Transcript Document
PSYC 2220 – HUMAN FACTORS IN DESIGN
Decision Making
Human Factors
PSYC 2200
Michael J. Kalsher
Department of
Cognitive Science
1
1
decision making
1 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
What is a Decision-Making Task?
One in which:
– a person must select among
more than one choice
alternative.
– some (but not all) information is
available for each alternative.
– the time frame is relatively
long (more than a second)
– the best choice is not
necessarily obvious (uncertainty
and risk present).
2
decision making
2 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
The 3 Phases of Decision-making
1. Acquiring/perceiving relevant information (decision cues)
2. Assessing the situation to determine how the
information we have relates to the decision at hand.
3. Planning and selecting choices (based on perceived costs and
benefits of each choice)
Note: Decision-making and problem-solving often go hand-in-hand -- not
easily separated.
Controlled vs. Automatic Decision-making
– Quick/automatic = “Intuitive decision-making”
– Slow/deliberate = “Analytic decision-making”
3
decision making
3 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Classical Decision Theory
(also termed Rational Decision Theory)
Assumes that if researchers could specify values
for the costs and benefits associated with
different choices, then mathematical models
could be applied to those values, yielding an
optimal choice (i.e., the one that maximizes benefit
and minimizes cost).
4
decision making
4 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Major Models of Decision-making
–Normative models
• Revolve around the concept of utility, or the overall
value of a choice to the decision-maker.
• Prescriptive -- they specify what people ideally
should do.
• Do not describe how people actually perform
decision making tasks.
–Descriptive models
• Attempt to describe and model actual human
decision-making.
5
decision making
5 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Normative Decision Models
• Translate the multi-dimensional space of
options into a single dimension reflecting the
overall utility (or value) of each option.
• Assume the overall value of a decision
option is the sum of the magnitude of each
attribute multiplied by the utility of each
attribute.
6
decision making
6 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Multi-attribute Utility Theory
U(v) =
a(i) =
u(i) =
n =
n
U(v)
= a(i)u(i)
i=1
overall utility of an option
magnitude of the option on the ith attribute
utility (importance) of the ith attribute,
number of attributes.
Utility of each
attribute (u)
Options
Attributes
U(v) or overall
utility or value
of each option
A1Price
A2Mileage
A3Insurance A4Stereo
A5Repairs
4
5
2
1
8
01Model 1
3 (12)
3 (15)
9 (18)
3 (3)
1 (8)
56
02Model 2
3 (12)
3 (15)
3 (6)
3 (3)
3 (12)
60
03Model 3
9 (36)
5 (25)
3 (6)
1 (1)
9 (72)
140*
04Model 4
1 (4)
3 (15)
9 (18)
9 (9)
9 (72)
118
•= Highest overall utility among the options available
7
decision making
7 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Normative Decision Models
Expected Value Theory
(addresses outcome uncertainty)
Applies to any decision that involves a “gamble” type
decision.
– Each choice has one or more outcomes with an associated worth and
probability.
• A .20 probability of winning $50 vs.
• A .60 probability of winning $20
Assumes that overall value of a choice is the sum of the
worth of each outcome multiplied by its probability.
– E(v) is the expected value of the choice
– p(i) is the probability of the ith outcome
– v(i) is the value of the
ith outcome
n
E(v) = ∑ p(i)v(i)
i=1
8
decision making
8 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Normative Decision Models:
Subjective Expected Utility Theory (SEU)
n
E(v) = ∑ p(i)v(i)
i=1
Worth
Component
E(v) is the expected value of the choice
p(i) is the probability of the ith outcome
v(i) is the value of the ith outcome
The worth component is considered to be subjective
and determined individually for each person.
– Assumes a person will select the action with the highest
overall subjective expected utility value.
– SEU useful for studying conditions in which humans make
decisions and for developing training and decision aids.
9
decision making
9 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Descriptive Decision Models
• Developed because human decision-making
frequently violates key assumptions of
normative models.
• Descriptive models attempt to capture how
humans actually make decisions.
– People tend to rely on simpler and less-complete means
of selecting among choices termed “heuristics.”
10
decision making
10 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Effects of Framing:
Prospect Theory
How a problem or situation is “framed” can affect the outcome of
choice problems in a way that violates assumptions of rational choice
theory (Tversky and Kahneman ,1981).
Gain-frame vs. Loss-frame
Example: If asked to choose between getting $1000 with certainty or
having a 50% chance of getting $2500, which would you choose?
