False Convictions and Their Numerous Overturns

False Convictions and Their Numerous Overturns Presented by: Mr. Freeman Date: Thursday 8 May 2025 Innumeracy In Action: The Misuse Of Probability In The 1995 Murder Trial Of OJ Simpson - Prosecution - OJ is more likely to be the killer because he was a domestic abuser Defense evidence - • 4.5 million cases of domestic abuse • 1800 women died of homicide • [4.5M/1,800] = "1 in 2500 chance OJ did it" Most errors in forensic medicine/science stem from confusion regarding probabilities and risk, because • Forensic analysis is only performed after the fact and... • When probabilistic reasoning is applied after the fact it is prone to hindsight bias due the need for cognitive closure • When risk is used after the fact it is prone to MSU The Monty Hall Problem Explained - Now what's the chance that you have picked the car? • - It's still 1 in 100 (1%) What is the chance you get the car if you do swap? • [100% -1%] = 99%!! Still Confused? You're Not Alone - All of the doors = all possibilities • probability of the first door] + [probability of all remaining doors] = 100% - The probability the car is behind the first door chosen stays the same no matter what happens with the remaining doors - The remaining door(s) equal [100% - (probability of the first door)] • If there's only 1 door left it represents all of the remaining probabilities of getting the car - How is this relevant to wrongful convictions and exonerations? - Counterfactual reasoning is the standard for a determination that a death was a homicide versus accident or natural, etc. • Also known as the "but-for" standard, as in but-for the intentional acts of another person, would the death have occurred? • The defense asserted that George Floyd's death was not due to the assault, prone restraint and associated compression asphyxia • Rather, they claimed that heart disease or illicit drug use caused the death How MHP Helps Us Recognize Hindsight Bias - The day before his murder, Mr. Floyd's risk of dying in a randomly selected 9 minute period of time was <1 in 15,000,000 (0.000007%) - After he died his risk of dying was 100% - His random 9 minute risk of dying didn't change after he died • Just like the first door choice probability - Thus, the risk of dying from the assault is [100% - 0.000007%] = 99.999993% Causation: An Essential Part Of Criminal Prosecutions Of Injury And Death All cases require medical evidence of physical harm • Injury and death, acute and chronic illness - All cases require evidence of wrongful action on the part of a defendant • Harm is due to actions committed or omitted - Causation is the (typically) medical opinion linking the harm and the wrongful action • Causation Is Consistently Confused With Diagnosis In Forensic Pathology A diagnosis can be seen • X-rays, MRI, CT Surgical/autopsy • There is direct evidence of diagnosis that can be observed - A cause cannot be seen • Always based on inference • There is only circumstantial evidence of causation that cannot be observed Risk Is Evaluated Using Epidemiologic Methods - Medical study of populations with similar injury or disease or exposure characteristics - - Purpose is to describe injury or disease • Injury Risk (Distribution) • Injury Risk Factors (Determinants) • Injury Risk Prevention (Deterrents) Epidemiologic concepts are woven into nearly all causation opinions Knowledge gap - Individual cause of death or injury determination depends on comparisons of risk Risk is assessed from populations (epidemiology) • Epidemiology is not directed at individuals - There is no standardized training in either individual (specific) causation or epidemiology in medicine - Can result in eminence-based practice (ipse dixit) that cannot be replicated by other experts Forensic Epidemiology - Forensic epidemiology is concerned with the investigation of specific causation using population-based information and techniques Cause And Manner Determination - Cause of death: Acute cardiac dysrhythmia during struggle • Positive toxicology for THC - Positive toxicology for THC - Mechanism of death: acute cerebral hypoxia - Manner of death: natural - Coroner determination, from ME report and investigation: • Minimal restraint prior to death • Natural MoD, death due to dysrhythmia (no mention of struggle) Post-Mortem Report: - No evidence of significant mechanical trauma - Chronic hypertrophic cardiomyopathy • Heart weight 540 g (normal for patient weight 350-400g) - Atherosclerotic cardiovascular disease • Focal 90% narrowing of LAD, multifocal 50% narrowing of LC coronary artery - Cardiac arrest prior to respiratory arrest • Multifocal pulmonary edema, onset at least hours prior to death FE Methods Provide An Evidence Basis For Opinion When Risk Is The Contended Issue - A&E nurse Benjamin Geen was accused of poisoning a number of patients and causing respiratory arrest • - 2 patients die of complications A criminal charge of GBH and multiple murder was brought based on the assertion that too many (17) patients suffered RA during Geen's shift, over a 42 day period • - No evidence of poisoning was found in any of the patients Geen was convicted on all 17 counts based entirely on the assertion that the only explanation for 17 RAs in 42 days is intentional poisoning - With no physical or direct evidence, counterfactual reasoning and hindsight bias based on the need for cognitive closure was the basis for the prosecution • The prosecution experts ruled out, as impossible, the explanation that the nurse did not intentionally cause the RAs and deaths Leaving only one explanation The defendant was sentenced to 17 life terms based on a purely statistical argument based on no statistics • Where do statistics on the risk of hospital death come from? Counterfactual Reasoning as Logical Fallacy • "When you have eliminated the impossible, whatever remains, however improbable, must be the truth." - Just because you say something is impossible doesn't make it so. MHP Reasoning: How Many RAs Would Be Expected During 42 Days Of Shifts At A&E If None Were Intentional? - Unless it was 0, the conviction is potentially unsafe - There are large and reliable US data sets that describe how often RAs occur in A&E • We would like to know 2 things: What's the chance that any nurse in an emergency department will encounter a patient with RA in 42 shifts? Given the admitting diagnoses among the victims, what was the chance that one or more of them would suffer an RA during their time in the ED? US National Emergency Department Data Analysis - 9 years of data, for a total of 197.8M ED admissions - During the 42 days during the months of interest (Dec-Feb) there were 10,109,698 admissions across 5,200 Eds • There were 145,328 episodes of RA (1.4% rate) • • This frequency equates to 28 RAs per ED per 42 days in December and February - If the average nurse sees 5 patients per shift, then there is a 95% chance of at least 1 RA during the 42 days - If they see 10 patients per shift, there is a 55% chance of at least 6 RAs, and an 8% (1 in 13) probability of at least 12. What About The Specific Patients Seen By The Nurse? - RA risk differs by admitting dx (e.g., pneumonia and sepsis will be higher than LE trauma) - Admitting diagnostic information was available for 12 of the 17 patients • A statistical model was build for each of the patients based on admitting dx, and extrapolated to the 5 patients with missing data Conclusions: - Prosecution theory • - - Zero background rate of RA during 42 days Evidence-based (epidemiologic) analysis • 1.4% among all hospital patients during 42 days • 13% among all patients with same diagnoses • 24% among THESE patients given their diagnoses If even only ¼ of the patients seen by the nurse were similar to these 12 during the 42 days, there was a 60% chance of at least 9 RAs