Human behavior is modulated by financial incentives, but it is not well understood what types of behavior are immune to incentive and why. The cognitive processes underlying behavior appear to create restrictions on the effect an individual's motivation will have on their performance. We investigate a classic category learning task for which the effect of financial incentives is still unknown (Shepard, Hovland, & Jenkins, 1961). Across four renditions of a category learning experiment, we find no effect of incentive on performance. On a fifth experiment requiring category recognition but not learning, we find a large effect on response time and small effect on task performance. Humans appear to selectively apply more effort in valuable contexts, but the effort is disproportionate with the performance improvement. Taken together, the results suggest that performance in tasks which require novel inductive insights are relatively immune to financial incentive, while tasks that require rote perseverance of a fixed strategy are more malleable.
What information guides individuals to trust an algorithm? We examine this question across three experiments that consistently found explanations and relative performance information increased trust in an algorithm relative to a human expert. Strikingly however, in only 23% of responses (414/1800) did an individual’s preferred agent for a task (e.g., driving a car) change from human to algorithm. Thus, initial preferences were ‘sticky’ and largely resistant to large shifts in trust. We discuss theoretical and practical implications of this work and identify important contributions to our understanding of how summaries of information can improve people’s willingness to trust decision aid algorithms.
The propensity for people to avoid mentally demanding tasks in the absence of reward is well documented. As a result, humans are often described as cognitive misers. This characterisation, while consistent with the psychological literature, contradicts everyday instances of effort being sought: reading, board games, and brain-teasing puzzles. Such examples however are markedly different from the types of tasks typically used in the mental effort literature (e.g., working memory tasks, demand selection tasks). The current set of experiments assessed whether the type of task (i.e., N-Back, Number Sequence Problems [NSP], or Anagrams) affects people’s aversion to, or desire for, increased effort. On average, across 3 experiments, participants showed an aversion to effort regardless of whether the effort required was more attentional (N-Back) or cognitive (NSP and anagrams) in nature, and were willing to forgo financial reward in order to avoid more difficult tasks. A minority of participants, however, sought more effortful tasks for equal or lesser reward.
Most theoretical accounts of non-instrumental information seeking suggest that the magnitude of rewards has a direct influence on the attractiveness of the information. Specifically, the magnitude of rewards is assumed to be proportional to the strength of information seeking (or avoidant) behaviour. In a series of experiments using numerical and pictorial stimuli, we explore the extent to which observed information seeking behaviour tracks these predictions. Our findings indicate a robust independence of information seeking from outcome magnitude and valence with preferences for information largely remaining constant across different reward valence and magnitudes. We discuss these results in the context of current computational models with suggestions for future theoretical and empirical work.
People’s desire to seek or avoid information is not only influenced by the possible outcomes of an event, but the probability of those particular outcomes occurring. There are competing explanations however as to how and why people’s desire for non-instrumental information is affected by factors including expected value, probability of outcome, and a unique formulation of outcome uncertainty. Over two experiments, we find that people’s preference for noninstrumental information is positively correlated with probability when the outcome is positive (i.e., winning money) and negatively correlated when the outcome is negative (i.e., losing money). Furthermore, at the aggregate level, we find the probability of an outcome to be a better predictor of information preference than the expected value of the event or its outcome uncertainty.
People intuitively distinguish between uncertainty they believeis potentially resolvable and uncertainty that arises from inher-ently stochastic processes. The vast majority of experimentsinvestigating decisions based on experience, however, have fo-cused exclusively on scenarios that promote a stochastic inter-pretation by representing options as images that remain iden-tical each time they are presented. In the current research,we contrasted this method with one in which the visual ap-pearance of options was subtly differentiated each time partic-ipants encountered them. We found that introducing this vari-ability to the appearance of options influenced the way peopleinterpreted uncertainty. Although there was little evidence ofan impact on exploration, these differences in interpretationmay reveal other limitations to the generalisability of previousdecision-making tasks.
Numerous experiments have suggested that extreme outcomes are disproportionately influential when we make decisions involving risk, but there is less consensus on what it actually means to be extreme. Existing accounts broadly fall into two categories: those that suggest that the best and worst outcomes are uniquely influential and those that suggest that outcomes become more influential with increasing deviation from the centre of the distribution. We conducted two experiments that aimed to tease apart these explanations. Although there was some evidence that the distance from the centre influences memory, neither account was able to fully explain the choices made by participants. This finding has implications for the viability of these explanations as well as for the generalisability of the effect and the interpretation of the method used to assess memory.
