Economic models of impulsivity

1
Economic models of impulsivity

• Don Ross
• U of Alabama at Birmingham / U of Cape Town

2
Three discounting functions
3
What do these three curves describe?

• Answer functions that represent the effects of
  temporal delay and uncertainty on valuation of
  (all-at-once) consumption of a reward at a
  particular time and following a particular
  interval or sequence of intervals after the
  moment of initial valuation.

4
Indeterminacies

• (1) Delay and uncertainty may exert different,
  and indeed conflicting, influences, even while
  both are best represented by similar functional
  forms. Furthermore, delay may imply uncertainty
  and (given instantaneous consumption) uncertainty
  imply delay.
• (2) We know nothing about how people and/or their
  brains distinguish between instantaneous and
  spread-out consumption.
• (3) There is some effect of interval variance on
  valuation, but it isnt the whole story (cf.
  different computational models of reward
  learning).
• (4) Parameters of the functions vary with
  different rewards. So what happens when different
  rewards are bundled?

5
Why do we entertain such objects?

• Answer because of the reason that impulsivity
  interests us in the first place.
• Illustrating this reason in the specific case of
  addiction We regard addiction as a problem
  because addicts (and their families and
  colleagues) say its one. The one they say it is
  is they consistently want to stop consuming or
  to consume less (of their target of addiction)
  but dont. This strikes psychologists as a
  problem because addicts are distressed. It
  strikes economists as a problem because addicts
  are inconsistent (so not agents).

6
Economic modeling alternatives

• Synchronic multiple selves with (i) conflicting
  utility functions (? Nash bargaining games) (ii)
  different time preferences (? dynamic
  optimization of aggregate utility relative to a
  welfare criterion)
• Diachronic multiple selves with (i) conflicting
  utility functions (? bargaining games solved by
  subgame-perfect or sequential equilibrium in
  finite-horizon cases ? in infinite-horizon
  cases) (ii) limited knowledge of one another (?
  signaling games)
• Unitary agents subject to exogenous shocks
  (temptations) (? standard consumer theory)

7
Apples and oranges

• Synchronic multiple-selves models (Class 1)
  typically read the functional forms as accounts
  of different brain regions (molecularism).
• Diachronic multiple-selves models (Class 2)
  interpret the functional forms as molar
  descriptions (typically of whole people, but this
  isnt necessary).
• Unitary agent models (Class 3) keep economics
  firmly separate from psychology / biology agents
  are abstract, but humans-as-agents are
  constrained by properties of specific bodies.

8
State of the race class 1 models

• Seem easiest to test because of their
  molecularism.
• If interpreted as implying different brain
  regions that directly compute different time
  preferences, are attracting confuting data
  (Glimcher Kable).
• Policy implications are unattractively
  paternalistic.
• If interpreted as molar-scale accounts of
  different neural agents (), they seem very
  helpful - e.g., opponent neurotransmitter system
  models of addiction.
• First full model of the type for the dopamine
  system recently developed by Caplin Dean
  (applause).

9
State of the race class 3 models

• Given what economists mainly do with individual
  agency, theyre going to prefer these come what
  may. Welfare implications are very attractive
  (based on peoples behavior). But this implies
  that the economist stays out of psychology. Where
  she doesnt - e.g., Loewenstein, Camerer - then
  there seems to be an arbitrary distinction drawn
  between agentic and non-agentic parts of the
  brain. This is too strongly cognitivist for those
  of us with behaviorist scruples.

10
State of the race class 2 models

• Heavy trade-off between flexibility richness of
  possible interpretations (?) and lack of clarity
  on testable implications (?).
• Do the functions represent discounting? Or the
  passage of time as a proxy for accumulating
  informational uncertainty?

11
A preference keep economics and psychology apart

• Organisms can be modeled as agents so can brain
  regions. Whole-organism agents arent composites
  of neural agents.
• Calling the functions of interest discounting
  functions invites confusion. Brain regions
  probably do compute discount rates at molecular
  scales, but these mechanisms lie under the hood
  for Class 2 models. Class 2 models are
  molar-scale profiles of the responses of learning
  systems to differences in reward rates across
  environments.

12
Implication for addiction research

• Hypothesis addiction results from continuous
  uncertainty about (learned) reward rates on
  micro-scales.
• Economists agents are approximations, units
  constructed as discrete and stationary
  representations of learning processes that are
  continuous. Agents model systems as if their
  learning were finished.
• Thus the tension between economic and
  psychological modeling of impulsivity isnt a
  pathology but also cant be resolved. Its a
  modeling duality and scientists must learn to
  flexibly switch gestalts.