Title: Attribute Based Stated Preference Methods by T. Holmes and W. Adamowicz
1Attribute Based Stated Preference Methodsby T.
Holmes and W. Adamowicz
- Carlos Mayen
- Ross Pruitt
- Christiane Schroeter
2Stated Preference
- Historically economists have relied on
historical data of preference - revealed
preference - In many cases there is no historical data, we
need to rely on what consumers say they will do
stated preference - Examples where stated preference data is
necessary - Organizations need to estimate demand for new
products with new features - Explanatory variables have little variability in
the market place - Observational data is time consuming expensive
to collect - The product is not traded in the real market
- Examples of SP tool include contingent
valuation and attribute based SP methods (ABM)
3History of Attribute Based Stated Preference
Methods
- Two Branches
- Conjoint analysis
- Hedonic method (Court, 1939 Griliches, 1961)
- Consumer Demand theory (Lancaster, 1966)
- Discrete Choice Theory
- Random Utility Theory (Thurstone, 1927)
- Integration of Random Utility theory with hedonic
analysis Conditional Logit model (MacFadden,
1974) - Integration of Conjoint Analysis and Discrete
Choice Theory - Experimental Designs (Louviere and Woodsworth,
1983) - Applications
- Marketing, Environmental Economics,
Transportation
4General Idea
- A good or service is decomposed into its relevant
attributes which in turn can have different
levels price (low, high) - Consumers are presented with several combinations
of attribute levels - Consumers choose, rate or rank such product
profiles - Econometric model estimates the utility function
depending on product attributes - Model Outputs relative importance of
attributes, willingness-to-pay, market shares,
5Conducting SP Research
- Define the problem
- What are the objectives?
- Identify and describe choice context, attributes,
levels, etc. - Design the experimental apparatus
- Develop the questionnaire
- Collect data
- Estimate the model
- Analyze the results
6Attributes of a Melon Product
7Experimental Design for CE
- Factorial Designs
- Fractional Factorial Designs
- Randomized Designs
- Advantages
- In big experiments it is likely that the entire
design space will be sampled - by randomly generated profiles
- All interactions can be estimated
- Application
- Attribute levels assigned randomly to each
attribute and brand - 4 brands 4 product alternatives per task with
inclusion of - no purchase alternative
8Survey
- Presented to target population through survey
- convenience sample- Meijer
- Four choice tasks presented to each respondent
in addition to demographic info - Pictures meant to expedite the response time
- Hypothetical scenario set up through
introduction of fresh-cut products and attributes
9Conditional Logit Model
The utility that the ith person obtains from
choosing the jth alternative is considered a
linear function of product attributes and a
stochastic term The probability that the ith
respondent chooses the jth alternative from
choice set C is the probability that the utility
for the jth choice is greater than the utility
for all other k choices in the choice
set. Assuming errors are iid with an extreme
value distribution, the probability that the ith
respondent chooses alternative j is
10Coding
- Two types
- Effects coding chosen attribute is coded 1, not
chosen is coded 0, and the last attribute is
coded -1 - Dummy variable coding chosen attribute is coded
1, and not chosen is coded 0
11Coefficient Estimates
12Model Outputs
- Relative Importance of Attributes
- Marginal WTP for changes in attribute levels
- WTP (Coeff. 1 Coeff.2)/ Coeff. Price
- WTP (moderate juice no juice)
- Marginal Effects of attribute levels
- Market shares
13IIA Property
- Conditional logit model based on the following
assumptions - Everyone in the population has same preference
structure - Ratio of probabilities between any two
alternatives unaffected by - other alternatives in choice set
- Relaxing the assumption of common preferences
- Including interaction effects with demographic
information - Estimating a Latent class/finite mixture model
- Utilizing a random parameters/mixed logit model
- Relaxing the IIA assumption
- Nested logit model
- Mixed multinomial logit models
14Advantages of ABM
- Useful in behavioral analysis, market research
- Experiment control
- Allows to obtain individual information
- Subjects think about tradeoffs
- Thorough examination of preferences
- Valuation of the product and its attributes
- Potential for combination with revealed
preference data
15Disadvantages of ABM
- Hypothetical bias
- Cognitively demanding for subjects
- Information provision challenging
- Strategic behavior
- Discrete nature of data problematic
- Cross-sectional/ time-series
16Summary
- ABMs are used to estimate value of attributes in
goods or services - As a stated preference method hypothetical bias,
strategic responses, etc. are still a concern - Requires good expertise on experimental designs