Memory-based models for predicting inferences about product quality

Simonyan, Y (2013) Memory-based models for predicting inferences about product quality. Doctoral thesis, University of London: London Business School.

Abstract

How are consumers' inferences about product quality related to brand information in memory? Prior literature suggests that, in vast majority of cases, consumers tend to assign higher quality to the products they have seen or heard of before than to those they do not recognize (Allison and Uhl 1964; Hoyer and Brown 1990). People's tendency to assign higher value to the objects they recognize has been documented in many areas outside consumer product domain as well (Gigerenzer and Goldstein 2011; Pachur and Hertwig 2006; Serwe and Frings 2006; Pachur and Biele 2007; Hertwig and Herzog 2011; Frosch, Beaman, and McCloy 2007; Richter and Sp{uml}ath 2006). However, people sometimes deviate from this tendency (Newell and Shanks 2004; Oppenheimer 2003; Pohl 2006; Richter and Sp{uml}ath 2006). While some reasons behind these deviation have been explored (Br{uml}oder and Eichler 2006; Gigerenzer and Goldstein 2011), others call for further investigation. For example, it still has not been explained why people assign higher criterion value to recognized objects less often when comparing unrecognized objects with "merely recognized" ones (Marewski et al. 2010), or why people report different levels of confidence for different pairs including a recognized and an unrecognized brand (Goldstein 1994). This work investigates the reasons behind these deviations and suggests a psychological model that builds on the idea that perceived product quality should be viewed not as a point estimate, but as a distribution of beliefs about quality. By modelling inferences, as well as confidence in inferences, via belief distributions, this thesis aims at explaining some unsolved phenomena regarding the relationship between quality perceptions, on one side, and recognition and other memory information, on the other. First, it tries to find out whether the belief distributions reflect the relationship between brand quality perceptions and recognition (as well as other memory cues), documented in the marketing literature. Second, it explores whether people infer that recognized brands associated with mediocre reputation are of higher quality than unrecognized brands. Third, using belief distributions, it attempts at explaining why people sometimes infer that an unrecognized brand is of higher quality than a recognized brand. Forth, it investigates whether the belief distributions predict inference and confidence in inferences better than existing models. In an attempt to answer these questions, I conducted two lab studies comprising over 35,000 individual inferences and collected field data concerning brands' frequency of mentions on the Internet. When predicting consumers' inferences about brand quality based on memory information, this thesis uses the following memory cues: recognition (whether or not consumers have seen or heard of a particular product brand), perceived frequency of encountering (how many times they think they have seen or heard of that brand), knowledge volume (how much they think they know about that product's quality) and knowledge valence (what proportion of that information suggests that the quality is high). In pursuit of externally valid and robust findings, I investigated these links for actual brands in five domains: refrigerators, vacuum cleaners, walking shoes, headphones and business schools.