Microsoft PowerPoint - consumer demand

Document Sample
Microsoft PowerPoint - consumer demand Powered By Docstoc
					     Modeling Consumer Choice


                         Vic Adamowicz
                         Rural Economy
                         University of Alberta


    Several Slides are based on the Course Material for
           “Modeling Human Choice behaviour”
       Joffre Swait, Vic Adamowicz and Ric Johnson
1
    Overview

     Remember when….consumer theory was
     simple
     The movement to disaggregate analysis
     Emergence of behavioural economics
     Development of econometric tools
     Can we link behavioural models and
     econometric models?
     Do we have the data?
     Is “food” different ?
2
    Consumer Theory
        The Traditional Model
    –     Individuals are perfect information processors
    –     A fixed, known choice set (brands, options)
    –     No time dimension (no habits, influences of the past
          or future, etc.)
    –     Full information about the goods (prices, attributes,
          etc.)
    –     All individuals have identical preferences
    –     The individuals trade off attributes and prices of the
          goods (compensatory decision strategy)
    –     They have no reference points and are not affected
          by framing
    –     The individuals’ choices are not influenced by
          anyone else



3
    Decomposing the Traditional Model:
    Individuals are perfect information
    processors
         Counter-evidence
          –   Simon, 1955, Q. J. of Econ.
          –   Heiner, 1983, Amer. Econ. Rev.
          –   Shugan, 1980, J. Cons. Res.
          –   DePalma et al, 1994, Amer. Econ. Rev.
          –   Swait et al, 2002, Mark. Letters
          –   Gigerenzer, 1999




4
    Missing Information
    When information is missing, people may
    –   Ignore it (depends on importance)
    –   Seek out the information (depends on cost)
    –   Infer the value of the missing attribute
           e.g. Johnson and Levin, 1985, J Cons. Rsch.


    Key point is that people use a superset, not a
    subset, of available information (see Simon),
    leading to a specification error




5
    Decomposing the Traditional Model:
    Individuals trade off attributes and
    prices of alternatives
         The assumption is that people use a
         compensatory decision strategy.
         Evidence that people do not use compensatory
         strategies in all cases:
          –   Payne, Bettman and Johnson, 1993, The Adaptive
              Decision Maker
          –   Elrod, Johnson and White, 2004
          –   Swait and Adamowicz, 2000, J. Cons. Res.
          –   Swait et al 2002, Mark. Letters




6
    Decomposing the Traditional Model:
    Individuals’ choices are not
    influenced by anyone else

        –   Manski, 2000, J. Econ. Perspectives
        –   Brock and Durlauf, Rev. Econ. Studies, 2001
        –   Becker and Murphy, 2000. Social Economics
        –   Browning and Chiaporri, 1998, Econometrica
        –   Fishbein and Ajzen, Psych Bull. 1977
        –   Argo, J. Cons. Rsch (Forthcoming)




8
    Social Interactions
    (Manski, 2000; Brock 2004)

     Agents with preferences, expectations and
     constraints make decisions/choices
     Interaction through constraints
      –   Congestion affecting time constraints (negative)
      –   R&D decision affected by others (positive)
     Interaction via expectations and information
      –   Observing actions and/or outcomes of others
      –   Information cascades
      –   Penalties for deviation from “norm”
     Preference interactions
      –   Preferences / utility depend on other’s actions
      –   Altruism, noncooperative game theory, etc.

9
     Emerging “Neuroeconomics” issues

      Choices are a complex combination of
      “processing” part of brain and “instinctive
      response” part of brain
      Provides explanations for “disgust”, time
      preference results, ultimatum game, heuristics,
      etc.
      Suggests 4 types of “utility” (remembering,
      anticipating, choice (RP) and experiencing –
      Kahneman, 1994)
      Comparable to Kahneman and Frederick “dual
      process theory”
       –    i) “System 1,” is quick, associative, and intuitive;
       –   ii) “System 2,” is slower, based on rules and reasoning
           (Kahneman and Frederick 2002)
10
     Source: Camerer, C. et al, 2004. Scan. J. Econ. P. 559
11
     “Context” Effects

      habit or experience       mental accounting,
      dependence effects,       choice bracketing,
      social                    motivation effects,
      interdependence,          decoy effects,
      accountability effects,   compromise effects,
      menu-dependence,          reference points
      chooser-dependence,       –   prospect theory
      framing                   complexity effects,
                                missing information


