Valuation 7: Contingent Choice Modelling Contingent choice modelling and its variants Steps and design stages for choice modelling Some econometrics Application.

Valuation 7:Contingent Choice

Modelling• Contingent choice modelling and

its variants• Steps and design stages for choice

modelling • Some econometrics• Application to green product

choice

Last two weeks

The contingent valuation method• History and welfare measures• Study design• Validity, reliability and biases• Application to non-use values including

the Exxon Valdez oil spill• Embedding and warm-glow• The EU environmental liability directive

Contingent Choice• (Contingent) choice modelling is similar to

contingent valuation in that it is a stated preference technique based on surveys

• The main difference is that instead to asking „how much are you willing to pay“, the question is „which situation would you prefer“

• The approach is based around the idea that any good can be described in terms of its attributes, or characteristics, and the levels that these take

• It originates in the market research and transport literature and has only recently been applied to areas such as the environment

Contingent Choice (2)

• Delivers answers to questions such as– Which attributes are significant determinants of

the value of non-market goods?– How are attributes ranked?– What is the value of changing more than one

attribute?

• Choice modelling comprises choice experiments, contingent ranking, contingent rating and paired comparisons

• Similar to conjoint analysis, apart from the interpretation of the results

Choice ExperimentsThe “good” to value is wildlife habitat on farms, defined as habitat areas and cost to the taxpayer.

Which of the following two schemes do you favour? Each would have a cost to your household. Alternatively you might neither scheme: taxes would not rise, but no areas would be protected.

Choice A Choice BNative woodland 500 ha 700 haHeather moorland 1200 ha 0 haLowland hay meadow 200 ha 300 haAdditional tax $25 $15

I would prefer: Choice A… Choice B… Neither…

Contingent RankingRank the alternative policy options below according to your preferences, assigning 1 to the most preferred.

Choice A Choice B Choice CNative woodland 500 ha100 ha 700 haHeather moorland 1200 ha 600 ha 0 haHay meadow 200 ha 0 ha 300 haAdditional tax $25 $5 $15

Your ranking: 1… 2… 3…

– Note that this does not correspond to typical market behaviour!

Contingent Rating

On the scale below, please show how strongly you would prefer the following policy option?

Characteristics Option 1Native woodland 500 haHeather moorland 1200 haLowland hay meadow 200 haAdditional tax $25

1 2 3 4 5 6 7 8 9 10Very low preference Very high preference

– Again, not common market behaviour!

Paired ComparisonsWhich of the two policy options described below would you be most in favour of? Indicate your preferences using the scale below

Choice A Choice BNative woodland 500 ha 700 haHeather moorland 1200 ha 0 haLowland hay meadow 200 ha 300 haAddition tax $25 $15

1 2 3 4 5 6 7 8 9 10Strongly prefer A Strongly prefer B

Choice Modelling• Provided that „do nothing“ is included,

choice experiments and contingent ranking can be used to estimate WTP or WTAC

• If „do nothing“ is not included, the set of options may be infeasible for the interviewee, resulting in nonsensical results

• Contingent rating does not yield WTP, as there is only one alternative

• Pairwise comparison is like a CVM referendum, but with shades of grey that are difficult to interpret

Implementation• Characterisation of the decision problem• Select attributes and levels• Experimental design• Choice sets• Develop questionnaire • Collect data• Estimate model• Apply model

Decision problem• The initial step is the identification of the

economic and environmental problem• What is the geographic and temporal scope of the

change?– Impact on single or multiple site– Possible spill-overs between changes– Instantaneously implemented

• What type of values are associated with the change?– Who will benefit– Will passive use values be affected– If use values are affected, what is the behaviour that

captures this value

Attributes and levels• Identifying the relevant attributes

– Have to be part of people’s preferences for the environmental change

– Attributes can be impacted by policy/project/management option choice

– A monetary cost should be one of the attributes, to allow the estimation of WTP

• Attribute levels should be realistic– They should span the range over which we

expect respondents to have preferences– They should be achievable

Experimental design• Complete factorial design

– Allows estimation of the full effects of the attributes– Often produces an impracticably large number of

combinations to be evaluated– An experiment with 4 attributes each at 3 levels

would lead to 3x3x3x3=81 (La) alternatives

• Reduce the number of scenario combinations– Identify subsets of the possible combinations of

attributes and levels that will “best” identify the attribute preferences

– Orthogonality is a desirable property– Include interaction effects?