•
Many people choose the certain $1000 instead of the uncertain chance of
getting $2500 even though the mathematical expectation of the uncertain
option is $1250. This attitude is described as risk-aversion.
•
Research also shows that the same people when confronted with a certain
loss of $1000 versus a 50% chance of no loss or a $2500 loss often choose
the risky alternative. This is called risk-seeking behavior.
11
decision making
11 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Examples of Framing:
The Asian Disease Problem
– Participants are asked to "imagine that the U.S. is
preparing for the outbreak of an unusual Asian
disease, which is expected to kill 600 people.”
– They are then asked to choose between two
alternative programs designed to combat the
disease.
– The consequences of the choices are posed as
probability statements.
12
decision making
12 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Effects of Framing: The Asian Disease Problem
One group of participants are presented with a choice between:
Program A: "200 people will be saved"
Program B: "there is a one-third probability that 600 people will
be saved, and a two-thirds probability that no people will be
saved"
72 percent preferred program A
28 percent preferred program B
• -------------------------------------------------------------------------------------Another group of participants are presented with the choice between:
Program C: "400 people will die"
Program D: "there is a one-third probability that nobody will die,
and a two-thirds probability that 600 people will die"
22 percent preferred program C
78 percent opting for program D
decision making
13
13 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Descriptive Models:
Satisficing (Simon, 1957)
Major Assumption
People don’t usually meet the goal of making
absolutely best/optimal decisions, but instead opt for
a choice that is “good enough.”
The rationale
Going beyond “good enough” has too little advantage to
be worth the effort.
Does it work?
- Reasonable approach given that people have limited
cognitive capacities and limited time.
- Sampling procedures are key to its success.
14
decision making
14 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Decision Models: A Summary
• Careful analysis of choices and their respective
utilities is desirable if:
– time is unlimited.
– the amount of available information is limited.
• Given limited time, too much information, and/or
stress, people tend to shift to simplifying
heuristics.
• Research shows that people can shift between
analytical and heuristic decision-making as
circumstances dictate.
15
decision making
15 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases:
What are they?
• Cognitive heuristics are usually very
powerful and efficient decision tools, but
their use does not guarantee the best
solution.
• Because they are simplifications,
heuristics sometimes lead to biases or
misperceptions.
16
decision making
16 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases:
Information-Processing Limitations in Decision Making
Figure 7.1 An Info Processing Model of Decision-making. The model highlights several cognitive
limits to effortful decision making that lead us to rely on heuristics, including: selective attention,
working memory, and retrieval from LTM.
17
decision making
17 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Information Processing:
Factors and
Limitations that Influence Decision-Making Quality
• Amount and quality of information brought into WM (e.g.,
workload; attentional resources)
• Working memory capacity limitations
• Amount of time available for decision making (e.g., medical
emergency; system failure).
• Amount and quality of knowledge a person holds in LTM
relevant to activity (“knowledge-in-the-head”).
• Ability to retrieve relevant information, hypotheses or actions
from LTM at the critical moment (Problem of inert knowledge)
People have the most difficulty with decisions made with too little or
erroneous information, extreme time stress, high cognitive workload,
changing dynamic informational cues, conflicting goals, and novel/unusual
circumstances--factors common in high-risk environments.
18
decision making
18 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases:
Information Processing Limits in Decision Making
• Cue Reception and Integration
– Relevant cues (pieces of info) are retrieved from environment and go into Working
Memory. Limits on the number of cues that can be considered.
• Hypothesis Generation, Evaluation and Selection
– Decision-makers make educated guesses as to the cues’ meaning.
– Meaning is derived by retrieving information from LTM and comparing it to the
cues; hypotheses are brought into WM and evaluated with respect to how likely
they are to be correct.
– Revise or generate a new hypothesis.
– Chosen hypothesis serves as the basis for course of action.
• Generating and Selecting Actions
– One or more possible actions generated in WM by retrieving possibilities from
LTM.
– Action selection is achieved by evaluating possible outcomes, likelihood of each
outcome, and the positive/negative factors associated with each.
19
decision making
19 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases:
Heuristics and Biases in Receiving and Using Cues
– Attention to a limited number of cues (limited by
constraints on working memory).
– Cue Primacy and Anchoring (first impressions are lasting).
