The Gap Between Stated Preferences and Real World Behavior

25 concepts24 papers1 proposal

Imagine asking someone if they would donate a biological sample to a biobank, install a privacy-first contact-tracing app, or adhere to a daily preventative medication. In the vacuum of a survey or a controlled interview, most people express a high willingness to cooperate. They want to be helpful, healthy, and socially responsible.

However, when faced with the actual real-world choice, a profound disconnect often emerges. This phenomenon—known as hypothetical bias or the intention-behavior gap—demonstrates that stated preferences are frequently poor predictors of actual human behavior. Across public health, medicine, and technology adoption, relying on what people say they will do rather than what they actually do leads to misallocated resources and failed interventions. Understanding why this gap exists is critical to designing systems that work in the wild.

The Illusion of Stated Intent

Why do surveys mislead us? When individuals evaluate a hypothetical scenario, they operate in a low-stakes environment where choices are costless. This abstraction invites social desirability bias: the subconscious desire to project an idealized version of oneself.

For instance, in public health infrastructure research, Mingtao Huang et al. observed a stark discrepancy between the public's stated willingness to participate in biobanks and their actual real-world enrollment rates. Respondents want to support scientific progress in theory, but that enthusiasm evaporates when faced with the real-world friction of physical donation.

A similar pattern emerges during technological rollouts. When evaluating the adoption of digital health tools, Samuel Altmann et al. and Tianshi Li et al. discovered that high baseline survey support for COVID-19 contact-tracing apps failed to translate into real-world installations. In the abstract, downloading an app seems like a minor civic duty; in reality, the action triggers immediate anxieties regarding government trust, privacy, and battery drain that surveys fail to capture.

The Frictions of Reality: DCEs and Clinical Settings

To quantify human preferences, researchers frequently use Discrete Choice Experiments (DCEs), which force participants to make trade-offs between hypothetical product or treatment attributes. While mathematically elegant, DCEs often fail because they cannot simulate the visceral frictions of real life.

In healthcare, these frictions include active disease pressure, financial strain, and logistical complexity:

  • Cognitive and Emotional Stress: Xiaolan Yu et al. and Jinsong Geng et al. utilized DCEs to evaluate preferences among hypertension and diabetes patients, showing that while patients state clear priorities regarding clinical efficacy and convenience in surveys, these models fail to capture the messy reality of long-term adherence. This discrepancy becomes acute in life-or-death contexts. Jingyi Meng et al. noted that while cancer patients might prioritize quality of life in a quiet interview setting, the emotional panic of active disease can radically alter their actual choices.
  • Immediate Material Constraints: In another study on lung cancer treatments, Jingyi Meng et al. demonstrated that out-of-pocket affordability becomes the single most dominant factor in real-world choices, overriding abstract preferences for marginal survival benefits. Similarly, Alexis Larson et al. found that cancer survivors' stated desires for comprehensive follow-up care frequently conflict with real-world constraints like insurance coverage and scheduling conflicts.
  • Logistical Realities: A DCE by Yuhan Chen et al. on elderly vaccine preferences showed that stated priorities often shift dramatically when individuals are confronted with actual logistical barriers and immediate health risks.

Risk Perception and Social Suppressors

The intention-behavior gap is further widened by how individuals navigate risk and social pressure in real time. In a hypothetical survey, risks are abstract percentages; in reality, they are visceral and immediate.

For example, Ho-Ming Chris Wong et al. found that while prostate cancer patients might express a preference for oral medications in surveys, they often choose long-term injections in practice to avoid the daily psychological burden of adherence, even when injections carry higher cardiovascular risks. Risk perception also shifts based on active experience: Peter J. Schulz et al. discovered that clinicians' self-reported willingness to adopt medical AI varies wildly when moving from survey evaluations to high-pressure, real-time clinical procedures where patient outcomes are immediately on the line.

Furthermore, scenario-based experiments struggle to replicate the intense pressure of real financial gain or loss. In high-stakes financial environments, Mingfeng Tang et al. demonstrated that measuring behavioral intentions through hypothetical scenarios fails to accurately predict loan defaults on peer-to-peer lending platforms.

