Nonmarket Valuation: the Challenge of Dealing with Preferences

1
Nonmarket Valuation the Challenge of Dealing
with Preferences

• Michael Hanemann
• University of California, Berkeley

2
TOPICS

• The complexity of modeling preferences for
  disaggregated commodities dealing with corner
  solutions
• The challenges remaining the context dependence
  of preference and choice

3
Preferences vs Behavior

• Starting around 1938, economics became firmly set
  in a behaviorist mold. Only behavior was the
  legitimate object of scientific study.
  Preferences were (i) of no inherent interest, and
  (ii) not susceptible of meaningful scientific
  study.
• In practice, this mean that one focused on the
  estimation of demand functions where
  uu(x1,..,xN) and xi hi(p1,..,pN,y),i 1,..,N.

4
Now

• This is no longer adequate as a complete
  description of what micro-economists do.
• We spend a significant amount of time thinking
  about preferences. Why?
• To conduct welfare valuation
• To provide a more meaningful account of demand
  behavior
• The modeling of demand for water-based outdoor
  recreation is an illustration of the latter.

5
Evolution of Travel Cost Method
6
(No Transcript)
7

• The next step was to incorporate quality site
  attributes explicitly into a demand function.
  This led to single site demand functions with
  quality
• x f(p,q,y)
• and then multiple site demand functions with
  quality
• xi fi(p1,..,pN, q1,..,qN,y) I 1,..,N.
• The question then arose
• How can one do welfare evaluations for a change
  in q how does consumers surplus apply?
• This was elegantly answered by Maler (1971, 1974)
  using the formulation u(x,q)

8
Attributes The Lancaster-Maler model

• The utility function is uu(x,q) where
• x is the quantities of individual commodities
  consumed,
• q is attributes/characteristics of these
  commodities
• While the individual selects the level of a,
  subject to a budget constraint, he takes q as
  given.
• More generally, q is anything that affects a
  persons utility but that he does not control
  fixed attributes of the commodities, supplies of
  public goods, fixed attributes of the individual.
• This model has become the workhorse of modern
  theory of environmental economics, but it also
  has an additional significance for consumer
  theory.

9
Disaggregated choice analysis

• The larger significance is that it was part of a
  general reshaping of what economists think of as
  a commodity
• Food, housing, shelter, versus an 8 ounce breast
  of organic, free-range chicken.
• That is, aggregate commodities versus commodities
  disaggregated
• Across individual, highly specific commodities
• Across individual consumers
• Across individual choice occasions.

10

• Such disaggregated commodities have become the
  focus of demand analysis in many areas of
  microeconomics, as well as in market research
  (e.g., scanner data), but this focus started
  first in economics with environmental economics.
• A major but inconvenient feature of disaggregate
  commodity data is the high frequency of zeroes
• There is a myriad of disaggregated commodities,
  but any individual consumers purchases a tiny
  handful of what is available.

11
The problem of zero consumption

• There are too many zeroes to be ignored.
• They are real (structural) zeroes, not just an
  artifact of data measurement error.
• They reflect a corner solution to the utility
  maximization problem.

12

• Interior solution
• Some of every good is consumed. The consumers
  optimum yields xi gt 0, all i.
• Corner solution
• Some goods are not consumed at the optimum
• Extreme Corner Only one at most of the discrete
  alternatives is consumed.
• What generates this the discrete alternatives
  are perfect substitutes or are mutually
  exclusive.
• General corner solution Some goods are not
  consumed, but more than one of the discrete
  alternatives is consumed.

13
Types of model

• Extreme corner solution
• Purely discrete choices (McFadden, 1974)
• Mixed discrete/continuous choices (Hanemann,
  1984)
• General corner solution
• Mixed discrete/continuous choices (Wales and
  Woodland, 1983 Hanemann, 1978 Lee and Pitt,
  1986 Phaneuf et al., 2000 vo Haefen et al 2004).

14
Pure discrete choice

• N discrete alternatives utility of ith given by
  ui vi(pi,q,y) ei
• The ith item is chosen iff ui uj all j.
• This leads to a N-1 dimensional probability
  integral which may or may not have a closed-form
  expression (e.g. extreme value does, while normal
  does not)

15
Ad hoc approach to combining discrete and
continuous choice
16
Utility-theoretic approach to discrete/continuous
choice model

• Given u(x,q,z), where x are disaggregated
  commodities with attributes q, while z is a
  Hicksian composite commodity standing for
  everything else, there are 2N possible
  solutions to the consumers problem, including an
  interior solution and 2N-1 corner solutions.
• There are two ways to proceed

17
Direct evaluation of all possible corner solutions

• Consider each of the possible corner solutions in
  turn and derive the indirect utility function
  that corresponds to it i.e. find the utility
  maximizing bundle of (x,z) subject to the
  constrain that the particular subset of
  commodities is non-zero.
• The probability that this particular solution is
  selected is Pr ui uj j 1, 2N-1.
• This leads to a 2N-1 dimensional probability
  integral which is unlikely to have a closed-form
  expression.

18
Approach based on Kuhn Tucker conditions
19
(No Transcript)
20

• The Kuhn-Tucker approach, introduced
  independently by Hanemann 1978 and Wales and
  Woodland 1983, has the advantage that there are
  is an (N-Q) dimensional probability integral,
  where Q is the number commodities consumed i.e.,
  one probability integral for each commodity not
  consumed. This is considerably less than 2N-1 .
• It still is complex. The trick is to come up with
  a specification of u(x,q,z,e) and a stochastic
  specification for the joint distribution of the
  es that is reasonably tractable.

