The power of first impressions

1
The power of first impressions

• Gitte Lindgaard
• Professor, NSERC/Cognos Chair in User-Centred
  Design
• Director, Human Oriented Technology Lab (HOTLab)
  Carleton University

Prepared for the National Taiwan University of
Science and Technology
2
When you first meet new people

• You know instantly if they make you feel
  comfortable or if youd rather not meet them
  again
• Next, you start looking for information that
  confirms your initial impression
• Interestingly, that first impression is often
  right but not always

3
The same is true even for descriptions

• Descriptions such as this
• The 32-year old Tom W. was sentenced to16 years
  imprisonment yesterday after a violent armed
  robbery last Thursday. Two people who were
  seriously injured during the robbery are still in
  a critical condition. Tom W. received his first
  sentence at age 10 for several shoplifting
  offences. He soon became an expert at
  break-and-entry crimes, creating a level of fear
  hitherto unheard of in the Ottawa community
• ..usually result in poor ratings of the person
  described

4
Primacy effect improves first impression

• After losing his father in a tragic accident
  when he was only eight years old, Tom W. bravely
  and single-handedly took care of his two younger
  sisters and as his mother whose increasingly
  crippling disease required ever more extremely
  expensive medication and treatment.
• The 32-year old Tom W. was sentenced to16
  years imprisonment yesterday after a violent
  armed robbery last Thursday. Two people who were
  seriously injured during the robbery are still in
  a critical condition. Tom W. received his first
  sentence at age 10 for several shoplifting
  offences. He soon became an expert at
  break-and-entry crimes, creating a level of fear
  hitherto unheard of in the Ottawa community.
• When positive information precedes the negative,
  ratings of the same description are significantly
  higher

5
But recency has little effect

• The 32-year old Tom W. was sentenced to16 years
  imprisonment yesterday after a violent armed
  robbery last Thursday. Two people who were
  seriously injured during the robbery are still in
  a critical condition. Tom W. received his first
  sentence at age 10 for several shoplifting
  offences. He soon became an expert at
  break-and-entry crimes, creating a level of fear
  hitherto unheard of in the Ottawa community
• After losing his father in a tragic accident when
  he was only eight years old, Tom W. bravely and
  single-handedly took care of his two younger
  sisters and as his mother whose increasingly
  crippling disease required ever more extremely
  expensive medication and treatment.
• But the same information presented at the end
  has
• no effect on ratings

6
Apparently, this is how we judge
Primacy effect
N words recalled
Recency effect
Serial position
7
And heres the proof
F(2,38) 7.34, p lt .01
8
Biologically speaking

• We are hard-wired to respond quickly
• Historically
• Speed
• Visual acuity
• Intuition
• Situational awareness
• Good health .determined survival

9
Proof the traditional visual pathway
10
More recent evidence
11
Summary

• A small bunch of neurons lead directly from the
  hypothalamus to the amygdala in a single synapse
• This by-passes the traditional visual pathway
  (via thalamus ? occipital cortex ? amygdala)
• Consequence at least some information is
  processed emotionally before the brain can
  determine what it means
• Thus, we react emotionally before we know what we
  are seeing/hearing/smelling/touching

12
Usability and user satisfaction

• ISO Standard 9241/11 says usability is about
• Effectiveness
• Efficiency
• Satisfaction
• Traditionally, satisfaction measured degree of
  comfort, attitude
• But the ISO standard is based on office machines
• Do peoples criteria for judging satisfaction
  differ between web sites and office machines?
• How I came to be interested in user satisfaction

13
(No Transcript)
14
Why such low satisfaction ratings?

• Is the concept of satisfaction meaningless in the
  context?
• The burden of reputation?
• Local Government site
• Colour combination?
• Known to affect website appeal (Knutson, 1998)
• Arousal
• Psychology of art (Kreitler Kreitler, 1982)
• Theory of aesthetics (Berlyne, 1971 1973)
• First impressions
• Neurophysiology (LeDoux, 1996)
• Somatic markers (Damasio, 2000)
• Mere exposure effect (Zajonc, 1980)
• Emotion (Norman, 2004 2005)

15
Arousal preferences and tension

• Two very bright, highly saturated primary colors
  juxtaposed create maximum tension (Kreitler
  Kreitler, 1972)
• What does maximum tension mean?

