The Behavioral and Computational Nature of Range Dependent Decisions – RANGE

The RANGE program developed an integrated experimental method to analyse range-normalization effects in risky choice. The objective is not only to record decisions, but to manipulate informational, symbolic and experiential context in a controlled way, in order to identify how individuals encode value and adjust preferences.

The method combines four complementary components.

1. Controlled manipulation of experienced ranges

In sampling tasks or lottery environments, participants encounter extreme outcomes (maximum gains/losses) whose dispersion is experimentally controlled. These extrema define an experienced range progressively incorporated into internal representations. We observe how this range influences subsequent valuation of identical prospects presented after different learning histories.

2. Crossed symbolic/experiential variation

Hybrid tasks combine symbolic description (text, probability, numeric cue, visual signal) and direct experience (repeated draws, feedback, sampling). Symbolic information acts as a contextual anchor, modifying the weighting of experienced value. This design isolates two mechanisms: divisive compression of experiential value and re-centering of utility around symbolic anchors.

3. Dynamic elicitation and endogenous preference instability

Preferences are not elicited once: they are measured at multiple stages of learning and choice, allowing us to detect the endogenous emergence of incompleteness, instability or apparent indifference. Instability is not treated as random noise: we test whether it correlates with contextual dispersion, meaning that incompleteness is produced by normalization, not by lack of information or cognitive limitation.

4. Strategic transmission and bilateral interaction

A subset of tasks extends manipulation into the strategic domain: participants evaluate, negotiate or trade assets after internal normalization. We measure how individual misvaluation becomes institutional, affecting expected surplus, bargaining outcomes and equilibrium allocations through range-dependent valuation scales.

Scientific Justification

This method turns RANGE into an experimentally testable theory: normalization is not assumed, but provoked, parameterized, tracked over time, and measured through its effects on preferences, choice and strategic interaction. The protocol identifies a central mechanism: subjective value is not absolute, but continuously recalibrated by recent ranges, symbolic cues and informational structure.Les décisions individuelles déterminent des enjeux majeurs, qu’ils soient financiers, sanitaires, environnementaux ou sociaux. La plupart de nos choix s’effectuent sous incertitude et exigent des arbitrages continus. Pourtant, les modèles économiques standard supposent que les préférences sont stables, complètes et définies sur une échelle absolue de valeur. Les résultats empiriques récents montrent au contraire que les individus évaluent les perspectives de manière relative au contexte

RANGE Project – Consolidated Results (2021–2024)

1. Context determines experiential value

In Garcia, Lebreton, Palminteri & Bourgeois-Gironde (2023, Nature Human Behaviour), hybrid choices reveal systematic underweighting of experiential value when a symbolic cue is present. Symbolic information acts as a contextual range-setter, compressing experiential valuation and re-centering utility. Subjective value is therefore range-relative, not absolute.

2. Incomplete preferences from contextual dispersion

In Arlegi, Bourgeois-Gironde & Hualde (2022, Journal of Economic Behavior & Organization), experimentally elicited preferences become incomplete or unstable when informational range widens or when attribute salience differs. Incompleteness correlates with contextual dispersion, indicating endogenous instability produced by normalization, not by lack of information.

3. Misvaluation propagates through interaction

In Bourgeois-Gironde & Czupryna (2022, Advances in Social Simulation), bilateral trade with endogenous misvaluation shows that early normalization biases amplify through interaction, altering expected surplus, bargaining outcomes and equilibrium allocations. Range-dependent valuation thus has institutional consequences, not just intrapersonal effects.

4. Epistemic inequality narrows valuation ranges

Work with Alda Mari (2024, submitted) shows that informational networks partition valuation windows. RANGE predicts higher incompleteness and misvaluation under epistemic inequality, because narrow informational ranges reduce discriminability and increase apparent indifference. This links normalization to network topology.

