Alessandra Cillo

Human-in-the-loop (HiL) systems integrate human judgment with algorithmic output, combining computational processing with human oversight and expertise. Despite their growing prevalence as sources of advice in organizational decision making, the extent to which decision makers value HiL advice remains underexplored, particularly compared to the extensive literature on purely algorithmic advice. Using the Judge Advisor System paradigm, we compare advice from three sources, HiL systems, algorithms, and human experts, and systematically manipulate the provision of explicit accuracy information. Our findings are threefold. First, participants hold higher expectations of accuracy for HiL systems relative to algorithms and human experts, yet this does not lead to higher utilization of HiL advice. Second, explicit accuracy information increases advice utilization for all sources: more accurate sources are used more frequently, regardless of type. Third, within any given source, higher perceived accuracy correlates with greater advice utilization. These results suggest a clear path forward for organizations: HiL systems will be valued when they deliver on their promise of superior accuracy by leveraging complementary human and algorithmic capabilities. Organizations should focus on designing HiL systems that combine the unique strengths of each source, such as algorithmic data processing with human judgment in ambiguous environments, and on transparently communicating accuracy information. Successful adoption of HiL systems requires empirical evidence of accuracy gains in specific contexts, not assumptions about decision maker preferences for HiL systems.

Catastrophic events, which are becoming increasingly frequent due to climate change, have significant negative impacts on housing. Yet, decision-makers remain puzzled by the low uptake rates for insurance coverage against these events. In this paper, we investigate whether this phenomenon is driven by individuals having miscalibrated beliefs about the likelihood of such events occurring. Additionally, we examine the impact of an information treatment designed to correct these perceived probabilities. Our experimental results show that participants in the treatment respond as expected, adjusting their beliefs accordingly. Moreover, these belief shifts influence participants’ demand for information on the topic and their interest in purchasing insurance. However, a follow-up survey conducted two months after the main experiment reveals that the effects of the treatment are short-lived, dissipating entirely within this period. Moreover, the information treatment shifts estimates about past mortality due to circulatory diseases, which are not closely related to the content of the information treatment, although more modestly. These two insights suggest ways of improving survey design in the field of information provision experiments. Our overall findings provide important insights for policymakers, highlighting the transient nature of the treatment-induced effects.

Many factors influence how individuals discount future payouts. A robust factor is the magnitude effect, whereby a small payout is discounted more than a larger one. But when individuals are confronted with more than one future payout, what drives the magnitude effect? Using a ceteris paribus design, we find that the discount rate decreases with the sum of cash flows and the highest cash flow. We also consider a separable model with cash-flow specific discount rates, but find mixed support for it. We also observe decreasing impatience, and propose various parametric models to predict present equivalents. To test these models out-of-sample, we design a retention bonus whose goal is to keep employees as highly motivated as possible during an extended period of time. All models that account for magnitude effect do a good out-of-sample job, whereas the models that merely account for decreasing impatience do not. Thus, our research can aid individuals, managers, and regulatory agencies better predict how individuals perceive financial offerings.

The attention paid to the role of money as a store of privacy is increasing. In a monetary transaction, full privacy protection coincides with anonymity. In such situations, an empirical question arises: Is anonymity relevant in shaping the demand for money? We attempt to answer this question through laboratory experiments. The results show that anonymity matters and increases the overall appeal of a medium of payment, and that this effect is stronger for risk-prone individuals. Moreover, the trade-off between the two properties of liquidity and return is relatively high - to accept higher illiquidity risks, individuals require a more-than-proportional increase in the expected return. In general, the experiments suggest that the future attractiveness of alternative currencies depends on whether the three properties of money are mixed in a way that is consistent with the individual's preferences.

In economics, models of decision-making under risk are widely investigated. Since many empirical studies have shown patterns in choice behavior that classical models fail to predict, several descriptive theories have been developed. Due to an evident phenotypic heterogeneity, obsessive compulsive disorder (OCD) patients have shown a general deficit in decision making when compared to healthy control subjects (HCs). However, the direction for impairment in decision-making in OCD patients is still unclear. Hence, bridging decision-making models widely used in the economic literature with mental health research may improve the understanding of preference relations in severe patients, and may enhance intervention designs. We investigate the behavior of OCD patients with respect to HCs by means of decision making economic models within a typical neuropsychological setting, such as the Cambridge Gambling Task. In this task subjects have to decide the amount of their initial wealth to invest in each risky decision. To account for heterogenous preferences, we have analyzed the micro-level data for a more informative analysis of the choices made by the subjects. We consider two influential models in economics: the expected value (EV), which assumes risk neutrality, and a multiple reference points model, an alternative formulation of Disappointment theory. We find evidence that (medicated) OCD patients are more consistent with EV than HCs. The former appear to be more risk neutral, namely, less sensitive to risk than HCs. They also seem to base their decisions on disappointment avoidance less than HCs.