Fractional Factorial Design

Attribute levels for a hypothetical wetland management scheme

Possible design suitable for four attributes each at three levels

Choice sets• The more familiar the subject the higher the

number of choice tasks a respondent can be asked to perform

• The fewer the number of attributes and levels, the higher the number of choice tasks that can be allotted

• If there are too many attributes and/or levels– Reduce their number if possible– Group the attributes into subsets– If the design needs to be divided in separate blocks

more interviews are needed

CM v CV• Generalisation of the binary discrete choice CV

study • Choice modelling allows for more nuanced

distinctions – but this also implies that more situations need to be assessed, that the questionnaire gets longer, and the interviewee may tire

• In choice modelling, money is less central (less protest votes), and preferences more

• In choice modelling, the econometrics is considerably more complicated

Econometrics• Ordinary least squares assumes that variables

are continuous• In choice modelling, the observations are

discrete, often 0-1• How to interpret such data?• Techniques are known as random utility,

discrete choice, logit ...• Random utility is the most common name in

economics• It start by saying that one would choose option

0 if its utility is higher than option 1

Econometrics (2)• One would choose option 0 if its utility is higher than

option 1: U0 > U1

• Utility consist of a determinisitc and a stochastic component U = Xb+v

• The vector X describes the attributes that influence utility• The b reflects the impact of changes in X• The v is the random component of utility• Prob(U0 > U1) = Prob(X0b0+ v0 > X1b1 +v1)

= Prob(X0b0- X1b1> v1 –v0)• The model predicts the probability that option 0 is chosen• Above, we do logit, but we could also use other

assumption on the distribution of v and other distance metrics

Example: Green product choice

• Product: Toilet paper; advantages: widely used; many varieties, some green, some not; close to reality

• Canberra, 1 supermarket, year unknown• Three focus groups as a preparation; to find out

important attributes, and to test questionnaire• 1100 questionnaires, 2 week period, 40%

response rate• Questionnaire made clear that this was not for

profit

Attributes• Price • Special on price• Number of rolls in pack (2, 4, 6, 8)• Number of ply and number of sheets as an

indicator for strength and overall quality• Colour/pattern on paper• Type of paper (standard, unbleached but not

recycled, both recycled and unbleached)• Brand; this is not really an attribute, but a

bundle of attributes; has to be included because of habit

Choices• With 6 attributes, the number of combinations

is quite astounding• An orthogonal fractional factorial design with

128 choice sets were created• Each interviewee was asked to choose

between 8 types, so that the sample was split in 16 subsamples

• Recall that only 440 surveys were returned• Sampling strategy was stratified for time of

day only, based on expert guesses

Results• The model predicts the probability of buying a

product, given its characteristics, and the characteristics of the buyer (green, dislike scent, cheap, decor, clean, age)

• Positive: Special, no. rolls, no. plys, white, coloured, unbleached and recycled

• Negative: Price, off-white, standard, unbleached and not recycled, scented

• Interactive effect between stated and actual greenness increases the unobserved utility

Prices• The model contains both price and

environmental attributes• That implies that one can compute the

increase in price that would correspond to a greener product

• The average respondent is willing to pay $0.66 extra to pay for recycled and unbleached toilet paper (this is for the average pack, other factors held constant)

• For greens, its $1.69, $0.39 for others

Yea-saying• Yea-saying is one of the biases in stated

preference methods; it means that the interviewee gives the answer that she thinks the interviewer wants to hear

• Harmony is preferred over conflict, and if harmony can be had by a little lie ...

• Consumer shopping survey (Green products)• By completing this survey, you will assist our

understanding of what people do when they go shopping (and how they go about choosing among products that claim to have different implications for the environment)

Prices (2)• First: The average respondent is willing to pay

$0.66 (0.33) extra to pay for recycled and unbleached toilet paper; For greens, its $1.69, $0.39 for others

• Second: The average respondent is willing to pay $0.61 extra to pay for recycled and unbleached toilet paper; for greens, its $1.57, $0.33 for others

• These differences are not-significant, but that may be because of the data

• The preference for green paper increased, as did the number of self-proclaimed greens; but so did price-sensitivity

Revealed preferences• Another nice thing about toilet paper is that it

is actually sold in all the varieties tested in the survey

• How do the results compare to actual sales data?

• Only aggregate sales are available, so one has to redo the stated preference model for the average consumer

• A formal test of differences between the models is tricky, because the buyers face different choices than do the interviewees, but the differences seem to be significant

Revealed vs stated preferences

• Stated preferences deviate from revealed ones, even for standard products like toilet paper, and even for such innocent things as its colour

• It appears that the choice model has generated significantly different taste parameters to those implicit in actual behaviour