21

• This proved tractable for maximum likelihood
  estimation when N 3 (Wales and Woodland) or N 4
  or 5. Prior to about 2002, the maximum had been N
  12.

22
Additional considerations

• So far we have focused on tractability of the
  likelihood for estimation of demands.
• In addition, once the underlying utility model
  has been estimated, one may want to use it
• To predict demands when (p,q) change
• To conduct welfare evaluation for changes in
  (p,q).
• These turn out to involve an evaluation of the
  full set of 2N-1 potential corner solutions. This
  became the binding constraint when N was large.
  

23
Welfare evaluation brute force evaluation
24
Welfare evaluation additively separable utility
25
Dealing with a large N

• Von Haefen applied his approach, based on
  additively separable utility, to N 55.
• Other researchers have worked with other
  additively separable utility functions.

26
The problem

• Being limited to an additively separable utility
  function in (x,z) is a serious constraint that
  significantly restricts the generality of the
  approach.
• Now Felipe Vasquez in his just completed
  dissertation at Berkeley has found a way to relax
  this constraint and solve large N problems with
  additively non-separable utility functions.

27
Vasquezs contribution
28
Application
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
Welfare measures
33
Implication

• The number of commodities is no longer a barrier
  to allowing for non-separable utility.
• It makes some difference whether or not one
  assumes additively separable utility.
• Substitution among sites, permitted in a
  non-additively separable utility model, reduces
  welfare loss from reduction in sportfishing
  opportunities.

34
Challenges

• Many other issues remain in the analysis of
  disaggregate choice behavior
• The set of salient attributes is inherently
  subjective
• There are likely to be non-attribute arguments of
  a utility function
• These are a powerful source of
• Heterogeneity in preferences
• Context dependence
• The bottom line we need to come to grips with
  the complexity of preferences!

35
Lancaster-Maler model introduces a subjective
element into preferences

• The attribute set which are the attributes that
  the consumer considers salient?
• What matters is the consumers subjective
  perception of the attributes he finds salient.
• There is also the possibility of non-attribute
  arguments of the utility function

36
Non-attribute arguments

• If a consumer cares about what she consumes, why
  might she also not care about how she consumes,
  including the process and circumstances.
• Not taking the last piece of cake, fitting in
  with the group, not wanting to be disrespectful
  of ones elders, doing the right thing, wanting
  to appear thoughtful or prudent, not wanting to
  overpay, not letting the other fellow get away
  with cutting in front of you in line these are
  all attributes that a person could take into
  consideration when making a choice to eat a piece
  of cake.
• If so, they become arguments of the utility
  function in an expanded Lancaster-Maler model.

37
Some implications

• The qs represent a potential source of
  inter-individual heterogeneity
• Also, a source of preference dynamics
• Rikers concept of heresthetics argument over
  which characteristics/issues are relevant to the
  issue at hand
• A hypothesis the coefficient on a given q is
  stable (sign does not change) but switching qs
  on/off depending on context generates behavior
  variation.
• Preference is a deeper structure, of which tastes
  are a particular manifestation

38
Context dependence

• Context dependence is a basic feature of human
  cognition. Perception, like so many other
  psychological processes is quintessentially
  contextual. Context can affect processes at
  every stage in early sensory transduction, in
  later perceptual encoding, in possible cognitive
  recoding, and in decision/response (Marks and
  Algom, 1998)

39

• Swait, et al. Context Dependence and Aggregation
  in Disaggregate Choice Analysis Marketing
  Letters, 2002.
• Context is not necessarily noise that washes out
  in the aggregate it can have a systematic impact
  on choice behavior.
• Present a conceptual framework and mathematical
  model to represent the sources of
  context-dependence in individual choice behavior.
• Argue for a shift in emphasis to mapping and
  measuring context dependence rather than assuming
  it away.

40
The nature of preference

• Underlying much of the discussion of preference
  anomalies is an outdated conception of mind. This
  is the stored-rule or filing cabinet concept
  of mind, which goes back to Hobbes, who conceived
  of cognition in terms of storing and retrieving
  slightly faded copies of sensory experiences.
• We now know from neuroscience that all cognition
  is a constructive process. People construct their
  memories, their attitudes, and their judgments.
  The real issue is not whether preferences are a
  construct, but whether they are a stable
  construct.

41

• Furthermore, stability has to be assessed in a
  probabilistic, not a deterministic, framework
  precisely because of the influence of affect and
  context on cognition.
• We know from the psychology literature that
  personality changes are rare, especially after
  the age of 30 (McCrae and Costa, 1990).
• The same may be somewhat true of preferences
  viewed as a deep structure.

42

• Economics is at a point where, having adopted a
  narrow and rigid concept of preferences, it is
  being forced to conclude that, as conceptualized
  thus, preferences do not exist.
• A preferable alternative is to redefine the
  concept based on the richer understanding we now
  have from neuroscience, behavioral economics,
  experimental economics, and stated-preference.

43

• Instead of writing preferences off, we need to
  take them seriously and come to grips with them.
• For this we need a new toolkit. Some of the tools
  will come from existing models in economics RUM
  models, models of preference change, the
  Lancaster-Maler model which have been
  underutilized. Needless to say, some of the tools
  will come from new advances based in part on
  approaches elucidated at this conference.