– Inattention to later cues (cues occurring later in time or ones
that change over time are ignored; attributable to attentional factors).
– Cue Salience (Loudest, brightest cues are more likely to attract
attention and are given more weight. The most salient cues aren’t always
the most diagnostic ones).
– Overweighting of unreliable cues (relative to more
reliable information).
20
decision making
20 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases:
Heuristics and Biases in Hypothesis Generation
After a set of cues are processed in WM, decision-makers generate
hypotheses by retrieving one or more from LTM: The following
heuristics/biases affect this process:
1. Generation of a limited number of hypotheses
– People generate only a small subset of hypotheses (1-4) due to WM constraints
and never consider all relevant ones. Stress exacerbates the problem.
– The first option considered by experts is likely to be reasonable, but not for
novices.
2. Availability heuristic
– People more easily retrieve hypotheses that have been considered recently or that
have been considered frequently; If something comes to mind easily, people
assume it is relatively common and therefore a good hypothesis
3. Representativeness heuristic
– Tendency to judge an event as likely if it represents features typical of its category.
4. Overconfidence
– People believe that they are more correct than they actually are; less likely to seek
out evidence for alternative hypotheses or prepare for the possibility they are
21
wrong.
decision making
21 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases: Heuristics and
Biases in Hypothesis Evaluation and Selection
Once the hypotheses have been brought into WM, additional cues
are sought to evaluate them. The process of considering additional
cue information is affected by the following cognitive limitations:
1. Cognitive Tunneling (functional fixedness; mental set).
– People tend to adopt and fixate on a single hypothesis, assume that it is
true, and then proceed with a solution consistent with the hypothesis.
– Fail to utilize subsequent cues.
2. Confirmation Bias
– People have a hypothesis they are trying to evaluate and seek
only confirming information in evaluating the hypothesis.
These limitations are exacerbated by high stress & mental workload
22
decision making
22 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Heuristics and Biases in Action Selection
1. Retrieve only a small number of actions
–
Limited number of plans can actually be retrieved and kept in
WM
2. Availability heuristic for actions
–
–
People retrieve most “available” actions from LTM
Availability of items in LTM is a function of recency, frequency,
and how strongly they are associated with hypothesis
3. Availability of possible outcomes
–
–
When more than one action is retrieved, must select based on
how well action will yield desirable outcomes
Evident when people make decisions and fail to foresee
outcomes that are readily apparent in hindsight
4. Framing bias.
--
Decisions affected by the way the situation is presented.
23
decision making
23 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Framing Bias: Additional Examples
1.
People are asked the price they would pay for a pound of ground meat
that is 10 percent fat or a pound that is 90 percent lean. People tend to pay
8.2 cents per pound more for the option presented as 90 percent lean.
2.
Students told either that they answered 80 percent of the questions on the
exam correctly or that they answered 20 percent of the questions
incorrectly.
Students more likely to feel they are performing better if they are told the former.
3.
People are told there is a 20 percent mortality rate associated with a
particular treatment or they are told there is an 80 percent chance the
treatment will save their life. People less likely to choose the treatment when
expressed in terms of mortality.
Sunk Cost Bias: The tendency to choose the riskier of two options when
the framed in terms of loss. (Hesitation to sell losing stocks, but
willingness to sell winning stocks to lock in a gain). Decisions should be
framed in terms of gains to counteract risk-seeking tendencies.
24
decision making
24 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Naturalistic Decision Making:
Making decisions in the real world
Real world decision making tasks tend to have
characteristics such as:
– Ill-structured problems
– Uncertain, dynamic environments
– Information-rich environments in which situational cues
change rapidly
– Cognitive processing that proceeds in iterative
action/feedback loops
– Multiple shifting and/or competing individual and
organizational goals
– High risk
– Time constraints & stress
– Many people involved in the decision
25
decision making
25 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
The Decision Making Context
Everyday decision-making is characterized by
cognitive complexity
– Multitude of factors affect everyday decisions
– Heuristics are accurate much of the time, but depends on
people having the appropriate information resources and an
ability to adapt them.
Traditional decision making research and more
recent naturalistic decision making are
complementary--not mutually exclusive
– Heuristics and biases discovered in laboratory research have
been validated in the “real world.”
