PSYCH 3CC3 Lecture 4: Eyewitness Testimony

639 views19 pages

amberdonkey208

19 Jun 2018

School

Department

Course

Professor

For unlimited access to Class Notes, a Class+ subscription is required.

Eyewitness Errors

Eyewitnesses are very unreliable.

People incorrectly identified by victim a lot.

The Innocence Project estimates ~ 73% of false convictions based on mistaken

eyewitness identification.

Dr. Donald M. Thompson:

Australian psychologist, lawyer and memory expert.

Detained by the police in 1975 for rape, based on the victim's description.

When the crime occurred, he was on tv program describing how one could

improve one's memory for faces alongside the chief of police.

The victim same him on the tv while she was being raped and that's why she

was able to identify him.

Memory as an Adaptation

It's not the case that memory is an infallible recording system - it has evolved to

have specific functions.

(1) We don’t have to remember everything.

No memory before ~ 2 years of age.

○

No information about repeated events.

E.g., We see coins everyday, but cannot accurately describe them.

○

Memory is a reconstruction from pieces.

○

(2) Memory need not be 100% accurate.

Blake, Nazarian and Castel (2015): tested UCLA students' recall,

recognition of the Apple logo.

Even apple users were only 50% accurate at identifying the correct

logo.

But, they are pretty confident in their responses.

○

Memory Processes

(1) Encoding

Putting information into a form that the memory system can store.

○

We try to leave it the same way we get it, but not always.

○

E.g., auditory sounds are not typically stored, but rather a verbal

representations.

○

(2) Storage

We don't store everything.

○

(3) Retrieval

Hardest part.

○

E.g., studying is encoding/storage, but when it comes to taking a test, it's

very difficult to recall the information.

○

Encoding Factors

Exposure duration

You're more likely to encode things you were exposed to longer.

○

In lineup studies, longer is better. .

○

Arousal level

U-shaped relationship between arousal and performance.

○

Moderate arousal is better than high or low.

○

Distraction

Typically we have to pay attention to things in a scene to encode it.

○

Inattentional Blindness: If you're busy focusing on something, you won't

pay attention to or remember the rest of the scene.

○

Change Blindness: aspects of a scene we're in may change, but we don't

notice these changes.

○

More perpetrators: it's harder to identify the offenders when the crime

was committed by multiple perpetrators.

○

Weapon focus: people pay more attention to a weapon (if the perp is

carrying one) than the offender's face.

○

Distinctiveness

The more distinct an event is, the more likely you are to encode it.

○

Flashbulb memories: very vivid and distinct memories of big/distinct

events.

E.g., JFK assassination, John Lennon killing, Challenger explosion,

9/11

Accuracy seems much more vivid that for other events.

But they're not necessarily accurate at all.

○

Experiment:

Right after the challenger explosion, they asked people what they

had done/where they were when it occurred.

2 years later, they asked the same people again.

Details of the two descriptions didn't match up.

○

Storage/Retrieval Factors

Labelling

Explicit or self-generated labels affect recall.

○

When we store something, we often encode things not in their original

format (e.g., when we encode images as verbal descriptions).

○

Experiment:

Shown one of three images, each group was told a different label

for each image (e.g, sun vs wheel)

The verbal label influence their recall later - people who were told

the image was a sun draw a more sun-like image later, and people

who were told the image was a wheel draw a more wheel-like

image.

○

We essentially remember our labelled version of an event.

○

Prejudices and biases

Buckhout (1974): show people image of a while and black man arguing

on a train; black man is well dressed, white man looks like a vlue caller

worker of some sort.

When recalling the details, people changed the detais so that the

black man was the angry one, the working and the white man was

the calm, well-dressed one.

○

Bahrick et al (1996): asked people to recall all of their grades over their

university career.

Self-serving bias: people remembered less of their poorer grades

and more of their good grades; self-serving because they're

remembering what makes them look/feel good and dropping out

the negative memories.

○

Inferences

Sometimes the contents of memory are inaccurate because you made

inferences.

○

Bower and Treyens (1981):

People waited for experiment in the "professor's office"

When they were finally brought in for the experiment, they were

just asked questions about the contents of the office they were in.

E.g., "Where there any books in his office?"

The experimenters had removed all of the books.□

But people assumed that there would be books in an

academic's office so they said "yes"

□

○

Interpolated retelling/testing:

In cases, individuals are often asked about what happened multiple

times.

○

Information from one retelling that is repeated and told again in

subsequent retelling will be remembered.

○

If material in an earlier retelling is NOT contained in subsequent

interviews, it will not be remembered as strongly.

○

Greatly affects final testimony in court.

○

Leading questions:

Leading questions bias and alter witness's memory of an event.

○

Harris:

Showed brief film of basketball game.

Asked questions about films -> different wording for different

groups.

E.g., "How tall was basketball player?" vs "How short was the

basketball player?"

□

The questions that determined the responses.

○

Loftus and Zanni (1975):

Written survey about headaches and use of headache products.

Differently worded questions about the same things resulted in the

same thing.

E.g., "How many products tried. 1, 2, 3?" vs "How many

products tried. 1, 5, 10?"

□

○

Loftus and Palmer (1974):

Asked questions about a car accident

Worded the questions differently for different groups

One week later, they asked them "was there glass on the ground at

the scene of the crime"

Their responses were correlated with whether or not they

were asked about the speed of the car.

□

○

Post-event Information (PEI):

Provided by leading questions..

○

Media reports (not so much in Canada)

○

Other witnesses who report things differently than the original event.

○

Witness might add this post-event information to their story.

Significant number of people do

○

Problem with interviews too

○

What actually happens to the original/correct information?

PEI that's incorporated into recollection erases old info??

OR PEI masks/covers up correct information and needs to be

controlled under certain conditions?

○

Why?

Go along with interviewer to seem cooperative.

OR they don't know where they got the information (source

confusion) so they assume it's right.

○

False Memories

E.g., recovered memories under hypnosis/strong suggestion

Women accused mal relatives (e.g., father/uncle) of childhood sexual

abuse.

○

By given how these memories were retrieved, are they accurate?

○

It is very easy to create a false memory.

Loftus and Pickrell (1995): Being lost in a mall

Suggested to participants that when they were kids, they were separated

from parents in the mall - even though they confirmed with their parents

that this had not happened.

○

Initially, they say they don't remember.

○

After 1 week of thinking, a large proportion of participants remembered

the event with significant details.

But these events never happened.

○

Hyman, Husband and Billings (1995): Causing trouble at a wedding

Porter, Yuille and Lehman (1999): attacked by a viscious animal

Heaps and Nash (2001): Saved from drowning by a lifeguard

Mazzoni and Memon (2003): Having a skin sample taken as a child

This became a huge issue in the 80s because kids were asked over and over if

they were molested and eventually, these kids came up with elaborate and

detailed descriptions.

E.g., California daycare accused of abusing and sacrificing kids and babies

based on the false memories of people who went to the daycare.

Started with one child and then the information spread through

leading questions and the inclusion of PEI.

○

Issue:

False accusations

○

Kids who think they were abused psychologically become kids who were

abused.

○

Careful guidelines for interviewing kids have been formed by the APA.

Wise et al (2009): Attorneys and Eyewitness Testimony

Statements of 75 prosecutors, and ~ 1200 defense attorneys

Completed 13-question survey

○

Defense attorneys know more about the nature of eyewitness testimony.

The lack of knowing among prosecutors is startling.

Offender Descriptions: Robbery and Rape Cases

Kuhn et al (1974):

Categorized information and the amount of information given by witness.

Top descriptors:

Gender and age -> not very telling because stats can usually tell us it is a

young-middle-aged man.

○

Height

○

Build

○

Race

○

Weight

○

Face -> most telling descriptor, but least commonly described.

