Measuring Consumer Preferences for Complex Products: A Compositional Approach BasedonPaired Comparisons

Abstract

Conjoint analysis has become a widely accepted tool for preference measurement in marketing research, though its applicability and performance strongly depend on the complexity of the product or service. Therefore, self-explicated approaches are still frequently used because of their simple design, which facilitates preference elicitation when large numbers of attributes need to be considered. However, the direct measurement of preferences, or rather utilities, has been criticized as being imprecise in many cases. Against this background, the authors present a compositional consumer preference measurement approach based on paired comparisons, otherwise known as PCPM. The trade-off character of paired comparisons ensures that the stated judgments are more intuitive than traditional self-explicated preference statements. In contrast to the latter, PCPM accounts for response errors and thus allows for the elicitation of more precise preferences. The authors benchmark PCPM against adaptive conjoint analysis and computer-assisted self-explication of multiattributed preferences to demonstrate its relative validity and predictive accuracy in two empirical studies using complex, high-involvement products. They find that PCPM yields better results than the benchmark approaches with respect to interview length, individual hit rates, and aggregate choice share predictions.

On the optimal number of scale points in graded paired comparisons

“…A more fine-grained assessment of consumers' preferences relies on a graded paired comparison (GPC, also referred to as constant sum paired comparison; see Day 1965;Netzer and Srinivasan 2011;Netzer, Toubia, Bradlow, Dahan, Evgeniou, Feinberg, Feit, Hui, Johnson, Liechty, Orlin and Rao 2008;Oishi, Hahn, Schimmack, Radhakrishan, Dzokoto and Ahadi 2005;Scholz, Meissner and Decker, 2010). In a GPC, individual consumers are instructed to rate the direction and intensity of their preference by dividing a particular sum of points among the two alternatives in every pair.…”

“…In addition, when individuals do not prefer one choice alternative over the other in a given pair, and are forced to assign a larger number to (at best, an arbitrarily chosen) choice alternative, GPC data are likely to suffer from logically 1 To date, no consensus exists as to what the optimal number of scale points is (Janhunen in press;Moors 2008). Some studies have suggested the optimal number of scale points to be seven: see Hofmans et al 2007;Lozano et al 2008;Weng 2004; few other studies suggested the optimal number of scale points to be higher than seven (see Alwin 1997;Scholz et al 2010), whereas other studies have argued that using five scale points is just as good as using seven or more scale points (see Böcker 1988;Churchill and Peter 1984). preference structures are not always logically consistent, and -even if an individual has a coherent preference structure -he/she may still provide preference data which are not completely in line with his/her preference structures.…”

Consumer Preference Elicitation of Complex Products Using Fuzzy Support Vector Machine Active Learning

“…The overall complexity of the digital camera category (with 30+ attribute levels) is parallel to product categories studied in prior research that elicits consumers' preferences for complex products (e.g., Park et al 2008;Netzer and Srinivasan 2011;Scholz et al 2010). The computer tablet category (with 70+ attribute levels) is considerably more complex than the ones used in extant methods, particularly in the context of decompositional preference elicitation methods.…”

“…Similar to prior studies in the literature (e.g., Scholz, et al 2010;Netzer and Srinivasan 2011), participants in all conditions first complete the preference measurement task, followed by an external validation task and a post-survey feedback task. Identical across the six experimental conditions, the external validation task comprises two choice questions, each including two camera profiles.…”

“…Conventional preference elicitation methods such as conjoint analysis often become infeasible in this context, because the number of questions required to obtain accurate estimates increases rapidly with the number of attributes (and/or attribute levels). Historically, researchers have mostly relied on compositional approaches to handle preference elicitation of such products (e.g., Srinivasan 1988;Scholz et al 2010;Netzer and Srinivasan 2011). Adaptive question selection algorithm has also been proposed for complex product preference elicitation, due to its ability to rapidly reveal consumer's product preferences with relatively few questions (e.g., Netzer and Srinivasan 2011).…”