Transcript P300 in Detection of Deception

ERPs in Deception,
Malingering, and False
Memory
J. Peter Rosenfeld
Psychology Department
Northwestern University
Evanston Illinois,USA
Principal Collaborators to
2008:
•
•
•
•
•
Joel Ellwanger
Tuti Reinhart Miller
Archana Rao
Matt Soskins
Greg Bosh
Ming Lui
• Many of the original ideas here were
theirs.
A simple neural code
Event-related potentials
P300 Attributes:
• An Endogenous, Event-Related
Potential (ERP)
• Positive polarity (down in Illinois).
• Latency range: 300-1000 msec
– varies with stimulus complexity/evaluation
time
• Typical Scalp Amplitude(Amp) Map
– Pz > Cz > Fz
• Amp = f(stim. probability, meaning)
P300 at 3 scalp sites
We are always wanting to
compare waves…
• ..that is, group or condition
• averages!
EVENT RELATED POTENTIAL
AVERAGING…..
…..from 4 sites on the scalp:
HERE IS WHAT
SPONTANEOUS EEG LOOKS
LIKE……
HERE IS A SINGLE SWEEP …
HERE ARE 3 TRIALS….
the class to the
movie…called, “ERP
Averaging”
REMINDER FOR PROF TO
TAKE
Since averages are so much
cleaner than single sweeps,
…and show the true stimulus-evoked
event that is time-locked to the
eliciting event, and are more noise
free,
….it obviously makes sense to
compare averages rather than single
sweeps, that is, to do analysis, like
t-tests on averages.
People did that in comparing group
ERPs or grand averages.
For example,
• The schizophrenic group average
versus the normal average
• or the well-trained group average
P300 vs. that of the untrained
group.
• Remember, in a group, each
subject has an average ERP.
• ….but within a single subject,
there are only single sweeps to
compare
In Bootstrapping…
• …..the original set of single
sweeps is repeatedly randomly
sampled –but with
replacement—
…yielding multiple averages in a
single subject.
• Let’s say there are 6 repetitions
of sampling of 18 single
sweeps:
EACH SET OF 18 SINGLE
SWEEPS IS AVERAGED
YIELDING 6 AVERAGES…
….that look like real average of
original set but with variations
P300 amplitude as recognition
index
• Autobiographical items (previous slide)
• Guilty Knowledge test items (Rosenfeld
et al., 1988)
• Antisocial/illegal acts in employee
screening (Rosenfeld et al., 1991).
• Tests of malingered cognitive deficits
with oddball paradigm. Do folks
recognize personal info? Start with
normal models….
3-stimulus protocol
• 1probe
• 2 irrelevant
• 3 target
Normals: autobiog. oddball
CHI patients: autobiog.
oddball
Individual detection rates for
various stimuli (normal
simulators).
E-Name forgetters(oddball is
dark line)
Screening example
Autobiographical paradigm
has limitations in detecting
malingerers
• Most malingerers are not so
unsophisticated as to verbally state that
they don’t recall, say, their birthdate,
when in fact they may have just filled
out a card in which they provided that
information.
•
Continuation…
• The behavioral “MDMT” was developed
as an entrapment test to catch these
people. It’s a simple matching-tosample test: A sample 3-digit number is
presented followed either by a match or
mismatch.
Simple MDMT paradigm:
• There is a 5-15 second interval between
sample and probe. This is an easy task,
yielding 100% performance even in
patients with moderate head injury-unless, oddly enough, they happen to be
in litigation !
• Where does one set the threshold for
diagnosis of malingering? 90%? (Some
non-litigating malingerers score well
below 90%, as we’ll see.)
Behavioral MDMT not reliable:
Some non-litigating pts. fail
P300-Souped-up MDMT:
simple version
• “Simple” means only one probe
stimulus per sample.
• P300 is recorded as soon as the probe -match or mismatch-- is presented.
• Match probability is kept low.
• RESULTS------------>
Match-To-Sample example
Computer-plotted data:
What would 75%-HITTING
plaintiff’s lawyer say?
• “Sure, my client scores 75% correct and
his P300 to matches is bigger than to
mismatches. But that’s because he
mostly DOES make the correct
discrimination--but 75% is still less than
normal. Therefore, give us the money
(me, one-third).”
•
Continuation…
• We did 2 experiments: 1) If a malingerer
aims to score 75% correct, whither
P300? 2) What happens to P300 with a
really tough discrimination?
Manipulated 75% “hit” rate
produces a larger P300….
100%
100%
Experiment 2: Difficult tasks: 7
and 9 digit numbers, match to
sample.
P300 wiped out in difficult
task, at 75%, even at
accuracy> 90%
Another View of same effect:
Simple P3-MDMT summary:
• If one fakes 75% hits, one’s P300 gets
bigger(or doesn’t change).
• If one has genuine difficulty--honest
75%--then P300 is totally removed.
• These findings should allow
discrimination of normals, malingerers,
real deficit(pts).
• BUT…diagnostic hit rate only 70% !!
Scalp Distribution
• For P300, Pz > Cz > Fz, usually, but…
• There are many ways that this can be
so:
SITE
AMP
Fz
Cz
Pz
Cz
Fz
Pz
lie
SITES
truth
AMP
Fz
Cz
Pz
Match-to-Sample Test:
advanced version
•
•
•
•
•
•
•
386 sample
212
457
386 (*)
789
325
123
Stimulus-Response Types
• Match(R) probe
– “Match” (RR--honest/correct)
– “Mismatch” (RW--dishonest/error)
• Mismatch(W) probe
– “Mismatch” (WW--honest/correct)
– “Match” (WR--dishonest/error)
ERPs in Liar Group to R and
W
Deception swamps out R/W
effect
“Profiles” of Deception
Truth vs Lie Groups
Deception overcomes
paradigm effects
Specificity (“Pinnochio”)
• Simple Truth vs. Lie Groups differ in
task demands.
• This is not relevant for practical field
detection.
• It is relevant for claims pertaining to a
specific lie response.
• How do you make a “perfect” control
group?
An imperfect(but not bad)
control
Two groups run in two trial blocks of
autobiog. oddball: [1. Phone #, 2. Bday]
• Lie Group
– Block 1 : Respond truthfully, repeat forwards.
– Block 2: Lie 50% of time, repeat forwards.
• Control Group
– Block 1: Respond truthfully, repeat forwards.
– Block 2: Respond truthfully, repeat
backwards(50%).
Only lying liars stick out.
Same result with simple truth
control
Lie Response<>Truth
Response; Psychopathy is
irrelevant(swamped).
Problems with these simple
oddball methods…
• (1) All the data I have shown --with
respect to scalp distribution-- were
based on group analysis, whereas in
deception detection, individual
diagnosis is the key and we never did
better than 73% accuracy, even with 32
electrodes!
• (2) Countermeasures…more later…
False(honestly
believed)memories:
• Deese/Roediger paradigm
– Presented words at study: sleep, bed,
dream,blanket,pajamas,dark….
– Not presented word: night.
• Test words:
– night-- a critical LURE--> possible
responses: “Old” or “New”
– bed-- an actual memory word “Old”
– table-- a completely new word “New”
Profiles depend on belief:
Replication data: almost ditto
P300 Latency is the
unconscious recognizer
Replication data: ditto !
What’s next?
• What does Malingered “false” memory
look like?
• Again, what happens as sites are
added?
• ________________________________
• [email protected]