confession appears to close the circle. A sequence of data seems to make everything coherent.
And it is precisely then that a different phase of the investigation should begin. Not the search for further confirmation, but the search for whatever might cause us to change our minds.
The real danger of tunnel vision is not merely making an interpretative error. It is reaching a point at which every new piece of information is used to protect the theory rather than to test it.
A rigorous investigation, therefore, is not one that constructs a convincing explanation as quickly as possible. It is one that preserves sufficient room to abandon that explanation.
Reliable justice does not arise from the idea of an infallible investigator. It arises from a method capable of challenging even its own most convincing explanation.
Perhaps this is precisely how the quality of an investigation should be measured: not by how quickly it reaches an answer, but by its capacity to continue searching even after it has found one. Because the day an investigation stops looking for what might disprove its conclusions, it is not necessarily moving closer to the truth. It may simply have stopped searching for it.
Essential bibliography
- Dror, I. E., “Cognitive and Human Factors in Expert Decision Making: Six Fallacies and the Eight Sources of Bias”, Analytical Chemistry.
- Elaad, E. (2022), experimental studies on tunnel vision and investigative assessment processes.
- Findley, K. A. – Scott, M. S., “The Multiple Dimensions of Tunnel Vision in Criminal Cases”, Wisconsin Law Review.
- Kassin, S. M. – Dror, I. E. – Kukucka, J., “The Forensic Confirmation Bias: Problems, Perspectives, and Proposed Solutions”, Journal of Applied Research in Memory and Cognition.
- Lenz, S. – Zohrevand, T. – Rassin, E. – Verschuere, B. (2026), study on investigative orientation, hypothesis falsification and scenario reconstruction, PLOS ONE.
- Rassin, E. – Eerland, A. – Kuijpers, I. (2010), studies on the influence of initial assessments of guilt on the selection of investigative activities.
- Italian Code of Criminal Procedure, Articles 192, 326 and 358.
- US Department of Justice, Office of the Inspector General, documentation and analysis concerning the Brandon Mayfield case.
- Methodological literature on the Scenario Reconstruction Method and investigative strategies developed within Dutch policing following the Schiedam Park case.
FIVE LATEST CONTRIBUTIONS
The Invisible Architectures of Coherence
Ceuta, When the Border Becomes a Weapon
Will Humans Govern AI, or Will AI Govern Humans? [VIDEO]
Ethica Societas is a free, non-profit review published by a social cooperative non-profit organization
Copyright Ethica Societas, Human&Social Science Review © 2026 by Ethica Societas UPLI onlus.
ISSN 2785-602X. Licensed under CC BY-NC 4.0