1 / 13

Mapping the WTP Distribution from Individual Level Parameter Estimates

Mapping the WTP Distribution from Individual Level Parameter Estimates. Matthew W. Winden University of Wisconsin - Whitewater S EA Conference – November 2012. Motivation.

wade-garza
Download Presentation

Mapping the WTP Distribution from Individual Level Parameter Estimates

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Mapping the WTP Distribution from Individual Level Parameter Estimates Matthew W. Winden University of Wisconsin - Whitewater SEA Conference – November 2012

  2. Motivation • Heterogeneity exists in respondents’ preferences, WTP, and error variances within the population (Lanscar and Louviere 2008) • Traditional Models Used in Non-Market Valuation Impose Distributional Assumptions About Preference Heterogeneity in the Population (Train 2009, Revelt and Train 1999) • Top-Down Modeling • Mixed Logit – Impose Continuous Distribution • Latent Class Logit – Impose Discrete Distribution • Misspecification May Lead to Bias in Parameter, Marginal Price (MP), and Willingness-To-Pay (WTP) Estimates • Leads to inefficient policy analysis and recommendations Matthew Winden, UW - Whitewater

  3. Individual Level Modeling As Solution • Louviere et al. (2008) estimate individual level parameters using conditional logit estimator (no welfare analysis) • Convergence issue 1: Collinearityof attributes • Convergence issue 2: Perfect Predictability • Cognitive Burden (Number of Questions/Attributes) • Louviere et al. (2010) • Best-Worst Scaling As Solution • Why Individual Level Models? •  “Bottom-Up Modeling Approach” Matthew Winden, UW - Whitewater

  4. Top-Down Versus Bottom-Up Modeling “Top-Down” “Bottom-Up” Assume Derive (, ) Estimate Derive Estimate Matthew Winden, UW - Whitewater

  5. Contributions • Objective 1: Use Monte-Carlo Simulation to Provide Evidence of the Validity of Individual Level Estimation Techniques • Objective 2a: Estimate Traditional and Individual Level Models on a Stated Preference Dataset • Eliminates Collinearity as a Convergence Problem • Objective 2b: Estimate Traditional and Individual Level Models on a Revealed Preference Dataset • Objective 3: Use Individual Level Estimates to Examine Potential Bias Resulting from Distributional Assumptions in Traditional Models Matthew Winden, UW - Whitewater

  6. Traditional Mixed Logit P(j|vi) = exp(Uji)/Σexp(Uji) Utility of choice j for respondent i: = αji + Βj+ ΦjZji + ΘjiWji where: αji= alternative-specific constant Βj= vector of fixed coefficients Χi= fixed individual characteristics Φj= vector of fixed coefficients Θj= vector of varying coefficients Zji& Wji = choice-varying attributes of choices Matthew Winden, UW - Whitewater

  7. Individual Level Simulation & Estimation Strategy • 3 Datasets (A, B, C) • Known parameter, attribute, and error distributions • 100 respondents, 100 choice scenarios • Face 3 attributes (X1 & X2 - Uniform, X3 – Zero, Status Quo) • Face 3 alternatives (Respondent Specific Error Term to Each Alternative) • Have 3 individual specific betas for each of the three attributes • Simulation A • Beta 1 = Normal, Beta 2 = Normal, Beta 3 = Normal • Simulation B • Beta 1 = Normal, Beta 2 = Normal, Beta 3 = Uniform • Simulation C • Beta 1 = Normal, Beta 2 = Normal, Beta 3 = Exponential Matthew Winden, UW - Whitewater

  8. Individual Level Model Simulation LL Values Indicate Stronger Fit Not Statistically Significantly Different Matthew Winden, UW - Whitewater

  9. Traditional and Individual Model Comparisons Table 34: Willingness-To-Pay Estimates ($/Gal) Matthew Winden, UW - Whitewater

  10. Individual Level Mapped Distribution • Kolmogorov-Smirnov tests reject the null hypothesis of the equivalence of normal and log-normal distributions Matthew Winden, UW - Whitewater

  11. Conclusions? (So-Far) • Result 1: Validity of Individual Estimation Demonstrated through Simulation  Kind Of... • Result 2: Individual Level Model Distributions, MPs, & WTPs Differ from Outcomes Using Traditional Models • Role of Including or Excluding Individuals with Statistically Significant (but possibly Lexicographic) Preferences on Estimates • Role of Including or Excluding Individuals with Statistically Insignificant values (Round to Zero?) • Result 3: Without knowing underlying distribution, may inadvertently choose incorrect mixing distribution based on LL • Although in the SP Case, the results are not statistically different Matthew Winden, UW - Whitewater

  12. Extensions • E1: True (Full) Monte-Carlo Simulation For Individual Level Specifcations • Vary Over Types of Non-Traditional Distributions and Number of Respondents, Choice Occasions, and Attributes • E2: Comparison using Revealed Preference Dataset • Introduces Potential Collinearity as a Convergence Issue • More Realistic Situation Under Which Heterogenity May Exist • E3: Significance Tests for Individual Level Models • E4: Compare Results Against Latent Class Models • E5: Scale Issues in Aggregation of Individual Respondents Matthew Winden, UW - Whitewater

  13. Thank You!

More Related