Assessing the Context of Evidence: A Bayesian Perspective on the Role of Publicity, Notoriety, and False Memory in the Blasey Ford Testimony to the United States Senate Confirmation Committee for Supreme Court Justice Brett Kavanaugh ()
1. Introduction/Purpose
Interpreting accusatory testimony usually requires an assessment of how it impacts the probability an accused individual committed an offense or a crime. Unfortunately, research suggests people are often not adept at assessing information probabilistically and that their intuition can lead them astray (Kahneman et al., 1982; Tversky & Kahneman, 1973) . Bayes’ Theorem can be useful in overcoming difficulties in interpreting evidence and can help answer questions such as: “How does this accusation influence the probability that the accused is guilty?”.
Almost four years ago, Brett Kavanaugh was sworn in as a United States (U.S.) Supreme Court justice following accusatory testimony that occurred during a contentious confirmation process. The accusation was the most highly publicized portion of the confirmation proceeding and was live broadcasted by all the major U.S. television networks. During that broadcast, Dr. Christine Blasey Ford accused Kavanaugh of making unwanted sexual advances and committing sexual assault (NPR Staff, 2018) . American viewers were cast into a role, not unlike jurors sitting on a criminal trial. They watched this witness testify that the assault had occurred approximately 35 years earlier at a private residence hosting an unsupervised high school party where alcohol was served. Although both the accuser and the accused were high school students at the time, Kavanaugh and Blasey Ford were enrolled at separate high schools and her testimony indicated that her only acquaintance with Kavanaugh had occurred during their mutual attendance at several such parties. This paper provides a Bayesian perspective on the accusatory evidence presented by Blasey Ford during the Supreme Court confirmation process. The approach documented here for evaluating this testimony could serve more generally as a useful example for assessing similar accusatory evidence that takes place within highly public hearings and legal proceedings.
The perceived motivations of an accusatory witness are important in judging the truthfulness of the testimony they provide (Bloom, 2002; Kassin et al., 2001) . In the case of the Blasey Ford testimony, the witness appeared to realize little personal gain by providing the testimony and likely experienced a loss of her personal privacy. For many who watched this witness, their intuitive analysis of her motivations and her expressed confidence in her memory reinforced their willingness to believe that she was being truthful. However, given that research indicates that the level of confidence in memory and its accuracy are not highly correlated, an assumption of truthfulness does not remove the necessity of evaluating the veracity of the memory (Simons & Chabris, 2011; Hirst et al., 2015; Hirst, 2009) .
Indications of the possible fallibility of Blasey Ford’s recollections were noted when she identified a close friend as having accompanied her to the party and an acquaintance of Brett Kavanaugh as having witnessed the alleged assault. Although these individuals were potential corroborating witnesses, neither had any recollection of that party and both denied any knowledge of an assault. In fact, in the various contexts in which she recounted the incident, there were inconsistencies regarding who attended the party and the circumstances surrounding it (Cleveland, 2018) . Ultimately, no witness came forward to acknowledge a recollection of that party and statements in interviews and within the testimony reflected inconsistencies regarding the events that surrounded the incident. Her inability to remember where the party had occurred, who had hosted the party, who attended it, her age when it occurred, and how she got to and from that party, led some to theorize that the testimony was based upon a false memory (McArdle, 2019) .
In judging the accuracy of testimony, it is important to consider the context and conditions in which the alleged incident and the testimony occurred. In this case, the context of the incident and the testimony did provide important clues regarding the accuracy of the witness’s memory. Unfortunately, the confirmation proceedings did not allow an opportunity for an expert forensic witness to testify regarding the reliability of memories formed under the conditions that characterized the incident and the testimony. In the Kavanaugh case, those conditions included 1) a stressful event, 2) an extended time since its occurrence, 3) a short acquaintance with the accused, 4) retrieval and rehearsal of the memory during psychological therapy, 5) a youthful victim, 6) an accusation against an individual of high notoriety, and 7) high publicity. All of these were potentially pertinent in judging the likely accuracy of that testimony.
