One of the most destabilizing findings in behavioral economics emerged from a simple observation: ask people to choose between two gambles, and they'll pick A. Ask them to price those same gambles, and they'll value B higher. This isn't noise. It's systematic, replicable, and profound in its implications.

Preference reversals violate the most basic axiom of rational choice theory—that preferences exist independently of how we ask about them. The reversal phenomenon, first documented by psychologists Lichtenstein and Slovic in the 1970s and later replicated in consequential market settings, suggests something unsettling: the preferences we measure may be partly created by our measurement procedures. This challenges the entire edifice of revealed preference theory.

For behavioral researchers and policy designers, preference reversals aren't just a curiosity—they're a window into how human valuation actually works. Understanding the mechanisms behind these reversals transforms how we interpret experimental data, design choice architectures, and think about the welfare implications of our interventions. The disagreement between choice and valuation isn't a bug in human cognition. It's a feature that reveals the constructed nature of preference itself.

The dominant explanation for preference reversals centers on what Slovic, Griffin, and Tversky called the compatibility principle: the weight of an attribute is enhanced when it is compatible with the response mode. When you ask someone to state a price, monetary attributes become more salient. When you ask them to choose, probability and other qualitative features gain weight.

Consider the classic P-bet versus $-bet paradigm. The P-bet offers a high probability of winning a modest amount—say, 35/36 chance of winning $4. The $-bet offers a low probability of winning a larger amount—say, 11/36 chance of winning $16. In choice, most people prefer the P-bet; its near-certainty dominates. In pricing, most people value the $-bet higher; its dollar amount anchors their valuation.

This isn't mere framing. The compatibility effect reflects deep features of how attention and weighting operate during judgment. Pricing tasks activate numerical comparison processes that privilege quantitative attributes expressed in compatible units. Choice tasks engage more holistic evaluation where qualitative features like probability can be directly weighed against each other.

The experimental evidence is remarkably robust. Tversky, Slovic, and Kahneman demonstrated that these reversals persist even when subjects are shown their inconsistency and given opportunities to revise. Market incentives don't eliminate them. Expertise doesn't eliminate them. The reversals emerge from the cognitive architecture of evaluation itself.

What makes this finding so consequential is its generality. Scale compatibility effects appear across domains far beyond gambling: in job candidate evaluation, consumer product assessment, environmental valuation, and legal damage awards. Wherever different elicitation methods make different attributes salient, systematic disagreement follows.

Takeaway

The method of asking doesn't just measure preferences—it actively shapes which attributes receive decisional weight, producing systematic divergence between choice and valuation.

If preferences were stable internal states waiting to be revealed, then different measurement methods should converge on the same underlying values. Preference reversals demonstrate that this convergence fails systematically. The alternative model—constructed preferences—holds that preferences are built during the elicitation process itself, not retrieved from some internal repository.

Slovic's constructive view treats preference as the output of a context-sensitive computation. When faced with a valuation task, people don't consult a pre-existing utility function. They assemble a response using whatever information, heuristics, and reference points the task makes available. Different tasks invoke different assembly processes, producing different outputs.

This framework explains why preference reversals are so resistant to correction. You cannot eliminate the construction process through instruction or incentive because construction is the process. Asking people to be more careful or consistent doesn't help if their cognitive system genuinely produces different preferences under different elicitation conditions.

The constructed preferences model has profound implications for how we interpret revealed preference data. Standard welfare economics assumes that choices reveal stable underlying preferences that can be used as the normative benchmark for policy evaluation. If preferences are constructed, this interpretation collapses. Which preference counts—the one revealed in choice or the one revealed in pricing?

Recent neuroeconomic research supports the construction hypothesis. Brain imaging studies show that different valuation tasks activate partially distinct neural circuits. Choice and pricing don't merely differ in their outputs; they differ in the underlying computational processes that generate those outputs. Preference isn't a single thing being measured differently—it's multiple computations producing genuinely different results.

Takeaway

Preferences aren't retrieved from memory like facts—they're computed on the fly using context-dependent processes, making the elicitation method a constitutive part of what gets 'revealed.'

Mechanism design theory traditionally assumes that agents have well-defined preferences that can be elicited through appropriately structured incentive schemes. Preference reversals challenge this assumption directly. If choice and pricing disagree, which response should the mechanism designer treat as the 'true' preference?

This question is not merely philosophical. Auction design, for instance, depends critically on assumptions about how bidders value items. If valuations are constructed during the bidding process itself, then different auction formats may not just elicit different bids—they may create different underlying valuations. The format becomes constitutive, not merely revelatory.

Consider contingent valuation methods used in environmental economics and legal damage assessment. These methods ask people to state willingness-to-pay for non-market goods like clean air or species preservation. The constructed preferences literature suggests that the numbers people provide are highly sensitive to question format, ordering effects, and anchoring. The 'value' of an environmental good may be as much a product of the survey instrument as of underlying preferences.

For policy design, this creates a genuine dilemma. We cannot simply abandon preference elicitation—we need some basis for evaluating interventions. But we must acknowledge that our measurements partly constitute what they measure. The solution isn't to find the 'right' elicitation method; it's to understand how different methods produce different constructions and design systems that account for this construction process.

Some researchers propose using multiple elicitation methods and examining the pattern of agreement and disagreement. Others suggest that choice should be privileged over stated valuation because choices have consequences that impose discipline. But neither approach fully resolves the normative puzzle. Mechanism design in a world of constructed preferences requires explicit theorizing about which constructions align with the welfare objectives we actually care about.

Takeaway

When preferences are constructed rather than revealed, mechanism design shifts from eliciting true values to choosing which construction process aligns with the welfare objectives being pursued.

Preference reversals expose a fundamental tension in how we think about human valuation. The gap between choice and pricing isn't a failure of rationality to be corrected—it's a window into how preferences actually form. They emerge from the interaction between cognitive processes and elicitation contexts, constructed rather than discovered.

For behavioral researchers, this demands methodological humility. Our measurements don't neutrally observe preferences; they participate in constructing them. For policy designers, it demands explicit normative theorizing about which constructions serve human welfare. The revealed preference framework offers no escape from these judgments.

The preference reversal literature ultimately points toward a richer science of valuation—one that takes the construction process seriously and designs institutions that work with, rather than against, the contextual nature of human preference. Better systems begin with honest acknowledgment of what preferences actually are.