People tend to place value on information even when it does not affect the outcome of a decision. Two competing accounts offer explanations for such non-instrumental information seeking. One account foregrounds the role of anticipation and the other focusses on uncertainty aversion. Both accounts make similar predictions for short cueoutcome delays and when outcomes are positively valenced, but they differ in their explanation of information preference at long delays with negative outcomes. We present a series of experiments involving both primary and secondary reinforcers that pit these accounts against each other. The results indicate a consistent preference for non-instrumental information even at long cue-outcome delays and no evidence for information avoidance with negative outcomes. This pattern appears to provide more support for the uncertainty-aversion account than one based on anticipation.
Interruptions are an inevitable part of every day life. Previous research suggests that interruptions can decrease performance and increase errors and response time. Additionally, there is evidence that providing a lag time prior to an interruption can mitigate some of the interruption costs. The goal of this paper is to investigate the effects of interruptions and interruption lags and explore possible strategies to attenuate interruption costs. A novel sequential decision-making paradigm was used, where the difficulty of the task and type of interruption were the two experimental manipulations. The results indicate that there is a potential benefit to including a lag time when presented with interruptions.
In uncertain environments we must balance our need to gather information with our desire to exploit current knowledge. This is further complicated in reactive environments where actions produce long-lasting change. In three experiments, we investigate how people learn to make effective decisions from experience in a dynamic four-armed bandit task. In contrast to the diminishing rewards found in most previous studies, options were framed as skills that developed greater rewards when chosen. We find that most individuals learn effective strategies for coping with reactive environments. We present a psychological model positing that decision makers move through three distinct processing phases, and show that it accounts for key behavioral patterns across experiments.
We replicate and extend work demonstrating that choice and probability estimation can be dissociated through the coexistence of contradictory reactions to rare events. In the context of experience-based risky choice, we find the simultaneous underweighting of rare events in choice and their overestimation in probability judgement. This tendency persisted in the presence of accurate descriptions of rare event occurrence (Experiment 1), but was attenuated by incentivizing accurate probability estimates (Experiment 2). The implications of these results for popular models of risky choice are briefly discussed.
Previous research on the effects of probability and delay on decision-making has focused on examining each dimension separately, and hence little is known about when these dimensions are combined into a single choice option. Importantly, we know little about the psychological processes underlying choice behavior with rewards that are both delayed and probabilistic. Using a process-tracing experimental design, we monitored information acquisition patterns and processing strategies. We found that probability and delay are processed sequentially and evaluations of risky delayed prospects are dependent on the sequence of information acquisition. Among choice strategies, directly comparing the values of each dimension (i.e., dimension-wise processing) appears to be most favored by participants. Our results provide insights into the psychological plausibility of existing computational models and make suggestions for the development of a process model for risky intertemporal choice.
We propose simple parameter-free models that predict how people learn environmental cue contingencies, use this information to measure the usefulness of cues, and in turn, use these measures to construct search orders. To develop the models, we consider a total of 8 previously proposed cue measures, based on cue validity and discriminability, and develop simple Bayesian and biased-Bayesian learning mechanisms for inferring these measures from experience. We evaluate the model predictions against people’s search behavior in an experiment in which people could freely search cues for information to decide between two stimuli. Our results show that people’s behavior is best predicted by models relying on cue measures maximizing short-term accuracy, rather than long-term exploration, and using the biased learning mechanism that increases the certainty of inferences about cue properties, but does not necessarily learn true environmental contingencies.
Utility based models are common in both the risky and intertemporal choice literatures. Recently there have been efforts to formulate models of choices which involve both risks and time delays. An important question then is whether the concept of utility is the same for risky and inter-temporal choices. We address this question by fitting versions of two popular utility based models, Cumulative Prospect Theory for risky choice, and Hyperbolic Discounting for inter-temporal choice, to data from three experiments which involved both choice types. The models were fit assuming either the same concept of utility for both, by way of a common value function, or different utilities with separate value functions. Our results show that while many participants seem to require the flexibility of different value functions, an approximately equal number do not suggesting they may have a single concept of utility. Furthermore for both choice types value functions were concave.
