Market segmentation under the choice modelling framework

This thesis examines market segmentation from an economic perspective by drawing on the choice modelling framework. The market segmentation literature focuses on providing information to marketers about customers, especially about their wants, needs and preferences regarding different products or services. A challenge confronting marketing researchers is to improve the estimation of market segments. The latent class model (LCM) has been extensively used to capture consumers’ heterogeneity and identify market segments. However, some questions remain unanswered. How reliable and effective is the LCM in capturing customer heterogeneity, especially unobserved heterogeneity? How are estimation results affected if customers do not consider all information provided in the choice tasks; that is, when attribute non-attendance (ANA) occurs? To answer these questions, this thesis assessed the LCM in two ways. First, it was compared with an alternative model––the mixed logit model (MLM)––and the role of individual-specific posterior distributions (ISPs) was evaluated to account for unobserved heterogeneity and identify market segments. Second, the thesis assessed the role of ANA in modelling and identifying market segments.

This thesis consists of three studies. The first study identifies market segments by examining customer heterogeneity from the ISP in the MLM and the LCM. When using the ISP in the LCM as the basis for segmentation, there is an explicit recognition that class membership is probabilistic. The identified market segments are compared in terms of the number of customers in each segment and their characteristics. The results suggest differences in both customer number and characteristics.

The second study performs a Monte Carlo simulation to determine the impact of consumer heterogeneity on the accuracy of the LCM and the MLM in identifying market segments. The design comprises four experiments with two levels of heterogeneity (low and high) and the presence or absence of small (niche) segments. The results showed that the accuracy of the models is contingent on the level of heterogeneity of individuals. Specifically, when heterogeneity is low, segments estimated by the LCM are more precise; however, when heterogeneity is high, the MLM outperforms the LCM. The results also suggest that using ISPs as the basis for segmentation in the LCM makes it possible to identify market segments more accurately when heterogeneity is high. In the presence of niche segments, there was no evidence of one model outperforming the other.  The third study accounts for ANA in identifying market segments using an LCM. Images were added to text, allowing consumers to visualise attribute levels. This was used to reduce ANA as a coping mechanism for complex tasks, thereby better capturing genuine consumer preferences. The results showed that inferred ANA combined with stated ANA in the LCM improves model performance. Moreover, using images to present attribute levels improves model performance when ANA is accounted for in the estimation.