The Generalized Stochastic Preference Choice Model

Abstract

We propose a new discrete choice model that generalizes the random utility model (RUM). We show that this model, called the Generalized Stochastic Preference (GSP) model can explain several choice phenomena that can't be represented by a RUM. In particular, the model can easily (and also exactly) replicate some well known examples that are not RUM, as well as controlled choice experiments carried out since 1980's that possess strong regularity violations. One of such regularity violation is the decoy effect in which the probability of choosing a product increases when a similar, but inferior product is added to the choice set.An appealing feature of the GSP is that it is non-parametric and therefore it has very high flexibility. The model has also a simple description and interpretation: it builds upon the well known representation of RUM as a stochastic preference, by allowing some additional consumer types to be non-rational.

Decision Forest: A Nonparametric Approach to Modeling Irrational Choice

“…They proposed a method for predicting revenues in which one predicts the worst-case revenue of an assortment, where the worst-case is with respect to all probability distributions over rankings that are consistent with the available data. Subsequent research on ranking-based models has studied other estimation approaches (van Ryzin and Vulcano 2014, Mišić 2016, Jagabathula and Rusmevichientong 2016, 2018, as well as methods for obtaining optimal or near-optimal assortments (Aouad et al 2015, 2018, Feldman et al 2018, Bertsimas and Mišić 2018. Another model is the Markov chain model of customer choice (Blanchet et al 2016).…”

“…Within operations management, examples of new choice models include the general attraction model (GAM) (Gallego et al 2014), the HALO-MNL model (Maragheh et al 2018) and the generalized stochastic preference (GSP) model (Berbeglia 2018). The main difference between our model and these prior models is in expressive power.…”

The Use of Binary Choice Forests to Model and Estimate Discrete Choice Models

“…There are a few experimental studies showing strong evidence that regularity may be violated (Simonson and Tversky, 1992). Several models are proposed to capture even more general behavior than RUM (Natarajan et al, 2009;Flores et al, 2017;Berbeglia, 2019;Feng et al, 2017). It is unclear if the estimation can be performed e ciently.…”

Random consideration and choice: A case study of “default” options

“…When modelling discrete choice behavior, it is natural to include a "default" option to capture situations where the decision maker selects none of the feasible alternatives. 1 While the idea dates back to Corbin and Marley (1974), 2 the practice of including a no-choice option only took off with the recent work of Manzini and Mariotti (2014) and others (Aguiar, 2015 and2017;Berbeglia, 2018;Brady and Rehbeck, 2016;Demirkany and Kimyaz, 2018;Echenique, Saito and Tserenjigmid, 2018;Zhang, 2016). 3 This is a welcome development.…”