Optimum arousal level
? Zone of maximum tension ?Point at which
operator resets internal criterion?
Asleep
16
Theory of aesthetics versus performance

• Berlyne Novelty complexity increases
  interestingness and pleasantness of
  meaningless stimuli
• Consistent with increases in arousal but not
  unpleasant
• Assume that
• resetting our internal preference/performance
  criterion is a protective survival mechanism
• We reset this criterion when the going becomes
  unpleasant
• then there is a close link between how I feel and
  how well I perform

17
Arousal performance and tension

• How it works
• At low task demand levels,
• human performance is aligned
• with task demands
• As demands increase beyond
• human capacity, the performer
• reduces his internal criterion.
• The gap between task demands
• and performance increases over
• time, but the human
• performer is unaware of this
• shift in performance criterion and
• continues to believe his
• performance is Ok
• (Moray,
  1983)

18
Unpacking the notion of satisfaction

• In several experiments we found user satisfaction
  to comprise five components
• likeability
• emotion
• aesthetics
• expectations
• Usability
• Our paradigm
• users browse for 10 minutes while talking about
  their experience
• Usability test
• Second interview

19
Overall experimental outcome
20
Proportion of positive aesthetics statements
21
Proportion of positive usability statements
22
Our results suggest

• Users
• form an opinion of a web site very quickly
• are not fooled by poor usability, but
• that does not influence their opinion of
  aesthetics
• this contradicts Tractinsky et al.s (2000)
  results claiming that if it is pretty, it is
  usable
• Once users have formed an opinion, they stick
  with it
• This suggests that the first impression is very
  important
• But how far can we trust our eyes?

23
(No Transcript)
24
Which circle in the middle is bigger?
25
Is this possible!?!
26
If you take a look at the following  picture ,
let me tell you ... it is not animated.  Your
eyes are making it move.  To test this, stare at
one spot for a couple seconds and everything will
stop moving.  Or look at the black center of each
circle and it will stop moving.  But move your
eyes to the next black center and the previous
will move after you take your eyes away from
it....  Weird 
27
So .

• We dont necessarily see what is there
• Our brains usually correct for impossible images
• We know in an instant if we like or dislike what
  we see.
• Next research question how long is an instant?
• Mere exposure effect in lt 5msec
• Effect increases with increasing number of
  exposures, but awareness does not .
• Starts to wane at 50msec

28
Experiment 500msec exposure time

• Participants (N 22)
• 100 home pages (good? bad) shown twice in
  different random orders
• 20 practice pages shown first
• 500msec exposure time
• Subject rated visual appeal (interval scale)
• Selected top 25 and bottom 25 for further
  investigation

29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
Results First versus second trial
R2 .94, plt.001
36
Next experiment

• Selected 25 best and 25 worst home pages
• N 31 500 msec exposure time
• Same method as before for first two trials
• Added trial 3
• Unlimited exposure time
• Subjects asked to rate
• Simple ? complex
• Interesting ? boring
• Good design ? bad design
• Clear ? confusing
• Good color ? bad color
• Good layout ? bad layout
• Imaginative ? unimaginative

37
Results First versus second trial
R2 .97, plt.001
38
Results Study 1 versus study 2
R2 .95, plt.001
39
Visual appeal versus visual characteristics

• Simple ? complex (R2 .80, p lt .01)
• Interesting ? boring (R2 .91, p lt .001)
• Good design ? bad design (R2 .92, p lt .001)
• Good layout ? bad layout (R2 .88, p lt .001)
• Clear ? confusing (R2 .39, plt.01)
• Good color ? bad color (R2 .90, p lt .001)
• Imaginative ? unimaginative (R2 .86, p lt .001)
  
• Multiple regression best linear combination, R2
  .94, p lt .001

40
Simple ? complex
Interesting? boring
Clear ? confusing
Good design ? bad design
Imaginative ? unimaginative
Good color ? bad color
Good layout? bad layout
41
Next experiment

• Same paradigm as before
• 50msec or 500msec exposure time
• Responded using numeric keys 1-9
• Comparison of medians in 500msec and 50msec
• first trial, R2 .83, p lt .001
• second trial R2 .85, p lt .001
• ? more variability among participants in 50msec
  than in 500msec condition (but R2 .90, p lt
  .001)

42
Conclusion

• First impression of web sites appears to be a
  mere exposure effect
• People decide how much they like a web site very
  quickly
• That first impression is a holistic, visceral
  (physiological) response it seems to stick

43
Returning to the primacy effect medical
decision making

• Primacy effects have been shown in many domains
• Do they occur in occupational relevant decisions
  too?
• To find out, we first had to quantify
  diagnosticity of signs and symptoms

44
Stimulus items and design

• Three lists generated with all symptoms belonging
  to a disease-pair
• 12 symptoms in each list
• 8 were disease-specific (4 confirmatory, 4
  disconfirmatory)
• 4 were equivocal (could occur in both diseases in
  a pair)
• Each list presented twice, with symptoms in
  different random orders (thus 6 orders)
• Two lists for same disease-pair never presented
  consecutively
• N 18