26
decision making
26 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Research on Decision-Making in Complex
Environments: Skill, Rule, Knowledge
Rasmussen’s SRK model describes 3 levels of cognitive control that might be
used during task performance: Skill-based behavior, Rule-based behavior, &
Knowledge-based behavior
27
decision making
27 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Rasmussen’s SRK Model:
Skill-Based Behavior
• If extremely experienced with task, information is
processed at the “skill-based level.”
– React to perceptual elements at automatic or subconscious
level.
– Performance governed by stimulus-response associations
developed at the neurological level.
• Errors in skill-based behavior usually caused by:
– Misdirected attention
– Paying too close attention to the task, which then may interrupt
an automated sequence of behavior (scripts).
28
decision making
28 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Rasmussen’s SRK Model:
Rule-Based Behavior
• If familiar with task, but limited experience,
information processed at the “rule-based
level.”
– Meaningful cues (signs) can trigger rules accumulated from
past experience.
– Rules are If-Then associations between cue sets and the
appropriate actions.
• Errors tend to result from misclassification of
the situation and subsequent application of the
wrong rule.
29
decision making
29 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Rasmussen’s SRK Model:
Knowledge-Based Behavior
• If situation is novel, person operates on knowledgebased level.
– Analytical processing using conceptual information.
– Person assigns meaning to cues and integrates this information
into a coherent “story” that describes what is happening.
– Information is processed with respect to goals in WM.
• Errors at knowledge-based level tend to result from
factors associated with analytical thinking
–
–
–
–
Limited WM
Biases in generating hypotheses/actions
Cognitive fixation
Incorrect mental models
30
decision making
30 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Cognitive Continuum Theory
• Decision making process assumed to occur along a
continuum from intuition to analysis
– Intuitive process—characterized by low levels of control and conscious
awareness, rapid processing, and high confidence in the answer.
– Analytical process—characterized by higher levels of cognitive control,
slow processing and lower confidence in answer.
• Use of Intuitive vs. Analytical processes is determined by
the nature of the task:
• Intuitive processing induced by tasks having a large number of
cues, simultaneous and brief display of cues, strong relationships
among the cues, and short time-frame for the decision.
• Analytical processing induced by tasks having fewer cues, high
confidence in the task, and long sequential availability of cues.
– Failure in the use of one type causes switching to the other.
31
decision making
31 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Situation Awareness
• Perception of the elements in the environment within
a volume of time and space, the comprehension of
their meaning and the projection of their status in the
near future (Micah Endsley, 1988).
• Levels of situation awareness (SA) and cognitive
complexity:
– Level I:
Perceiving the status, attributes, and dynamics of
relevant elements in the environment
– Level II: Comprehending relevant cues in light of one’s
goals
– Level III: Projecting the future activity of the elements in the
environment
32
decision making
32 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Situation Awareness:
Cognitive Processing Requirements
– The integration of cues into complex mental
representations of a system accomplished by
using pre-existing knowledge to interpret and
give meaning to cues.
– SA may also require evaluation of factors such
as risk and time available for decision.
In times of high mental workload and stress,
people seem to “lose” situation awareness.
33
decision making
33 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Recognition-Primed Decision Making:
Studying Experts Making Decisions Under Time Stress
• According to Klein (1989), in most instances, experts
simply recognize a pattern of cues and recall a single
course of action which is then implemented.
• 3 Assumptions of RPDM:
– People use experience to generate a plausible option the first
time around.
– If the decision makers are experts, time pressure should not
cripple performance because of rapid pattern matching.
– Experienced decision makers can adopt a course of action
without comparing and contrasting possible courses of action.
34
decision making
34 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Schemas, Stories & Mental Models
• Schemas
People use previous knowledge to comprehend and integrate
the situational cues into a dynamic model of situations they are
trying to evaluate.
• Explanation-based decision making
(Pennington & Hastie, 1988; 1993)
– Consists of 3 activities
• Receiving info & constructing a causal story that can
account for the information
• Generating possible actions
• Determining actions that best fit the story via a matching
process
– Constructing a causal explanation is pivotal!
decision making
35
35 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Mental Models
When applied to decision-making, suggests that
people construct mental models of the relevant
system or environment and use it to run simulations
throughout the decision-making process.
Simulation is used to:
1. generate expectations for other cues not previously
considered.
2. guide observation of changes in system variables.
