PSYC 162 Week 2 (lec. 4-8)

Lie Detection

People vs. Instruments

  • Can people tell when others are lying?

    • The average person is not very good at detecting lies

    • Certain professional groups are noticeably better (e.g. Secret Service agents)

  • Can instruments detect lies?

    • Polygraph (i.e., lie-detector test)

    • fMRI

  • People, in general, are poor lie detectors

    • People fare only slightly better than a coin toss at detecting lied

    • A meta-analysis of 253 studies showing that overall accuracy was approx. only 53%

    • Dr. Paul Ekman, one of the foremost experts on deception detection

      • Emotions show on our faces

Paul Ekman’s Approach to Studying Lie Detection

  • His theory is universal, all facial expressions are the same regardless of culture

  • Assumption: any ability to detect a lie requires that the lie arouse emotion (i.e., the liar has to be lying about something important)

  • “Low stakes” lies (e.g., lying about visiting a museum or about whether or not you like the taste of a drink) do not elicit emotion, so the lies would be hard to detect

    • Nothing to detect

    • It doesn’t matter

  • “High stakes” lies arouse emotions that create facial expressions that “leak through” attempts to mask them (potentially detectable)

    • Lies that are important

    • Hard to conceal your emotions, they’ll leak through and show on your face, those small movements are

      • Micro-expressions

  • Do facial expressions differ in any measurable way when someone is lying compared to when they are telling the truth?

  • Many studies using low stakes lies say “no”

  • Using high stakes lies, Ekman, Friesen & O’Sullivan (1988) found measurable differences using the Facial Action Coding System

    • Subtle facial movements associated with being happy occurred more often in the truthful condition

    • Subtle facial movements associated with negative emotions occurred more often in the lying condition

    • When you’re lying about something important, your facial expressions are different than if you’re telling the truth about something important

  • A general approach to studying whether or not people can detect when someone is lying

  • Ten confederates either lie or tell the truth in each condition (the lies are important, not trivial, and the liars are motivated to deceive) - they are high stakes lies

  • Five confederates are lying and five are telling the truth

  • Test-takers watch each person on tape and indicate whether or not the person on the tape is lying

    • Guilty trials - 5 guilty of lying

    • Innocent trials - 5 telling the truth

    • They could’ve interrogated people, could’ve used the false evidence ploy

  • Three categories of high-stakes lies in this study

    • Emotion Deception Judgment Task

      • A confederate talks about watching a nature video while watching an upsetting video (lying) - 5 lying

      • A confederate talks about watching a nature video while watching a nature video (telling the truth) - 5 telling truth

    • Opinion Deception Judgment Task

      • Express an opinion (e.g., about the death penalty) that goes against your strong beliefs (lying)

      • Express an opinion that agrees with your strong beliefs (telling the truth)

    • Crime Deception Judgment Task

      • Deny stealing $50 (lying)

      • Deny stealing $50 (telling the truth)

  • Secret service agents were better at detecting lies

Facial Action Coding System

  • Facial motion movements associated with emotions

  • Ex: face of anger

    • Eyebrows pulled down and together

    • Eyes opened wide, staring hard

    • Margins of lips rolled inward, pressed together

Wizards of Lie Detection

  • O’Sullivan & Ekman (2004)

  • They found 50 individuals (out 20,000 screened) who performed very well on all three high-stakes lie detection tests (opinions, a mock crime, and emotion)

  • A concern is that, when so many people are tested, some will perform well merely due to random chance, not because they are good at detecting lies (Bond & Uysal, 2007)

  • Whether or not that is true, some “wizards” made the most of their special status

  • Some of these “wizards” make a career out of it

    • Teaching people how to find liars

Ekman in Later Years

  • Trained law enforcement and TSA in lie detection

  • Stopped publishing scientific research on lie detection (to keep the information from hostile nations and terrorists)

  • In the absence of new science, skeptics doubt his body of work

  • His work has been the most influential and left a lasting legacy

  • Who killed my relative? Police officers’ ability to detect real-life high-stake lies

    • The police officers in the study correctly detected lying 50% of the time (i.e. hit rate = .50)

    • At random chance if the false alarm rate is 50%

    • False alarm rate is important!!