○

Mean = 7.2 descriptors

○

Sporer (1992):

Similar results

People didn't provide very good descriptions

Top descriptors:

Gender

○

Age

○

Height

○

Build

○

Race

○

Weight

○

Clothing (31%)

○

Face (30%)

○

Mean = 9.7 descriptors

○

Lindsay et al (1994): Kingston ON

Real Crime:

Average of 4 descriptors

Likely because in real crime, more stress, paying less attention

Therefore, poorer memory for events

○

Clothing (60%)

Gender (96%)

Hair colour (38%)

Race (25%)

Face (< 10%)

Mock Crime:

Average of 7.35 descriptors

Clothing (99%)

Hair colour (90%)

Height (86%)

Eyes (43%)

Race, age, sex (< 50%)

Other face (< 25%)

Van Koppen and Lochun (1997): Robbery Cases

24 permanent features (gender, age, skin colour)

Median = 5, inner face = < 5%

○

Eye colour = 35% correct

○

Nose = 35% correct

○

Mouth = 40% correct

○

Chin = 38%

○

19 temporary features (clothing, mask, hair)

Median = 2, mostly clothing

○

Hats = 51% mentioned

○

Hat colour = 31%

○

Jackets et al = ~ 27%

○

Jacket colour = ~ 25%

○

Yuille and Cutshal (1986): not the typical findings; witnesses actually accurate

21 shooting witnesses

391 action details recalled (82% accurate)

78 object details (89% accurate)

180 person descriptions (76% accurate)

Height, weight, age (77% accurate)

Hair, clothing style + colour (82% accurate)

Facial hair (<10% accurate)

Interesting finding.

○

Might think that this would've been one of the easiest and most obvious

thing to remember.

○

It just tells you how little attention people pay to the offender's face.

○

Elaborate, accurate descriptions even 5 months later.

Surprising, outlier

○

Few studies report this kind of detail/accuracy for eyewitness testimony.

○

Height and Weight Estimates

In convenience stores and banks, there's usually an indicator of height in the

stores so that if they were ever robbed, they could hopefully accurately

indicate the height of the perpetrator.

Not sure if this actually helps.

○

Flin and Shepherd (1986):

588 Ps estimate height, weight of 14 male targets with companion who asked for

directions.

Correlations between estimated and actual height and weight not high.

Witnesses use own height and weight as anchors in estimating others'.

E.g., "heavier than me"

○

Witnesses tend to underestimate height and weight.

Regress to mean: overestimate light/short people, underestimate tall/heavy

people.

Accuracy overall was not good.

Facial Composites

Seen demonstrated in a number of tv programs and movies.

Police artist's drawing

Out of date

○

Useful to show this, but not widely used today.

○

Used as early as 1910, as recently as 1995 in Oklahoma bombing

○

Little research on accuracy or utility.

○

Very rare today.

○

Only 18 full-time police artists in 500 U.S. police departments as of 2004.

○

Drawing from photo often no better than drawings from memory.

In research context.

Facial composites pretty poor.

○

E.g., Unabomber

Tried to use facial composites from police drawings.

Facial composite didn't resemble the actual offender v well.

Composite played no role in catching him.

○

Mechanical systems

Physically arrange pieces of a f

○

Individual takes pieces of a face and arranges them

physically/mechanically to arrange a composite of the offender's face.

○

E.g., Identi-kit and Photofit

○

Identi-Kit

Developed in California police offer Hugh MacDonald in 1959.

Originally consisted of ~ 600 drawings of facial features on clear

acetate sheets.

You can pick different features and stack them until you have a full

face.

Now available as a computer program.

○

Photofit:

Developed in UK by inventor Jacques Penry in 1970.

Originally consisted of ~ 560 photos of facial features (including lots

of hair!) on this cards.

Chosen features are fitted together in a frame to produce likeness

of offender's face.

Now available as a computer program.

Produces more photo-realistic reconstructions; not line drawings

like identi-kit.

○

Computer systems

Instead of physically making a face, we use computer programs to

essentially build faces.

○

E.g., E-fit, Mac-a-Mug, FACES, Facette, Identi-Kit, Photofit

○

This is the way it is typically done.

○

Lab Tests of Photofit

Ellis et al (1975): Judges try to determine which of 36 phots was used to create

Photofit likeness. Likeness images were made with the actual photo of the person

right beside them for reference.

Only 12.5% of 1st choices were correct

25% of 1st-3rd choice were correct.

Terrible accuracy

Ellis et al (1978): No difference in resemblance between Photofit images from

memory or from photo directly.

You would assume that with a photograph right in front of you that you would

be able to create a more accurate face than from memory, but this isn't the

case - there was no difference in the judges accuracy to pick out the photo that

the photofit image was based off of (memory vs picture).

Racial familiarity effect

We are much better at discriminating people from our own individual

group.

○

If the individual creating the photofit likeness was of the same racial

background of the person they were creating an image of, they were

better than if they were of a different racial background.

○

Christie & Ellis (1981): again, comparing judges ability to pick out the face being

described by photofit likeness or by a verbal description.

Verbal descriptions were better guide to likeness than Photofit composites.

Seems very counterintuitive.

Field Tests of Photofit

People in law enforcement were actually asked about the accuracy and

usefulness of photofit likeness images.

Kitson et al (1997): In 150 solved cases (of 729) how useful was the photofit

composite in solving the crime/identifying the individual

Composite critical - 5%

Very useful -17%

Useful -33%

But no use at all - 25%

Not very useful - 20%

Almost 50/50 for usefulness and uselessness

Levi (1997): In 54 convictions out of 243 cases in which Identikit was used by the

Israeli police …

Only 5 (10%) significantly aided by Identikit (drawing) composite.

Note: it appears that these facial composites are not very useful

Two issues: (1) are these things accurate? Nope. (2) are they useful? Nope.

Tests of Mac-a-Mug

Koehn & Fisher (1997): Judges evaluate Mac-a-Mug likeness of stranger.

69% rated 1 or 2 out of 10

Only 10% of Mac-a-Mug likenesses matched to correct photo of 6.

Less than chance performance

Not only did the likeness not look like the photograph, it was distinctively

unlike the photo.

○

Kovera et al (1997): Composites for former teachers, classmates judged by fellow

students.

Only 3 of 167 names correct.

All knew who the people were when they were told so it wasn't that they

simply forgot the names of people; when told they were like "oh, I know that

guy … that doesn't look like him."

EFIT-V

Very much like Photofit except it comes with instructions and training.

The idea, therefore, is that it should be more accurate because people have

been trained to use it.

Photographic quality features.

Davies et al (2000): Judges identify composites made by either trained operators or

witnesses themselves.

35% of operator composites correctly identified.

25% of composites produced by witnesses.

Not a huge difference despite training.

Davies and Oldman (1999): Witness makes composite of famous face, both from

photo and memory; faces everyone knows - instantly recognizable.

Judges name only 10% of composites from photo.

6% of composites produced from memory.

How could this be this poor?

Evo-Fit

Go holistic - give up deconstructing faces into constituent parts.

We don't see faces in pieces the way other facial composite programs

expect us to.

○

We are holistic processors of faces.

○

Principal component analysis producing 'eigenfaces'

Eigenfaces are the principal patterns of lightness and darkness.

○

Can combine 'eigenfaces' to produce any of the faces in the face space.

E.g., add 10% eigenface 1, 20% eigenface 2, etc.

○

Can also create brand new faces.

Genetic Algorithm

Randomly generate several composites from face space using eigenfaces.1.

Ask which one of these faces looks most like the perp.

Witness chooses closest likeness.2.

Create random variations around likeness leads to new set of

composites.

Ask again which one of these faces looks most like the perp.

Repeat from #2 until people have the closest person.4.

