An Epistemology for Forensic Psychiatry

Uncertainty as an Epistemic Obstacle

The trajectory toward an authoritative opinion, what Kenneth Weiss and I have called the path from inquiry to knowledge,⁵ requires epistemology to overcome a series of uncertainties. These range from definitions that differ among authorities to varying conceptualizations of justice itself. Jurisdictions in the United States, for example, define insanity differently depending on their commitment to the American Law institute standard, to the historic M’Naughten case, or to the absence of an insanity standard altogether.⁶ Disability, another common forensic assessment, differs across institutions, with the Social Security Administration focusing on the ability to work, the Centers for Disease Control and Prevention highlighting functional impairment, and the federal government, as a whole, defining civil rights protections under the Americans with Disabilities Act.⁷ Such variation affects how experts study, conceptualize, and testify on fundamental forensic topics.

This extends to conceptualizations of justice. Plato famously identified justice as a characteristic of both harmonious institutions and individuals, whereas Thomas Aquinas described it as a specific virtue or habit.⁸ In more recent times, John Rawls defined justice as fairness,⁹ while Amartya Sen updated Rawls’s idealized state to frame justice as a remediable practical matter, a problem that had to be corrected in individual cases.¹⁰ Modern thinking on justice has begun to consider the disparate treatment of marginalized groups who are exposed to greater police violence, higher bail, and longer sentences.^11,12

Such perspectives affect forensic analysis. Akin to prosecutors deciding when to bring a case, experts must decide how much trust to place in different sources of information, including police reports, witness statements, and medical records. They must assess when information is valid and credible and when it contributes meaningfully to an opinion. Moreover, experts have a perspective on justice and its machinery that may resemble a broad commitment to Rawls’s general fairness or to Sen’s specific requirement for a just outcome. Forensic professionals adopting Sen’s standard would judge success by a case’s outcome, while those adopting Rawls may simply accept a result that follows the system’s imperfect procedural design.

Experts must likewise consider how their view of justice informs the judgment of a defendant from a minoritized community or fraught cases in general (e.g., sex crimes, child custody). Other considerations include the relationship with the retaining party (e.g., defense, prosecution) or how the expert weighs different kinds of information (e.g., primary source, third-party, quantitative, qualitative). Outcomes that differ by philosophy, jurisdiction, defendant, retaining party, legal charge, or data type will require a stout framework for generating consistency and knowledge.

The greatest uncertainty forensic epistemology must overcome is likely the application of general information to specific cases. Known in forensic and clinical medicine as the nomothetic versus idiographic problem, this conceptual challenge attempts to apply research findings from groups of people to the unique circumstances of the individual.¹³ Epistemic tools are necessary for this connection because the specific case may not align well with the study population, the condition, behavior, or demographics being considered.

In the assessment of adjudicative competence, for example, this has been a problem in predicting medication effectiveness among persons diagnosed with delusional disorder. Delusional disorder is difficult enough to treat in clinical settings without speculating on its responsiveness in forensic ones.^14,15 Relevant forensic studies are small and poorly designed, so the connective reasoning must be tight for Sell hearings where the probability of attaining competence is a cardinal element of testimony.¹⁶

One particular intellectual tool may consequently be useful, not only to make this epistemic connection to the individual evaluee but also to illustrate how a forensic epistemology overcomes uncertainty. Numerous American Academy of Psychiatry and the Law (AAPL) colleagues and I have written on the importance of Occam’s razor, the parsimony of logic that cuts through extraneous thinking to yield valid forensic knowledge.^17,18 Named for philosopher and theologian William of Occam (c.1287-1347), the Razor elevates the simplest and most direct reasoning to confront overly inclusive thinking in expert opinions. It reminds analysts not to ascribe more causes to an event than absolutely necessary, to choose the simplest among predictive theories. The assertion is that logic (tight, parsimonious, razor-sharp) provides not only the skills that overcome uncertainty and its idiographic example but also reveals the presence of a coherent epistemology.