At the core of every decision-making task are two simple features; outcome values and probabilities. Over the past few decades, many models have developed from von Neumann’ and Morgenstern’s (1945) Expected Utility Theory to provide a thorough account of people’s subjective value and probability weighting functions. In particular, one such model that has been largely successful in both Psychology and Economics is Cumulative Prospect Theory (CPT; Tversky & Kahneman, 1992). While these models do fit people’s choice behavior well, few models have attempted to provide a psychological account for subjective value, probability weighting, and resulting choice behavior. In this paper, we focus on a memory confusion process as described in Hawkins et al.’s (2014) exemplar-based model for decisions from experience, the Exemplar Confusion (ExCon) model, and adapt it to account for biased probability estimates in decisions from description. Using Bayesian model selection techniques, we demonstrate that it is able to account for real choice data from a Rieskamp (2008) study using gains, losses, and mixed description-based gambles, and performs at least as well as CPT.
How do people solve the explore-exploit trade-off in a changing environment? In this paper we present experimental evidence in an “observe or bet” task, comparing human behavior in a changing environment to their behavior in an unchanging one. We present a Bayesian analysis of the observe or bet task and show that human judgments are consistent with that analysis. However, we find that people’s behavior is most consistent with a Bayesian model that assumes a rate of change that is higher than the true rate in the task. We argue that this tendency is the result of asymmetric consequences: assuming that the world changes more often than it really does is not very costly, whereas assuming a too-low rate of change can carry much more severe consequences.
In most everyday decisions we learn about the outcomes of alternative courses of action through experience: a sampling process. Current models of these decisions from experience do not explain how the sample outcomes are used to form a representation of the distribution of outcomes. We overcome this limitation by developing a new and simple model, the Exemplar Confusion (ExCon) model. In a novel experiment, the model predicted participants’ choices and their knowledge of outcome probabilities, when choosing among multiple-outcome gambles in sampling and feedback versions of the task. The model also performed at least as well as other leading choice models when evaluated against benchmark data from the Technion Prediction Tournament. Our approach advances current understanding by proposing a psychological mechanism for how probability estimates arise rather than using estimates solely as inputs to choice models.
A basic challenge in decision-making is to know how long to search for information, and how to adapt searchprocesses as performance, goals, and the nature of the task environment vary. We consider human performance on two experiments involving a sequence of simple multiple-cue decision-making trials, which allow search to be measured, and provide feedback on decision accuracy. In both experiments, the nature of the trials changes, unannounced, several times. Initially minimal search is required, then more extensive search is required, and finally only minimal search is again required to achieve decision accuracy. We find that people, considered both on aggregate, and as individuals, are sensitive to all of these changes. We discuss the theoretical implications of these findings for modeling search and decision-making, and emphasize that they show adaptation to an external error signal must be accompanied by some sort of internal self-regulation in any satisfactory account of people’s behavior.
An experiment examined two aspects of performance in a multi-attribute inference task: i) the effect of stimulus presentation format (image or text) on the adoption of decision strategies; and ii) the ability of an evidence accumulation model, which unifies take-the-best (TTB) and rational(RAT) strategies, to explain participants’ judgments. Presentation format had no significant effect on strategy adoption at a group level. Individual level analysis revealed large intraparticipant consistency, including some participants who consistently changed the amount of evidence considered for a decision as a function of format, but wide inter-participant differences. A unified model captured these individual differences and was preferred to the TTB or RAT models on the basis of the minimum description length model selection criterion.
The extent to which a low probability event can be imagined appears to increase the weight attached to the possibility of that event occurring. Two experiments tested contrasting accounts of how this ‘imagability’ of events is enhanced. The experiments used negative (e.g. suffering the side effect of a vaccine) and positive (e.g. winning a lottery) low probability events. Both experiments found strong support for the frequency format account, whereby imagability is enhanced through the use of frequency formats for conveying statistical information (e.g., 20 out of 2000). However, only limited support was found for ‘exemplar-cuing theory’ (J.J. Koehler & L. Macchi, 2004), which proposes two distinct mechanisms for the generation of instances. Overall, the results support the claim that the imagability of outcomes plays a role in thinking about low probability events, but question the underlying mechanisms specified by exemplar cuing theory for mediating such effects.
Previous studies suggest improved learning when participants actively intervene rather than passively observe the stimuli in a judgment task. In two experiments the authors investigate if this improvement generalizes to multiple cue judgment tasks where judgments may be formed from abstract knowledge of cue-criterion relations or exemplar memory. More specific hypotheses were that intervention in learning should improve performance over observation, and that improvement should be associated with a relative shift from exemplar memory to cue abstraction. In contrast to previous studies, in a multiplecue judgment task with binary cues and continuous criterion, there was poorer learning with intervention than observation, and participants actively experimenting more produced poorer judgments. The results suggest that intervention may distract from efficient exemplar encoding and improvement may be limited to tasks efficiently addressed by cue-abstraction.