Genetic Algorithm #2

Randomly generate several composites from face pace using eigenfaces.1.

Witness chooses two closest likenesses2.

'Mating' of these two leads to new generation of 'child faces'3.

Repeat from #2 until you have the closest likeness. 4.

Mugbook Selections

The accuracy of these selections depends greatly on where the photo is located

in the mugbook; flawed.

McCallister, Stewart and Loveland (2003): placed 69 preceding foils and perpetrator's

photo at various locations in computerized mugbook (locations 1-70, 71-140,

141-210).

Later placement led to fewer correct identifications.

Later placement led to fewer false positive choices of foils.

Probability of correct identification decline the more pictures the witness

viewed (i.e., the further in the book the actual perp was).

In the case of real police identification and they aren't actually aware of who

the perp is yet, we cannot make assumptions about accuracy.

Mugbook Exposure Effect

Problem: if we haven't been exposed to a person, but we've selected them as

the perp from a mugbook, when shown them in a lineup later, we're more

likely to pick the person we were already exposed to even if the actual perp

was in the lineup.

We can only get error rates from lab studies; these results are disturbingly high.

Brown, Deffenbacher & Sturgill (1977): first demonstration of 'mugshot exposure

effect'. Three experiments

See 5 suspects for 25 secs each, knowing they will later pick them from

mugshots and from lineup.

1.5 hours later, 15 mugshots (15 secs each): 5 suspects and 5 people to appear

in later lineup. Asked whether each a suspect and their confidence in that

judgement.

72% of suspects were correctly identified.

○

45% false alarms (non-suspect called a suspect).

○

No correlation between confidence and accuracy; equal confidence in

accurate and inaccurate judgments.

○

1 week later, see 4-5 person lineups, with 1-3 'suspects'. Asked whether each

person a 'suspect', or appeared only in mugbook and their confidence in their

judgement.

20% of individual who appeared only in mugshots named as suspects.

○

Again, no correlation between confidence and accuracy; equal

confidence in accurate and inaccurate judgements.

○

Just seeing pictures in the mugbook biases later decisions.

Why do we see this effect?

Familiarity:

Sense of familiarity when seeing foil in lineup that was seen in mugshot.

○

We mistake that sense of familiarity for presence at crime scene.

○

Essentially a case of source confusion.

○

E.g., Dr. Daniel Thompson case

Psych prof mistakenly identified as rapist because his interview was

on tv in the room while the victim was being raped.

○

Commitment:

Incorrect foil chosen from mugshots; witness wants to be consistent, so

picks same foil from lineup.

○

Should be no incorrect lineup choice when foil seen, but not picked, from

mugshots.

○

Therefore, making the incorrect choice in earlier stages leads to incorrect

choices when faced with the lineup.

○

There is evidence for the accuracy of both processes.

Might both be involved to some extent and in different cases. '

○

Lineup Accuracy Measures:

We will establish a diagnosticity ratio.

Take the ratio of hits to false alarms.

○

Can only be calculated in lab studies - not in the real world.

○

Four possible choices:

Hit, miss, false alarm, correct rejection

○

○(see slide)

Problem with the DR as a measure of accuracy.

Accuracy = individuals ability to discriminate perpetrators from non-

perpetrators.

○

It does to some degree assess accuracy, but it also assesses individual's

confidence criterion in making these judgements.

○

○(see slide)

Red curve = confidence in choice for non-perpetrators, green curve =

confidence in choice for perpetrators

The difference between the means is our discriminability.

○

Problem is that discriminability is not the only basis on which individuals base

their choice.

Also, response criterion.

A point where you will say beyond this point, I will say that an

individual is the perpetrator and below this point I will say that the

individual is not the perpetrator.

○

In every case, we have this criterion for the amount of confidence we

have to have in order to make either choice.

○

E.g., response criterion = 25% confidence

DR messed up; ~75% FA and ~ 100% H

§(see slide)

○

E.g., Response criterion = 50% confidence

Discriminability hasn't changed, but DR will change

DR much higher because less FA

§(see slide)

By this criteria, we're more accurate even though our

discriminability hasn't changed.

○

This the problem with the DR as a measure of accuracy: it confounds the role of

actual discriminability from the role of response criterion in determining

accuracy.

We want to know what the real discriminability is.

○

To what extent can people discriminate non-perpetrators from

perpetrators.

○

This is the underlying question that the DR cannot answer.

○

Solution: ROC Curve

Uses signal detection theory - set of mathematical procedures designed to

disentangle actual discriminability from response criterion.

Result looks like this:

○(see slide)

In some way, similar to DR

○

Shows the ratio of H to FA as the response criterion changes.

○

Get the ROC curve: how is the receiver doing at balancing H and FA given

the response criterion.

○

Measure we use is d' (d prime) - a measure of discriminability

Distance between the diagonal line that represents the total inability to

make any discriminations ever to the most distant point of the curve.

○

If we had a point

○

As discriminability decreases, the curve gets closer and closer to the

diagonal.

○

We can also look at the area under the curve (the area between the curve and

the diagonal line)

If the curve went all the way to top left corner, the area would be 1.00

○

As the curve gets closer and close to the diagonal line, the area under the

curve gets smaller and smaller and discriminability thus gets smaller and

smaller.

○

Lineup Identification Issues

Disguises or changed appearance

Facial hair, hair colour, etc.

Play a large role in the accuracy of identification because these are things

that can be easily changed between time of the crime and identification.

Cross-race effect

We're bad at discriminating faces from other ethnic groups because we

don't have as much experience making these discriminations.

Can lead us to making errors in judgements of people if the perpetrator

was outside of our ethnic group.

Cross-age bias

We're also not good a discriminating between people outside of our age

range.

Cross-gender bias

males are better at discriminating between other males and females are

better at discriminating between other females.

Smaller effect than cross-race or cross-age discriminations.

Retention interval

E.g., How long has it been since you say this suspect?

Source confusion becomes an issue the longer you wait.

Verbal overshadowing

If you give a verbal description before picking out suspect, you will do

worse.

Retrieval interference (verbal vs visual)

Inhibition of non-verbal processes

Engaging local vs global processing

We remember a face holistically; global processing.

But when we give it a verbal description, we're using local

processing.

Memory transference

Like source confusion

Memory from one context confused with a memory from another

context.

Context reinstatement

E.g., Imagine you are at the crime scene.

Generally, improves recall.

But for witness identification, the data is mixed.

Lineup Factors

Social Factors

Presumes presence of offender in lineup.

Problem: in a significant proportion of lineups, the actual suspect isn't

there.

This leads to a lot of false identifications because people think they

have to pick someone.

Lineup Instructions:

Biased: "See if you recognize the offender in the lineup."

Unbiased: "See if you recognize the offender, who may not be in the

lineup."

90% still assume presence of the offender in the lineup.

Lineup Composition:

How many foils?

3 plausible foils minimum; accuracy is no better beyond that.

How are foils chosen?

Match to description: foils chosen to match description given by

the witness.

This is the way it's typically done; better of the two according

to the literature.

□

Match to offender: foils chosen to match the actual appearance of

the suspect/offender.

Sometimes foils are simply pulled off of the street.

Lineup Presentation Mode:

Simultaneous Presentation:

A lot of foils come out together, stand together at the same time

(works with photos too).

Problem: witness makes judgement of relative similarity rather

than identity

E.g., which one is the most similar to what I remember?□

Sequential Presentation:

One person/picture is presented at a time.

No relative comparison; judgments are made before the next

person is seen.

Choice ends the lineup.

Witness doesn’t know how many are in the lineup.

Simultaneous vs Sequential Lineups?

Lindsay and Wells (1985): Argued that sequential better than

simultaneous, because witnesses make absolute rather than

relative judgements of similarity in sequential.

