PSYCH 3CC3 Lecture 3: Deception Detection
19 Jun 2018
School
Department
Course
Professor
The literature shows that the things that interviewers are looking for in initial
BAI (Reid) are not strongly associated with deception.
-
In some cases, they're looking at cues more strongly associated with truth
telling.
-
We spend a lot of time lying.
Approximately 4 times per day.
○
Mostly harmless.
○
-
To what extent are we able to detect deception in the legal context?
-
Bottom-line = we aren't very good at it.
-
How Good At Detecting Deception?
Ekman and O'Sullivan (1991):
Ekman is the leading expert on facial expressions of emotion.
○
TV show, Lie to Me, based on him and his work.
○
Ss correctly or incorrectly report positive feelings while watching
pleasurable or gruesome film.
○
Truth judged by Secret Service, FBI, judges, police, psychiatrists, university
students, adult taking on-day course on deceit.
○
Videos previously found to contain nonverbal cues to lying: masking
smiles and higher pitch.
○
Judges were asked to rate both general lie-detection ability and their
performance in the task.
○
Results:
All but one group performed at chance levels.
§
Only U.S. Secret Service agents judge better than chance and than
all other groups (64% accuracy)
§
Accuracy uncorrelated with ratings of general ability or task
performance (except for deceit group).
§
Adult deceit course group accurately judged their accuracy.
§
○
Possible shortcomings:
Several earlier studies found no evidence that "experts" better at lie
detection than students.
§
Forensic relevance unclear: most forensic lies not about currently
felt emotions.
Police don't ask about feelings, they ask about specific aspects
of a crime.
□
So the extent of the relevancy is an issue□
§
○
-
Ekman, O'Sullivan and Frank (1999):
Participants make true or false statements about their opinions on
videotape.
○
U.S. law enforcement official and psychologists judge honesty while
viewing videotapes.
○
Overall accuracy ranges from 51% (mixed law enforcement officers) to
73% (federal officers)
One of the few studies that finds a significantly different accuracy.
§
○
Lie accuracy (% of correctly identified lies) ranges from 48% (mixed law
enforcement) to 80% (federal officers).
○
Leaves 20% error even in the very best group.
This is not good enough in the forensic context.
§
Minimal p value of statistical significance is 5% chance of error.
§
Here we're accepting 20% error over someone's life … not good
enough.
§
○
Truth accuracy (% correctly identified truths) ranges from 54% (mixed law
enforcement) to 66% (federal officers)
Not as good at detecting truthful statements.
§
This is bias we are concerned about.
§
In the legal system, we'd rather let a criminal go than convict an
innocent person.
§
○
Overall accuracy.
More accurate to judge a lie as a lie.
§
Only Academic psych's had same accuracy across the board.
§
Interestingly, people who were the best at detecting lies were not
that good at detecting true statements.
They have a bias towards calling things lies when they are
actually truths.
□
Generally, in the literature we see a truth bias instead -
people are more inclined to deem false statements as being
truthful when it really isn't.
□
§
○
-
Bond and DePaulo (2006):
Meta-analysis of 206 studies and 24000 judges = 54% overall accuracy.
○
This has been the finding overall in the literature -- suggests that we are
not much better than chance.
○
47 % accuracy in detecting lies.
○
61% accuracy in detecting truths.
Truth bias: we have a tendency to rate more statements as true
than false.
§
We tend to believe what people say.
§
○
Greater accuracy when hearing than seeing and hearing liars and truth-
tellers.
People are less accuracy when they see + hear statement.
§
Suggests that we are more deceived by a person's facial expressions
and nonverbal cues than we are enlightened by them.
§
We are looking at the wrong things when looking for physical cues.
§
○
The more significant, more important a lie is makes it easier to detect.
Motivation makes liars easier to detect.
§
In lab studies, we see low-risk lies so motivation to be a successful
liar is relatively low.
§
In the forensic/legal context, the motivation is high, but the
motivation makes it more likely that you'll be detected.
§
Likely because you're more anxious/nervous.
§
○
-
Whelan, Wagstaff and Wheatcroft (2015):
Used 36 videos of relatives asking for information about missing or dead
relatives.
○
Half of speakers were criminally involved in the case (liars), the other half
were truth-tellers and had nothing to do with the person's
disappearance/death.
○
Note - these are high-stakes lies.
○
How good were people at detecting who the liars are?
○
Participants:
Firearms officers in UK (n=37)
§
CID officers in UK (n=33)
§
Undergraduates (n=37)
§
○
Findings (detection)
Note that only undergraduates show the truth bias typically
reported in the literature.
§
Accuracy is still significantly above chance, but a pretty large error
rate still.
§
Confirms that highly motivated liars are easier to detect.
§
○
Findings (confidence)
Confidence rated on 5-point scale from 1 (not at all confident) to 5
(very confident)
§
All above average pretty much.
§
People are more confident when they've correctly detected a lie,
less confident when they've correctly identified a truth.
§
** not as important.
§
○
-
Whelan, Wagstaff and Wheatcroft (2014);
2 successful raters viewed 9 videos of people pleading for help in locating
missing relative.
5 honest, 4 deceptive.
§
○
Indicated which cues using to detect deceit.
Very difficult for most people to do.
§
Often, we 'know' people are lying or telling the truth because we
have a certain feeling about it
§
○
Deceptive appeals:
Fake emotion
§
Lack of emotion
§
Distancing
§
Implausibility
§
Negative personal reaction
Turns out to be a very good predictor of whether or not we're
right about our judgement.
□
§
○
Honest appeals:
Genuine sadness emotion
§
Genuine/heartfelt
§
Containing emotion
§
Sympathetic personal reaction
No indication of what it was about the person that made you
feel positively, but it is still a good indicator for whether or
not your judgment was correct.
□
§
○
-
Facial/Body Cues to Deception?
Vrij (1998)
Asked people on the street and law enforcement officials about what it is they are
looking for when determining if someone is lying or not.
Avoiding eye contact.
Strongly considered.a.
We tend to assume that this is an indication of lying, but the literature
doesn't support this.
b.
1.
More smiling and laughter.2.
Higher rate of eye blinking.3.
Nervous fidgeting.4.
More illustrative gestures.5.
More movement of legs, feet, hands. 6.
More body, head movements.7.
More shrugging.8.
Virtually none of these things are strongly associated with deception.
No consistent, reliable association.
○
But this is what normal people AND LAW ENFORCEMENT say they're
looking for to determine if someone is telling the truth.
○
-
"the meta-analytic literature fails to identify any pattern of facial or body
movement that generally signals deception. However, some studies designed to
develop indicators based on these movements show some ability to
discriminate lying from truth-telling … these studies suggest that the right
measures of facial and motion feature can offer accuracy better than chance for
the detection of deception from demeanor in somewhat realistic situations."
Not a v strong statement.
○
But for the most part, this acknowledges the fact that our ability to detect
deception using these cues is about chance.
○
-
Meta-Analysis
Looked at around 150 cues of deception.
-
But there's nothing in the list consistently associated with truth-telling or lying.
-
No aspect of facial or body movement that can reliably help us identify lying.
-
This would explain why people are better at identifying a false statement when
they are listening rather than looking.
When you are looking, you're watching for all of these incorrect cues
which are no reliably indicative of lying.
○
-
Working Cues to Deception
Liars less forthcoming.
They give you less information.a.
1.
Liars tell less compelling stories.2.
Liars leave a more negative impression.
Recall study above.a.
We don't really understand the reasons, but it's an excellent predictor of
deception.
b.
3.
Liars are more tense.4.
Liars make fewer errors, unusual content.5.
The Language of Deception
Low verbal immediacy
Distancing of the individual from the things they're talking about.