While research does address the accuracy of memories recalled 1) under conditions of negative emotions and trauma (Loftus & Pickrell, 1995; Deffenbacher et al., 2004; Kaplan et al., 2016) , 2) after an extended time interval since occurrence (Shapira & Pansky, 2019) , 3) with limited exposure to the individual(s) involved in the crime (Reynolds & Pezdek, 1992) , 4) in the context of psychotherapy (Loftus, 1993; Otgaar et al., 2022) , and 5) during adolescence (Meusel et al., 2012) , factors related to 6) notoriety and 7) publicity have not been carefully considered in the literature. Although the large volume of research literature pertaining to the first five factors precludes a complete review, it is sufficient in this context to acknowledge that the research does suggest that these factors are relevant in assessing the accuracy of a recalled memory. However, the sixth and seventh factors related to notoriety and publicity have not been adequately addressed in the literature and this paper presents an analysis of how notoriety and its attendant publicity impact accuracy. Using the Blasey Ford appearance at those hearings as an example, a presentation of how the accuracy of memories about a famous individual can influence testimony in highly public proceedings. Lastly, the paper uses a Bayesian weighting of the testimonial evidence to guide judgements regarding the likelihood of guilt.
The remainder of this paper considers how factors related to notoriety and publicity impact the probability of false memories and/or false accusations and how these, in turn, influence the likelihood of Kavanaugh’s guilt. Section 2 in this paper contains four subsections that describe aspects of notoriety and publicity and how they influence the probability of false accusations. Subsection 2.1 estimates the number of Brett Kavanaugh’s personal acquaintances, Subsection 2.2 documents the effectiveness of the media in notifying potential character witnesses, Subsection 2.3 discusses the impact of notoriety on the likelihood of being accused, and Subsection 2.4 considers how the methods of witness selection relate to the likelihood of a negative false memory or baseless accusation. Section 3 describes how these influences can be formally represented in a probabilistic Bayesian estimate of guilt. Section 4 discusses the implications and limitations of this research.
2. Notoriety, Publicity, and False Memories
This discussion of notoriety and publicity uses the Blasey Ford appearance at the U. S. Senate confirmation hearing as an example. There is evidence to suggest that testimony in cases featuring accusations against individuals of high notoriety within a highly public forum influences how a witness is selected and the nature of the testimony. In this regard, there are four features of this particular testimony that should be considered. These features are, 1) the estimated number of personal acquaintances from which to draw a character witness, 2) the effectiveness of the media in notifying those acquaintances of the opportunity to testify, 3) the impact of notoriety on the likelihood of being accused, and 4) the estimated frequency of false memories and false accusations. After briefly describing each influence, a discussion is presented regarding how they governed who was selected to be a character witness against Brett Kavanaugh, what that witness was likely to remember, and ultimately the likelihood of finding an acquaintance who possessed an inaccurate or negative false memory of him.
Before beginning a discussion of the influence of notoriety and publicity, it should be noted that public figures are frequently accused of sexually scandalous behavior. While it is possible that the behavioral attributes of famous people differ somewhat from the public at large, it seems likely that notoriety itself plays a causative role in the likelihood of being accused of sexual impropriety. This section investigates why being a famous individual might be associated with an enhanced probability of being accused of sexual indiscretion and why notoriety and its attendant publicity may be a causative influence in the occurrence of such accusations.