Gronlund (2004): Found simultaneous-sequential lineup difference

due mostly to more conservative response criterion in sequential.

In sequential, response criterion is higher, so they're less

likely to make a false choice and they're more confident in

their accuracy.

□

Gronlund et al (2012): ROC analysis shows sequential lineups never

more accurate than simultaneous.

Gronlund, Wixted & Mickes (2014): "Not clear which lineup

procedure will prove to be generally superior, but … ROC analysis is

the only way to make that determination."

Identification Characteristics

Identification Latency

Shorter latency - choices likely to be more accurate.

Faster = more accurate (r= -.20 to -.40)

Not a high correlation, but it's there.

'Sweet spot' at 10-15 seconds?

Point of highest accuracy.

Identification Confidence

Low confidence-accuracy correlations

But evidence for good calibration

Calibrated - ideally, the accuracy would be exactly the same as

confidence.

We do get a pretty good ratio though.

Ideal scenario: (see slide)

What we actually get: (see slide)

Relationship between confidence depends on what proportions of

the lineups actually have the perp present - lab setting only.

There's a sweet spot … when 15-25% of trials do not contain the

perp, we're the closest to the callibration line.

Whenever we get a deviation, it's because we have higher

confidence than is warranted.

Dotted line - source of false convictions

Witnesses are 100% certain someone is the guy when in fact

it isn't

□

Jurors think that there's a strong correlation between accuracy and

confidence so eyewitnesses have big weight in court.

It's very common for the crown attorney to ask the witness how

confident they are.

Confidence seems pretty compelling for jurors.

But, as we can see it's not that strong- we're making a significant amount

of errors in judgment at least under lab conditions.

Post-Identification Feedback

Positive feedback increases expressed confidence

Positive feedback from other witnesses (i.e., "I picked 3 too!") or

officers (i.e., "thank you for your identification.")

Negative feedback reduced expressed confidence

Negative feedback from other witnesses (i.e., "you thought it was #

1? I thought it was #2?") or from officers (i.e., "thank you for

trying").

Therefore, we should take confidence estimates immediately.

But they don't normally do this -normally they aren't asked until at

trial on the witness stand.

Lineup Recommendations

Only one suspect per lineup1.

At least 4 foils. (even though the max is 3)2.

Match foils to witness descriptions (standard procedure).3.

Indicate that offender may not be present.4.

Indicate that confidence will be assessed.5.

Have police unaware of suspect's identity.6.

NRC (2014): Recommendations re: Eyewitness ID

Train all law enforcement officers in eyewitness ID>

Typically, law enforcement has much greater faith than appropriate in

eyewitness testimony.

Use double-blind lineup and photo array procedures.2.

Develop and use standardized witness instructions.3.

Document witness confidence judgements.4.

Videotape witness identification process.5.

Make basic judicial inquiries when eyewitness ID offered. 6.

Make juries aware of prior identifications.7.

Use expert scientific testimony on eyewitness ID.8.

Use jury instructions as alternate way to convey info re: eyewitness ID.9.

Establish national research initiative on eyewitness ID.10.

Additional research on system and estimator variables.

Estimator variables effect accuracy of witness - things/conditions at crime

scene effecting description.

System variables - part of the system in which we get the information.

11.

Eyewitness Testimony

Tuesday, February 13, 2018

10:46 AM

Unlock document

This preview shows pages 1-3 of the document.
Unlock all 19 pages and 3 million more documents.

Already have an account? Log in

Eyewitness Errors

Eyewitnesses are very unreliable.

People incorrectly identified by victim a lot.

The Innocence Project estimates ~ 73% of false convictions based on mistaken

eyewitness identification.

Dr. Donald M. Thompson:

Australian psychologist, lawyer and memory expert.

Detained by the police in 1975 for rape, based on the victim's description.

When the crime occurred, he was on tv program describing how one could

improve one's memory for faces alongside the chief of police.

The victim same him on the tv while she was being raped and that's why she

was able to identify him.

Memory as an Adaptation

It's not the case that memory is an infallible recording system - it has evolved to

have specific functions.

(1) We don’t have to remember everything.

No memory before ~ 2 years of age.

○

No information about repeated events.

E.g., We see coins everyday, but cannot accurately describe them.

○

Memory is a reconstruction from pieces.

○

(2) Memory need not be 100% accurate.

Blake, Nazarian and Castel (2015): tested UCLA students' recall,

recognition of the Apple logo.

Even apple users were only 50% accurate at identifying the correct

logo.

But, they are pretty confident in their responses.

○

Memory Processes

(1) Encoding

Putting information into a form that the memory system can store.

○

We try to leave it the same way we get it, but not always.

○

E.g., auditory sounds are not typically stored, but rather a verbal

representations.

○

(2) Storage

We don't store everything.

○

(3) Retrieval

Hardest part.

○

E.g., studying is encoding/storage, but when it comes to taking a test, it's

very difficult to recall the information.

○

Encoding Factors

Exposure duration

You're more likely to encode things you were exposed to longer.

○

In lineup studies, longer is better. .

○

Arousal level

U-shaped relationship between arousal and performance.

○

Moderate arousal is better than high or low.

○

Distraction

Typically we have to pay attention to things in a scene to encode it.

○

Inattentional Blindness: If you're busy focusing on something, you won't

pay attention to or remember the rest of the scene.

○

Change Blindness: aspects of a scene we're in may change, but we don't

notice these changes.

○

More perpetrators: it's harder to identify the offenders when the crime

was committed by multiple perpetrators.

○

Weapon focus: people pay more attention to a weapon (if the perp is

carrying one) than the offender's face.

○

Distinctiveness

The more distinct an event is, the more likely you are to encode it.

○

Flashbulb memories: very vivid and distinct memories of big/distinct

events.

E.g., JFK assassination, John Lennon killing, Challenger explosion,

9/11

Accuracy seems much more vivid that for other events.

But they're not necessarily accurate at all.

○

Experiment:

Right after the challenger explosion, they asked people what they

had done/where they were when it occurred.

2 years later, they asked the same people again.

Details of the two descriptions didn't match up.

○

Storage/Retrieval Factors

Labelling

Explicit or self-generated labels affect recall.

○

When we store something, we often encode things not in their original

format (e.g., when we encode images as verbal descriptions).

○

Experiment:

Shown one of three images, each group was told a different label

for each image (e.g, sun vs wheel)

The verbal label influence their recall later - people who were told

the image was a sun draw a more sun-like image later, and people

who were told the image was a wheel draw a more wheel-like

image.

○

We essentially remember our labelled version of an event.

○

Prejudices and biases

Buckhout (1974): show people image of a while and black man arguing

on a train; black man is well dressed, white man looks like a vlue caller

worker of some sort.

When recalling the details, people changed the detais so that the

black man was the angry one, the working and the white man was

the calm, well-dressed one.

○

Bahrick et al (1996): asked people to recall all of their grades over their

university career.

Self-serving bias: people remembered less of their poorer grades

and more of their good grades; self-serving because they're

remembering what makes them look/feel good and dropping out

the negative memories.

○

Inferences

Sometimes the contents of memory are inaccurate because you made

inferences.

○

Bower and Treyens (1981):

People waited for experiment in the "professor's office"

When they were finally brought in for the experiment, they were

just asked questions about the contents of the office they were in.

E.g., "Where there any books in his office?"

The experimenters had removed all of the books.□

But people assumed that there would be books in an

academic's office so they said "yes"

□

○

Interpolated retelling/testing:

In cases, individuals are often asked about what happened multiple

times.

○

Information from one retelling that is repeated and told again in

subsequent retelling will be remembered.

○

If material in an earlier retelling is NOT contained in subsequent

interviews, it will not be remembered as strongly.