○
Past tense, passive voice.
○
-
Fewer details in story.
-
Impressions of verbal uncertainty.
Not clear or definitive enough.
○
Could be the product of not having enough time to come up with a story.
○
-
Impressions of nervousness.
Very subjective.
○
An innocent/truthful person could also appear extremely nervous.
○
-
Lack of logical structure to story
-
Lack of plausibility to story.
-
Raised pitch of speech.
-
NOTE: None of these can be used to reliably identify a liar.
There's a lot of variation.
○
Does not actually discriminate a liar from a truth-teller.
○
-
Matsumoto and Hwang (2014)
Used word count program Linguistic Inquiry and Word Count (LIWC) to analyze true
and false statements, verbal and written, about mock crime - "theft' of $100 cheque.
Monetary reward given for escaping detection.
True and false accounts differed significantly in word usage, based on previous
LIWC research.
-
Word usage predicted truth from lies with 69% accuracy for written statements.
-
Word usage predicted truth from lies with 71% accuracy for verbal interview
statements.
-
For written statements, liars used:
Fewer words overall
○
Fewer words related to money
○
Fewer words related to negation, motion, time
○
More tentative, equivocation words
○
More positive emotion words
○
-
For interview statements, liars used:
Fewer words related to motion
○
More words related to first person ('I', 'me', 'my', etc.)
○
-
Hauch et al (2014)
Meta-analysis indicated that liars:
Experienced more cognitive load.
-
Expressed more negative emotion.
-
Distanced themselves more from events.
-
Used fewer sensory-perceptual words.
-
Referred less often to cognitive processes.
-
Expressed as much certainty as truth-teller.
Inconsistent with other studies.
○
-
Detection Easiest When:
Lies have high personal relevance
-
Stakes of deception are high
-
Liar has little chance to rehearse
-
Problem: we can't test our ability to detect lies because we have no what of
knowing who is a real liar.
-
Polygraph vs Lie Detector
Polygraph: a machine to record voltages
Greek meaning "many writings"
○
-
Lie detector = application of a polygraph.
Sometimes used to detect lies.
○
-
Polygraph record for lie detection measures …
Blood pressure
○
Heart rate
○
Respiration (breathing)
○
Galvanic skin response (sweating; conductance/resistance)
○
-
Many kinds of polygraph tests, the most common is the CQT.
-
Control Question Test (CQT)
Relevant questions: about crime
E.g., "Did you shoot Sam Smith?"
○
-
Irrelevant questions: about nothing
E.g., "Do you live in Toronto?"
○
-
Control questions:
E.g., "Have you ever injured anyone?"
○
-
Compare control vs relevant
When telling the truth, we expect arousal to be higher for control
questions than for relevant questions.
Because there's no reason for them to get upset about relevant
questions because they know they didn't do it.
§
○
When lying, much more arousal is expected for relevant questions than
control questions.
○
-
Compare three pairs of control + relevant questions.
-
CQT Procedure
Pre-test
Convince subject of test accuracy
Often done through a card trick.
§
○
Learn personal details
○
Develop irrelevant questions
○
-
Test
Ask and score questions
○
10 questions: 3 pairs of control and relevant + 4 irrelevant
○
-
Post-test
Sum scores, interpret results
○
Score difference between pairs: 1, 2, 3 or -1, -2, -3 or o
○
Add them up …
If large enough in a positive direction - truth
§
If large enough in a negative direction - liar
§
Close to zero = inconclusive
§
○
-
Alternatives to CQT
Directed Lie Test (DLT)
Broad control Q's
○
Subject instructed to lie to all.
This way, you have a true comparison of arousal.
§
○
-
Positive Control Test
Relevant questions asked twice.
○
Tell truth once, lie once.
○
-
Critique of CQT
Truth-teller may respond more to relevant that to control questions.
But relevant question itself is more arousing when you're innocent
because you're getting upset about being accused.
○
Makes truth-tellers look like liars.
○
-
Liars may respond more to control than to relevant questions if they have been
asked before.
They have practice saying "No, I didn't do it."
○
-
Not admissible as evidence in court because it's not totally reliable.
-
Guilty Knowledge Test
Used primarily in Europe.
-
Not really about truth-telling because it's about how much guilty knowledge you
have.
No lie detection; knowledge detection.
○
-
GSR response to important versus unimportant information.
-
5-alternative multiple-choice questions.
GSR arousal measured when examiner says the option they know to be
the truth.
○
Don't have to actually answer the question.
○
One of the choice is known only to the perpetrator and police.
This is why it's not used in the U.S.
§
It's very difficult to keep case details out of the media/from the
public.
§
○
-
Pros and Cons of GKT
Pro: It does not require subject to respond to questions.
-
Con: Usable only if some aspects of the crime are unknown to the public.
-
Testing Polygraph
Not very good
-
U.S. Supreme Court ruled it inadmissible in 1988.
Because it takes away jury's right to decide on guilt.
○
Not even because it's inaccurate.
○
-
Mock crime studied: subjects commit 'mock crime', then get tested.
-
Field studies: comparing original with 'blind' scoring.
We take a bunch of people who have taken polygraph (some liars, some
truth-tellers based on court rulings)
○
Polygraph tech has to look at just the polygraph records and determine
truth/lie.
○
Look at the accuracy of their blind judgments.
○
-
Field analogue studies: subject commits detectable crime, then tested.
-
Honts and Raskin (1988): How good is polygraph lie detection?
Looked at the accuracy of the CQT (DLT specifically); field test where they looked at
the polygraph recordings of innocent and guilty individuals (no confession vs
confession).
Looked at the originial judgements of the individuals who conducted the
polygraph.
'Actually' Innocent
91% correctly judged as innocent
§
0% judged as guilty
§
9% judged as inconclusive
§
○
'Actually' Guilty
8% incorrectly judged as innocent
§
92% correctly judges as guilty
§
0% judges as inconclusive
§
○
-
Looks pretty good for polygraphy …
-
Gave the records to blind polygraph examiners who weren't present at the time
of the polygraph test.
'Actually' Innocent
62% correctly judged as innocent
§
14% incorrectly judges as guilty
§
23% judged as inconclusive
§
MUCH WORSE
§
○
'Actually' Guilty
Results exactly the same.
§
○
-
This suggests that polygraph records are not that reliable in the sense that
individual expert judgement is relied on to determine guilt/innocence and
whenever judgement is in play, there's always some error.
We obviously don't want this much error.
○
-
Also suggests that the original polygrapher was using something else, some
other cue, in the presence of the suspect.
They were adding something they saw/heard/felt about the examiner into
their judgements other than simply their interpretation of the polygraph
record.
○
-
We know that seeing someone tell a lie is misleading.
-
Patrick and Iacono (1991)
Polygraph records from the RCMP; looked at the records of innocent and guilty people
as determined by confession and/or court conviction.
Original examiner's results:
'Actually' Innocent
73% correctly judged as innocent
§
8% incorrectly judges as guilty
§
19% judged as inconclusive
§
○
'Actually' Guilty
0% incorrectly judged as innocent
§
98% correctly judged as guilty
§
2% judged as inconclusive
§
○
-
What we see here is that in general, CQT lie detection tests are better at
detecting liars that they are at clearing truth-tellers -- widely replicated finding .
-
Blind examiner's results:
'Actually' Innocent
30% correctly judges as innocent
§
24% incorrectly judged as guilty
§
46% judged as inconclusive
§
○
'Actually' Guilty
2% incorrectly judged as innocent
§
92% correctly judged as guilty
§
6% judged as inconclusive
§
○
-
Again, we see the bias - we're much better at identifying liars with polygraph
CQT than we are at clearing truth-tellers.