Bias

A connection drawn directly from the forensic literature ties established tools of forensic logic to its epistemology. This link of forensic thinking to epistemology was heralded by Charles Scott’s 2013 AAPL presidential address.¹⁹ Entitling his discourse “Believing doesn’t make it so: forensic education and the search for truth,” Scott argued that forensic expertise depends on confidence in scientifically reliable and unbiased information. Using Paul Appelbaum’s model of objective and subjective truth as a guide (i.e., objective scientific evidence and the expert’s subjective belief in it),²⁰ Scott focused on identifying and overcoming the many kinds of bias that interfere with an expert’s belief in valid information. Education and methodologic training could surmount insidious influences from anchoring and attribution bias to confirmation and hindsight bias (Ref. 19, p 27), all familiar to scientists and practitioners developing complex information.

Identifiable in all corners of science, this connection between belief and knowledge was most notably explored by social and cognitive psychology pioneers Daniel Kahneman and Amos Tversky. Directly relevant to forensic discussions of risk and credibility, the duo’s decades-long studies of bias stressed the availability of recent emotion-laden information as one of the most significant influences on people’s beliefs.^21,22 The availability of crime coverage in the news, for example, or the prominent reporting of airline disasters sway audiences to consider crime and plane crashes more prevalent than they are. This availability bias is recognized among patients as well, where recent emotional experiences affect the medical decision-making of cancer and diabetes patients, among many others.^23,–,25

Research on cognitive bias across scientific fields supports Scott’s idea that forensic conclusions are shaped not only by evidence but also by how experts interpret, evaluate, and respond to it. The research confirms the basis for Kahneman and Tversky’s far-reaching antibias guidance for “Thinking Fast [emotionally] and Slow [cognitively]”²⁶ and cements the epistemic connection between forensic belief and knowledge.

As significant as identifying and overcoming bias was for science in general and forensic psychiatry in particular, it was Scott’s invocation of truth and belief that tied forensic knowledge to epistemology. In epistemology, the classic relationship between truth and belief has been the equation Truth + Justified Belief = Knowledge.^27,28 This time-honored combination of the actual state of things (truth) and a level of certainty about it (belief) builds confidence through the collection of high-quality information (justification). In the sciences, this is historically the empiricism of Isaac Newton or the falsifiability of Karl Popper.^29,–,31 Newton’s observational methods influenced researchers for generations, producing ground-breaking knowledge of gravity and planetary motion, whereas Popper used empirical challenges to disconfirm hypotheses, inspiring the hypothesis-testing of modern grant-writing and study design. For forensic psychiatry, Scott recognized that unbiased scientific methods build belief in true, justifiable information: an epistemological process that reverberates through history and science.

The Error That Proves Logic’s Importance

The error that best illustrates logic’s importance to forensic epistemology is the prosecutor’s fallacy.^32,–,37 This error of critical thinking focuses on the evidence rather than the defendant’s guilt or innocence, the ultimate focus of the criminal legal process. It illustrates how misusing basic logic can result in frank injustice. The prosecutor’s fallacy specifically, and erroneously, equates the probability of the evidence (Prob E) given someone’s innocence (I) to the probability of someone’s innocence given the evidence. It is simply not the same to say that the probability of seeing the evidence in an innocent person is identical to the probability the person is innocent given that evidence. In statistics, this is described as Prob [E|I] = Prob [I|E]. It is a problem both in the direction of the logic (the probabilities remain inexplicably equated even when the variables are reversed) and in its context (namely, the base rates that affect the probability; see below).

In United States criminal jurisprudence, it is the California Supreme Court case People v. Collins that most prominently illustrates this error.^35,38 In 1964 Los Angeles, Mr. and Mrs. Collins were arrested for robbery after matching characteristics described by an eyewitness. A local statistician testified using numerous rates provided by the prosecution, including the probability of a woman having blonde hair, of an interracial couple in a yellow car, of a Black man with a beard, and of a man with a mustache. The statistician multiplied the rates to arrive at a 12,000,000:1 probability of the couple’s guilt.

