Scholarlink

<p><?tk 2?>Nowadays, choice modelling based on stated choice data is widely used for assessing consumer preferences, with the hierarchical Bayes multinomial logit (HB-MNL) model being typically embedded as choice model for preference estimation. The HB-MNL model accounts for individual preference heterogeneity and is therefore able to weaken the Independence of Irrelevant Alternatives property (IIA) the standard MNL model suffers from. The IIA property states that the ratio of choice probabilities of two alternatives remains constant independent of the availability of further alternatives and hence implies an often unrealistic proportional substitution pattern among alternatives. The IIA property can also be addressed by applying nested multinomial logit (NMNL) models, which explicitly allow for a representation of different degrees of similarity between subsets of alternatives (nests). In this paper, we consider a HB-NMNL model which combines both the idea of nested market structures and heterogeneous consumer preferences with the expectation to handle the IIA property even better. We propose an extensive simulation study in which we compare the performance of the established HB-MNL model to the HB-NMNL model under varying experimental conditions for model fit, prediction accuracy, and parameter recovery. Our findings from this Monte Carlo study are twofold: first, both types of models perform rather close with regard to predictive validity (and model fit). This result is quite surprising and provides strong support for the use of the less complex HB-MNL model. Second, as could be expected, the HB-NMNL model shows advantages in terms of parameter recovery which is an important criterion for product design decision-making. We further present results from applying and comparing both models in an empirical study for summer tires. Model estimation is carried out using the publicly available <i>R</i> software package <i>RSGHB</i>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Identifying nested preference structures in choice models based on stated choice data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月1日

查看全文
  • Nils Goeken,
  • Peter Kurz,
  • Winfried J. Steiner

摘要

Nowadays, choice modelling based on stated choice data is widely used for assessing consumer preferences, with the hierarchical Bayes multinomial logit (HB-MNL) model being typically embedded as choice model for preference estimation. The HB-MNL model accounts for individual preference heterogeneity and is therefore able to weaken the Independence of Irrelevant Alternatives property (IIA) the standard MNL model suffers from. The IIA property states that the ratio of choice probabilities of two alternatives remains constant independent of the availability of further alternatives and hence implies an often unrealistic proportional substitution pattern among alternatives. The IIA property can also be addressed by applying nested multinomial logit (NMNL) models, which explicitly allow for a representation of different degrees of similarity between subsets of alternatives (nests). In this paper, we consider a HB-NMNL model which combines both the idea of nested market structures and heterogeneous consumer preferences with the expectation to handle the IIA property even better. We propose an extensive simulation study in which we compare the performance of the established HB-MNL model to the HB-NMNL model under varying experimental conditions for model fit, prediction accuracy, and parameter recovery. Our findings from this Monte Carlo study are twofold: first, both types of models perform rather close with regard to predictive validity (and model fit). This result is quite surprising and provides strong support for the use of the less complex HB-MNL model. Second, as could be expected, the HB-NMNL model shows advantages in terms of parameter recovery which is an important criterion for product design decision-making. We further present results from applying and comparing both models in an empirical study for summer tires. Model estimation is carried out using the publicly available R software package RSGHB.