Huge difference in this case.
○
Horrible for forensic purposes.
○
This is why polygraphs are not trusted in Canada.
○
-
Also indicates again that the original polygraphers had to be using something
else to get their results.
Something from the suspect themselves.
○
-
Granhag and Vrij (2005)
Survey of the literature; looked at other studies which compared the accuracy of
detecting truth vs deception in CQT; all had relatively small numbers of participants
(n).
Innocent = prop of individuals correctly judged to be innocent.
Not that good.
○
Numbers as high as 73%, but number mostly around chance.
○
Philosophy is that it's better to let a guilt person go than to convict an
innocent person and that doesn't seem to be happening here.
○
-
Guilty = prop of individuals correctly judged as being guilty.
Much higher numbers (in the 80s)
○
-
This is the bias the CQT typically has: it is better at detecting lies than it is at
clearing truth-tellers.
-
It appears that the original examiner is not just using the polygraph record to
make judgements.
They're looking at other cues.
○
Looks like he sees guilt where is does exist than they see innocent.
○
-
Review of the Guilty Knowledge Test
Individual in GKT doesn't have to answer any questions at all; three studies (Ben-
Shakhar and Elaad (2003), Elaad (1990), Elaad, Ginton and Jugnman (1992)).
Correctly judged as innocent.
95%
○
98%
○
94%
○
GKT much better at detecting individuals telling the truth.
○
-
Correctly judged as guilty.
85%
○
42%
○
76%
○
GKT much worse at detecting individuals in possession of guilty
knowledge.
○
Opposite bias from CQT.
○
-
NRC (2003): Polygraph Review
Theoretical rationale re differential arousal is weak.
There's no really strong data to support the expected effect.
○
Basic assumption is very questionable.
○
-
Response measured not unique to deception.
HR, increased BP, etc. can all be indicators of many other things.
○
-
Test overestimate field accuracy.
Accuracy in field is not as high as some of the studies, which even aren't
that high.
○
-
Accuracy may be degraded by countermeasures.
You can defeat polygraph measures by using countermeasures.
○
-
Computer analysis may improve accuracy.
Why not make a computer program that can more accurately detect
lies/truths.
○
If they were 100% accurate, however, the likelihood of people agreeing to
take them would go down.
○
-
May promote admissions, instill confidence, deter crime
It gives the public greater confidence to detect criminals prior to trials.
○
It may lead to many things.
○
Weak rationale, but the NRC wanted to be as balanced as possible and
find some positive points.
○
-
P300 Brain Response
When a stimulus is presented to us, we evoke an ERP.
-
There are several components to an ERP.
They are described in terms of whether changes occur in a positive or
negative direction and how long after the stimulus presentation they
occur.
○
-
One of the changes that occurs happens 300 ms after the presentation of the
stimuli and occurs in a positive direction - P300 brain response
-
Green line = any stimulus, red line = relevant stimulus
The difference between the two is what's supposed to indicate what a
person knows (consider rationale of GKT)
○
○
-
"Brain Fingerprinting" popularized by Dr. Farwell
He didn't develop it, but he made it into a company.
○
-
There's another possible measure - N400
-
NRC (2003): P300-GKT
May not be better than traditional polygraph.
-
Would polygraph + P300-GKT be better?
We have no evidence to support this.
○
-
There might be effective countermeasures.
There might be a way to beat this.
○
-
fMRI
Suggested that the areas of the brain that are active when a person is telling a
lie are different from the areas active when a person is telling the truth.
-
There is some evidence for this, but it's not strong.
-
So far, there's never been a real life use of fMRI to detect deception in a
forensic setting (other than research).
Because it is extremely time consuming and expensive.
○
Other lie detection is much cheaper.
○
-
There's also some question about whether the difference in brain area activity
during lies/truths are reliable measures.
E.g., is it the same for all people?
○
-
Facial Thermography
Involves taking the temperature of the skin on the face.
-
When we're anxious or embarrassed, we blush and the skin on our face/cheeks
gets very warm.
-
This is actually in real life use in a number of places (e.g., airport security)
-
Beauty of this is that the person being recorded doesn't even know it's
happening.
Scanner can be mounted on a wall and secretely scan face.
○
-
Won't be convicted on this basis, but if you do have a temperature spike during
questioning, you might have your bags checked or more questions asked.
-
Other Measures?
None of the ones that have emerged have been allowed to use in court.
-
In US is has nothing to do with the reliability, but they'll all be inadmissible
because they take away the jury's right to determine guilt on their own.
-
It may one day be allowed in Canada.
-
Summary
Looking at physiological measures is not terribly good.
-
Even human judgement.
-
There are studies that groups of individuals are better than individuals at
detecting truths based on evidence.
-
Problem is that there are a lot of times where cases are only seen by judges.
Suggests that judges are probably going to be less good at detecting
deception on their own. s
○
-
Deception Detection
Sunday, February 4, 2018
7:01 PM
The literature shows that the things that interviewers are looking for in initial
BAI (Reid) are not strongly associated with deception.
-
In some cases, they're looking at cues more strongly associated with truth
telling.
-
We spend a lot of time lying.
Approximately 4 times per day.
○
Mostly harmless.
○
-
To what extent are we able to detect deception in the legal context?
-
Bottom-line = we aren't very good at it.
-
How Good At Detecting Deception?
Ekman and O'Sullivan (1991):
Ekman is the leading expert on facial expressions of emotion.
○
TV show, Lie to Me, based on him and his work.
○
Ss correctly or incorrectly report positive feelings while watching
pleasurable or gruesome film.
○
Truth judged by Secret Service, FBI, judges, police, psychiatrists, university
students, adult taking on-day course on deceit.
○
Videos previously found to contain nonverbal cues to lying: masking
smiles and higher pitch.
○
Judges were asked to rate both general lie-detection ability and their
performance in the task.
○
Results:
All but one group performed at chance levels.
§
Only U.S. Secret Service agents judge better than chance and than
all other groups (64% accuracy)
§
Accuracy uncorrelated with ratings of general ability or task
performance (except for deceit group).
§
Adult deceit course group accurately judged their accuracy.
§
○
Possible shortcomings:
Several earlier studies found no evidence that "experts" better at lie
detection than students.
§
Forensic relevance unclear: most forensic lies not about currently
felt emotions.
Police don't ask about feelings, they ask about specific aspects
of a crime.
□
So the extent of the relevancy is an issue
□
§
○
-
Ekman, O'Sullivan and Frank (1999):
Participants make true or false statements about their opinions on
videotape.
○
U.S. law enforcement official and psychologists judge honesty while
viewing videotapes.
○
Overall accuracy ranges from 51% (mixed law enforcement officers) to
73% (federal officers)
One of the few studies that finds a significantly different accuracy.
§
○
Lie accuracy (% of correctly identified lies) ranges from 48% (mixed law
enforcement) to 80% (federal officers).
○
Leaves 20% error even in the very best group.
This is not good enough in the forensic context.
§
Minimal p value of statistical significance is 5% chance of error.
§
Here we're accepting 20% error over someone's life … not good
enough.
§
○
Truth accuracy (% correctly identified truths) ranges from 54% (mixed law
enforcement) to 66% (federal officers)
Not as good at detecting truthful statements.
§
This is bias we are concerned about.
§
In the legal system, we'd rather let a criminal go than convict an
innocent person.
§
○
Overall accuracy.
More accurate to judge a lie as a lie.
§
Only Academic psych's had same accuracy across the board.
§
Interestingly, people who were the best at detecting lies were not
that good at detecting true statements.
They have a bias towards calling things lies when they are
actually truths.