○

Greatly affects final testimony in court.

○

Leading questions:

Leading questions bias and alter witness's memory of an event.

○

Harris:

Showed brief film of basketball game.

Asked questions about films -> different wording for different

groups.

E.g., "How tall was basketball player?" vs "How short was the

basketball player?"

□

The questions that determined the responses.

○

Loftus and Zanni (1975):

Written survey about headaches and use of headache products.

Differently worded questions about the same things resulted in the

same thing.

E.g., "How many products tried. 1, 2, 3?" vs "How many

products tried. 1, 5, 10?"

□

○

Loftus and Palmer (1974):

Asked questions about a car accident

Worded the questions differently for different groups

One week later, they asked them "was there glass on the ground at

the scene of the crime"

Their responses were correlated with whether or not they

were asked about the speed of the car.

□

○

Post-event Information (PEI):

Provided by leading questions..

○

Media reports (not so much in Canada)

○

Other witnesses who report things differently than the original event.

○

Witness might add this post-event information to their story.

Significant number of people do

○

Problem with interviews too

○

What actually happens to the original/correct information?

PEI that's incorporated into recollection erases old info??

OR PEI masks/covers up correct information and needs to be

controlled under certain conditions?

○

Why?

Go along with interviewer to seem cooperative.

OR they don't know where they got the information (source

confusion) so they assume it's right.

○

False Memories

E.g., recovered memories under hypnosis/strong suggestion

Women accused mal relatives (e.g., father/uncle) of childhood sexual

abuse.

○

By given how these memories were retrieved, are they accurate?

○

It is very easy to create a false memory.

Loftus and Pickrell (1995): Being lost in a mall

Suggested to participants that when they were kids, they were separated

from parents in the mall - even though they confirmed with their parents

that this had not happened.

○

Initially, they say they don't remember.

○

After 1 week of thinking, a large proportion of participants remembered

the event with significant details.

But these events never happened.

○

Hyman, Husband and Billings (1995): Causing trouble at a wedding

Porter, Yuille and Lehman (1999): attacked by a viscious animal

Heaps and Nash (2001): Saved from drowning by a lifeguard

Mazzoni and Memon (2003): Having a skin sample taken as a child

This became a huge issue in the 80s because kids were asked over and over if

they were molested and eventually, these kids came up with elaborate and

detailed descriptions.

E.g., California daycare accused of abusing and sacrificing kids and babies

based on the false memories of people who went to the daycare.

Started with one child and then the information spread through

leading questions and the inclusion of PEI.

○

Issue:

False accusations

○

Kids who think they were abused psychologically become kids who were

abused.

○

Careful guidelines for interviewing kids have been formed by the APA.

Wise et al (2009): Attorneys and Eyewitness Testimony

Statements of 75 prosecutors, and ~ 1200 defense attorneys

Completed 13-question survey

○

Defense attorneys know more about the nature of eyewitness testimony.

The lack of knowing among prosecutors is startling.

Offender Descriptions: Robbery and Rape Cases

Kuhn et al (1974):

Categorized information and the amount of information given by witness.

Top descriptors:

Gender and age -> not very telling because stats can usually tell us it is a

young-middle-aged man.

○

Height

○

Build

○

Race

○

Weight

○

Face -> most telling descriptor, but least commonly described.

○

Mean = 7.2 descriptors

○

Sporer (1992):

Similar results

People didn't provide very good descriptions

Top descriptors:

Gender

○

Age

○

Height

○

Build

○

Race

○

Weight

○

Clothing (31%)

○

Face (30%)

○

Mean = 9.7 descriptors

○

Lindsay et al (1994): Kingston ON

Real Crime:

Average of 4 descriptors

Likely because in real crime, more stress, paying less attention

Therefore, poorer memory for events

○

Clothing (60%)

Gender (96%)

Hair colour (38%)

Race (25%)

Face (< 10%)

Mock Crime:

Average of 7.35 descriptors

Clothing (99%)

Hair colour (90%)

Height (86%)

Eyes (43%)

Race, age, sex (< 50%)

Other face (< 25%)

Van Koppen and Lochun (1997): Robbery Cases

24 permanent features (gender, age, skin colour)

Median = 5, inner face = < 5%

○

Eye colour = 35% correct

○

Nose = 35% correct

○

Mouth = 40% correct

○

Chin = 38%

○

19 temporary features (clothing, mask, hair)

Median = 2, mostly clothing

○

Hats = 51% mentioned

○

Hat colour = 31%

○

Jackets et al = ~ 27%

○

Jacket colour = ~ 25%

○

Yuille and Cutshal (1986): not the typical findings; witnesses actually accurate

21 shooting witnesses

391 action details recalled (82% accurate)

78 object details (89% accurate)

180 person descriptions (76% accurate)

Height, weight, age (77% accurate)

Hair, clothing style + colour (82% accurate)

Facial hair (<10% accurate)

Interesting finding.

○

Might think that this would've been one of the easiest and most obvious

thing to remember.

○

It just tells you how little attention people pay to the offender's face.

○

Elaborate, accurate descriptions even 5 months later.

Surprising, outlier

○

Few studies report this kind of detail/accuracy for eyewitness testimony.

○

Height and Weight Estimates

In convenience stores and banks, there's usually an indicator of height in the

stores so that if they were ever robbed, they could hopefully accurately

indicate the height of the perpetrator.

Not sure if this actually helps.

○

Flin and Shepherd (1986):

588 Ps estimate height, weight of 14 male targets with companion who asked for

directions.

Correlations between estimated and actual height and weight not high.

Witnesses use own height and weight as anchors in estimating others'.

E.g., "heavier than me"

○

Witnesses tend to underestimate height and weight.

Regress to mean: overestimate light/short people, underestimate tall/heavy

people.

Accuracy overall was not good.

Facial Composites

Seen demonstrated in a number of tv programs and movies.

Police artist's drawing

Out of date

○

Useful to show this, but not widely used today.

○

Used as early as 1910, as recently as 1995 in Oklahoma bombing

○

Little research on accuracy or utility.

○

Very rare today.

○

Only 18 full-time police artists in 500 U.S. police departments as of 2004.

○

Drawing from photo often no better than drawings from memory.

In research context.

Facial composites pretty poor.

○

E.g., Unabomber

Tried to use facial composites from police drawings.

Facial composite didn't resemble the actual offender v well.

Composite played no role in catching him.

○

Mechanical systems

Physically arrange pieces of a f

○

Individual takes pieces of a face and arranges them

physically/mechanically to arrange a composite of the offender's face.

○

E.g., Identi-kit and Photofit

○

Identi-Kit

Developed in California police offer Hugh MacDonald in 1959.

Originally consisted of ~ 600 drawings of facial features on clear

acetate sheets.

You can pick different features and stack them until you have a full

face.

Now available as a computer program.

○

Photofit:

Developed in UK by inventor Jacques Penry in 1970.

Originally consisted of ~ 560 photos of facial features (including lots

of hair!) on this cards.

Chosen features are fitted together in a frame to produce likeness

of offender's face.

Now available as a computer program.

Produces more photo-realistic reconstructions; not line drawings

like identi-kit.

○

Computer systems

Instead of physically making a face, we use computer programs to

essentially build faces.

○

E.g., E-fit, Mac-a-Mug, FACES, Facette, Identi-Kit, Photofit

○

This is the way it is typically done.

○

Lab Tests of Photofit

Ellis et al (1975): Judges try to determine which of 36 phots was used to create

Photofit likeness. Likeness images were made with the actual photo of the person

right beside them for reference.

Only 12.5% of 1st choices were correct

25% of 1st-3rd choice were correct.

Terrible accuracy

Ellis et al (1978): No difference in resemblance between Photofit images from

memory or from photo directly.

