Market Segmentation

Definition

Market segmentation is the identification of distinct subgroups of consumers with homogeneous preferences within each group and heterogeneous preferences across groups. In the random utility framework, segmentation models assume the population preference distribution is a finite mixture: $p(\beta) = \sum_{k=1}^K \pi_k \, \delta_{\beta_k}$ (latent class model) or a continuous distribution such as $p(\beta) \sim \mathcal{N}(\bar\beta, V_\beta)$ (hierarchical Bayes / random-coefficient model).

Key Ideas

• Latent class approach: assigns each consumer to one of $K$ segments with segment-specific preference vectors $\beta_k$; class membership is estimated from the data. See Latent Class Model.
• Continuous heterogeneity: the hierarchical Bayes (HB) approach models $\beta_i \sim \mathcal{N}(\bar\beta, V_\beta)$, enabling individual-level posterior estimates $E[\beta_i | y_i]$ — more flexible than latent classes when the population distribution is unimodal.
• Targeting via tail areas: segment identification is a tail-area problem; the covariance matrix $V_\beta$ determines what fraction of consumers exceed any profitability threshold. See Conjoint Analysis for the "extremes argument."
• Post-hoc vs. a priori segmentation: a priori segmentation partitions on observable demographics; post-hoc (model-based) segmentation recovers unobserved preference types from choice behaviour.

How It Works

HB models for segmentation use a three-block Gibbs sampler: (1) individual parameters $\beta_i | \bar\beta, V_\beta, y_i$; (2) population mean $\bar\beta | \{\beta_i\}, V_\beta$; (3) covariance $V_\beta | \{\beta_i\}, \bar\beta$. See Bayesian Hierarchical Model.

Related

• Conjoint Analysis
• Consumer Heterogeneity
• Multinomial Probit
• Latent Class Model
• Mixture Model
• Nonhomothetic Preferences