□
Generally, in the literature we see a truth bias instead -
people are more inclined to deem false statements as being
truthful when it really isn't.
□
§
○
-
Bond and DePaulo (2006):
Meta-analysis of 206 studies and 24000 judges = 54% overall accuracy.
○
This has been the finding overall in the literature -- suggests that we are
not much better than chance.
○
47 % accuracy in detecting lies.
○
61% accuracy in detecting truths.
Truth bias: we have a tendency to rate more statements as true
than false.
§
We tend to believe what people say.
§
○
Greater accuracy when hearing than seeing and hearing liars and truth-
tellers.
People are less accuracy when they see + hear statement.
§
Suggests that we are more deceived by a person's facial expressions
and nonverbal cues than we are enlightened by them.
§
We are looking at the wrong things when looking for physical cues.
§
○
The more significant, more important a lie is makes it easier to detect.
Motivation makes liars easier to detect.
§
In lab studies, we see low-risk lies so motivation to be a successful
liar is relatively low.
§
In the forensic/legal context, the motivation is high, but the
motivation makes it more likely that you'll be detected.
§
Likely because you're more anxious/nervous.
§
○
-
Whelan, Wagstaff and Wheatcroft (2015):
Used 36 videos of relatives asking for information about missing or dead
relatives.
○
Half of speakers were criminally involved in the case (liars), the other half
were truth-tellers and had nothing to do with the person's
disappearance/death.
○
Note - these are high-stakes lies.
○
How good were people at detecting who the liars are?
○
Participants:
Firearms officers in UK (n=37)
§
CID officers in UK (n=33)
§
Undergraduates (n=37)
§
○
Findings (detection)
Note that only undergraduates show the truth bias typically
reported in the literature.
§
Accuracy is still significantly above chance, but a pretty large error
rate still.
§
Confirms that highly motivated liars are easier to detect.
§
○
Findings (confidence)
Confidence rated on 5-point scale from 1 (not at all confident) to 5
(very confident)
§
All above average pretty much.
§
People are more confident when they've correctly detected a lie,
less confident when they've correctly identified a truth.
§
** not as important.
§
○
-
Whelan, Wagstaff and Wheatcroft (2014);
2 successful raters viewed 9 videos of people pleading for help in locating
missing relative.
5 honest, 4 deceptive.
§
○
Indicated which cues using to detect deceit.
Very difficult for most people to do.
§
Often, we 'know' people are lying or telling the truth because we
have a certain feeling about it
§
○
Deceptive appeals:
Fake emotion
§
Lack of emotion
§
Distancing
§
Implausibility
§
Negative personal reaction
Turns out to be a very good predictor of whether or not we're
right about our judgement.
□
§
○
Honest appeals:
Genuine sadness emotion
§
Genuine/heartfelt
§
Containing emotion
§
Sympathetic personal reaction
No indication of what it was about the person that made you
feel positively, but it is still a good indicator for whether or
not your judgment was correct.
□
§
○
-
Facial/Body Cues to Deception?
Vrij (1998)
Asked people on the street and law enforcement officials about what it is they are
looking for when determining if someone is lying or not.
Avoiding eye contact.
Strongly considered.a.
We tend to assume that this is an indication of lying, but the literature
doesn't support this.
b.
1.
More smiling and laughter.2.
Higher rate of eye blinking.3.
Nervous fidgeting.4.
More illustrative gestures.5.
More movement of legs, feet, hands. 6.
More body, head movements.7.
More shrugging.8.
Virtually none of these things are strongly associated with deception.
No consistent, reliable association.
○
But this is what normal people AND LAW ENFORCEMENT say they're
looking for to determine if someone is telling the truth.
○
-
"the meta-analytic literature fails to identify any pattern of facial or body
movement that generally signals deception. However, some studies designed to
develop indicators based on these movements show some ability to
discriminate lying from truth-telling … these studies suggest that the right
measures of facial and motion feature can offer accuracy better than chance for
the detection of deception from demeanor in somewhat realistic situations."
Not a v strong statement.
○
But for the most part, this acknowledges the fact that our ability to detect
deception using these cues is about chance.
○
-
Meta-Analysis
Looked at around 150 cues of deception.
-
But there's nothing in the list consistently associated with truth-telling or lying.
-
No aspect of facial or body movement that can reliably help us identify lying.
-
This would explain why people are better at identifying a false statement when
they are listening rather than looking.
When you are looking, you're watching for all of these incorrect cues
which are no reliably indicative of lying.
○
-
Working Cues to Deception
Liars less forthcoming.
They give you less information.a.
1.
Liars tell less compelling stories.2.
Liars leave a more negative impression.
Recall study above.a.
We don't really understand the reasons, but it's an excellent predictor of
deception.
b.
3.
Liars are more tense.4.
Liars make fewer errors, unusual content.5.
The Language of Deception
Low verbal immediacy
Distancing of the individual from the things they're talking about.
○
Past tense, passive voice.
○
-
Fewer details in story.
-
Impressions of verbal uncertainty.
Not clear or definitive enough.
○
Could be the product of not having enough time to come up with a story.
○
-
Impressions of nervousness.
Very subjective.
○
An innocent/truthful person could also appear extremely nervous.
○
-
Lack of logical structure to story
-
Lack of plausibility to story.
-
Raised pitch of speech.
-
NOTE: None of these can be used to reliably identify a liar.
There's a lot of variation.
○
Does not actually discriminate a liar from a truth-teller.
○
-
Matsumoto and Hwang (2014)
Used word count program Linguistic Inquiry and Word Count (LIWC) to analyze true
and false statements, verbal and written, about mock crime - "theft' of $100 cheque.
Monetary reward given for escaping detection.
True and false accounts differed significantly in word usage, based on previous
LIWC research.
-
Word usage predicted truth from lies with 69% accuracy for written statements.
-
Word usage predicted truth from lies with 71% accuracy for verbal interview
statements.
-
For written statements, liars used:
Fewer words overall
○
Fewer words related to money
○
Fewer words related to negation, motion, time
○
More tentative, equivocation words
○
More positive emotion words
○
-
For interview statements, liars used:
Fewer words related to motion
○
More words related to first person ('I', 'me', 'my', etc.)
○
-
Hauch et al (2014)
Meta-analysis indicated that liars:
Experienced more cognitive load.
-
Expressed more negative emotion.
-
Distanced themselves more from events.
-
Used fewer sensory-perceptual words.
-
Referred less often to cognitive processes.
-
Expressed as much certainty as truth-teller.
Inconsistent with other studies.
○
-
Detection Easiest When:
Lies have high personal relevance
-
Stakes of deception are high
-
Liar has little chance to rehearse
-
Problem: we can't test our ability to detect lies because we have no what of
knowing who is a real liar.
-
Polygraph vs Lie Detector
Polygraph: a machine to record voltages
Greek meaning "many writings"
○
-
Lie detector = application of a polygraph.
Sometimes used to detect lies.
○
-
Polygraph record for lie detection measures …
Blood pressure
○
Heart rate
○
Respiration (breathing)
○
Galvanic skin response (sweating; conductance/resistance)
○
-
Many kinds of polygraph tests, the most common is the CQT.
-
Control Question Test (CQT)
Relevant questions: about crime
E.g., "Did you shoot Sam Smith?"
○
-
Irrelevant questions: about nothing
E.g., "Do you live in Toronto?"
○
-
Control questions:
E.g., "Have you ever injured anyone?"
○
-
Compare control vs relevant
When telling the truth, we expect arousal to be higher for control
questions than for relevant questions.
Because there's no reason for them to get upset about relevant
questions because they know they didn't do it.
§
○
When lying, much more arousal is expected for relevant questions than
control questions.
