Brand loyalty for frequently bought goods

Studying brand and category purchasing patterns in near-steady state markets produces generalisable and predictable results. Brand loyalty is broadly defined as an ongoing propensity to purchase the brand. The main theoretical basis for this approach is the NBD-Dirichlet model of choice in competitive situations. The empirical evidence, and also the model predictions, show that competing brands differ little in their levels of loyalty. Any differences that do occur are mostly related to market share. This thesis extends and critically probes the Dirichlet approach. The aim is to: (a) identify and model generalisable discrepancies, (b) refine existing measures of brand loyalty to produce additional ones, (c) test the new measures across several product categories in different countries, (d) test if a number of loyalty-based measures are linked. Analyses of individual household level panel data have led to a number of substantive findings: 1. Over a year, repeat-purchase loyalty erodes by an average of 15%. 2. Erosion of brand loyalty differs little by weight of purchase. 3. Erosion is lower for brand leaders. 4. About one-third of a brand's buyers have that brand as their favourite. 5. A favourite brand is bought twice as often as the average brand, and accounts for just over half of its buyers' category requirements and two-thirds of the brand's sales. 6. The purchase distributions for favourite brand buyers are much the same for light and heavy category buyers. 7. Heavy brand buyers give a significantly higher share, than predicted, of their category purchases to the brand. 8. High share-of-category purchase is a weak predictor of a brand's repeatpurchase erosion. 9. At the individual level, behavioural and attitudinal loyalty-related measures show very little correlation. 10. Price-related promotions are used by existing customers of the brand; promotions do not make customers more loyal. 11. Competitive brands do not segment the market. This is consistent with the lack of segmentation variables in the Dirichlet. The implications of these findings are discussed