45
Instructions and task
Your task is to estimate approximately how often
you would expect to observe each fact, sign and
symptom listed below in a pool of 100 patients
who have all been diagnosed as having the
disease indicated. If you were to draw a random
sample of 100 patients who had all been diagnosed
as having Peritonitis, in how many of these 100
patients would you expect to find the
following Abdominal pain increasing when asked
to cough Light abdominal
colics Pallor Sudden
increase in pulse and temperature Hyper-per
istalsis Keloidal scar
tissue Sudden onset of
pain Vomiting No
peristalsis Abdominal area is very
tender to touch Rigidity of abdominal
muscles
46
Expected results
47
Actual results
Mean subjective ratings for Angina Pectoris and
Glomerulonephritis
judged under Angina Pectoris
judged under Glomerulonephritis
48
Symptoms selected
High- and low-diagnostic Symptoms for GLO
High- and low-diagnostic Symptoms for ANP
49
Bayes Theorem
50
Bayes Theorem

• When the individuating information is
  nondiagnostic i.e. P(D? H) P(D? H)
• then P(H? D) P(H)/P(H) P(H)
• Since P(H) P(H) 1, P(H? D) P(H)

51
Example of a case history
Mr. Brown refused both lunch and afternoon tea
because he is very Nauseated and has vomited
once in the past hour. He is distinctly cyanotic
His lips and nails are blue, and his hands and
feet are cold. He says that he Feels as if he is
about to have one of his frequent bouts of
diarrhoea. Of the 100 patients to be evacuated,
72 (28) were Angina Pectoris cases and 28 (72)
were Glomerulonephritis patients. How likely do
you think it is that Mr. Brown is one of the 72
(28) Angina Pectoris cases? _________________
____________ 0 100
52
Serial position effects (nurses)
Main effect, serial position p lt .001
53
Serial position effects (physicians)
Main effect, serial position, p lt .001
54
Conclusion

• First impressions are extremely powerful
• We judge situations, people, diseases very
  quickly, sometimes too quickly
• Most of the time this tendency serves us well
• If we did not do it, we would probably not have
  survived as a species

55
So, keeping in mind the human mind.

• The human brain has certain limitations
• We must take these into account when we design
  interactive technology
• Human Computer Interaction (HCI) is a field of
  study that does this

56
What do we know

• Subjects do not consider P(D? H)
• No evidence that nondiagnosticity was detected
• Poor understanding of diagnosticity
  
• 
  
  (Fischhoff Beyth-Marom, 1983)
• Disconfirmatory evidence is seen to be less
  important than confirmatory evidence
• But no unequivocal evidence for a confirmation
  bias

57
Primacy effects

• Directed impression an interactive cognitive
  process in which an impression established by the
  initial item affects the meaning of later items
• Meaning of later presented items changes due to
  impression of first item.. Cannot be entirely
  excluded
• Inconsistency discounting a decrease in weight
  of later items due to inconsistency with earlier
  presented items. Meaning is constant across all
  items
• In diagnostic HL and LH cases, probability
  estimates higher when A was high and D was low
  than in reverse cases. Weightings do not change
  as a function of serial position of symptoms
• Attention decrement weighting of later items
  decreases due to progressive reduction in
  attention over several items. Independent of
  relationships among the items
• Progressive decrease in attention over items.
  Estimates were closer together in 3rd serial
  position than in the other two

58
Diagnosticity revisited

• (a) P(D? H) 0.85 P(D? H) 0.83
• (b) P(D? H) 0.04 P(D? H) 0.02
• (c) P(D? H) 0.85 P(D? H) 0.20
• If subjects do not understand concept of
  diagnosticity, they rely on the absolute
  frequency of D under H, P(H? D) should be equal
  for (a) and (c)
• If they use the difference in frequency of D
  under H and H, P(H? D) should be equal for (a)
  and (b) but higher for (c)
• If P(H? D) is calculated normatively, considering
  both the absolute and relative frequency of
  occurrence, (c) should be highest, followed by
  (b) and (a)

59
Implications for training

• Medical personnel could benefit from a computer
  system that would
• Provide information about diseases under which
  the symptom configuration observed in a given
  patient may occur
• Yield objective estimates of the relative
  frequency of occurrence of individual symptoms
  under a range of diseases
• Calculate posterior probabilities based on
  information provided by a diagnostician

60
What could we do about it?

• Database of cross-referenced medical texts to
  raise awareness of range of diseases under which
  symptoms occur
• Database of actual patient data
• Generate checklists that would allude to missing
  information
• Teach formal decision theory in medical and
  nursing schools

61
Databases

• Problems
• Many missing data
• Errors occurring in the data entry phase
• Physicians will not enter data regularly
• and possible solutions
• Novel paper product uses conventional forms
• Allow speech and pen-based data entry
• (Cohen McGee, 2004)

62
Barriers

• Culture of shame and blame
• Misguided professionalism
• Lack of organizational control
• Lack of organizational commitment to safety

63
Acknowledgements

• this talk included work done with
• Cathy Dudek
• Dick Dillon
• Gary Fernandes
• Judy Brown

64
Thank you for your kind attention
Any questions