○
-
Compare three pairs of control + relevant questions.
-
CQT Procedure
Pre-test
Convince subject of test accuracy
Often done through a card trick.
§
○
Learn personal details
○
Develop irrelevant questions
○
-
Test
Ask and score questions
○
10 questions: 3 pairs of control and relevant + 4 irrelevant
○
-
Post-test
Sum scores, interpret results
○
Score difference between pairs: 1, 2, 3 or -1, -2, -3 or o
○
Add them up …
If large enough in a positive direction - truth
§
If large enough in a negative direction - liar
§
Close to zero = inconclusive
§
○
-
Alternatives to CQT
Directed Lie Test (DLT)
Broad control Q's
○
Subject instructed to lie to all.
This way, you have a true comparison of arousal.
§
○
-
Positive Control Test
Relevant questions asked twice.
○
Tell truth once, lie once.
○
-
Critique of CQT
Truth-teller may respond more to relevant that to control questions.
But relevant question itself is more arousing when you're innocent
because you're getting upset about being accused.
○
Makes truth-tellers look like liars.
○
-
Liars may respond more to control than to relevant questions if they have been
asked before.
They have practice saying "No, I didn't do it."
○
-
Not admissible as evidence in court because it's not totally reliable.
-
Guilty Knowledge Test
Used primarily in Europe.
-
Not really about truth-telling because it's about how much guilty knowledge you
have.
No lie detection; knowledge detection.
○
-
GSR response to important versus unimportant information.
-
5-alternative multiple-choice questions.
GSR arousal measured when examiner says the option they know to be
the truth.
○
Don't have to actually answer the question.
○
One of the choice is known only to the perpetrator and police.
This is why it's not used in the U.S.
§
It's very difficult to keep case details out of the media/from the
public.
§
○
-
Pros and Cons of GKT
Pro: It does not require subject to respond to questions.
-
Con: Usable only if some aspects of the crime are unknown to the public.
-
Testing Polygraph
Not very good
-
U.S. Supreme Court ruled it inadmissible in 1988.
Because it takes away jury's right to decide on guilt.
○
Not even because it's inaccurate.
○
-
Mock crime studied: subjects commit 'mock crime', then get tested.
-
Field studies: comparing original with 'blind' scoring.
We take a bunch of people who have taken polygraph (some liars, some
truth-tellers based on court rulings)
○
Polygraph tech has to look at just the polygraph records and determine
truth/lie.
○
Look at the accuracy of their blind judgments.
○
-
Field analogue studies: subject commits detectable crime, then tested.
-
Honts and Raskin (1988): How good is polygraph lie detection?
Looked at the accuracy of the CQT (DLT specifically); field test where they looked at
the polygraph recordings of innocent and guilty individuals (no confession vs
confession).
Looked at the originial judgements of the individuals who conducted the
polygraph.
'Actually' Innocent
91% correctly judged as innocent
§
0% judged as guilty
§
9% judged as inconclusive
§
○
'Actually' Guilty
8% incorrectly judged as innocent
§
92% correctly judges as guilty
§
0% judges as inconclusive
§
○
-
Looks pretty good for polygraphy …
-
Gave the records to blind polygraph examiners who weren't present at the time
of the polygraph test.
'Actually' Innocent
62% correctly judged as innocent
§
14% incorrectly judges as guilty
§
23% judged as inconclusive
§
MUCH WORSE
§
○
'Actually' Guilty
Results exactly the same.
§
○
-
This suggests that polygraph records are not that reliable in the sense that
individual expert judgement is relied on to determine guilt/innocence and
whenever judgement is in play, there's always some error.
We obviously don't want this much error.
○
-
Also suggests that the original polygrapher was using something else, some
other cue, in the presence of the suspect.
They were adding something they saw/heard/felt about the examiner into
their judgements other than simply their interpretation of the polygraph
record.
○
-
We know that seeing someone tell a lie is misleading.
-
Patrick and Iacono (1991)
Polygraph records from the RCMP; looked at the records of innocent and guilty people
as determined by confession and/or court conviction.
Original examiner's results:
'Actually' Innocent
73% correctly judged as innocent
§
8% incorrectly judges as guilty
§
19% judged as inconclusive
§
○
'Actually' Guilty
0% incorrectly judged as innocent
§
98% correctly judged as guilty
§
2% judged as inconclusive
§
○
-
What we see here is that in general, CQT lie detection tests are better at
detecting liars that they are at clearing truth-tellers -- widely replicated finding .
-
Blind examiner's results:
'Actually' Innocent
30% correctly judges as innocent
§
24% incorrectly judged as guilty
§
46% judged as inconclusive
§
○
'Actually' Guilty
2% incorrectly judged as innocent
§
92% correctly judged as guilty
§
6% judged as inconclusive
§
○
-
Again, we see the bias - we're much better at identifying liars with polygraph
CQT than we are at clearing truth-tellers.
Huge difference in this case.
○
Horrible for forensic purposes.
○
This is why polygraphs are not trusted in Canada.
○
-
Also indicates again that the original polygraphers had to be using something
else to get their results.
Something from the suspect themselves.
○
-
Granhag and Vrij (2005)
Survey of the literature; looked at other studies which compared the accuracy of
detecting truth vs deception in CQT; all had relatively small numbers of participants
(n).
Innocent = prop of individuals correctly judged to be innocent.
Not that good.
○
Numbers as high as 73%, but number mostly around chance.
○
Philosophy is that it's better to let a guilt person go than to convict an
innocent person and that doesn't seem to be happening here.
○
-
Guilty = prop of individuals correctly judged as being guilty.
Much higher numbers (in the 80s)
○
-
This is the bias the CQT typically has: it is better at detecting lies than it is at
clearing truth-tellers.
-
It appears that the original examiner is not just using the polygraph record to
make judgements.
They're looking at other cues.
○
Looks like he sees guilt where is does exist than they see innocent.
○
-
Review of the Guilty Knowledge Test
Individual in GKT doesn't have to answer any questions at all; three studies (Ben-
Shakhar and Elaad (2003), Elaad (1990), Elaad, Ginton and Jugnman (1992)).
Correctly judged as innocent.
95%
○
98%
○
94%
○
GKT much better at detecting individuals telling the truth.
○
-
Correctly judged as guilty.
85%
○
42%
○
76%
○
GKT much worse at detecting individuals in possession of guilty
knowledge.
○
Opposite bias from CQT.
○
-
NRC (2003): Polygraph Review
Theoretical rationale re differential arousal is weak.
There's no really strong data to support the expected effect.
○
Basic assumption is very questionable.
○
-
Response measured not unique to deception.
HR, increased BP, etc. can all be indicators of many other things.
○
-
Test overestimate field accuracy.
Accuracy in field is not as high as some of the studies, which even aren't
that high.
○
-
Accuracy may be degraded by countermeasures.
You can defeat polygraph measures by using countermeasures.
○
-
Computer analysis may improve accuracy.
Why not make a computer program that can more accurately detect
lies/truths.
○
If they were 100% accurate, however, the likelihood of people agreeing to
take them would go down.
○
-
May promote admissions, instill confidence, deter crime
It gives the public greater confidence to detect criminals prior to trials.
○
It may lead to many things.
○
Weak rationale, but the NRC wanted to be as balanced as possible and
find some positive points.
○
-
P300 Brain Response
When a stimulus is presented to us, we evoke an ERP.
-
There are several components to an ERP.
They are described in terms of whether changes occur in a positive or
negative direction and how long after the stimulus presentation they
occur.