Leaving aside the unempirical origin of the rates, the analysis treated the variables as independent when they clearly were not: a man with a beard probably has a mustache as well and a Black man is necessarily part of an interracial couple when his partner is of another race. The statistician also used a sampling probability when he should have used an inferential one (see Ref. 35, pp 87-104). The probability of matching sampling variables to a couple is different from inferring their guilt, the inferential probability. These are two conceptually distinct ideas. Equating them is an error that conditions probability on different assumptions: one assesses an observation (an interracial couple in a car) whereas the other assesses a hypothesis (the defendants are guilty).

Finally, inferring probability depends on the base rates of interracial couples in the area (five to six million by two accounts),^34,35 not simply multiplying possible matches to highly specific variables. Considering base rates diminishes the probability of guilt dramatically.

This problem may be more familiar from the use of DNA data. Matching DNA from a crime scene to someone in a city comes up against the probability of false positives, so even the importance of an individual match must be contextualized by the false positives arising from the base rate; say nine false matches in a city of a million, a nine out of 10 false discovery rate.³⁷ This is nine to one in favor of innocence, not guilt. The probability of a single match is tempered by the presence of false positives; logic now flows in the direction of innocence. This is why the prosecutor’s fallacy is also called base rate neglect.

The logical process of inferring is a mathematical and conceptual one; another reason Weiss and I see this as an epistemic path from inquiry to knowledge.^5,35 It takes into account knowledge of existing and prior probabilities (base rates) as well as alternatives (innocence and guilt) to infer an outcome. This should resonate for forensic practitioners who know that base rates of violence and suicide are low so they must testify to risk factors rather than calculations of certainty. In Collins, the California Supreme Court recognized these effects in overturning the conviction, writing:We observe that the prosecution’s theory of probability rested on the assumption that the witnesses…had conclusively established that the guilty couple possessed the precise characteristics relied upon by the prosecution. But no mathematical formula could ever establish beyond a reasonable doubt that the prosecution’s witnesses correctly observed and accurately described the distinctive features which were employed to link defendants to the crime (Ref. 38, p 9; inferring a legal outcome from a sampling probability).The prosecution attempted to compute the probability that a random couple would include a bearded Negro, a blonde girl with a ponytail, and a partly yellow car; the prosecution urged that this probability was no more than one in 12 million. Even accepting this conclusion as arithmetically accurate, however, one still could not conclude that the Collinses were probably the guilty couple. On the contrary…the prosecution’s figures actually imply a likelihood of over 40 percent that the Collinses could be “duplicated” by at least one other couple (Ref. 38, pp 9-10; recognizing base rate neglect, false positives).

Probability

This exploration demonstrates the partnership of probability and logic in creating a way of thinking about forensic analysis. Probability is consequently the other pillar of forensic epistemology. It can help forensic psychiatrists evaluate evidence and uncertainty more realistically. So far, its importance is indicated by its use in scientific logic (i.e., in Collins), in applying prior probabilities (base rates), and in considering the problematic probability equation Prob [E|I] = Prob [I|E] (the prosecutor’s fallacy). But probabilistic logic already has a lengthy pedigree in the philosophy of science and in forensic scholarship.

Readers may recognize that the combination of probabilities and alternatives reflects the work of 18^th century English minister and mathematician Thomas Bayes (1702-1761). Bayes offered theologians, scientists, and philosophers of his time an innovative framework for considering the quality of knowledge: the probability of evidence before gathering data (base rates), the probability of the evidence being offered (Prob E), and the relative probability of the solution given other explanations (guilt and innocence as above).^39,–,41 Bayes Theorem, now a foundation of statistics and medical research, derives the probability of a hypothesis A given B, (P(A|B)) from a number of relevant probabilities: the probability of B given A, (P(B|A)) and the probability of A and B on their own, (P(A), P(B)). To logicians and statisticians, the probability of the hypothesis given the evidence is technically known as posterior probability, determined from evidence and alternatives that are updated with each new data point. In practical terms, Bayesian reasoning updates working conclusions as new information becomes available. The stronger and more reliable the evidence, the greater confidence in the result. This posterior probability therefore consists of prior probabilities and its updates, a statistical analogy to Truth + Justified Belief = Knowledge.