5. Normalization affects mechanism design

Joint work with Daniel Benoliel (2023–24, accepted/under review, Illinois Law Review; Yale International Law Journal) shows that valuation ranges influence disclosure and screening outcomes. When risks or externalities are unevenly known, internal normalization biases generate strategic distortions, relevant for innovation and sustainability incentives.

Unified Implication

Across hybrid valuation (Garcia et al. 2023), incomplete elicitation (Arlegi et al. 2022), bilateral misvaluation (Czupryna & BG 2022), epistemic segmentation (Mari & BG 2024) and regulatory screening (Benoliel & BG 2023–24), the same regularity emerges: utility is dynamically range-normalized. Context, information structure, temporal history and interaction shape subjective valuation, producing systematic weighting biases, endogenous incompleteness, welfare-relevant misvaluation and regulatory distortions. RANGE is therefore a behavioural, institutional and regulatory principle supported by empirical work in experimental economics, decision theory, social simulation and legal-economic mechanism design.

The RANGE programme now provides convergent evidence that valuation is implemented through context-sensitive normalization mechanisms, operating at multiple scales: within individual decision processes, across informational networks, and in strategic and institutional environments. The next stage is to develop an integrated theory of range-normalized preferences capable of explaining not only isolated behavioural anomalies, but also systemic distortions of markets, welfare, learning, and regulation. A first research axis concerns the mechanistic foundations of range normalization. Existing models (divisive normalization, adaptive utility scaling, Bayesian evidence accumulation) remain partly descriptive. The RANGE results suggest that history-dependence and symbolic anchoring must be made endogenous to representations and learning dynamics, rather than treated as exogenous framing. A complete model should incorporate: (i) temporal aggregation of experienced extrema, (ii) symbolic or linguistic priors, and (iii) attentional allocation as a dynamic scaling mechanism. The integration of reinforcement learning and normalization theory is especially promising for capturing dynamic preference instability, incomplete ordering, and endogenous valuation drift. A second research axis concerns the extension of RANGE mechanisms to institutional design. The empirical and modeling results already show that internal normalization biases propagate through bilateral and strategic interaction (Bourgeois-Gironde & Czupryna 2022) and that agents’ endogenous valuation scales influence disclosure, reporting and screening outcomes (Benoliel & Bourgeois-Gironde 2023–24). This motivates a systematic exploration of mechanisms, contracts, and regulatory instruments that remain robust under range-normalized valuation. We need to understand how information design, default rules, risk disclosure, and adaptive scoring systems could compensate for endogenous valuation drift. This opens a major research avenue for risk-reflexive governance, where regulatory instruments are explicitly calibrated to the behavioural structure of normalized utility, not to idealized absolute preferences. A third perspective is network-theoretic. If epistemic networks partition informational range (Mari & Bourgeois-Gironde 2024), then preference completeness, discriminability and welfare depend on network topology. RANGE predicts that agents embedded in narrow informational neighbourhoods will exhibit higher incompleteness, lower discriminability and stronger susceptibility to symbolic anchors. This provides a direct empirical link between cognitive normalization and epistemic inequality. A natural next step is to study how altering the topology of information access (connectivity, strategic transparency, public signals) stabilizes or destabilizes subjective valuation ranges, and how interventions on informational structure can substitute for costly monetary or contractual instruments.

The present project stems from two fundamental observations concerning economic decision-making: 1) the subjective value of an option is deeply affected by the other options presented simultaneously or in the recent past (context-dependence). 2) the subjective value of an option is different if the relevant information is explicitly communicated or implicitly inferred by trial-and-error (experience/description gap). The goal of the present project is to describe the behavioral and computational mechanisms underlying context-dependent decision-making using a combination of behavioral economics and computational cognitive science. Our hypotheses are that: 1) context-dependence in economic decision-making can be formalized as a range-adaptation process where the subjective value of an outcome is normalized as a function of the minimum and maximum outcomes encountered in a given situation 2) this process is stable across explicit information- and implicit information-based decision.