Rational Choice Theory

Rational Choice Theory: A Forum for
Exchange of Ideas between the Hard and
Social Sciences in Predictive Behavioral
Modeling
Sun-Ki Chai
Dept. of Sociology
University of Hawai‘i
The Social Sciences and Predictive
Behavioral Modeling
• Unprecedented interest among “hard” scientists
in study of human behavior
• Remarkably little use of existing social science
theory or method, even those that adopt formal,
positivist approaches
• Why?
– Technical inadequacy?
– Lack of familiarity?
– Methodological incompatibility?
Agent-Based Computational Modeling and
Rational Choice
• ABCM is dominant approach to predictive behavioral
modeling among computer scientists and engineers.
• Rational Choice is by far the dominant theoretical
approach to predictive behavioral model in the social
sciences.
• Until recently, the literatures developed without much
dialogue or cross-citation.
• Social scientists have recently taken greater interest in
ABCM.
• Still seen as alternative approaches rather than
complements
Why is Rational Choice Theory Important to
Hard Scientists?
• subject to greater formal development and elaboration
than any other approach
• applied to a much wider range of empirical behavioral
phenomena
• spatiality, biology, and yes, culture, can be incorporated
and indeed enhance rational choice models
• contrary to conventional wisdom, compatible with either
analytic or computational solutions
• is both more general (in assumptions) and broader (in
types of application) than conventional agent-based
modeling
Positive, Formal Rational Choice as One Type of
Agent-Based Modeling?
• both approaches take, often incompatible, multiple
forms, however . . .
• like computational agent-based modeling, an individuallevel approach that seeks to predict system-level
outcomes through complex processes of aggregation
(emergence as contested term)
• common roots in axiomatic models of behavior (Von
Neumann/Morgenstern)
• game theory a formalization common to both
approaches, though tends to be used differently
Qualitative Rational Choice as a Bridge to the
Social Sciences and Humanities
• Rational choice is used qualitatively as well as
quantitatively
– comparative historical rational choice
– institutional rational choice
– “folk psychology”
• Is used positive, normative, and interpretively
– prescriptive models of justice and ethics
– rational interpretation of personal narratives
Basic Assumptions of Rational Choice
• Though based on common set of axioms, there is no one
“single” rational choice model. “Thin” version includes:
– logically consistent beliefs that do not violate laws of probability
– “well-behaved” utility – strict order, completeness, asymmetry,
and transitivity
– actors choosing in order to maximize utility given beliefs
• “Thick” version adds:
– self-regarding, materialistic (money, power, health), isomorphic
utility
– information-based (observation and valid inference) beliefs,
common knowledge of rationality
Introducing Culture into “Thin” Rational
Choice
Conventional Rationality-based approach:
single model generalizable to multiple, even novel contexts
theories can be cumulated into larger whole
tends to produce falsifiable predictions (though often anomalous)
Conventional Culture-based approach:
sensitive to social differences and personal development
deeper and more nuanced depiction of social process
avoids predictive anomalies (because it avoids prediction)
Main Steps and Hurdles to Integration
•
specifying dimensions of culture in general fashion
•
retaining simplicity and analytical tractability
•
formalizing in way that is compatible with choicetheoretic models of action across full-range of
environments
•
modeling cultural change algorithmically
•
combining generality and predictive determinacy
Conventional cultural typologies, e.g. (modern vs. traditional,
Hofstede and “comparative capitalisms”) tend to focus on first two
points but do not provide general implication for behavior.
Objectives
• Development of a general, predictive
model of cultural change
• Integration with choice-theoretical model
of action
• Software implementation into decisionsupport and simulation environments
Approach
• Representation of culture through gridgroup framework
• Modeling of cultural change through
coherence model
• Implementation in simulation and decisionsupport systems
Grid-Group Framework for Representation
of Culture
ABSTRACT DIMENSIONS
Grid =
extent to which social rules prescribe and restrict action
Group =
extent to which identity is directed towards others
Widely used in Cult./Soc. Anthropology and Political Science: Douglas
1970, 1978; Douglas and Wildavsky 1982; Wildavsky et al. 1990.
Adapted for choice-theoretic models in Chai and Wildavsky 1993;
Chai and Swedlow 1998.
Attributes of Grid-Group Framework
• More abstract than competing frameworks for representing
cultural differences
• Operationalization methods straightforward and well-tested
• Works well as front-end to “thin” rational choice models of
decision-making
• Fits with abstract dimensions of social organization found in
social theories, e.g. regulation and integration
• Decomposes into four major cultural types
• individualist – low grid / low group
• fatalist – high grid / low group
• hierarchical – high grid / high group
• egalitarian – low grid / low group
Grid-Group Transformations within Defined
Group Boundaries
Groupness-transformed payoff:
yi = (j<>i gi xj ) + xi
Gridness-transformed payoff:
ui = yi (ord(ai = oi) + (1 –hi) ord(ai <> oi))
where gi and hi are group and grid coefficients for individual
i, ai is her action, xj is untransformed payoff, and oi her
specified operation under standard procedures.
Concepts and Assumptions of
Coherence Model
Expected Regret: subjective probability-weighted difference between maximal utility
possible in a particular state of the environment and the utility provided by a chosen
set of actions
Coherence: expected regret of zero
PREFERENCE AND BELIEF ASSUMPTIONS OF MODEL
• Meta-optimization
• Environment constrains Beliefs
• No “Yogic Utility”
Parametric form, but not parametric values, determined by exposure to social
communication
Forms considered in order of message prevalence of communications describing
such forms, but parameter weightings can be accepted or rejected.
c.f. Chai 2001.
Intuitions behind Model
actors are engaged in a collective process of constructing
their own identities
this process is aimed at creating an individual and
collective sense of self that is both positive and
consistent
preferences and beliefs are not mere precursors to action,
but there is a mutually causative relationship between these
entities
Coherence (preference-based):
adjustment of g, h to minimize d
Expected Regret (single-period, individual form):
d = s (u(s,a*(s)) – u(s,a))) p(s) ds
where
a*(s)=argmax aA u(s,a)
a=argmax aA, sS s u(s,a) p(s) ds
s states of the environment, a actions, u utility
function, and p subjective probabilities
Some non-intuitive implications of
coherence model. . .
Means will become ends (functional autonomy of motives) iff
there exists there exists perception of some state of environment
where alternative actions superior
Sour grapes / forbidden fruit effect caused by actions that are
perceived to preserve / alter the status quo more than alternatives
Wishful / unwishful thinking strongest when an individual adopts
actions that are subject to more / less variation in comparison to
alternatives
Effects depend on and magnify in proportion to subjective probability
and extent to which chosen action will be suboptimal
Some implications linking structure to culture
Mutual altruism will be generated in groups engaging in repeated
collective action, particularly where public goods are generated more
reliably than private goods
Materialistic culture will be generated by clearly defined structures of
mobility in which the relative returns to vocational choices is not
circumstance-dependent
Explicit ideologies will be adopted by groups whose members face
incoherence with regards to a similar set of action choices.
“Cultural” implications built into conventional
economic models
Risk seeking or aversion: x = f(w); f’’(w) > 0
f’’(w) < 0 implies risk aversion;
f”(w) > 0 implies risk seeking
“Subjective” material payoffs are not a linear function of the quantity of
goods
Time discounting: U = ∑Tδt u
Cumulative utility is a function of period-specific utilities multiplied by a
“discount factor” representing devaluation of deferred utility