Bayesian reasoning not only supplies the normative standard against which Kahneman and Tversky’s biases were measured (and whose deliberate application can correct them)^21,26 but also fits two major analytic approaches in forensic psychiatry.^42,–,45 For one, Thomas Gutheil’s team at the Massachusetts Mental Health Center described a decision-making approach to malpractice and other cases that was quintessentially probabilistic.⁴² It acknowledged the common uncertainties of observation, measurement, and analysis of forensic phenomena, namely the science, behaviors, and motivations of forensic cases. These were recognized as more subjective than traditional approaches that sought specific answers to scientific questions. The approach to forensic decision-making consequently reflected the probabilities and values that entered forensic analysis through the perspective of different jurisdictions, professions, and data types. Involuntary commitment statutes, for example, weigh values of community safety and patient autonomy along a spectrum, judges and experts balance social and scientific factors according to their roles, and all consider objective and subjective input in various ways (Ref. 42, pp 91-93, 102-106, 133). Forensic reasoning is consequently a judgment call that considers probabilities and uncertainty.

Second, Douglas Mossman of the University of Cincinnati specifically connected probabilistic thinking to Bayesian analysis.^43,–,45 In exploring the evolution of violence risk assessment,⁴³ Mossman described the landscape largely as one of binary predictions before the landmark Tarasoff ruling.⁴⁶ Afterward, however, a sea-change produced varying levels of confidence, with assessments increasingly accounting for base rates and false positives. Mossman demonstrated that this expansion of thinking could be described precisely by Bayes Theorem. For forensic psychiatry, violence risk assessments had become estimates of the probability of violence.

Not satisfied with the mere connection to Bayesian analysis, Mossman created a specific series of equations for the assessment of malingering.^44,45 Considering malingering base rates from a population matched to the individual (thereby overcoming the nomothetic versus idiographic problem) as well as to the individual’s test scores, Mossman derived Bayesian equations for individual test-takers. These considered base rates (described as pretest estimates) and the test scores themselves to produce posttest (recall posterior) probabilities. It was a concrete way to move beyond imprecise forensic descriptions of an evaluee’s behavior, namely being “characteristic of persons who are malingering” (Ref. 45, p 762). These were specific statistical solutions for the probability of malingering given probabilities before testing and then after specific testing outcomes. To characterize a forensic epistemology, it is sufficient to demonstrate, as Mossman did, the extent that Bayesian thinking can be applied to specific conditions without requiring practitioners to derive mathematical equations.

The Raven Paradox

An exercise in logic applied to the case of Norwegian terrorist Anders Breivik illustrates how probabilities lead logically toward a scientific conclusion. The contention is that Mr. Breivik did not use logic like a critical thinker who weighs evidence, recognizes truth, and builds knowledge. There would be no consideration of base rates and probabilities in his thinking.

In 2011, Mr. Breivik bombed the prime minister’s office building in Oslo and assassinated dozens of young people at a Labor Party youth camp outside the city. Despite disagreement among experts at trial, the prevailing theory was that Mr. Breivik acted on Internet-fueled rumors of the great replacement theory, a supremacist scheme that promotes a conspiracy among political and social elites to substitute Muslims and people of color for white European Christians. This creates the continent of Eurabia (Europe + Arabia).^47,–,49 European supremacists consequently misuse population data to promote this conspiratorial replacement stratagem.

A famed exercise in logic called the Raven or Confirmation Paradox demonstrates both how experts develop knowledge incrementally and how they can unravel conspiracy thinking at the same time (Ref. 5, chap. 1). Conspiracy theories are notoriously unparsimonious and improbable, requiring the nefarious coordination of multiple groups or actors when simpler reasoning is available: for the European refugee crisis, this is the chaotic global migration following war and famine in Africa and the Middle East.^50,51

A construct from World War II era philosopher Carl Gustav Hempel provides a model for the confirmation of information.^52,–,54 Hempel argued that if

All ravens are black, and

Anything nonblack is nonraven, then

Nonblack nonravens confirm that ravens are black.