You would assume that with a photograph right in front of you that you would

be able to create a more accurate face than from memory, but this isn't the

case - there was no difference in the judges accuracy to pick out the photo that

the photofit image was based off of (memory vs picture).

Racial familiarity effect

We are much better at discriminating people from our own individual

group.

○

If the individual creating the photofit likeness was of the same racial

background of the person they were creating an image of, they were

better than if they were of a different racial background.

○

Christie & Ellis (1981): again, comparing judges ability to pick out the face being

described by photofit likeness or by a verbal description.

Verbal descriptions were better guide to likeness than Photofit composites.

Seems very counterintuitive.

Field Tests of Photofit

People in law enforcement were actually asked about the accuracy and

usefulness of photofit likeness images.

Kitson et al (1997): In 150 solved cases (of 729) how useful was the photofit

composite in solving the crime/identifying the individual

Composite critical - 5%

Very useful -17%

Useful -33%

But no use at all - 25%

Not very useful - 20%

Almost 50/50 for usefulness and uselessness

Levi (1997): In 54 convictions out of 243 cases in which Identikit was used by the

Israeli police …

Only 5 (10%) significantly aided by Identikit (drawing) composite.

Note: it appears that these facial composites are not very useful

Two issues: (1) are these things accurate? Nope. (2) are they useful? Nope.

Tests of Mac-a-Mug

Koehn & Fisher (1997): Judges evaluate Mac-a-Mug likeness of stranger.

69% rated 1 or 2 out of 10

Only 10% of Mac-a-Mug likenesses matched to correct photo of 6.

Less than chance performance

Not only did the likeness not look like the photograph, it was distinctively

unlike the photo.

○

Kovera et al (1997): Composites for former teachers, classmates judged by fellow

students.

Only 3 of 167 names correct.

All knew who the people were when they were told so it wasn't that they

simply forgot the names of people; when told they were like "oh, I know that

guy … that doesn't look like him."

EFIT-V

Very much like Photofit except it comes with instructions and training.

The idea, therefore, is that it should be more accurate because people have

been trained to use it.

Photographic quality features.

Davies et al (2000): Judges identify composites made by either trained operators or

witnesses themselves.

35% of operator composites correctly identified.

25% of composites produced by witnesses.

Not a huge difference despite training.

Davies and Oldman (1999): Witness makes composite of famous face, both from

photo and memory; faces everyone knows - instantly recognizable.

Judges name only 10% of composites from photo.

6% of composites produced from memory.

How could this be this poor?

Evo-Fit

Go holistic - give up deconstructing faces into constituent parts.

We don't see faces in pieces the way other facial composite programs

expect us to.

○

We are holistic processors of faces.

○

Principal component analysis producing 'eigenfaces'

Eigenfaces are the principal patterns of lightness and darkness.

○

Can combine 'eigenfaces' to produce any of the faces in the face space.

E.g., add 10% eigenface 1, 20% eigenface 2, etc.

○

Can also create brand new faces.

Genetic Algorithm

Randomly generate several composites from face space using eigenfaces.1.

Ask which one of these faces looks most like the perp.

Witness chooses closest likeness.2.

Create random variations around likeness leads to new set of

composites.

Ask again which one of these faces looks most like the perp.

Repeat from #2 until people have the closest person.4.

Genetic Algorithm #2

Randomly generate several composites from face pace using eigenfaces.1.

Witness chooses two closest likenesses2.

'Mating' of these two leads to new generation of 'child faces'3.

Repeat from #2 until you have the closest likeness. 4.

Mugbook Selections

The accuracy of these selections depends greatly on where the photo is located

in the mugbook; flawed.

McCallister, Stewart and Loveland (2003): placed 69 preceding foils and perpetrator's

photo at various locations in computerized mugbook (locations 1-70, 71-140,

141-210).

Later placement led to fewer correct identifications.

Later placement led to fewer false positive choices of foils.

Probability of correct identification decline the more pictures the witness

viewed (i.e., the further in the book the actual perp was).

In the case of real police identification and they aren't actually aware of who

the perp is yet, we cannot make assumptions about accuracy.

Mugbook Exposure Effect

Problem: if we haven't been exposed to a person, but we've selected them as

the perp from a mugbook, when shown them in a lineup later, we're more

likely to pick the person we were already exposed to even if the actual perp

was in the lineup.

We can only get error rates from lab studies; these results are disturbingly high.

Brown, Deffenbacher & Sturgill (1977): first demonstration of 'mugshot exposure

effect'. Three experiments

See 5 suspects for 25 secs each, knowing they will later pick them from

mugshots and from lineup.

1.5 hours later, 15 mugshots (15 secs each): 5 suspects and 5 people to appear

in later lineup. Asked whether each a suspect and their confidence in that

judgement.

72% of suspects were correctly identified.

○

45% false alarms (non-suspect called a suspect).

○

No correlation between confidence and accuracy; equal confidence in

accurate and inaccurate judgments.

○

1 week later, see 4-5 person lineups, with 1-3 'suspects'. Asked whether each

person a 'suspect', or appeared only in mugbook and their confidence in their

judgement.

20% of individual who appeared only in mugshots named as suspects.

○

Again, no correlation between confidence and accuracy; equal

confidence in accurate and inaccurate judgements.

○

Just seeing pictures in the mugbook biases later decisions.

Why do we see this effect?

Familiarity:

Sense of familiarity when seeing foil in lineup that was seen in mugshot.

○

We mistake that sense of familiarity for presence at crime scene.

○

Essentially a case of source confusion.

○

E.g., Dr. Daniel Thompson case

Psych prof mistakenly identified as rapist because his interview was

on tv in the room while the victim was being raped.

○

Commitment:

Incorrect foil chosen from mugshots; witness wants to be consistent, so

picks same foil from lineup.

○

Should be no incorrect lineup choice when foil seen, but not picked, from

mugshots.

○

Therefore, making the incorrect choice in earlier stages leads to incorrect

choices when faced with the lineup.

○

There is evidence for the accuracy of both processes.

Might both be involved to some extent and in different cases. '

○

Lineup Accuracy Measures:

We will establish a diagnosticity ratio.

Take the ratio of hits to false alarms.

○

Can only be calculated in lab studies - not in the real world.

○

Four possible choices:

Hit, miss, false alarm, correct rejection

○

○(see slide)

Problem with the DR as a measure of accuracy.

Accuracy = individuals ability to discriminate perpetrators from non-

perpetrators.

○

It does to some degree assess accuracy, but it also assesses individual's

confidence criterion in making these judgements.

○

○(see slide)

Red curve = confidence in choice for non-perpetrators, green curve =

confidence in choice for perpetrators

The difference between the means is our discriminability.

○

Problem is that discriminability is not the only basis on which individuals base

their choice.

Also, response criterion.

A point where you will say beyond this point, I will say that an

individual is the perpetrator and below this point I will say that the

individual is not the perpetrator.

○

In every case, we have this criterion for the amount of confidence we

have to have in order to make either choice.

○

E.g., response criterion = 25% confidence

DR messed up; ~75% FA and ~ 100% H

§(see slide)

○

E.g., Response criterion = 50% confidence

Discriminability hasn't changed, but DR will change

DR much higher because less FA

§(see slide)

By this criteria, we're more accurate even though our

discriminability hasn't changed.

○

This the problem with the DR as a measure of accuracy: it confounds the role of

actual discriminability from the role of response criterion in determining

accuracy.

We want to know what the real discriminability is.

○

To what extent can people discriminate non-perpetrators from

perpetrators.

○

This is the underlying question that the DR cannot answer.

○

Solution: ROC Curve

Uses signal detection theory - set of mathematical procedures designed to

disentangle actual discriminability from response criterion.