○
-
One of the changes that occurs happens 300 ms after the presentation of the
stimuli and occurs in a positive direction - P300 brain response
-
Green line = any stimulus, red line = relevant stimulus
The difference between the two is what's supposed to indicate what a
person knows (consider rationale of GKT)
○
○
-
"Brain Fingerprinting" popularized by Dr. Farwell
He didn't develop it, but he made it into a company.
○
-
There's another possible measure - N400
-
NRC (2003): P300-GKT
May not be better than traditional polygraph.
-
Would polygraph + P300-GKT be better?
We have no evidence to support this.
○
-
There might be effective countermeasures.
There might be a way to beat this.
○
-
fMRI
Suggested that the areas of the brain that are active when a person is telling a
lie are different from the areas active when a person is telling the truth.
-
There is some evidence for this, but it's not strong.
-
So far, there's never been a real life use of fMRI to detect deception in a
forensic setting (other than research).
Because it is extremely time consuming and expensive.
○
Other lie detection is much cheaper.
○
-
There's also some question about whether the difference in brain area activity
during lies/truths are reliable measures.
E.g., is it the same for all people?
○
-
Facial Thermography
Involves taking the temperature of the skin on the face.
-
When we're anxious or embarrassed, we blush and the skin on our face/cheeks
gets very warm.
-
This is actually in real life use in a number of places (e.g., airport security)
-
Beauty of this is that the person being recorded doesn't even know it's
happening.
Scanner can be mounted on a wall and secretely scan face.
○
-
Won't be convicted on this basis, but if you do have a temperature spike during
questioning, you might have your bags checked or more questions asked.
-
Other Measures?
None of the ones that have emerged have been allowed to use in court.
-
In US is has nothing to do with the reliability, but they'll all be inadmissible
because they take away the jury's right to determine guilt on their own.
-
It may one day be allowed in Canada.
-
Summary
Looking at physiological measures is not terribly good.
-
Even human judgement.
-
There are studies that groups of individuals are better than individuals at
detecting truths based on evidence.
-
Problem is that there are a lot of times where cases are only seen by judges.
Suggests that judges are probably going to be less good at detecting
deception on their own. s
○
-
Deception Detection
Sunday, February 4, 2018 7:01 PM
The literature shows that the things that interviewers are looking for in initial
BAI (Reid) are not strongly associated with deception.
-
In some cases, they're looking at cues more strongly associated with truth
telling.
-
We spend a lot of time lying.
Approximately 4 times per day.
○
Mostly harmless.
○
-
To what extent are we able to detect deception in the legal context?
-
Bottom-line = we aren't very good at it.
-
How Good At Detecting Deception?
Ekman and O'Sullivan (1991):
Ekman is the leading expert on facial expressions of emotion.
○
TV show, Lie to Me, based on him and his work.
○
Ss correctly or incorrectly report positive feelings while watching
pleasurable or gruesome film.
○
Truth judged by Secret Service, FBI, judges, police, psychiatrists, university
students, adult taking on-day course on deceit.
○
Videos previously found to contain nonverbal cues to lying: masking
smiles and higher pitch.
○
Judges were asked to rate both general lie-detection ability and their
performance in the task.
○
Results:
All but one group performed at chance levels.
§
Only U.S. Secret Service agents judge better than chance and than
all other groups (64% accuracy)
§
Accuracy uncorrelated with ratings of general ability or task
performance (except for deceit group).
§
Adult deceit course group accurately judged their accuracy.
§
○
Possible shortcomings:
Several earlier studies found no evidence that "experts" better at lie
detection than students.
§
Forensic relevance unclear: most forensic lies not about currently
felt emotions.
Police don't ask about feelings, they ask about specific aspects
of a crime.
□
So the extent of the relevancy is an issue□
§
○
-
Ekman, O'Sullivan and Frank (1999):
Participants make true or false statements about their opinions on
videotape.
○
U.S. law enforcement official and psychologists judge honesty while
viewing videotapes.
○
Overall accuracy ranges from 51% (mixed law enforcement officers) to
73% (federal officers)
One of the few studies that finds a significantly different accuracy.
§
○
Lie accuracy (% of correctly identified lies) ranges from 48% (mixed law
enforcement) to 80% (federal officers).
○
Leaves 20% error even in the very best group.
This is not good enough in the forensic context.
§
Minimal p value of statistical significance is 5% chance of error.
§
Here we're accepting 20% error over someone's life … not good
enough.
§
○
Truth accuracy (% correctly identified truths) ranges from 54% (mixed law
enforcement) to 66% (federal officers)
Not as good at detecting truthful statements.
§
This is bias we are concerned about.
§
In the legal system, we'd rather let a criminal go than convict an
innocent person.
§
○
Overall accuracy.
More accurate to judge a lie as a lie.
§
Only Academic psych's had same accuracy across the board.
§
Interestingly, people who were the best at detecting lies were not
that good at detecting true statements.
They have a bias towards calling things lies when they are
actually truths.
□
Generally, in the literature we see a truth bias instead -
people are more inclined to deem false statements as being
truthful when it really isn't.
□
§
○
-
Bond and DePaulo (2006):
Meta-analysis of 206 studies and 24000 judges = 54% overall accuracy.
○
This has been the finding overall in the literature -- suggests that we are
not much better than chance.
○
47 % accuracy in detecting lies.
○
61% accuracy in detecting truths.
Truth bias: we have a tendency to rate more statements as true
than false.
§
We tend to believe what people say.
§
○
Greater accuracy when hearing than seeing and hearing liars and truth-
tellers.
People are less accuracy when they see + hear statement.
§
Suggests that we are more deceived by a person's facial expressions
and nonverbal cues than we are enlightened by them.
§
We are looking at the wrong things when looking for physical cues.
§
○
The more significant, more important a lie is makes it easier to detect.
Motivation makes liars easier to detect.
§
In lab studies, we see low-risk lies so motivation to be a successful
liar is relatively low.
§
In the forensic/legal context, the motivation is high, but the
motivation makes it more likely that you'll be detected.
§
Likely because you're more anxious/nervous.
§
○
-
Whelan, Wagstaff and Wheatcroft (2015):
Used 36 videos of relatives asking for information about missing or dead
relatives.
○
Half of speakers were criminally involved in the case (liars), the other half
were truth-tellers and had nothing to do with the person's
disappearance/death.
○
Note - these are high-stakes lies.
○
How good were people at detecting who the liars are?
○
Participants:
Firearms officers in UK (n=37)
§
CID officers in UK (n=33)
§
Undergraduates (n=37)
§
○
Findings (detection)
Note that only undergraduates show the truth bias typically
reported in the literature.
§
Accuracy is still significantly above chance, but a pretty large error
rate still.
§
Confirms that highly motivated liars are easier to detect.
§
○
Findings (confidence)
Confidence rated on 5-point scale from 1 (not at all confident) to 5
(very confident)
§
All above average pretty much.
§
People are more confident when they've correctly detected a lie,
less confident when they've correctly identified a truth.
§
** not as important.
§
○
-
Whelan, Wagstaff and Wheatcroft (2014);
2 successful raters viewed 9 videos of people pleading for help in locating
missing relative.
5 honest, 4 deceptive.
§
○
Indicated which cues using to detect deceit.
Very difficult for most people to do.
§
Often, we 'know' people are lying or telling the truth because we
have a certain feeling about it
§
○
Deceptive appeals:
Fake emotion
§
Lack of emotion
§
Distancing
§
Implausibility
§
Negative personal reaction
Turns out to be a very good predictor of whether or not we're
right about our judgement.
□
§
○
Honest appeals:
Genuine sadness emotion
§
Genuine/heartfelt
§
Containing emotion
§
Sympathetic personal reaction
No indication of what it was about the person that made you
feel positively, but it is still a good indicator for whether or
not your judgment was correct.