Although this follows the rules of logic it seems counter-intuitive; the reasoning suggests that a red couch, a brown lectern, or anything else nonblack confirms that ravens are black.^5,54 Yet the paradox illustrates that even seemingly unrelated observations can provide small amounts of support for a hypothesis. The key point is not the example itself, but the realization that evidence contributes to a hypothesis incrementally and probabilistically rather than absolutely. For students of ornithology, the red couch is certainly less meaningful than a red robin, but it nonetheless moves the probability (slightly) toward its conclusion. American philosopher Marc Lange puts this in terms of the quality and validity of the data it takes to confirm a hypothesis,^53,54 bringing the epistemic discussion full circle (recall Truth + Justified Belief = Knowledge). It is a Bayesian approach that, like forensic psychiatry’s Gutheil and Mossman, considers information of different probabilities (here, strengths) as it builds toward knowledge. Confirmation consequently supports probability in revealing an epistemic path to knowledge.

Collecting and evaluating uncertain confirmatory data are familiar to forensic experts, who commonly validate data using collateral sources of unclear value, from contradictory witness statements and grainy camera footage to jargon-filled or cloned medical records.⁵ Likewise for Bayesian thinkers, confirmation grows toward knowledge from the strength of the background information (namely its probability) as well as the probability of the accumulating data and alternative hypotheses. Probability quantifies one’s growing belief.⁵

Experts analyzing conspiracy thinking can use logic and probability to forge a Bayesian path exposing faulty reasoning. The probability of Mr. Breivik’s grand replacement theory, for example, would depend on:The probability of the evidence before gathering data, namely the base rate probability (the low prior probability) of a multifaceted plot among coordinated governments and officials;The higher likelihood of the migration pattern under chaotic global alternatives (war, famine) than under the coordinated plot; andThe final (posterior) probability dominated by the realistic alternatives (war, famine).

One systematic review of 25 studies on the interventions that counteract conspiracy theories demonstrated that teaching critical thinking was in fact the most effective strategy for overcoming them.⁵⁵ Inoculating people before exposure to misinformation (e.g., “You may run into information that…”) provides resistance, as does “priming” (encouraging) critical thinking skills and education on the scientific method itself (recall Newton and Popper). This kind of analysis is useful therefore not only for identifying flawed reasoning but for transcending it as well. Warning patients about false information has already proven successful during the coronavirus disease (COVID) pandemic, so it is no stretch to credit these epistemic strategies for countering illogic and false attribution (conspiracy) more broadly.^56,–,58

It bears mentioning that probabilistic logic may also address the vague reasonable medical certainty standard followed by U.S. courtroom experts.^59,60 Disagreement persists over whether this testimonial standard is clinical or legal: it may be a legal standard of preponderance or clear and convincing evidence or a clinical threshold of quality and credibility. Disagreement notwithstanding, reasonable medical certainty requires a level of confidence in one’s information that may benefit from an epistemology of logic and probability that follows a Bayesian path to knowledge.

The explicit nature of this pathway offers another benefit. When forensic experts systematically identify the data and reasoning that create knowledge, they will necessarily expose bias in their views of justice as well. How they build probability and justify and confirm data will become part of an assessment of values as well as science. This is because their confidence in how justice serves society will be exposed by a systematic, transparent process. Crediting or endorsing one party or kind of data over another may consequently expose bias, because the choices are not well justified or confirmed. What are exposed are any idiosyncratic perspectives on values like justice itself.

One hopeful study that may carry this epistemic analysis into the future is an investigation using artificial intelligence (AI) to counter conspiracy beliefs.⁶¹ U.S. researchers used an AI chatbot to generate specific arguments tailored to respondents’ beliefs. Possibly because the chatbot’s reasoning was personalized and dynamic, conspiracy theorists durably reduced their beliefs in a range of theories from the assassination of President Kennedy to the Illuminati. This use of targeted critical reasoning may provide further justification for an epistemology of logic and probability among experts who have long considered such unparsimonious and improbable thinking to be fact- and logic-resistant.