Consider-Then-Choose Models in Decision-Based Design Optimization

Abstract

This article describes an advance in design optimization that includes consumer purchasing decisions. Decision-Based Design optimization commonly relies on Discrete Choice Analysis (DCA) to forecast sales and revenues for different product variants. Conventional DCA, which represents consumer choice as a compensatory process through maximization of a smooth utility function, has proven to be reasonably accurate at predicting choice and interfaces easily with engineering models. However the marketing literature has documented significant improvement in modeling choice with the use of models that incorporate non-compensatory (descriptive) and compensatory (predictive) components. The non-compensatory component can, for example, model a “consider-then-choose” process in which potential customers first narrow their decisions to a small set of products using heuristic screening rules and then employ a compensatory evaluation to select from this set. This article demonstrates that ignoring consider-then-choose behavior can lead to sub-optimal designs, and that optimality cannot be “recovered” by changing marketing variables alone. A new computational approach is proposed for solving optimal design problems with consider-then-choose models whose screening rules are based on conjunctive (logical “and”) rules. Computational results are provided using three state-of-the-art commercial solvers (matlab, KNITRO, and SNOPT).

Sensitivity of Vehicle Market Share Predictions to Alternative Discrete Choice Model Specifications

“…The more flexible mixed and nested logit models did not offer meaningful improvements for the new vehicle market using the types of covariates that have been used in the literature, although we only examined mixed logit models with diagonal covariance matrices. We caution that these types of models may not be well suited to predict vehicle shares for new design or policy evaluation [7,[9][10][11] and that these same limitations may apply to other product design domains which use discrete choice models [4][5][6]. Reducing the data set to include only midsize vehicles improved the predictive capabilities of the model.…”

“…Furthermore, MacDonald et al suggest that consumers' preferences for attributes do not exist a-priori but rather that products are evaluated on a case by case basis [56]. Morrow et al [10] and others have suggested that vehicle choice behavior may be better represented by a consider-then-choose model where consumers do not maximize their utility over the universe of available choices but rather screen all choices to narrow them down to a reasonable subset over which utility maximization applies. And more broadly, the Lucas critique warns against use of aggregated historical data to predict outcomes in counterfactual future scenarios [70].…”

“…Motivated by a call to base design decisions explicitly on predictions of their downstream consequences for the firm [1], researchers have proposed a variety of methods to predict the influence of design decisions on firm profit. The majority of these efforts apply discrete choice methods [2] to predict consumer choice as a function of product attributes and price, using choice predictions to guide or even optimize design decisions [3][4][5][6][7][8][9][10][11]. Such methods rely on the accuracy of choice predictions: uncertainty in choice predictions creates uncertainty about which designs are best [5,12].…”

Market-System Design Optimization With Consider-Then-Choose Models

“…As described in Ref. [54], this technique does not easily generalize to other problems. The contours of the objectives, the budget constraint, and the optimal solutions for Scenarios (1) and (2) are illustrated in Fig.…”

“…(22)- (28) or with other techniques; e.g., see Ref. [54]. Because of this flexibility in solution technique, it is useful to state important technical results in terms of Eq.…”