□
§
○
-
Facial/Body Cues to Deception?
Vrij (1998)
Asked people on the street and law enforcement officials about what it is they are
looking for when determining if someone is lying or not.
Avoiding eye contact.
Strongly considered.a.
We tend to assume that this is an indication of lying, but the literature
doesn't support this.
b.
1.
More smiling and laughter.2.
Higher rate of eye blinking.3.
Nervous fidgeting.4.
More illustrative gestures.5.
More movement of legs, feet, hands. 6.
More body, head movements.7.
More shrugging.8.
Virtually none of these things are strongly associated with deception.
No consistent, reliable association.
○
But this is what normal people AND LAW ENFORCEMENT say they're
looking for to determine if someone is telling the truth.
○
-
"the meta-analytic literature fails to identify any pattern of facial or body
movement that generally signals deception. However, some studies designed to
develop indicators based on these movements show some ability to
discriminate lying from truth-telling … these studies suggest that the right
measures of facial and motion feature can offer accuracy better than chance for
the detection of deception from demeanor in somewhat realistic situations."
Not a v strong statement.
○
But for the most part, this acknowledges the fact that our ability to detect
deception using these cues is about chance.
○
-
Meta-Analysis
Looked at around 150 cues of deception.
-
But there's nothing in the list consistently associated with truth-telling or lying.
-
No aspect of facial or body movement that can reliably help us identify lying.
-
This would explain why people are better at identifying a false statement when
they are listening rather than looking.
When you are looking, you're watching for all of these incorrect cues
which are no reliably indicative of lying.
○
-
Working Cues to Deception
Liars less forthcoming.
They give you less information.a.
1.
Liars tell less compelling stories.2.
Liars leave a more negative impression.
Recall study above.a.
We don't really understand the reasons, but it's an excellent predictor of
deception.
b.
3.
Liars are more tense.4.
Liars make fewer errors, unusual content.5.
The Language of Deception
Low verbal immediacy
Distancing of the individual from the things they're talking about.
○
Past tense, passive voice.
○
-
Fewer details in story.
-
Impressions of verbal uncertainty.
Not clear or definitive enough.
○
Could be the product of not having enough time to come up with a story.
○
-
Impressions of nervousness.
Very subjective.
○
An innocent/truthful person could also appear extremely nervous.
○
-
Lack of logical structure to story
-
Lack of plausibility to story.
-
Raised pitch of speech.
-
NOTE: None of these can be used to reliably identify a liar.
There's a lot of variation.
○
Does not actually discriminate a liar from a truth-teller.
○
-
Matsumoto and Hwang (2014)
Used word count program Linguistic Inquiry and Word Count (LIWC) to analyze true
and false statements, verbal and written, about mock crime - "theft' of $100 cheque.
Monetary reward given for escaping detection.
True and false accounts differed significantly in word usage, based on previous
LIWC research.
-
Word usage predicted truth from lies with 69% accuracy for written statements.
-
Word usage predicted truth from lies with 71% accuracy for verbal interview
statements.
-
For written statements, liars used:
Fewer words overall
○
Fewer words related to money
○
Fewer words related to negation, motion, time
○
More tentative, equivocation words
○
More positive emotion words
○
-
For interview statements, liars used:
Fewer words related to motion
○
More words related to first person ('I', 'me', 'my', etc.)
○
-
Hauch et al (2014)
Meta-analysis indicated that liars:
Experienced more cognitive load.
-
Expressed more negative emotion.
-
Distanced themselves more from events.
-
Used fewer sensory-perceptual words.
-
Referred less often to cognitive processes.
-
Expressed as much certainty as truth-teller.
Inconsistent with other studies.
○
-
Detection Easiest When:
Lies have high personal relevance
-
Stakes of deception are high
-
Liar has little chance to rehearse
-
Problem: we can't test our ability to detect lies because we have no what of
knowing who is a real liar.
-
Polygraph vs Lie Detector
Polygraph: a machine to record voltages
Greek meaning "many writings"
○
-
Lie detector = application of a polygraph.
Sometimes used to detect lies.
○
-
Polygraph record for lie detection measures …
Blood pressure
○
Heart rate
○
Respiration (breathing)
○
Galvanic skin response (sweating; conductance/resistance)
○
-
Many kinds of polygraph tests, the most common is the CQT.
-
Control Question Test (CQT)
Relevant questions: about crime
E.g., "Did you shoot Sam Smith?"
○
-
Irrelevant questions: about nothing
E.g., "Do you live in Toronto?"
○
-
Control questions:
E.g., "Have you ever injured anyone?"
○
-
Compare control vs relevant
When telling the truth, we expect arousal to be higher for control
questions than for relevant questions.
Because there's no reason for them to get upset about relevant
questions because they know they didn't do it.
§
○
When lying, much more arousal is expected for relevant questions than
control questions.
○
-
Compare three pairs of control + relevant questions.
-
CQT Procedure
Pre-test
Convince subject of test accuracy
Often done through a card trick.
§
○
Learn personal details
○
Develop irrelevant questions
○
-
Test
Ask and score questions
○
10 questions: 3 pairs of control and relevant + 4 irrelevant
○
-
Post-test
Sum scores, interpret results
○
Score difference between pairs: 1, 2, 3 or -1, -2, -3 or o
○
Add them up …
If large enough in a positive direction - truth
§
If large enough in a negative direction - liar
§
Close to zero = inconclusive
§
○
-
Alternatives to CQT
Directed Lie Test (DLT)
Broad control Q's
○
Subject instructed to lie to all.
This way, you have a true comparison of arousal.
§
○
-
Positive Control Test
Relevant questions asked twice.
○
Tell truth once, lie once.
○
-
Critique of CQT
Truth-teller may respond more to relevant that to control questions.
But relevant question itself is more arousing when you're innocent
because you're getting upset about being accused.
○
Makes truth-tellers look like liars.
○
-
Liars may respond more to control than to relevant questions if they have been
asked before.
They have practice saying "No, I didn't do it."
○
-
Not admissible as evidence in court because it's not totally reliable.
-
Guilty Knowledge Test
Used primarily in Europe.
-
Not really about truth-telling because it's about how much guilty knowledge you
have.
No lie detection; knowledge detection.
○
-
GSR response to important versus unimportant information.
-
5-alternative multiple-choice questions.
GSR arousal measured when examiner says the option they know to be
the truth.
○
Don't have to actually answer the question.
○
One of the choice is known only to the perpetrator and police.
This is why it's not used in the U.S.
§
It's very difficult to keep case details out of the media/from the
public.
§
○
-
Pros and Cons of GKT
Pro: It does not require subject to respond to questions.
-
Con: Usable only if some aspects of the crime are unknown to the public.
-
Testing Polygraph
Not very good
-
U.S. Supreme Court ruled it inadmissible in 1988.
Because it takes away jury's right to decide on guilt.
○
Not even because it's inaccurate.
○
-
Mock crime studied: subjects commit 'mock crime', then get tested.
-
Field studies: comparing original with 'blind' scoring.
We take a bunch of people who have taken polygraph (some liars, some
truth-tellers based on court rulings)
○
Polygraph tech has to look at just the polygraph records and determine
truth/lie.
○
Look at the accuracy of their blind judgments.
○
-
Field analogue studies: subject commits detectable crime, then tested.
-
Honts and Raskin (1988): How good is polygraph lie detection?
Looked at the accuracy of the CQT (DLT specifically); field test where they looked at
the polygraph recordings of innocent and guilty individuals (no confession vs
confession).
Looked at the originial judgements of the individuals who conducted the
polygraph.