12
     Many new theories of behaviour

      Do these apply to food, nutrition, etc.?
      –   Repetitive choice
      –   Low cost (generally)
      –   Mistakes easily corrected (?)
      –   Many substitutes
      –   Large choice sets
      –   Heterogeneity in tastes, perceptions, etc.
      How do we (or can we) develop
      empirical models of demand for
      these approaches to choice?
14
     Basic Choice Model
                                                   Price       Attributes

                                                               Demographics
     i alternatives
     n individual     U in = V ( M − Pin , ain ; sn ) + ε in         Error

            Pr(i ) = Pr( Vin + ε in ≥ V jn + ε jn ) ∀ i, j ∈Cn
     OR
              Pr(i ) = Pr( U in = max U jn ) ∀ i, j ∈Cn

                              exp[(µ )(Vi )]         Assuming Type I error
          π n (i) =       J
                                                     Distribution:
                                                     Conditional Logit
                        ∑ exp[(µ )(V )]
                         j =1
                                               j
                                                     Model
16
      The “Scale” or “Variance” Factor
                                            1




                              Decreasing
                                           0.8
                               Variance


                                           0.6




                                           0.4




                                           0.2




                                            0
                 -3      -2          -1          0   1    2        3



               Effect of Variance on Choice Probability


   See Brock 2004 for a dynamic decision theory interpretation
18 See Anderson et al for an error-affected game theoretic interpretation
     Behavioural theories can be
     represented by:

      Clever modification of the utility specification
       –   Reference points
       –   Noncompensatory behaviour
       –   Strategy selection
       –   Interdependence
       –   Intertemporal choice
       –   Information and updating
      Choice set formation models
       –   Strategy selection
       –   Heuristics
      Modeling variances
      Heterogeneity
      More and better data?
19
      Prospect Theory
      Individuals’ preference
      depend on a reference          v(X)
      point
      –   Risk averse for gains,
          risk loving for losses
      –   Losses loom larger than
          gains; hence “loss
          aversion” or “status quo
                                            X
          bias”




     VP= ∑π ( p) v( X )
20
     Modeling Reference Points

      Hu et al (2004)
      –   Elicit “reference points”
      –   Model choice as a function of reference
          points, attributes and prices
      –   Reference point effects are stochastic /
          heterogeneous and are affected by
          demographic factors
            Higher income – less averse to losses
            Older – less averse to losses (“experience effect”?)
      –   Reference point effects appear to be “larger”
          than variability (scale) effects for this
          empirical case.
21
     Noncompensatory behaviour
      Significant evidence that people employ cutoff-based
      heuristics to simplify decision making.
      –   “I can’t see paying more than $X for this.”
      –   Klein and Bither (1987), Huber and Klein (1991)
      –   Models w/ Cutoffs: ACA, EBA
      Swait (2001b)
      –   An extension of the full info availability, full info processing,
          utility maximizing preference model that further allows for
      –   person-specific attribute upper & lower cutoffs.
      –   Can be extended to other kinds of constraints – see Swait
          (2001b).
      Gilbride and Allenby (2004)
      –   Screening alts. via attribute thresholds
      –   Conjunctive, disjunctive and compensatory evaluation rules
      Elrod, Johnson and White (2004)
      –   Mathematical forms used to capture non-compensatory
          behaviour
             General non-rectangular hyperbola (GNH)
22
     Strategies (Heuristics, Complexity)

      Swait and Adamowicz (2002)
       –   Cognitive burden constraints
       –   Changes in processing strategy when
           complexity or cognitive burden increases
       –   Ordered latent class model
      DeShazo
       –   Error variance a function of complexity




24
     Interdependence

      Brock and Durlauf, Brock 2004
      –   Brock (2004) includes cost of deviation from
          “norm”
      Signorino and Yilmaz (interdependence
      via strategic voting behaviour)
      Boxall et al

      Dosman and Adamowicz
            V (p, x, δ ) = max δ u a (q a , q b , Q) + (1 − δ ) u b (q a , q b , Q)
                            q a ,q b , Q