Result looks like this:

○(see slide)

In some way, similar to DR

○

Shows the ratio of H to FA as the response criterion changes.

○

Get the ROC curve: how is the receiver doing at balancing H and FA given

the response criterion.

○

Measure we use is d' (d prime) - a measure of discriminability

Distance between the diagonal line that represents the total inability to

make any discriminations ever to the most distant point of the curve.

○

If we had a point

○

As discriminability decreases, the curve gets closer and closer to the

diagonal.

○

We can also look at the area under the curve (the area between the curve and

the diagonal line)

If the curve went all the way to top left corner, the area would be 1.00

○

As the curve gets closer and close to the diagonal line, the area under the

curve gets smaller and smaller and discriminability thus gets smaller and

smaller.

○

Lineup Identification Issues

Disguises or changed appearance

Facial hair, hair colour, etc.

Play a large role in the accuracy of identification because these are things

that can be easily changed between time of the crime and identification.

Cross-race effect

We're bad at discriminating faces from other ethnic groups because we

don't have as much experience making these discriminations.

Can lead us to making errors in judgements of people if the perpetrator

was outside of our ethnic group.

Cross-age bias

We're also not good a discriminating between people outside of our age

range.

Cross-gender bias

males are better at discriminating between other males and females are

better at discriminating between other females.

Smaller effect than cross-race or cross-age discriminations.

Retention interval

E.g., How long has it been since you say this suspect?

Source confusion becomes an issue the longer you wait.

Verbal overshadowing

If you give a verbal description before picking out suspect, you will do

worse.

Retrieval interference (verbal vs visual)

Inhibition of non-verbal processes

Engaging local vs global processing

We remember a face holistically; global processing.

But when we give it a verbal description, we're using local

processing.

Memory transference

Like source confusion

Memory from one context confused with a memory from another

context.

Context reinstatement

E.g., Imagine you are at the crime scene.

Generally, improves recall.

But for witness identification, the data is mixed.

Lineup Factors

Social Factors

Presumes presence of offender in lineup.

Problem: in a significant proportion of lineups, the actual suspect isn't

there.

This leads to a lot of false identifications because people think they

have to pick someone.

Lineup Instructions:

Biased: "See if you recognize the offender in the lineup."

Unbiased: "See if you recognize the offender, who may not be in the

lineup."

90% still assume presence of the offender in the lineup.

Lineup Composition:

How many foils?

3 plausible foils minimum; accuracy is no better beyond that.

How are foils chosen?

Match to description: foils chosen to match description given by

the witness.

This is the way it's typically done; better of the two according

to the literature.

□

Match to offender: foils chosen to match the actual appearance of

the suspect/offender.

Sometimes foils are simply pulled off of the street.

Lineup Presentation Mode:

Simultaneous Presentation:

A lot of foils come out together, stand together at the same time

(works with photos too).

Problem: witness makes judgement of relative similarity rather

than identity

E.g., which one is the most similar to what I remember?□

Sequential Presentation:

One person/picture is presented at a time.

No relative comparison; judgments are made before the next

person is seen.

Choice ends the lineup.

Witness doesn’t know how many are in the lineup.

Simultaneous vs Sequential Lineups?

Lindsay and Wells (1985): Argued that sequential better than

simultaneous, because witnesses make absolute rather than

relative judgements of similarity in sequential.

Gronlund (2004): Found simultaneous-sequential lineup difference

due mostly to more conservative response criterion in sequential.

In sequential, response criterion is higher, so they're less

likely to make a false choice and they're more confident in

their accuracy.

□

Gronlund et al (2012): ROC analysis shows sequential lineups never

more accurate than simultaneous.

Gronlund, Wixted & Mickes (2014): "Not clear which lineup

procedure will prove to be generally superior, but … ROC analysis is

the only way to make that determination."

Identification Characteristics

Identification Latency

Shorter latency - choices likely to be more accurate.

Faster = more accurate (r= -.20 to -.40)

Not a high correlation, but it's there.

'Sweet spot' at 10-15 seconds?

Point of highest accuracy.

Identification Confidence

Low confidence-accuracy correlations

But evidence for good calibration

Calibrated - ideally, the accuracy would be exactly the same as

confidence.

We do get a pretty good ratio though.

Ideal scenario: (see slide)

What we actually get: (see slide)

Relationship between confidence depends on what proportions of

the lineups actually have the perp present - lab setting only.

There's a sweet spot … when 15-25% of trials do not contain the

perp, we're the closest to the callibration line.

Whenever we get a deviation, it's because we have higher

confidence than is warranted.

Dotted line - source of false convictions

Witnesses are 100% certain someone is the guy when in fact

it isn't

□

Jurors think that there's a strong correlation between accuracy and

confidence so eyewitnesses have big weight in court.

It's very common for the crown attorney to ask the witness how

confident they are.

Confidence seems pretty compelling for jurors.

But, as we can see it's not that strong- we're making a significant amount

of errors in judgment at least under lab conditions.

Post-Identification Feedback

Positive feedback increases expressed confidence

Positive feedback from other witnesses (i.e., "I picked 3 too!") or

officers (i.e., "thank you for your identification.")

Negative feedback reduced expressed confidence

Negative feedback from other witnesses (i.e., "you thought it was #

1? I thought it was #2?") or from officers (i.e., "thank you for

trying").

Therefore, we should take confidence estimates immediately.

But they don't normally do this -normally they aren't asked until at

trial on the witness stand.

Lineup Recommendations

Only one suspect per lineup1.

At least 4 foils. (even though the max is 3)2.

Match foils to witness descriptions (standard procedure).3.

Indicate that offender may not be present.4.

Indicate that confidence will be assessed.5.

Have police unaware of suspect's identity.6.

NRC (2014): Recommendations re: Eyewitness ID

Train all law enforcement officers in eyewitness ID>

Typically, law enforcement has much greater faith than appropriate in

eyewitness testimony.

Use double-blind lineup and photo array procedures.2.

Develop and use standardized witness instructions.3.

Document witness confidence judgements.4.

Videotape witness identification process.5.

Make basic judicial inquiries when eyewitness ID offered. 6.

Make juries aware of prior identifications.7.

Use expert scientific testimony on eyewitness ID.8.

Use jury instructions as alternate way to convey info re: eyewitness ID.9.

Establish national research initiative on eyewitness ID.10.

Additional research on system and estimator variables.

Estimator variables effect accuracy of witness - things/conditions at crime

scene effecting description.

System variables - part of the system in which we get the information.

11.

Eyewitness Testimony

Tuesday, February 13, 2018 10:46 AM

Unlock document

This preview shows pages 1-3 of the document.
Unlock all 19 pages and 3 million more documents.

Already have an account? Log in

Eyewitness Errors

Eyewitnesses are very unreliable.

People incorrectly identified by victim a lot.

The Innocence Project estimates ~ 73% of false convictions based on mistaken

eyewitness identification.

Dr. Donald M. Thompson:

Australian psychologist, lawyer and memory expert.

Detained by the police in 1975 for rape, based on the victim's description.

When the crime occurred, he was on tv program describing how one could

improve one's memory for faces alongside the chief of police.

The victim same him on the tv while she was being raped and that's why she

was able to identify him.

Memory as an Adaptation

It's not the case that memory is an infallible recording system - it has evolved to

have specific functions.

(1) We don’t have to remember everything.

No memory before ~ 2 years of age.

○

No information about repeated events.

E.g., We see coins everyday, but cannot accurately describe them.

○

Memory is a reconstruction from pieces.

○

(2) Memory need not be 100% accurate.

Blake, Nazarian and Castel (2015): tested UCLA students' recall,

recognition of the Apple logo.

Even apple users were only 50% accurate at identifying the correct

logo.

But, they are pretty confident in their responses.

