Quantum & Context Effects

Uncertainty can lead to violations of both “rational” decision-making (Tversky & Kahneman, 1974) and the laws of classical probability (CP; Busemeyer et al., 2011). One type of violation – interference effects – occurs when a marginal decision distribution depends on the presence or absence of a preceding category judgment (Wang & Busemeyer, 2016). Interestingly, recent studies have shown that interference effects emerge under some preceding categorization conditions but not others, resulting in a critical asymmetry (Busemeyer et al., 2009; Wang & Busemeyer, 2016). Models based on the formalism of quantum probability (QP) provide a good accounting of this unusual pattern in the data (Busemeyer et al., 2009; Wang & Busemeyer, 2016). More recently, we have defined models based on CP augmented with ancillary mechanisms that can account for interference effects in the data, in general, but not necessarily the critical asymmetry (anonymous 1; anonymous 2). Given these varied resolutions to violations of CP, an important question warrants further investigation: what is the generalizability of these models and the efficacy of their mechanisms under conditions of novel stimulus types and decisions? In the current study, we probe questions about generalizability by using non-human-stimuli to control for the potential confounding effects of pre-existing associations with human faces within the categorization-decision paradigm used by Wang and Busemeyer (2016). We further assess the influence of bias on interference effects by leveraging a pre-existing bias based on preferences for sophisticated, in comparison to simplistic, mind-types (Almaraz et al, 2018; Dennett, 1996; Epley & Eyal, 2019; Waytz et al., 2010) and a resource allocation task with a clear link to mind-type preferences (Dennett, 1996; Waytz et al., 2010). We expect our study to identify the extent to which interference effects, along with the critical asymmetry, emerge in the new data reflecting a pre-existing bias, and to determine whether three proposed models (using quantum cognition, ACT-R, and a multinomial processing tree) can account for the new findings. Ultimately, we expect our study to inform the degree to which a set of current models of interference effects generalize for biased beliefs and, by extension, to provide insights into processes underlying interference effects in other contexts.

This is an in-person presentation on July 19, 2023 (11:20 ~ 11:40 UTC).

Contextuality in systems of random variables has been originally formulated in quantum physics in terms of hidden variable models (HVMs). These formulations are contingent on the systems of random variables being consistently connected, which means that any two variables answering the same question in different contexts have the same distribution. Outside quantum physics, and specifically in systems of random variables describing behavioral phenomena, consistent connectedness is virtually never observed. This has necessitated, in the last decade, an extension of the notion of contextuality to arbitrary systems of random variables, resulting in the theory called Contextuality-by-Default. However, for inconsistently connected systems the possibility of interpreting contextuality in terms of HVMs is lost, and this is considered by some a major problem. It can be shown, however, that any inconsistently connected system can be recast as a consistently connected one, so that the two systems describe precisely the same empirical or theoretical situations, and they are contextual or noncontextual together. The consistently connected rendering of a system is amenable to formulation in terms of HVMs. The similarities of this formulation with and differences from the HVM representations of the traditionally considered quantum-mechanical systems elucidate the subtle interplay of the mathematical and the empirical in describing phenomena by systems of random variables. [1] Kujala, J.V., Dzhafarov, E.N., & Larsson, J.-A. (2015). Necessary and sufficient conditions for extended noncontextuality in a broad class of quantum mechanical systems. Physical Review Letters 115, 150401. [2] Cervantes, V.H., & Dzhafarov, E.N. (2018). Snow Queen is evil and beautiful: Experimental evidence for probabilistic contextuality in human choices. Decision 5, 193-204. [3] Dzhafarov, E.N. (2022). Contents, contexts, and basics of contextuality. In Shyam Wuppuluri and Ian Stewart (Eds). From Electrons to Elephants and Elections, The Frontiers Collection. pp. 259-286. Cham, Switzerland: Springer.

This is an in-person presentation on July 19, 2023 (11:40 ~ 12:00 UTC).

Context effects, including attraction, similarity, and compromise effects, have been widely studied. These effects occur when choices among existing alternatives are impacted by adding new alternatives to the choice set. Sometimes the addition of a new alternative impacts the relative choice share (RCS) for one alternative compared to another. In other cases, adding a new alternative simply increases the absolute choice share (ACS) one alternative receives. Here we report a meta-analysis of all three effects asking how reliably, across 23 papers with 29,538 observations, these effects impact the RCS and ACS. The results revealed that these three context effects robustly impacted the RCS of an option. While the attraction and compromise effects only weakly impacted the ACS. Results further showed that the context effects depend on the configuration of attributes across the choice set, yet nearly all the studies to date have focused on a very specific configuration. Furthermore, simulations with leading choice models that predict context effects like MDFT and MLBA make very different predictions about how different configurations of attributes give rise to these effects. Altogether our results establish a great need to map out how these context effects change over a much larger configuration of alternatives.

This is an in-person presentation on July 19, 2023 (12:00 ~ 12:20 UTC).

Decades of research have been dedicated to understanding context effects (attraction, compromise, and similarity) in multi-alternative, multi-attribute choice. Most studies have used laboratory tasks with artificial stimuli. For example, choices among apartments with attributes represented numerically (i.e., 10 miles to work). However, when participants are shown more naturalistic stimuli (e.g., photos of apartments), the effects often disappear (Frederick, Lee, & Baskin, 2014). Thus, researchers have argued that context effects are an artifact of artificial stimuli and do not occur in naturalistic choices (Frederick et al., 2014; Yang & Lynn, 2014; Trendl, Stewart, & Mullett, 2021). However, the absence of context effects does not imply the absence of contextual sensitivity. Context-dependent behavior occurs whenever the evaluation of an option is dependent on the other options, often defined as a violation of simple scalability. We take a joint experimental and computational modeling approach to address whether naturalistic decisions demonstrate contextual sensitivity. One of the critical limitations for using computational cognitive models to study naturalistic decision making is that these models require quantitative representations of stimuli. In the past, obtaining representations of naturalistic stimuli has been challenging. I will describe how machine learning models can be coupled with cognitive models to overcome this limitation and help resolve the issue of contextual sensitivity in naturalistic multi-alternative choice.

This is an in-person presentation on July 19, 2023 (12:20 ~ 12:40 UTC).