25          s.t. p(q a + q b + Q) = x
     Intertemporal Choice Models

      Habits (low cognitive cost; norms)
      Example: Swait, Adamowicz and van Bueren
      (2004)
      1. Initial conditions,
      2. Temporal dependence on past attribute level via past
        utilities, and
      3. State dependence (via inclusion of past choices in the
        information set It),
      4. Time-varying scales and covariance structures, and
      5. Time-varying tastes (identification restrictions apply,
        however).
      Impact decay (long term response to change).
26
     Information

      Erdem and Keane (1996)
      –   Model utility from attributes that have “uncertain”
          levels (variances)




      –   Consumers update based on experience, advertising,
          etc.
      Cameron (2004): similar model
      –   Includes effects of ambiguity, Bayesian-ish updating,
          significant heterogeneity with different sources of
          information
27
     Health “Risk”

      From the health valuation literature (e.g.
      Cameron and DeShazo, 2004)
      –   Risk processing
      –   Morbidity (sources, context)
      –   Mortality (sources, context)
      –   Latency
      –   Discounting
      –   Current health status
      Generally increasing risk reduction values

28
     Choice set formation

                              Universal Set (M)

                              Evoked Set
                              Consideration Set
                               Choice Set
                            Choice set:
                            Subset of all alternatives
                            in universal set that are
                            available at time of
                            choice and have non-
                            zero probability of
                            being chosen.

29
     Choice Set Formation

      Swait (various)
      –   Independent availability model
      –   GENL
      Pap et al 2005
      –   Probabilistic consideration set formation
          model
            Probability of being in consideration set a function of
            demographics and state dependence




30
                                             IAL Prediction of Brand Inclusion &
                                             Choice Set Size (Swait, 1984; 2002)
     Figure 10.5                                    Predicted Probability of Inclusion in Some Choice Set
        Pr(Brand Being in Some Choice Set)




                                              1


                                             0.8


                                             0.6


                                             0.4


                                             0.2


                                              0
                                                   CK     Gap      Lee     Levis   Wrangler   None of       Figure 10.4                Predicted Choice Set Size Distribution
                                                                                              These

                                                                     Brand
                                                                                                                             0.4




                                                                                                            Pr(Set Size=k)
                                                                                                                             0.3


                                                                                                                             0.2


                                                                                                                             0.1


                                                                                                                              0

                                                                                                                                   1           2         3         4            5   6
                                                                                                                                                        Set Size (k)




32
     Heterogeneity

      Amazing advances in computational
      ability
      –   Preference heterogeneity, conditional on
          exogenous factors
      –   Wide variety of parametric forms
      –   Individual parameter estimation
      Pattern analysis – facilitates assessment
      of strategies
      Typically assume homoskedasticity.
      Misspecification concerns
33
     Do we have the data?

      Disaggregate choice data?
      –   What level of aggregation is appropriate for
          studying food / nutrition / information?
      –   Health data? Choice set information?
      –   Norms?, Neighborhoods?, Risk Perceptions?
      Experiments?
      Stated Preference Data?
      Fusion?
      Calibration?
34
     Policy Analysis and Consumer
     Behaviour

     “The existence of underlying preferences is a vital scientific
       question for economists. If the answer is affirmative, then
       the evidence from cognitive psychology implies only that
       economists must look through the smoke screen of rules
       to discern deeper preferences that are needed to value
       economic policies. This is a difficult task but not an
       impossible one. If the answer is negative, then economists
       need to seek a foundation for policy analysis that does not
       require that the concept of ‘the greatest good for the
       greatest number’ be meaningful. I am guardedly
       optimistic that the question has an affirmative answer.”
       (McFadden, D. 2000. p. 345-346).
38
     Conclusions
      Behavioural economics is helping us understand
      the richness, complexity and variability of
      disaggregate choice behaviour
      Econometric tools combined with creativity can,
      to a certain extent, model such behaviour
      –   Identification problems are horrific
      –   Misspecification can lead to dramatically different
          insights
      There is considerable work yet to be done
      –   Information acquisition, updating, risk perception
      –   Interdependence – Information linkages / Norms
      –   Choice set analysis
      –   Habits, variety seeking, intertemporal analysis
      Data may be our most limiting factor
      –   Fusion and calibration
40

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:2
posted:9/22/2012
language:English
pages:30