Foundations of Decision Theory: From Utility to Prospect Theory

The quest to understand the biological substrates of choice has fundamentally reshaped economic theory and neuroscience alike, birthing the interdisciplinary field of neuroeconomics. At its core, this endeavor seeks to replace the classical “black box” model of the mind—where inputs (stimuli) and outputs (behavior) are observed but internal processes remain hidden—with a mechanistic, biologically plausible account of decision computation. This transformation hinges on a critical re-examination of the very theories that have long governed our understanding of rational choice. Normative models, particularly expected utility (EU) theory, have provided a gold standard for rational behavior for over seven decades. However, a wealth of systematic violations of its axioms, meticulously documented by cognitive psychologists, revealed that human choice is not merely a noisy implementation of Euclidean optimization but is instead governed by distinct, lawful principles that deviate from normativity in predictable ways. The descriptive alternative, prospect theory (PT), stands as the most influential framework, offering a psychologically rich account centered on reference dependence, loss aversion, and probability weighting. These formal theories are not merely abstract mathematical constructs; they generate precise, testable hypotheses about the neural algorithms that implement valuation, learning, and selection. This chapter establishes the essential economic and psychological bedrock upon which neuroeconomics is built. We will critically dissect the axioms and mathematical structure of EU theory, contrast it with the key tenets of prospect theory, and explicitly outline how each theory’s components map onto potential neural substrates and experimental paradigms. By tracing the historical paradigm shift from purely behavioral economics to the neurobiological interrogation of choice, and by providing concrete case studies that re-analyze classic data and design model-contrasting experiments, we will set a rigorous foundation for the neuroscientific inquiries that define the remainder of this book.

The Normative Benchmark: Expected Utility Theory and Its Axioms

Expected utility theory, formalized by von Neumann and Morgenstern (1944) and built upon earlier work by Bernoulli, provides the canonical model of rational choice under risk. Its power derives from its axiomatic foundation: if a decision-maker’s preferences satisfy a set of intuitively appealing consistency rules, then those preferences can be represented by an expected utility function. This representation theorem elevates EU from a prescriptive rule to a necessary consequence of rationality itself. The theory operates in a well-defined space: a finite set of outcomes, $X = \{x_1, x_2, ..., x_n\}$, and lotteries (prospects) $L$ that assign probabilities $p_i$ to these outcomes, where $\sum p_i = 1$. The core primitive is a preference relation $\succeq$ over lotteries.

The four key axioms are: 1. Completeness: For any two lotteries $L_a$ and $L_b$, either $L_a \succeq L_b$, $L_b \succeq L_a$, or both (indifference). 2. Transitivity: If $L_a \succeq L_b$ and $L_b \succeq L_c$, then $L_a \succeq L_c$. 3. Continuity: If $L_a \succeq L_b \succeq L_c$, then there exists a probability $p \in [0,1]$ such that a lottery offering $L_a$ with probability $p$ and $L_c$ with probability $1-p$ is indifferent to $L_b$. 4. Independence (or Sure-Thing Principle): If $L_a \succeq L_b$, then for any third lottery $L_c$ and any $\alpha \in (0,1)$, the compound lottery $\alpha L_a + (1-\alpha) L_c \succeq \alpha L_b + (1-\alpha) L_c$. This axiom is the most behaviorally consequential; it states that mixing a common outcome ($L_c$) should not affect the preference between $L_a$ and $L_b$.

If these axioms hold, there exists a utility function $u: X \rightarrow \mathbb{R}$, defined up to positive affine transformations, such that for any lottery $L = \{(x_i, p_i)\}$, its utility is $U(L) = \sum p_i u(x_i)$. The decision-maker chooses the lottery with the highest $U(L)$. The linearity in probabilities is a direct mathematical consequence of the independence axiom. This linear aggregation is the hallmark of EU theory: the value of a prospect is a probability-weighted sum of the utilities of its possible outcomes.

Neurological Implications of the EU Framework

From a neurocomputational perspective, EU theory implies a specific two-stage architecture: 1. Outcome Valuation: The brain assigns a scalar value $u(x_i)$ to each possible outcome $x_i$. This value is assumed to be stable and context-independent. The neural substrate for this could be a distributed representation encoding the objective magnitude (e.g., monetary amount) and perhaps some intrinsic desirability, potentially involving ventromedial prefrontal cortex (vmPFC) or orbitofrontal cortex (OFC) which encode economic value signals. 2. Probability Integration: These outcome values are multiplied by their explicit, stated probabilities $p_i$ and summed. This linear weighting suggests a neural mechanism that treats probabilities as multiplicative factors on value signals. One candidate mechanism is a divisive normalization circuit where the activity representing a stimulus’s value is scaled by a signal representing its probability of occurrence, though the linearity constraint is strict.

The independence axiom, in particular, has profound neural consequences. It demands that the value computation for an option be separable from the context of other options. In neural terms, this implies that the value signal for a given outcome-probability pair should be invariant to the set of other lotteries being considered—a form of “value independence” at the representational level. Violations of independence, as we will see, suggest that value integration is not linear and that context modulates the perceived weight of probabilities or the subjective value of outcomes.

However, EU theory’s descriptive validity collapsed under the weight of experimental evidence. Two famous paradoxes, in particular, demonstrated systematic violations of its core axioms.

The Allais Paradox (1953) directly violates the independence axiom. Consider two pairs of choices:

• **Choice 1**: $A$: \$1M with certainty vs. $B$: \$5M with 0.1, \$1M with 0.89, \$0 with 0.01.
• **Choice 2**: $C$: \$1M with 0.11, \$0 with 0.89 vs. $D$: \$5M with 0.1, \$0 with 0.9.

Many people prefer $A$ (certainty) over $B$ in Choice 1, but prefer $D$ over $C$ in Choice 2. This is inconsistent. The two pairs are identical except for the common outcome (the \$0 in the second pair replaces the \$1M with 0.89 probability in the first). By independence, the preference between the risky options ($B$ vs. $D$) should be the same as between their modified versions ($A$ vs. $C$), but it is not. The presence of a sure thing in Choice 1 makes the risky option $B$ unattractive, a phenomenon known as the certainty effect.

The Ellsberg Paradox (1961) violates the assumption that probabilities are known and treated linearly. People exhibit a preference for known risks over unknown risks (ambiguity aversion). In the classic urn problem, an urn contains 90 balls: 30 red, and the remaining 60 are either black or yellow (unknown distribution). Subjects prefer a bet on “red” (known probability 1/3) over a bet on “black” (unknown probability) even though the expected number of black and yellow balls is equal. This suggests that subjective decision weights are not simply the stated probabilities but are influenced by the perceived ambiguity or source of those probabilities.

These paradoxes, among many others, revealed that human decision-making is reference-dependent (outcomes are valued as gains or losses relative to a reference point, not in absolute terms) and that probabilities are non-linearly transformed before being combined with outcomes. The normative edifice of EU theory,