○

Memory Processes

(1) Encoding

Putting information into a form that the memory system can store.

○

We try to leave it the same way we get it, but not always.

○

E.g., auditory sounds are not typically stored, but rather a verbal

representations.

○

(2) Storage

We don't store everything.

○

(3) Retrieval

Hardest part.

○

E.g., studying is encoding/storage, but when it comes to taking a test, it's

very difficult to recall the information.

○

Encoding Factors

Exposure duration

You're more likely to encode things you were exposed to longer.

○

In lineup studies, longer is better. .

○

Arousal level

U-shaped relationship between arousal and performance.

○

Moderate arousal is better than high or low.

○

Distraction

Typically we have to pay attention to things in a scene to encode it.

○

Inattentional Blindness: If you're busy focusing on something, you won't

pay attention to or remember the rest of the scene.

○

Change Blindness: aspects of a scene we're in may change, but we don't

notice these changes.

○

More perpetrators: it's harder to identify the offenders when the crime

was committed by multiple perpetrators.

○

Weapon focus: people pay more attention to a weapon (if the perp is

carrying one) than the offender's face.

○

Distinctiveness

The more distinct an event is, the more likely you are to encode it.

○

Flashbulb memories: very vivid and distinct memories of big/distinct

events.

E.g., JFK assassination, John Lennon killing, Challenger explosion,

9/11

Accuracy seems much more vivid that for other events.

But they're not necessarily accurate at all.

○

Experiment:

Right after the challenger explosion, they asked people what they

had done/where they were when it occurred.

2 years later, they asked the same people again.

Details of the two descriptions didn't match up.

○

Storage/Retrieval Factors

Labelling

Explicit or self-generated labels affect recall.

○

When we store something, we often encode things not in their original

format (e.g., when we encode images as verbal descriptions).

○

Experiment:

Shown one of three images, each group was told a different label

for each image (e.g, sun vs wheel)

The verbal label influence their recall later - people who were told

the image was a sun draw a more sun-like image later, and people

who were told the image was a wheel draw a more wheel-like

image.

○

We essentially remember our labelled version of an event.

○

Prejudices and biases

Buckhout (1974): show people image of a while and black man arguing

on a train; black man is well dressed, white man looks like a vlue caller

worker of some sort.

When recalling the details, people changed the detais so that the

black man was the angry one, the working and the white man was

the calm, well-dressed one.

○

Bahrick et al (1996): asked people to recall all of their grades over their

university career.

Self-serving bias: people remembered less of their poorer grades

and more of their good grades; self-serving because they're

remembering what makes them look/feel good and dropping out

the negative memories.

○

Inferences

Sometimes the contents of memory are inaccurate because you made

inferences.

○

Bower and Treyens (1981):

People waited for experiment in the "professor's office"

When they were finally brought in for the experiment, they were

just asked questions about the contents of the office they were in.

E.g., "Where there any books in his office?"

The experimenters had removed all of the books.

□

But people assumed that there would be books in an

academic's office so they said "yes"

□

○

Interpolated retelling/testing:

In cases, individuals are often asked about what happened multiple

times.

○

Information from one retelling that is repeated and told again in

subsequent retelling will be remembered.

○

If material in an earlier retelling is NOT contained in subsequent

interviews, it will not be remembered as strongly.

○

Greatly affects final testimony in court.

○

Leading questions:

Leading questions bias and alter witness's memory of an event.

○

Harris:

Showed brief film of basketball game.

Asked questions about films -> different wording for different

groups.

E.g., "How tall was basketball player?" vs "How short was the

basketball player?"

□

The questions that determined the responses.

○

Loftus and Zanni (1975):

Written survey about headaches and use of headache products.

Differently worded questions about the same things resulted in the

same thing.

E.g., "How many products tried. 1, 2, 3?" vs "How many

products tried. 1, 5, 10?"

□

○

Loftus and Palmer (1974):

Asked questions about a car accident

Worded the questions differently for different groups

One week later, they asked them "was there glass on the ground at

the scene of the crime"

Their responses were correlated with whether or not they

were asked about the speed of the car.

□

○

Post-event Information (PEI):

Provided by leading questions..

○

Media reports (not so much in Canada)

○

Other witnesses who report things differently than the original event.

○

Witness might add this post-event information to their story.

Significant number of people do

○

Problem with interviews too

○

What actually happens to the original/correct information?

PEI that's incorporated into recollection erases old info??

OR PEI masks/covers up correct information and needs to be

controlled under certain conditions?

○

Why?

Go along with interviewer to seem cooperative.

OR they don't know where they got the information (source

confusion) so they assume it's right.

○

False Memories

E.g., recovered memories under hypnosis/strong suggestion

Women accused mal relatives (e.g., father/uncle) of childhood sexual

abuse.

○

By given how these memories were retrieved, are they accurate?

○

It is very easy to create a false memory.

Loftus and Pickrell (1995): Being lost in a mall

Suggested to participants that when they were kids, they were separated

from parents in the mall - even though they confirmed with their parents

that this had not happened.

○

Initially, they say they don't remember.

○

After 1 week of thinking, a large proportion of participants remembered

the event with significant details.

But these events never happened.

○

Hyman, Husband and Billings (1995): Causing trouble at a wedding

Porter, Yuille and Lehman (1999): attacked by a viscious animal

Heaps and Nash (2001): Saved from drowning by a lifeguard

Mazzoni and Memon (2003): Having a skin sample taken as a child

This became a huge issue in the 80s because kids were asked over and over if

they were molested and eventually, these kids came up with elaborate and

detailed descriptions.

E.g., California daycare accused of abusing and sacrificing kids and babies

based on the false memories of people who went to the daycare.

Started with one child and then the information spread through

leading questions and the inclusion of PEI.

○

Issue:

False accusations

○

Kids who think they were abused psychologically become kids who were

abused.

○

Careful guidelines for interviewing kids have been formed by the APA.

Wise et al (2009): Attorneys and Eyewitness Testimony

Statements of 75 prosecutors, and ~ 1200 defense attorneys

Completed 13-question survey

○

Defense attorneys know more about the nature of eyewitness testimony.

The lack of knowing among prosecutors is startling.

Offender Descriptions: Robbery and Rape Cases

Kuhn et al (1974):

Categorized information and the amount of information given by witness.

Top descriptors:

Gender and age -> not very telling because stats can usually tell us it is a

young-middle-aged man.

○

Height

○

Build

○

Race

○

Weight

○

Face -> most telling descriptor, but least commonly described.

○

Mean = 7.2 descriptors

○

Sporer (1992):

Document Summary

The innocence project estimates ~ 73% of false convictions based on mistaken eyewitness identification. Detained by the police in 1975 for rape, based on the victim"s description. When the crime occurred, he was on tv program describing how one could improve one"s memory for faces alongside the chief of police. The victim same him on the tv while she was being raped and that"s why she was able to identify him. It"s not the case that memory is an infallible recording system - it has evolved to have specific functions. (1) we don"t have to remember everything. No memory before ~ 2 years of age. Memory is a reconstruction from pieces. (2) memory need not be 100% accurate. E. g. , we see coins everyday, but cannot accurately describe them. Blake, nazarian and castel (2015): tested ucla students" recall, recognition of the apple logo. Even apple users were only 50% accurate at identifying the correct logo.

PSYCH 3CC3 Lecture 4: Eyewitness Testimony

Document Summary

Get access

Related Documents

PSYCH 3CC3 Study Guide - Final Guide: Innocence Project, Oklahoma City Bombing, Facial Composite

PSYCH 3CC3 Lecture Notes - Lecture 15: Eyewitness Identification, Australian Psychological Society, Innocence Project

PSYCH 3CC3 Lecture Notes - Lecture 4: Australian Psychological Society, Innocence Project, Detection Theory