2.1. The Number of Personal Acquaintances
Published research aimed at determining the size of social acquaintance networks has utilized various methodologies for estimating the average number of persons an individual knows (Killworth et al., 1990; Marsden, 2005; Knoke & Yang, 2021) . Methods used to estimate the number of acquaintances include survey methods, summing acquaintances within categories (e.g. family, coworker etc.), reverse small-world methods, scale-up methods, diary logs, and cued responses using last names from a phone book. While each method has strengths and weaknesses, an early seminal paper by Pool and Kochen (1978) helped define the statistical issues relevant to the various methodologies for estimating that number. Although a complete review of that literature is beyond the scope of this paper, the conclusions from that research do provide enhanced precision and a reasonable consensus regarding the estimated average number of acquaintances. A conservative interpretation of that research would suggest that an individual has on average more than 500 acquaintances (McCormick et al., 2010; Killworth et al., 1990) . While there is variance in the number of personal acquaintances depending upon life circumstances, personality, and other variables, it is clear that the majority of individuals living in modern urban environments have a very large number of personal acquaintances whom they have met, interacted with, recognize, and know by name. For purposes of estimating the number of Brett Kavanaugh’s personal acquaintances from which to select a character witness, an estimate of 500 is chosen here as an estimate.
2.2. The Effectiveness of the Media in Informing Acquaintances of the Opportunity to Testify
Because of the perceived importance of the Supreme Court appointment, it was a prominent story in the U.S. news media for over two months. This meant that the vast majority of Americans in general and Brett Kavanaugh’s acquaintances in particular, where aware of his judicial nomination. In fact, it became one of the most watched stories of the decade. An NPR/PBS NewsHour/Marist poll found that approximately 6 in 10 Americans (58%) followed the proceedings closely or very closely and many more Americans were aware of the ongoing confirmation proceedings (Montanaro, 2018) . An indication of the attention the testimony received was reflected by its television ratings. Despite being televised during a workday on a Thursday morning, it had a viewership of more than 20 million—nearly as high as a weekend broadcast of an NFL playoff game (News Staff, 2018) . Clearly, the American public and Judge Kavanaugh’s acquaintances in particular had been effectively informed of his nomination and the impending hearings.
2.3. The Notoriety and the Probability of Accusations
Statistical evidence suggests that being a famous man is associated with an enhanced likelihood of being accused of sexual assault. For example, in the U.S., during the year prior to Blasey Ford’s Senate testimony, the so-called #MeToo year, there were 425 highly publicized sexual assault accusations against famous men (Griffin et al., 2018) . Although fame exists at various levels, a common metric can be defined by the level of population-wide recognition. Using one such metrics, it has been estimated that roughly 1 in 10,000 individuals in the U.S. can be regarded as famous (Arbesman, 2013) . This implies that there are approximately 16,450 famous males in the U.S. and that they had an approximate 2.5% chance of being accused of sexual harassment or assault that year. It is interesting to compare this chance of being accused with the chance of being accused as a non-famous male. The number of sexual assaults nationally in the U.S. during that year was 463,634, of which 90% were female (Department of Justice, Office of Justice Programs, Bureau of Justice Statistics, 2020) . Even with the assumption that each is committed by a different identified male perpetrator and that each had a male identified, on average, a U.S. male had an approximately. 28% (463,643/164,500,000 = .0028) chance of being accused that year. While this a very rough estimate, it suggests that a famous male had an approximately nine times greater likelihood of being accused of a sexual assault.
2.4. The Likelihood of One or More Negative False Memories or Accusations
While even with a complete review of the literature it is not possible to accurately estimate the probability of a false memory or false accusation, the relevant research evidence does inform decision-makers in a general way about the likelihood of its occurrence in particular circumstances. Since the goal here is to calculate an estimate of the probability that the accused is guilty, it is useful to observe how the differences in the subjective estimate of the occurrence of a false memory impacts the probability of guilt. In this instance, relevant evidence that needs to be considered regarding the possibility of a false memory or false accusation are the long time interval since the memory’s formation, the witness’s inability to recall events surrounding the incident, the traumatic nature of the event, the fact that it was discussed during psychological therapy, and the publicity associated with notoriety. On the other hand, the subjective belief that it was an accurate portrayal is supported by the perceived credibility and truthfulness of the witness.