In public health, immediate social and physical frictions frequently suppress stated intent. Yvonne J. Yamanaka et al. documented how individuals might express a strong theoretical willingness to seek HIV treatment, yet retreat when faced with the real-world social stigma of a diagnosis. This pattern repeats across diverse interventions: Li Ye et al. highlighted a massive gap between high hypothetical interest in HIV Pre-Exposure Prophylaxis (PrEP) and the complex social realities of daily adherence, while Chuanyi Ning et al. found that public health interventions could successfully drive actual uptake of medical procedures like circumcision without necessarily altering participants' long-term stated willingness in surveys.

Even when individuals support an idea in theory—such as home-based malaria testing evaluated by Shamika Ranasinghe et al.—it remains highly uncertain whether they will accurately execute the medical tasks or follow treatment protocols when faced with a real-world illness.

The Danger of Relying on Proxy Metrics

When researchers and policymakers rely on stated intent as a proxy for behavior, they risk designing interventions that fail in the wild. This is particularly true when measuring digital engagement or public compliance.

In a randomized trial exploring how smokers interact with health information on social media, Jessica Liu et al. noted that the intent to "like" or "share" a post serves as a weak proxy that does not correlate with actual, long-term health behavior changes or smoking cessation. In the context of public health compliance, an analysis of Canadian vaccine hesitancy trends published in BMJ 2022 emphasized that self-reported willingness to comply with health guidelines is deeply susceptible to social desirability bias, making verified behavioral outcomes (such as actual vaccination records) essential for accurate modeling.

This speculative bias is not unique to healthcare. Even in market research for emerging technologies, the gap can distort industry expectations. Writing at the dawn of the natural language processing industry, Katharina Morik showed that market research conducted on non-existent technologies relies so heavily on hypothetical scenarios that a massive gap inevitably emerges between an enterprise's stated investment intent and its actual adoption behavior.

[Hypothetical Survey State] ───(Social Desirability Bias)───► High Stated Intent
                                                                     │
                                                          (Real-World Frictions)
                                                                     ▼
[Real-World Action State]   ◄───(Stigma, Cost, Logistics)────── Low Actual Behavior

Bridging the Gap in Research Design

To build more predictive models, modern research is shifting away from abstract surveys toward designs that either simulate real-world trade-offs or measure verified actions directly.

One approach is to refine how we model human decision-making by incorporating naturalistic cost-benefit trade-offs. Qianying Wu et al. developed a computational framework demonstrating that prosocial helping behavior cannot be accurately captured by imagined scenarios; instead, models must account for immediate physical risks and the bystander effect to predict real-world actions.

Another strategy involves changing the intervention medium itself to close the cognitive gap between imagination and reality. Clara Vandeweerdt et al. demonstrated that using immersive virtual reality to let hesitant individuals "experience" the community benefits of herd immunity was significantly more effective at shifting actual behavioral intentions than traditional text-and-image surveys.

Ultimately, the most reliable solution is to design studies that prioritize the measurement of actual behavior over intent from the ground up. For example, in evaluating adolescent mental health programs, Dan I. Lubman et al. structured their trial protocol to measure verified, real-world help-seeking actions rather than relying on self-reported willingness. By acknowledging that saying and doing are governed by entirely different psychological architectures, researchers can move past the illusion of stated preferences and build interventions that truly work in the wild.

Key Foundational Work

Religion and HIV in Tanzania: influence of religious beliefs on HIV stigma, disclosure, and treatment attitudes (2009) — contributed 1 concept elements to this direction.
Demand and Requirements for Natural Language Systems - Results of an Inquiry (1983) — contributed 1 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 24 papers
  2. Version 1 · 2026-10-04 · 39 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_hypothetical_bias_b49eda69,
  title = {hypothetical bias},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/hypothetical-bias/rdr_b49eda69d8da41429effb562e5103d2f},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 2, frozen text):

@misc{lacuna_hypothetical_bias_b49eda69_v2,
  title = {hypothetical bias},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/hypothetical-bias/rdr_b49eda69d8da41429effb562e5103d2f/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake