Eliciting Consumer Preferences Using Robust Adaptive Choice Questionnaires

We propose a framework for designing adaptive choice-based conjoint questionnaires that are robust to response error. It is developed based on a combination of experimental design and statistical learning theory principles. We implement and test a specific case of this framework using Regularization Networks. We also formalize within this framework the polyhedral methods recently proposed in marketing. We use simulations as well as an online market research experiment with 500 participants to compare the proposed method to benchmark methods. Both experiments show that the proposed adaptive questionnaires outperform existing ones in most cases. This work also indicates the potential of using machine learning methods in marketing.

1 Introduction

A central problem in marketing is the understanding of consumers' preferences. Various methods have been developed for this purpose with conjoint analysis [9], [8], [18], [19], [47] being among the most widely used ones. The approach is based on asking respondents questions about their preferences among products (for example, “do you prefer product A or product B?”) and estimating their preferences for the features (attributes) of the products based on their responses. A key issue in conjoint analysis is the design of the questionnaires. The traditional approach has been to use nonadaptive questionnaires: The questions are determined in advance and are not influenced by the respondents' answers. With the advent of online marketing research, researchers and practitioners have become increasingly interested in adaptive questionnaires [21]: Each question for every individual is designed in real time based on his or her responses to earlier questions. Despite their attractiveness, adaptive questionnaires can be subject to endogeneity (as will be illustrated in Section 3.5): Questions are influenced by the response errors to earlier answers [22]. As a result, although recently proposed adaptive methods tend to outperform nonadaptive benchmarks when response error is low, they typically do not perform as well when response error is high [43], [44] (as is often the case in online environments). This suggests the need for questionnaires that retain the benefits of adaptive interviews while being robust to response error.