Incorporating consumer inattentiveness into discrete choice analysis via heteroskedastic adjustment = 이분산성 조정을 통한 소비자 부주의성의 이산선택분석 반영

국문 초록 (Abstract)

전통적인 이산선택모형은 개인의 의사결정이 완전한 주의와 일관된 선호를 반영한다는 가정에 기반하고 있다. 그러나 실제 응답자의 행동을 관찰하면, 이들이 종종 주의력이 결핍된 상태에서 선택을 내리는 사례가 발견되며, 이는 효용 추정치의 왜곡과 진술선호 데이터에 대한 신뢰도 저하로 이어질 수 있다. 본 연구는 간단하면서도 효과적인 방식으로 이러한 부주의성을 혼합로짓모형 구조에 반영하여 고전적 방법론을 보완하는 접근법을 제안한다. 제안된 방법론은 엔트로피 기반의 행동지표인 “부주의성 지표”를 정의하여 속성효용의 분산을 개인 수준에서 조절하고 이를 통해 주의력에 따른 속성효용 추정치의 불확실성을 포착한다. 부주의성 지표는 진술선호 외적인 개인 특성이나 설문장치에 의존하지 않고 관측된 선택 데이터만을 기반으로 계산된다. 이를 통해 이산선택실험의 설계와 무관하게 일관된 방식으로 개인별 부주의성을 수치화할 수 있다. 본 연구는 제안된 지표와 모형의 성능을 시뮬레이션 분석과 실증분석을 통해 검증하였으며, 특히 부주의성이 존재하는 상황에서 속성 한계효용의 참값 복원력과 모형 적합도 측면에서 기존 혼합로짓 모형과의 비교를 수행하였다. 실증분석에서는 에너지 정책 분야의 다양한 데이터셋을 활용하여 부주의성이 의사결정에 미치는 복합적 영향을 해석하였다. 제안된 부주의성 지표와 이를 반영한 이분산 혼합로짓 모형은 간결하고 유연한 구조를 통해 전통적 이산선택모형의 특성을 유지하면서도, 그간 상대적으로 간과되어온 의사결정 참여도와 주의력의 문제를 효과적으로 다루는 데 기여하였다.

다국어 초록 (Multilingual Abstract)

Traditional discrete choice models assume that individuals make decisions with full attention and stable preferences. Yet in practice, respondents often show signs of inattentiveness, which can distort utility estimates and reduce the reliability of stated preference data. This study proposes a simple but effective way to incorporate such inattentiveness into the mixed logit framework. The proposed approach adjusts the conventional model by including an entropy-based inattention measure, the Inattentiveness Index, as a variance control for marginal utility estimates, introducing an interpretable means to reflect such behavioral inconsistencies. The Inattentiveness Index is calculated using only information from the observed choices, without relying on external traits or post-survey responses. This design keeps the method broadly applicable in discrete choice experiments while making inattentiveness interpretable within the model. Simulation and empirical analyses show that the inattention-adjusted model preserves parameter recovery and improves model fit, especially in decision contexts that are cognitively demanding. In particular, applications to energy policy datasets demonstrate that inattentiveness plays a non-trivial role in shaping stated preferences. The combined use of the Inattentiveness Index and Inattention-Adjusted Mixed Logit offers a simple yet flexible strategy for handling inattentiveness in discrete choice experiments, allowing behavioral uncertainty to be formally integrated into existing models without compromising their structure or interpretability.

목차 (Table of Contents)

• Abstract iii
• Contents v
• List of Tables ix
• List of Figures xi
• Chapter 1. Introduction 1
• 1.1 Research Background 1
• 1.2 Research Objective 5
• 1.3 Research Outline 8
• Chapter 2. Literature Review 10
• 2.1 Literature on Discrete Choice Model with Utility Maximization 10
• 2.1.1 Conditional Logit Model 10
• 2.1.2 Mixed Logit Model 12
• 2.1.3 Logit Model Variations 15
• 2.2 Literature on Inattentive Choice Behaviors 19
• 2.2.1 Decision Avoidance 19
• 2.2.2 Attribute Non-Attendance 22
• 2.2.3 Inattention 26
• 2.3 Limitations of Discrete Choice Literature 31
• Chapter 3. Methodology 33
• 3.1 Methodological Motivation 33
• 3.2 Formulation of Inattentiveness Index 36
• 3.2.1 Introduction to Shannon Entropy 36
• 3.2.2 Pattern-Specific Entropy 37
• 3.2.3 Attribute-Specific Entropy 41
• 3.2.4 Inattentiveness Index 48
• 3.3 Inattention-Adjusted Mixed Logit 52
• 3.3.1 Mixed Logit with Heteroskedastic Control for Random Coefficients 52
• 3.3.2 Reduced Form of Greene and Hensher’s Mixed Logit Specification 55
• Chapter 4. Model Validation 62
• 4.1 Simulation Analysis 63
• 4.1.1 Scenario Setup 63
• 4.1.2 Results of Simulations 66
• 4.2 Validation with Empirical Data 76
• 4.2.1 Comparative Estimation: MXL vs. IAML Across Datasets with Different Backgrounds 76
• 4.2.2 External Validation of the INI Using Conventional Attention-State Screening Measures 78
• 4.3 Summary of the Model Validation 84
• Chapter 5. Empirical Analyses 87
• 5.1 Empirical Analysis 1: Standard MXL vs. IAML 88
• 5.1.1 Experiment on the Hydrogen Portfolio Standard 88
• 5.1.2 Experiment on Renewable Energy Mix with Profit-Sharing Scheme 96
• 5.1.3 Experiment on Alternative Energy Mix with Nudge Treatments 102
• 5.2 Empirical Analysis 2: Interpretation on the Distribution of the INI 110
• 5.2.1 Sensitivity of Variance to INI Levels 111
• 5.2.2 Comparison of the Mean Marginal Willingness-to-Pay 124
• 5.3 Empirical Analysis 3: Alternative Specifications of the INI 133
• 5.3.1 IAML Performance Comparison between the INI and Individual-Specific Variables 134
• 5.3.2 IAML Performance Comparison between Different INI Specifications 141
• 5.4 Summary of Empirical Analyses 147
• Chapter 6. Summary and Conclusion 151
• 6.1 Concluding Remarks and Contributions 151
• 6.1.1 Summary of the Study 151
• 6.1.2 Methodological Contribution 154
• 6.1.3 Practical Implications 156
• 6.2 Limitations and Suggestions for Future Research 158
• Bibliography 165
• Appendix 1: Full Result of the Simulation Analysis 186
• Appendix 2: Data-generating Process for the Simulation 187
• Appendix 3: Survey Questionnaires for Empirical Datasets 188
• Appendix 4: Descriptions on Personal Traits 205
• Appendix 5: Details on the Latent Class Logit 213
• Appendix 6: Changes of the Variance Term According to the INI 217
• Appendix 7: Histogram of WTP Estimates Across 5% Bins of the Sample’s INI Distribution 220
• Appendix 8: Estimation Results for the Media and Electric Vehicle Datasets 223
• Appendix 9: Estimation Results for Alternative Model Specifications 228
• Abstract (Korean) 233