  • Follow up study they actually have the false alarm rate but they focus more on the accuracy of liars and those who tell the truth

Wright Whelan, Wagstaff & Wheatcroft (2015)

  • Participants:

    • 70 police officers & 37 undergraduates

  • Stimuli:

    • Videos of people making public appeals for help with missing relatives (high stakes)

    • 18 were honest, 18 deceptive

  • Participants decided if they were telling the truth or lying

  • Results:

  • Anytime hit rate = false alarm rate it = chance, there’s no information

  • This study is consistent with Paul’s idea that they have to be high-stakes lies to detect liars

  • Undergrads had a lower hit rate and a higher false alarm rate than police officers

  • If you were asked which is better police have hit rate of .70 and false alarm rate of .26, or the undergrads having hit rate of .60 and false alarm rate of .20

    • You shouldn’t be able to tell easily which is better

    • You just see one false and hit alarm rate for the police officers and one for the undergrads

    • You want to plot them on the ROC and then you’ll see which one is higher and which one is better

  • Hidden false alarm and hit rates/ Confidence scale

    • What’s hidden is that there’s an arbitrary decision that has been made in order to call things hits and false alarms

    • Okay to thing of the underlying thing as confidence

    • You’re picking an arbitrary point on a confidence scale

    • Ex: when you’re watching a video, you automatically make the decision if they’re lying or telling the truth

      • You have some degree of confidence

    • The farther you are from an arbitrary threshold, the more confident you are

    • Matters because there is nothing special about that point, if you move it, it changes both your false alarm and hit rates

    • Continous scale - can be 1 to whatever

Underlying Confidence (Neutral)Declaring more confidence to call them a liar

  • Those who previously ended in 5.1, 5.2, etc. were previously called liars, but now that the threshold changed, you call them truth-tellers because you changed your standard to require more confidence to call them liars

  • You’ll have fewer hits and fewer false alarms, both rates will go down

  • Calling more people liars

  • False alarm and hit rates both go up

ROC Analysis

  • Plot hit and false alarm rate

  • The black points more down to the left is conservative

  • The black points more up and to the right are more liberal

  • To get these points, you split the people into four groups, but have the same scale as before, some being more liberal and others more conservative

    • Four groups and you tell them how conservative or liberal they should be about calling someone a liar

  • You give the same instructions, and get 4 ROC points again, but before you do you run them through a Paul Ekman training to get better at detecting liars

  • If the training is effective you’ll see a higher ROC

  • Higher ROC is objectively better

Actively Prompting Information to Detect Lies

  • In most deception-deception experiments, all judges have to go on is “sender“ demeanor, which makes sense because theories posit that deception is detectable based on demeanor (e.g. microexpressions of emotion)

  • However, police do not merely passively watch demeanor, but instead actively and strategically question a suspect (e.g. they compare what the person says to what is known about the crime)

  • In short, what makes an expert adept at lie detection is not the passive observation of sender demeanor, but the active prompting of diagnostic information (i.e. interrogation)

  • Tested by using a confederate to induce some participants to cheat on a test involving answering hard trivia questions

    • You induce some people to cheat - guilty group

    • Don’t induce others to cheat - innocent group

    • Try to figure out when they deny cheating if they are lying or if they are telling the truth

  • The interviewers were five federal agents with substantial polygraph and interrogation experience

  • The interviews (3-18 minutes) were completely unscripted and varied substantially in style and substance from expert to expert

  • The hit rate was 39/40 = .98 and the false alarm rate was 1/49 = .02

Criticism of Levine et al. (2014)

  • The trivia questions were almost impossible for anyone to answer

    • Therefore, if someone answered several of them correctly, it was already easy to tell that they were probably cheating

    • Too easy to tell the liars

  • The fact that they were cheating would quickly become obvious if you asked them what else they knew about the trivia question (nothing)

    • They only knew the answer to the question, asked them something similar and they wouldn’t get it

  • Study didn’t accomplish getting on a higher ROC

A Cognitive Approach to Lie Detection

  • Vrij, Fisher, & Blank (2017)

  • Method to pass lie detection in a police interrogation

    • Impose cognitive load (because lying is more mentally demanding than truth-telling), such as asking interviewee to tell their story in reverse order

      • Harder to tell a lie than to tell the truth

      • Make interviewee engage in more mental labor

    • Encourage interviewee to say more (because liars tend to “weave a tangled web” the more they talk)

      • Harder to talk for a liar

    • Ask unexpected questions (because liars prepare themselves for anticipated interviews), such as general opening questions (expected) versus spatial questions (unexpected)

      • They think it’ll start with the beginning of the story

  • Results

    • Better overall accuracy for the cognitive vs. the standard approach

    • Standard approach: hit rate = .47, false alarm rate = .43

      • Almost equal means nothing, chance

    • Cognitive approach: hit rate = .67, false alarm rate = .33

      • Don’t need ROC analysis because hit rate is a good amount larger

Summary

  • Passive lie-detection is really hard

  • High-stakes lies are necessary for it to be possible (so that emotion leaks through in the form of microexpressions)

