he objective of analytics and data science is to use data, analytical techniques, and explanatory and predictive models to support or automate human decision making. Given the key objective is to support and automate decision making, it is important for analytics and data science professionals to have a good understanding of the human decision-making process and the role data plays in that process.
This workshop gives an introduction to the latest research into how people in organisations make decisions. It also shows how human decision-making is often subject to biases that can lead to poor decision outcomes and how to overcome those biases. The workshop concludes with recommendations on how to best match analytics and data science activities with the level and type of decisions so the best decisions can be made, and how to ensure scarce analytics and data science resources are used effectively to support decision-making.
2. Agenda
• Why should we understand decision making?
• Common decision-making approaches
• Cognitive bias
• Business decisions and decision support
• Decision-centric analytics
• Implications for analytics in organisations
3. Analytics is the extensive use of data, statistical
and quantitative analysis, explanatory and
predictive models, and fact-based management
to monitor organisation performance and
support or automate decision making
Why should we understand decision-
making?
Adapted from Davenport and Harris, 2009
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 3
4. Human Decision Making
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 4
Decision making is a mental or behavioural commitment to a course of action.
Research
Approaches
Prescriptive
(“should do”)
Rational
Descriptive
(“does do”)
Bounded
Rationality
Dual Process
Cognition
5. 12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 5
How did you make the
decision to attend the
presentation tonight?
6. Rational Decision Making
■ Often associated with economic modelling
– “Homo Economicus” – Rational Man
■ Assumes people…
– …are consistently rational
– …always have perfect information
– …always have sufficient time
– …always seek to maximise utility
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 6
7. Rational Decision Making
The Typical Process
1. Define the problem
2. Identify the decision criteria
3. Weight the decision criteria
4. Generate alternatives
5. Rate each alternative against criteria
6. Compute optimal decision
The Common Problems
■ Is the problem perfectly defined?
■ Are all decision criteria identified?
■ Are all decision criteria weighted fairly?
■ Are all alternatives considered?
■ Are all alternatives rated fairly?
■ Is the best alternative actually chosen?
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 7
Bazerman and Moore, 2009
Humans rarely make purely rational decisions.
8. Bounded Rationality
■ Human cognition is limited or “bounded”
– Information is not always readily
available or is too expensive (time or
money)
– All options cannot easily be identified
or accurately considered
– Subjective viewpoints cannot be
eliminated
■ Usually a satisfactory or “good-enough-
under-the-circumstances” decision is made
■ A rough model of human decision making
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 8
11. Dual Process Cognition
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 11
System1
• “Fast”
• Unconscious
• High Capacity
• Automatic
• Effortless
• Heuristics-Based
• Skilled
• Error-Prone
Kahneman, 2011
System2
• “Slow”
• Conscious
• Low Capacity
• Controlled
• Effortful
• Rule-Based
• Process-Following
• Predictable
12.
13. Cognitive Biases
■ A consequence of applying the heuristics of System 1 thinking
■ The Availability Heuristic – Over reliance on readily available information
– Retrievability Bias – Unintended employment discrimination
■ The Representativeness Heuristic – Over reliance perceptions of stereotypical groups
– Ignoring Base Rate Bias – Over-optimistic start-up founders
■ The Confirmation Heuristic – Over emphasise importance of information we agree with
– Anchoring and Adjustment Bias – Post graduate salary negotiations
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 13
Bazerman and Moore, 2009
A systematic deviation from rational decision making outcomes
14. Decision Making Exercise
The Scenario
You have been diagnosed with a very serious illness, which,
if left untreated, will become fatal.
Your doctor has presented you with data that describes the
outcome statistics for each treatment option.
Your doctor has asked you to read the outcome statistics and
decide which treatment option you would choose.
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 14
15. Survival Framing
Treatment A
Of 100 people having
Treatment A, 90 live through
the post-treatment period,
68 are alive at the end of the
first year and 34 are alive at
the end of five years.
Treatment B
Of 100 people having
Treatment B, all live through
the post-treatment period, 77
are alive at the end of the first
year and 22 are alive at the
end of five years.
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 15
16. Mortality Framing
Treatment A
Of 100 people having
Treatment A 10 die during
the treatment or in the post-
treatment period, 32 die by
the end of the first year and
66 die by the end of five
years.
Treatment B
Of 100 people having
Treatment B none die during
treatment, 23 die by the end
of the first year and 78 die by
the end of five years.