'Actually' Innocent
91% correctly judged as innocent
§
0% judged as guilty
§
9% judged as inconclusive
§
○
'Actually' Guilty
8% incorrectly judged as innocent
§
92% correctly judges as guilty
§
0% judges as inconclusive
§
○
-
Looks pretty good for polygraphy …
-
Gave the records to blind polygraph examiners who weren't present at the time
of the polygraph test.
'Actually' Innocent
62% correctly judged as innocent
§
14% incorrectly judges as guilty
§
23% judged as inconclusive
§
MUCH WORSE
§
○
'Actually' Guilty
Results exactly the same.
§
○
-
This suggests that polygraph records are not that reliable in the sense that
individual expert judgement is relied on to determine guilt/innocence and
whenever judgement is in play, there's always some error.
We obviously don't want this much error.
○
-
Also suggests that the original polygrapher was using something else, some
other cue, in the presence of the suspect.
They were adding something they saw/heard/felt about the examiner into
their judgements other than simply their interpretation of the polygraph
record.
○
-
We know that seeing someone tell a lie is misleading.
-
Patrick and Iacono (1991)
Polygraph records from the RCMP; looked at the records of innocent and guilty people
as determined by confession and/or court conviction.
Original examiner's results:
'Actually' Innocent
73% correctly judged as innocent
§
8% incorrectly judges as guilty
§
19% judged as inconclusive
§
○
'Actually' Guilty
0% incorrectly judged as innocent
§
98% correctly judged as guilty
§
2% judged as inconclusive
§
○
-
What we see here is that in general, CQT lie detection tests are better at
detecting liars that they are at clearing truth-tellers -- widely replicated finding .
-
Blind examiner's results:
'Actually' Innocent
30% correctly judges as innocent
§
24% incorrectly judged as guilty
§
46% judged as inconclusive
§
○
'Actually' Guilty
2% incorrectly judged as innocent
§
92% correctly judged as guilty
§
6% judged as inconclusive
§
○
-
Again, we see the bias - we're much better at identifying liars with polygraph
CQT than we are at clearing truth-tellers.
Huge difference in this case.
○
Horrible for forensic purposes.
○
This is why polygraphs are not trusted in Canada.
○
-
Also indicates again that the original polygraphers had to be using something
else to get their results.
Something from the suspect themselves.
○
-
Granhag and Vrij (2005)
Survey of the literature; looked at other studies which compared the accuracy of
detecting truth vs deception in CQT; all had relatively small numbers of participants
(n).
Innocent = prop of individuals correctly judged to be innocent.
Not that good.
○
Numbers as high as 73%, but number mostly around chance.
○
Philosophy is that it's better to let a guilt person go than to convict an
innocent person and that doesn't seem to be happening here.
○
-
Guilty = prop of individuals correctly judged as being guilty.
Much higher numbers (in the 80s)
○
-
This is the bias the CQT typically has: it is better at detecting lies than it is at
clearing truth-tellers.
-
It appears that the original examiner is not just using the polygraph record to
make judgements.
They're looking at other cues.
○
Looks like he sees guilt where is does exist than they see innocent.
○
-
Review of the Guilty Knowledge Test
Individual in GKT doesn't have to answer any questions at all; three studies (Ben-
Shakhar and Elaad (2003), Elaad (1990), Elaad, Ginton and Jugnman (1992)).
Correctly judged as innocent.
95%
○
98%
○
94%
○
GKT much better at detecting individuals telling the truth.
○
-
Correctly judged as guilty.
85%
○
42%
○
76%
○
GKT much worse at detecting individuals in possession of guilty
knowledge.
○
Opposite bias from CQT.
○
-
NRC (2003): Polygraph Review
Theoretical rationale re differential arousal is weak.
There's no really strong data to support the expected effect.
○
Basic assumption is very questionable.
○
-
Response measured not unique to deception.
HR, increased BP, etc. can all be indicators of many other things.
○
-
Test overestimate field accuracy.
Accuracy in field is not as high as some of the studies, which even aren't
that high.
○
-
Accuracy may be degraded by countermeasures.
You can defeat polygraph measures by using countermeasures.
○
-
Computer analysis may improve accuracy.
Why not make a computer program that can more accurately detect
lies/truths.
○
If they were 100% accurate, however, the likelihood of people agreeing to
take them would go down.
○
-
May promote admissions, instill confidence, deter crime
It gives the public greater confidence to detect criminals prior to trials.
○
It may lead to many things.
○
Weak rationale, but the NRC wanted to be as balanced as possible and
find some positive points.
○
-
P300 Brain Response
When a stimulus is presented to us, we evoke an ERP.
-
There are several components to an ERP.
They are described in terms of whether changes occur in a positive or
negative direction and how long after the stimulus presentation they
occur.
○
-
One of the changes that occurs happens 300 ms after the presentation of the
stimuli and occurs in a positive direction - P300 brain response
-
Green line = any stimulus, red line = relevant stimulus
The difference between the two is what's supposed to indicate what a
person knows (consider rationale of GKT)
○
○
-
"Brain Fingerprinting" popularized by Dr. Farwell
He didn't develop it, but he made it into a company.
○
-
There's another possible measure - N400
-
NRC (2003): P300-GKT
May not be better than traditional polygraph.
-
Would polygraph + P300-GKT be better?
We have no evidence to support this.
○
-
There might be effective countermeasures.
There might be a way to beat this.
○
-
fMRI
Suggested that the areas of the brain that are active when a person is telling a
lie are different from the areas active when a person is telling the truth.
-
There is some evidence for this, but it's not strong.
-
So far, there's never been a real life use of fMRI to detect deception in a
forensic setting (other than research).
Because it is extremely time consuming and expensive.
○
Other lie detection is much cheaper.
○
-
There's also some question about whether the difference in brain area activity
during lies/truths are reliable measures.
E.g., is it the same for all people?
○
-
Facial Thermography
Involves taking the temperature of the skin on the face.
-
When we're anxious or embarrassed, we blush and the skin on our face/cheeks
gets very warm.
-
This is actually in real life use in a number of places (e.g., airport security)
-
Beauty of this is that the person being recorded doesn't even know it's
happening.
Scanner can be mounted on a wall and secretely scan face.
○
-
Won't be convicted on this basis, but if you do have a temperature spike during
questioning, you might have your bags checked or more questions asked.
-
Other Measures?
None of the ones that have emerged have been allowed to use in court.
-
In US is has nothing to do with the reliability, but they'll all be inadmissible
because they take away the jury's right to determine guilt on their own.
-
It may one day be allowed in Canada.
-
Summary
Looking at physiological measures is not terribly good.
-
Even human judgement.
-
There are studies that groups of individuals are better than individuals at
detecting truths based on evidence.
-
Problem is that there are a lot of times where cases are only seen by judges.
Suggests that judges are probably going to be less good at detecting
deception on their own. s
○
-
Deception Detection
Sunday, February 4, 2018 7:01 PM
Document Summary
The literature shows that the things that interviewers are looking for in initial. Bai (reid) are not strongly associated with deception. In some cases, they"re looking at cues more strongly associated with truth telling. Bottom-line = we aren"t very good at it. Ekman is the leading expert on facial expressions of emotion. Tv show, lie to me, based on him and his work. Ss correctly or incorrectly report positive feelings while watching pleasurable or gruesome film. Truth judged by secret service, fbi, judges, police, psychiatrists, university students, adult taking on-day course on deceit. Videos previously found to contain nonverbal cues to lying: masking smiles and higher pitch. Judges were asked to rate both general lie-detection ability and their performance in the task. All but one group performed at chance levels. Only u. s. secret service agents judge better than chance and than all other groups (64% accuracy) Accuracy uncorrelated with ratings of general ability or task performance (except for deceit group).