  • Using videos of high-stakes expressions of true or false claims about missing relatives, people are not bad at detecting lies (and police are a little better than that)

  • Lie-detection may be much better for an interviewer who is able to actively question the individual (though more work is needed to verify these controversial new findings)

The Polygraph

  • A polygraph (poly = many, graph = write) is an instrument that simultaneously records changes in physiological processes such as heart rate, blood pressure, respiration and electrical resistance of the skin (galvanic skin response)

  • The underlying theory of the polygraph is that when people lie they also get measurably nervous about lying. Heart rate increases, blood pressure goes up, breathing rhythms change, perspiration increases, etc

History of Polygraph

  • William Moulton Marston (1893-1947)

    • Student of Hugo Münsterberg at Harvard

    • Discovered correlation between blood pressure and arousal during lying

    • Invented Wonder Woman

  • John Augustus Larson

    • Rookie police officer in the Berkeley, CA, police department

    • Ph.D. in physiology from UC

      • Read Marston’s article “Physiological Possibilities of the Deception Test”

    • Improved Marston’s test through continuous recording of blood pressure

  • First real-world application

    • Berkeley sorority house - 1921

    • Money and an expensive ring had been stolen from rooms

  • Larson asked suspects questions while hooked up

    • “Are you interested in math?”

    • “Did you steal the money?”

    • “The test shows you stole it. Did you spend it?”

  • Suspect = a woman named Helen Graham

    • Her record showed a large change in blood pressure, skipped heartbeats and a halt in her breathing

    • She was arrested but later convinced Larson she was innocent

How to Question Suspect Taking Polygraph Test

  • Relevant/Irrelevant Test

    • Ask some irrelevant questions (“Are you interested in math?”)

    • Ask questions relevant to the crime (“Did you steal the money?”)

    • If arousal(relevant ) > arousal(irrelevant ) = guilty of lying

  • Comparison Question (control question) Test

    • Add a probable-lie comparison question (“Have you ever stolen anything?”)

    • Arousal(relevant ) > Arousal(comparison) = guilty of lying

  • Concealed Information (i.e., Guilty Knowledge) Test

    • Present one true and several false details of an incident that has not been publicized, so the true answer would be known only to the perpetrator

    • Innocent suspects will not be more aroused in response to the true item, but guilty suspects will be

    • Arousal(true detail) > Arousal(false details) = guilty of lying

    • Some people believe this approach is the best

Comparison Question Test

  • Imagine that the examiner rates the difference on each physiological measure using a 0 → 3 scale (summed range 0 → 9)

  • Further imagine that the examiner concludes that a subject is lying whenever the summed score across three measures is 5 or more

  • This test-taker gets a score of 0 + 3 + 2 = 5 and is judged to be guilty of lying

  • Is it a hit or a false alarm?

Example of Guilty Knowledge Test

  • Carmel et al. (2003)

  • Participants were instructed to enter the office of a named TA in Psychology and steal a CD with a red case containing an exam

  • Told to convince the examiner that they are innocent

  • Hooked up to polygraph and given “guilty knowledge“ questions

    • Color of the CD case (blue, green, red, purple) +1

    • Name of TA (name 1, name 2, name 3, name 4) +3

    • Subject of the exam (Anthro, Psych, Bio, Physics) +2

  • Scoring

    • For each question, if the “guilty knowledge” item elicits the largest response, +3 points; if 2nd largest response, +2 points; if 3rd largest response, +1 point, otherwise 0

    • Values are then summed up across all questions to produce a single score between 0 and 9 (this test-taker gets a score of 1 + 3 + 2 =6)

    • Decision rule: if score > 5 → “guilty of lying“, so this test-taker is judged guilty

    • Is it a hit or a false alarm?