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 16
17. Framing and Decision Making – Jakarta
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 17
64%
36%
Treatment A Treatment B
Decision with Survival Frame
57%
43%
Treatment A Treatment B
Decision with Mortality Frame
18. Framing and Decision Making – Research
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 18
37%
63%
Treatment A Treatment B
Decision with Survival Frame
61%
39%
Treatment A Treatment B
Decision with Mortality Frame
19. Minimising Cognitive Bias
■ Use decision support and automation tools
– Identify decisions prone to bias
– Analyse the decision process
– Build tool to support or make decision
■ Beware of bias being built into tool
■ Acquire More Domain Expertise
– Learn when System 1 fails you
– Move from System 2 to System 1
■ Requires long-term, accurate and
immediate feedback on outcomes
■ Debias Judgement
– Unfreeze – Show person their biases
– Change – Examples with explanations
– Refreeze – Practice and review
■ Review Decision Making Processes
– Good for approach strategic decisions
– Focus on identifying bias in advisers
■ Use systematic approach to identify
biases in advice provided
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 19
The effect of cognitive biases can never be eliminated, only minimised
Bazerman and Moore, 2009 | Kahneman, 2011
20. Business Decisions
Types
■ System 2 Only
– Decisions with clear context and
known process to solve
■ System 1 and System 2
– Decisions where intuitive responses are
reviewed and revised with processes
■ System 1 Only
– Decisions that can only be made with
skill and intuition (i.e. no process)
Levels
■ Strategic Planning
– Set long-term goals and determine
resources required
■ Management Control (a.k.a. “Tactical”)
– Obtain resources and allocate them
efficiently and effectively
■ Operational Control
– Ensure tasks are completed efficiently
and effectively
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 20
Arnott, Lizama & Song, 2017
21. Decision-Centric Analytics
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 21
Operational Control Management Control Strategic Planning
UnsuitableHigh Medium Low
Analytics
Suitability:
System 2 Only
System 1 & 2
System 1 Only
Adapted from Arnott, Lizama & Song, 2017
22. Types of Analytics
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 23
Descriptive
• Standard
Reports
• Dashboards
• Ad Hoc Reports
• Query / Drill
Down
• Alerts
Explanatory
• Statistical
Analysis
• Numerical
Analysis
Predictive
• Forecasting
• Propensity
• Segmentation
• Network
Analysis
Prescriptive
• Optimisation
• Machine
Learning
• Experiments
Davenport, 2017
← “Traditional”?, “Low-Value”? → ← “Modern”?, “High-Value”? →
23. Decision-Centric Analytics
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 24
Operational Control Management Control Strategic Planning
System 2 Only
System 1 & 2
System 1 Only
Prescriptive
Predictive
Predictive
Explanatory
Explanatory
Descriptive
Descriptive
Adapted from Arnott, Lizama & Song, 2017
24. The Decision-Centric Approach
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 25
•What is the
type of
decision?
•What is the
level of
decision?
Decision
Task
•Who is the
decision
maker?
•How do they
think?
Decision
Maker
•What
information
is required?
•How should
it be
presented?
Analytics
System
•Is the data
available?
•What quality
is the data?
Data
We need to reverse the way we think about analytics – Start with the decision
25. Summary of Human Decision Making
■ Humans rarely (~5% of the time) make pure rational decisions
■ Bounded Rationality is a rough approximation of the decision making process
■ Dual Process Cognition is the most accurate model of human decision making
– System 1: “Fast”, Unconscious, High Capacity, Effortless, Heuristics-Based, Error-Prone
– System 2: “Slow”, Conscious, Low Capacity, Controlled, Rule-Based, Predictable
■ Dominance of System 1 often leads to cognitive biases, especially when cognitive load is high.
■ Cognitive biases mean human decision making is subject to personal preference
– Same decision task can lead to different decision for different people
■ Inherently ambiguous, especially for strategic decisions – No one right process or answer.
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 26
26. Implications for Analytics Programs
■ Business Analytics should be decision-driven and business-owned
– Ensures output from analytics function is more closely aligned with business needs
■ There are no traditional/modern or high-value/low-value types of analytics
– All type or analytics are valuable.
– Match the type of analytics to the type and level of decision being made
■ Intuitive decision-making is not intrinsically good or bad
– Both System 1 and System 2 thinking have their place in organisation decision-making
■ Just having good analytics systems and technically competent teams is not enough
– Data-driven decisions require both managers and analytics teams to understand decision-
making and biases
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 27
27. Want to Learn More?
12/02/2018 Understanding human decision making and bias: The foundation of effective analytics and data science 28