Influence of memory processes on choice-consistency

Choice-consistency is considered a hallmark of rational value-based choice. However, because the cognitive apparatus supporting decision-making is imperfect, real decision-makers often show some degree of choice inconsistency. Cognitive models are necessary to complement idealized choice axioms with attention, perception and memory processes. Specifically, compelling theoretical work suggests that the (imperfect) retention of choice-relevant memories might be important for choice-consistency, but this hypothesis has not been tested directly. We used a novel multi-attribute visual choice paradigm to experimentally test the influence of memory retrieval of exemplars on choice-consistency. Our manipulation check confirmed that our retention interval manipulation successfully reduced memory representation strength. Given this, we found strong evidence against our hypothesis that choice-consistency decreases with increasing retention time. However, quality controls indicated that the choice-consistency of our participants was non-discernable from random behaviour. In addition, an exploratory analysis showed essentially no test–retest reliability of choice-consistency between two observations. Taken together, this suggests the presence of a floor effect in our data and, thus, low data quality for conclusively evaluating our hypotheses. Further exploration tentatively suggested a high difficulty of discriminating between the choice objects driving this floor effect.

Influence of memory processes on choice consistency

Choice-consistency is considered a hallmark of rational value-based choice. However, because the cognitive apparatus supporting decision-making is imperfect, real decision-makers often show some degree of choice inconsistency. Cognitive models are necessary to complement idealized choice axioms with attention, perception and memory processes. Specifically, compelling theoretical work suggests that the (imperfect) retention of choice-relevant memories might be important for choice-consistency, but this hypothesis has not been tested directly. We used a novel multi-attribute visual choice (MAVC) paradigm to experimentally test the influence of memory retrieval of exemplars on choice-consistency. Our manipulation check confirmed that our retention interval manipulation successfully reduced memory representation strength. Given this, we found strong evidence against our hypothesis that choice-consistency decreases with increasing retention time. However, quality controls indicated that the choice-consistency of our participants was non-discernable from random behavior. In addition, an exploratory analysis showed essentially no test-retest reliability of choice-consistency between two observations. Taken together, this suggests the presence of a floor effect in our data and, thus, low data quality for conclusively evaluating our hypotheses. Further exploration tentatively suggested a high difficulty of discriminating between the choice objects driving this floor effect.

How robust is rational choice?

Neoclassic economic choice theory assumes that decision-makers make choices as if they were rational agents. This assumption has been critically challenged over the last decades, yet systematic aggregation of evidence beyond single experiments is still surprisingly sparse. Here, we asked how robust choice-consistency, as a proxy for rationality, is to endogenous and exogeneous factors. To this end, we conducted a systematic quantitative literature research, reviewing 5327 articles, identifying 44 as relevant that contained hypothesis tests on possible influence factors of choice-consistency. To assess the evidential value of any effect of such influence factors on choice- consistency, we conducted a robust p-curve analysis. Our results indicate that choice-consistency is affected by endogenous or exogeneous factors. This result holds for multiple testing procedures and a robustness check. However, due to the breadth of the contemporary research agenda, the lack of replications and the unavailability of original data in the field of choice-consistency, it is currently not possible to draw meaningful conclusions regarding specific influence factors. Despite this lack of specificity, our results implicate that people’s decisions might be a noisier and more biased indicator of their underlying preferences than previously thought. Hence, we provide systematic evidence for the wide-spread belief that rationality cannot be assumed unconditionally.

Inconsistently consistent: Rationality is not reliable

Contemporarily, experimental investigations of revealed preference choice consistency utilize different tasks interchangeably. However, the reliability of choice consistency measurements among (inter-method) and within tasks (test-retest) has not been determined so far. Hence, it is unclear whether estimations of choice consistency fulfill a basic requirement of valid psychometric measures. Further, it is unclear how far results from different studies using different methodologies are comparable. In the study described here, we investigated the reliability of two established and one novel choice consistency tasks in an online-experiment under non-incentivized conditions in the choice domain of social decisions. Our results confidently indicate generally poor inter-method reliability and at best moderate test-retest reliability for the two indices, the Critical Cost Efficiency Index (CCEI) and the Houtman-Maks-Index (HMI), with the CCEI being the tentatively more reliable measure. This is especially concerning, since the full experiment (including test and retest measurement) lasted on average less than 45 minutes. Hence, it appears that estimations of choice consistency do not fulfill a basic requirement of valid psychometric measures. Further, results from different studies using different methodologies should not be compared without caution. Future work should investigate the impact of incentivization as well as the choice domain generality of our results.

Transient Value Refinements during Deliberation Facilitate Choice

After deciding among options, decision makers exhibit an increased positive evaluation of the chosen option coupled with a decreased evaluation for those rejected, resulting in a spreading of alternatives (SoA). There has been a longstanding debate as whether SoA results from post-choice dissonance reduction, or whether it plays an instrumental role in reaching a decision. Here we introduce a novel procedure that measures SoA implicitly during the process of choice. Participants provided simultaneous value ratings for the options on offer on each trial before finalizing their choice. The results clearly demonstrate that SoA occurs before choices are finalized. Furthermore, SoA is instrumental to the choice process: it enhances choice consistency and confidence and reduces response time. The SoA generated during choice is transient in nature, partially regressing toward baseline shortly afterwards.

Neural correlates of individual differences in aversion to risk and choice inconsistencies

Most people tend to prefer smaller certain gains to large uncertain gains when making financial choices (risk aversion). However, attitudes toward risk vary greatly between individuals, and over time within individuals. Consistent behavior may reflect the adoption by the individual of a simple or automatized heuristic which reduces the subject's uncertainty about the outcome of a behavioral choice. In contrast inconsistent behavior may reflect the adoption of a "fuzzy" logic, likely leaving high levels of uncertainty in the participant making the choice. Therefore, inconsistent behavior may often be associated with greater risk aversion. The use of simple/automatized heuristics may also lead to increased reliance on fast brain processes, whereas fuzzy heuristic may lead to lingering uncertainty. These two modes of processing may therefore lead to different brain dynamics. To examine these dynamics we recorded event-related brain potentials (ERPs) from 22 adults participants engaged in a task requiring choices between certain (but often smaller) gains and an uncertain (but often bigger) gains. Behavioral analyses allowed us to quantify choice consistency and risk aversion for each individual. Choice consistency was related to the amplitude of P200; risk aversion was related to modulation of the medial frontal negativity (MFN) as a function of choice uncertainty, to the amplitude of a late positive potential (LPP). These findings are consistent with the idea that differences in individuals' behavior when making financial choices may reflect variations in the type of heuristics they adopt, which in turn may may be reflected in differences in brain dynamics.

Multi-parameter utility and drift-rate functions conflate attribute weights and choice consistency

Standard decision models include two components: subjective-value (utility) functions and stochastic choice rules. The first establishes the relative weighting of the attributes or dimensions and the second determines how consistently the higher utility option is chosen. For a decision problem with M attributes, researchers often estimate M-1 utility parameters and separately estimate a choice-consistency parameter. Instead, researchers sometimes estimate M parameters in the utility function and neglect choice consistency. I argue that while these two approaches are mathematically identical, the latter conflates utility and consistency parameters, leading to ambiguous interpretations and conclusions. At the same time, behavior arises from the interaction of utility and consistency parameters, so for choice prediction they should not be considered in isolation. Overall, I advocate for a clear separation between utility functions and stochastic choice rules when modeling decision-making, and reinforce the notion that researchers should use M-1 parameters for M-attribute decision problems.