3. evaluate goals, actions & plans and to make predictions
useful in monitoring actions & consequences in the system
or environment
36
decision making
36 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Integrated Model:
Adaptive Decision Making
37
decision making
37 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Integrated Model
• Information enters system and is
processed at one of three levels:
– Automatic skill-based
– Intuitive rule-based
– Analytical knowledge-based
• If situation is difficult or complex and
time allows, decision maker utilizes
more complex evaluative processes
38
decision making
38 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Improving Human Decision Making
• “Human Error” focuses on cognitive
errors (poor decision making) rather than
behavioral errors (hand slipped)
• Possibilities for improvement
– Redesign for performance support
– Training
– Decision Aids
39
decision making
39 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Redesign for Performance Support
Improve the quality of information
provided by the external environment
(e.g., focus on improving the quality of “knowledge-in-theworld”).
Doing so supports better decision
making, thereby eliminating the need
to change the person making the
decisions
40
decision making
40 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Training
• Train people to overcome heuristics/biases
– Focus on counteracting specific types of biases
– Allow natural use of varying strategies, but teach people
when to use them and the shortcoming of each
• Highlight the value of metacognition by training
people to:
– Recognize and use appropriate/adequate cues that facilitate
situation awareness
– Check situation assessments or explanations for
completeness and consistency with cues
– Analyze data that conflict with situation assessment
– Recognize when too much conflict exists between the
explanation or assessments and the cues
41
decision making
41 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Training
• At intuition rule-based level, provide training
to enhance perceptual and pattern
recognition skills
• Focus on situation assessment
– Trainees learn to recognize critical situational cues
& to improve accuracy of their time available & risk
judgments
• At automatic level, focus relevant cues in raw
data form
– Works only for situations where a cue set
consistently maps onto a particular action
42
decision making
42 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Training
• Decision maker should receive
feedback, preferably for each cognitive
step
• Training does not overcome memory
limitation problems
• Large amount of knowledge may remain
inert and unretrieved
43
decision making
43 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Decision Aids
Decision tables
– Used to list possible outcomes, probabilities &
utilities of action alternatives
– Deflects the load placed on WM
– Similar to a decision tree used for representing
decisions that involve sequence of decisions &
possible outcomes
• Branching point used for possible consequences &
associated probabilities
44
decision making
44 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Decision Aids
Expert systems
– Computer programs designed to capture one or
more experts’ knowledge and provide answers in
a consulting type role
– Developed to help with wide variety of tasks
– Take situational cues as input and provide either a
diagnosis or suggested action as output
– Have not yet been successful in complex decision
environments
• Limited ability to collaborate/communicate with expert
system
45
decision making
45 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Cognitive Support:
Decision Support Systems
– Designed to improve decision making by extending
the user’s cognitive decision making abilities
• e.g., utilize computers to support WM and perform calculations
– Front-end analysis of the task is critical to
determine what information should be provided or
what calculations/modeling needs to be performed
– Usability testing is also needed, especially for
advanced features
– Success depends upon:
• Users’ ability to control and/or redirect the subsystem
• The extent to which the user/subsystem have common or
shared representations of the state and problem status
46
decision making
46 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Problem Solving:
Characteristics
• Occurs when the “problem-solver” does not have the
method to do a task stored in memory.
• Problem solving is difficult because:
– Limitations on WM; Lack of sufficient relevant system knowledge
to solve problem; Person has sufficient system knowledge, but is
disconnected/disorganized and cannot access it from LTM
• Solution path usually a set of “subroutines” that are
combined to solve the problem.
• Knowledge based decision-making shares similar
cognitive processes as problem solving.
47
decision making
47 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Problem Solving:
Requirements
• Relies extensively on generating actions and
planning.
• Requires large amounts of relevant knowledge,
good strategies for generating solutions, and
effective mental models to offset memory
limitations, lack of knowledge in LTM and a lack
of good memory retrieval strategies.
48
decision making
48 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Problem Solving:
Errors and Biases
• First type of difficulty caused by the way
in which people represent the problem
- If overly constrained, omits constraints, or only
allows one view of things, a solution will be less
likely to be generated
• Failure to generate correct solution plan
- Due to fixation on previous plans that worked in
the past
- Prone to functional fixedness
49
decision making
49 of 49
PSYC 2220 – HUMAN FACTORS IN DESIGN
Errors in Problem Solving
• Failure to develop solution caused by
limitations of WM
– Often a long sequence of action “packets”
must be composed into a “plan”
– Cognitive simulation must be carried out to
evaluate the plan
• Frequently involves too many bits of info to be
handled in working memory
50
decision making
50 of 49