  • Remember hit or false alarm rate involves an arbitrary criterion that is hidden from you

  • To find out if guilty knowledge or comparison question test, do the process for the ROC analysis for both

    • Look at the other false and hit rates

    • Split into conservative, liberal, and neutral groups

    • Plot the points

    • Higher ROC is better

    • More conservative → hit and false alarm rates go down

    • More liberal → hit and false alarm rates go up

Polygraph Accuracy

  • Many believe it is worst for the relevant-irrelevant method and best for the guilty knowledge method

  • The data suggest that all methods are about equally effective

Polygraph Accuracy Might be Lower in Practice

  • Analogy: medical tests seldom perform as well in general field use as their performance in initial evaluations seem to promise

    • Initial evaluations are typically conducted on people whose disease status is uncomplicated by other conditions that might interfere with test accuracy

    • Samples are drawn, tests conducted, and results analyzed under optimal conditions

  • Similar issues could pose problems for regular use of the polygraph

    • In experimental tests, polygraph instrumentation and maintenance and examiner training and proficiency are typically well above field situations

    • Testing is undertaken soon after the crime, so limited forgetting or “emotional distracting“ occurs

    • Countermeasures might be more of a problem (e.g. biting one’s tongue)

    • If polygraph examiners have preexisting knowledge of evidence against the suspect (e.g. fingerprint evidence), it can bias the interpretation of their lie-detection test

The Green River Killer

  • On July 15, 1982, the body of Wendy Lee Coffield was found in the Green River in a rural area of the City of Kent in King County, Washington. Another body was found in the river within a month. While surveying the riverbank near two bodies on August 15th, detectives discovered another body.

  • Thus began one of the longest and largest serial murder investigations in United States history. Eventually, the deaths of at least 48 women would be linked the Green River killer.

  • In the early months of 1987, investigators had a new suspect in relation to the Green River murders. Previously known to police, the newest suspect had been picked up for attempting to solicit an undercover police officer posing as a prostitute in May 1984. However, the man was released after he successfully passed a lie detector test.

  • They might’ve been to conservative with the ROC and that’s why they think he wasn’t guilty/he passed the test

    • That’s why it’s so important going on a higher ROC

  • They ended up catching him because of DNA

A World of Difference

  • Polygraph testing applied to a suspect in a criminal investigation

    • vs.

      • Polygraph testing used for screening purposes (e.g., screening applicants for a job in law enforcement)

Lie Detection in Connection with a Single Incident vs. Mass Screening

  • Polygraph testing has been used as a general screening device (e.g., for CIA applicants)

  • A much more problematic use of the polygraph

  • No specific incident, so only general questions can be asked (“have you ever had any unauthorized foreign contacts?” “Have you ever smoked marijuana?”)

  • Base rate of guilt (i.e., percent of examinees who are lying) is very low → more false alarms than hits

Base Rate Problem

  • this is what a good test looks like because the hit rate is way higher than the false alarm rate

  • Even though it’s a good test but if you test a lot of innocent people, people are still going to fail

  • Using a polygraph test in the "probable cause" scenario

    • The police already have reason to suspect a particular person of a crime (e.g., Gary Ridgeway)

    • For every 200 suspects given a polygraph, ~100 will be innocent and ~100 will be guilty

  • Using a polygraph test to search for a suspect (screening)

    • Almost everyone tested will be innocent

    • For every 200 employees given a polygraph, ~198 will be innocent and ~2 will be guilty

  • What’s the problem with screening?

    • Far more of the examinees are innocent than guilty (i.e., base rate of innocent far exceeds base rate of guilty)

    • Therefore, even a good test (high hit rate, low false alarm rate) will wrongly identify far more innocent people than guilty people

Legal Status of the Polygraph

  • Polygraph results are sometimes admissible in court

  • 1993 Supreme Court decision (Daubert v. Merrell Dow Pharmaceuticals)

    • Admissibility of scientific evidence determined case-by-case based on evidentiary hearings

    • Judge makes the call (after considering if there is a scientific consensus, will the jury understand the results, etc.)

  • 1998 Justice Clarence Thomas expressed concerns about polygraph testing

    • Lack of consensus about scientific validity and reliability

    • Concern that polygraph evidence usurps jury role

Polygraph as Coercion

  • Polygraphs are not used only to detect lies

  • Used to induce confessions

  • Presented as opportunity to prove innocence

  • “The results of a lie detector are not admissible in court, but if you confess during the course of interrogation, that’s admissible. The lie detector is essentially used in practice as a way to get people to confess to crimes” (Ken Alder, professor of history at Northwestern University and author of The Lie Detectors: The History of an American Obsession).

  • Even the lie detector’s harshest critics concede it can be a useful interrogation tool

Jurors and the Polygraph

  • Jurors generally find results persuasive

  • Results can change outcome of a trial

  • Experts are more skeptical than general public

Lies in the Brain

  • Functional magnetic resonance imaging (fMRI) detects “where” lying occurs in the brain

  • Does not measure physiological arousal, so it might even be able to detect low-stakes lies

  • Some areas of the brain began to light up when you told more low stake lies

  • fMRI shows you group differences but nothing about the individual that’s why it doesn’t help you