Risk analysts are often concerned with identifying key safety drivers, that is, the systems, structures, and components (SSCs) that matter the most to safety. SSCs importance is assessed both in the design phase (i.e., before a system is built) and in the implementation phase (i.e., when the system has been built) using the same importance measures. However, in a design phase, it would be necessary to appreciate whether the failure/success of a given SSC can cause the overall decision to change from accept to reject (decision significance). This work addresses the search for the conditions under which SSCs that are safety significant are also decision significant. To address this issue, the work proposes the notion of a 0-importance measure. We study in detail the relationships among risk importance measures to determine which properties guarantee that the ranking of SSCs does not change before and after the decision is made. An application to a probabilistic safety assessment model developed at NASA illustrates the risk management implications of our work.

Risk-informed decision making is often accompanied by the specification of an acceptable level of risk. Such target level is compared against the value of a risk metric, usually computed through a probabilistic safety assessment model, to decide about the acceptability of a given design, the launch of a space mission, etc. Importance measures complement the decision process with information about the risk/safety significance of events. However, importance measures do not tell us whether the occurrence of an event can change the overarching decision. By linking value of information and importance measures for probabilistic risk assessment models, this work obtains a value-of-information-based importance measure that brings together the risk metric, risk importance measures, and the risk threshold in one expression. The new importance measure does not impose additional computational burden because it can be calculated from our knowledge of the risk achievement and risk reduction worth, and complements the insights delivered by these importance measures. Several properties are discussed, including the joint decision worth of basic event groups. The application to the large loss of coolant accident sequence of the Advanced Test Reactor helps us in illustrating the risk analysis insights.

This work addresses the early phases of the elicitation of multiattribute value functions proposing a practical method for assessing interactions and monotonicity. We exploit the link between multiattribute value functions and the theory of high dimensional model representations. The resulting elicitation method does not state any a-priori assumption on an individual's preference structure. We test the approach via an experiment in a riskless context in which subjects are asked to evaluate mobile phone packages that differ on three attributes. (C) 2015 Elsevier B.V. and Association of European Operational Research Societies (EURO) within the International Federation of Operational Research Societies (IFORS). All rights reserved.

This paper reports a new test of intransitive choice using individual measurements of regret-and similarity-based intransitive models of choice under uncertainty. Our test is tailor-made and uses subject-specific stimuli. Despite these features, we observed only a few intransitivities. A possible explanation for the poor predictive performance of intransitive choice models is that they only allow for interactions between acts. They exclude within-act interactions by retaining the assumption that preferences are separable over states of nature. Prospect theory, which relaxes separability but retains transitivity, predicted choices better. Our data suggest that descriptively realistic models must allow for within-act interactions but may retain transitivity.

We study a mean-risk model derived from a behavioral theory of Disappointment with multiple reference points. One distinguishing feature of the risk measure is that it is based on mutual deviations of outcomes, not deviations from a specific target. We prove necessary and sufficient conditions for strict first and second order stochastic dominance, and show that the model is, in addition, a Convex Risk Measure. The model allows for richer, and behaviorally more plausible, risk preference patterns than competing models with equal degrees of freedom, including Expected Utility (EU), Mean-Variance (M-V), Mean-Gini (M-G), and models based on non-additive probability weighting, such as Dual Theory (DT). In asset allocation, the model allows a decision-maker to abstain from diversifying in a positive expected value risky asset if its performance does not meet a certain threshold, and gradually invest beyond this threshold, which appears more acceptable than the extreme solutions provided by either EU and M-V (always diversify) or DT and M-G (always plunge). In asset trading, the model provides no-trade intervals, like DT and M-G, in some, but not all, situations. An illustrative application to portfolio selection is presented. The model can provide an improved criterion for mean-risk analysis by injecting a new level of behavioral realism and flexibility, while maintaining key normative properties.