3. A Bayesian Perspective on Estimating the Probability of the Accused Is Guilty
It is important to consider how to logically combine the information related to the testimony and what it implies regarding the likelihood of guilt. Since even expert forensic scientist can mistake the probability that certain evidence would be present if the accused was guilt [P (E|G)] with the probability that the accused is guilty given the evidence [P (G|E)], a formal model can improve decisions and conclusions (Robertson & Vignaux, 1995) . In the evidence presented against Kavanaugh, the probability of any one randomly selected acquaintance having an inaccurate memory of him is much smaller than the probability of one or more acquaintance from the pool of Kavanaugh’s acquaintances possessing a false memory. This distinction is especially relevant here as this witness was selected by those aligned against Kavanaugh’s confirmation and the highly public nature of this case meant that the search for witnesses was quite broad. Clearly, the implicit criteria for witness selection included being a verifiable acquaintance of Kavanaugh and being willing to recount a negative experience about him. As the notoriety of this case implies that a large majority of his acquaintances were aware of the opportunity to testify, a consideration of the accuracy of a chosen witness must include a probabilistic consideration of the likelihood that one or more of Brett Kavanaugh’s 500 personal acquaintances might harbor a negative false memory and making a false accusation. In calculating a formal probability, we should avoid confusing the probability of a particular acquaintance possessing an inaccurate or false memory of the accused with the probability of one or more of Kavanaugh’s acquaintances having an inaccurate or false memory. While the exact values of these probabilities are impossible to know, it is clear that the nature of the memory recounted in Blasey Ford’s testimony was susceptible to false memory formation and that the notoriety of the accused made him vulnerable to false accusations. A statistical representation of this uncertainty would be useful in judging the evidence and in helping answer the question: How does the evidence offered by Blasey Ford impact the probability that the accused is guilty of this sexual assault?
To derive a statistical estimation of the probability that Kavanaugh was guilty of the alleged assault, it is useful to employ a Bayesian estimation that adjusts the probability of guilt as new evidence is presented. In this case, the new evidence was the Blasey Ford testimony. In its simplest form, Bayes Theorem can be represented as:
(1)
The prior odds reflect the subjective probability that Kavanaugh was guilty before the introduction of the new evidence (i.e. Blasey Ford’s testimony). It is calculated as the ratio of the probability that something is true (p) divided by the probability that it is false (1 – p) as: p/(1 – p). In the derivation of the prior odds, it is necessary to consider the evidence of guilt that had existed prior to the testimony. In this instance, all we know as viewers prior to this testimony is that a past acquaintance is willing to testify against him. Viewers had not been provided other evidence to suggest Kavanaugh was a sexual predator. No evidence had been presented or published regarding any previous such behavior and no previous legal findings suggested Kavanaugh had engaged in aggressive or assaultive behavior at other times. Since estimating a prior probability is inherently subjective, it is possible that some observers felt that Kavanaugh was prone to commit such crimes and that the investigative and legal systems were not effective in making this known. On the other hand, others might consider it unlikely that he would be involved in this behavior. So, the subjective estimation of the prior odds is likely to vary between individuals who watched those proceedings. Given this, the derivation of the final posterior probabilities will use a range of values reflecting differing perspectives on the prior likelihood of guilt.
The likelihood ratio (LR) in Equation (1) is calculated using the predictive information contained in the new evidence. In this case, the new evidence is the testimony given by Blasey Ford. Therefore, the LR for this evidence can be represented by the ratio:
(2)
where P (E|G) is the probability of the evidence given the accused is guilty and P (E|I) is the probability of the evidence given the accused is innocent. In this case, P (E|G) might be regarded as the hit rate (i.e. the evidence accurately implies guilt) and P (E|I) is the false alarm rate (i.e. the evidence inaccurately implies guilt). A maximum value of P (E|G) = 1.0 is used in this analysis to indicate the belief that if the accused was guilty of this crime, a witness would certainly come forward to offer testimony regarding the offense. This will yield higher values for the final probability of guilt by assuming that such a witness would certainly come forward if they experienced this assault. P (E|I) reflects the possibility of false memory or false accusation due to conditions surrounding the incident and testimony. If these influences were viewed as substantial, it would indicate assigning a higher probability to P (E|I). A higher probability would affirm that the characteristics of the incident made it vulnerable to the formation of a false memory and that high notoriety and publicity increased the likelihood of false accusations by selecting from a large pool of potential witnesses.
While Equation (2) provides an odds ratio, Equation (3) transforms the odds ratio into a conditional probability. Equation (3) calculates the probabilistic answer to the question: What is the probability of guilt given the evidence [P (G|E)]? In Table 1, values for P (G|E) are displayed assuming two values for P (E|I), two subjective prior probabilities of guilt [P (G)], and with P (E|G) consistently set to p = 1.0.
(3)
Figure 1 graphically displays the P (G|E) outcomes shown in Table 1 based upon two prior estimates of guilt [Prior G] and two estimates of the probability of the testimony given innocence P (E|I). For example, supposing that before hearing the testimony, the viewer had a strong feeling that Kavanaugh was guilty of sexual assault [Prior G = .50] and also believed that such accusations had a small chance of occurring in the absence of guilt [P (E|I) = .25]. The [P (G|E)] would equal p = .80. While this is still below the standard of ‘beyond a reasonable doubt’, it indicates an 80% certainty of guilt. However, if the viewer believes there is an appreciable chance of a false memory or false accusation [P (E|I) = .50] and that there was little prior reason to think he was guilty [Prior G = .25], the probability of guilt [P (G|E)] would equal p = .40.
![]()
Table 1. Table of the probability of guilt given the evidence under two conditions of prior probabilities of guilt and two conditions of probabilities of evidence given innocence.
![]()
Figure 1. Graph of the probability of guilt given the evidence under two conditions of prior probabilities of guilt and two conditions of probabilities of evidence given innocence.
Figure 1 shows that both P (E|I) and Prior G are influential in determining the probability of guilt. Numerically documenting our beliefs by formalizing these values is one way to reduce unintended bias and to transparently represent the logic behind our reasoning. For example, if this evidence was being weighed in a court of law, fairness considerations would imply that prior probability should be set low to accommodate a reasonable presumption of innocence and that conviction should meet a standard of beyond a reasonable doubt. On the other hand, if viewers of the broadcast are weighing the evidence to draw their own personal conclusions outside the legal framework, there might be some reasonable justification for setting a larger prior probability. For example, an individual who had specific experiences with high school parties might reach the conclusion that such behaviors are common and that memories of such events are consistently accurate. Other viewers might adopt a view that there was an appreciable chance of false memory and/or false testimony and have little reason to suspect guilt prior to the testimony. It is clear, however, that the values obtained from each of these perspectives indicate significant levels of doubt regarding whether the accused was guilty.
4. Implications and Limitations
The probability of guilt values obtained in this study allows certain conclusions. Survey research based upon responses from 1200 United States judges demonstrates that 2 out 3 judges believe that the judicial concept of “beyond a reasonable doubt” represents a probability of at least a 95% chance of guilt (Simon, 1969) . Using this legal standard, all the obtained results fall below this “beyond a reasonable doubt” standard. Some were far below that standard. In this case, understanding the non-random nature of the witness selection process is critical to accurately judging the implications of the evidence. While it may be true that any randomly chosen acquaintance is highly likely to have an accurate memory of Brett Kavanaugh, the non-random nature of the witness selection process dramatically impacted this probability estimate. What we know objectively is that the witness selection process was focused on finding negative evidence and identifying an individual from among his acquaintances to offer testimony and that it was based upon a memory that appeared to be vulnerable to false memory formation. The Bayes decision model incorporates and makes explicit the context of the evidence presented in this case. Conditions under which the memory was formed and how a witness was identified were represented in assessing the probability of guilt.
A limitation of this research is that it does not supply a complete consideration of all the research that might be relevant to the formation of memories. It is also true that estimates of impact yielded only rough approximations of the relevant probabilities. However, even with these approximations, the study does serve as an informative psychological lesson on the importance of considering the context of evidence in decision-making.