Behavioral Bias Lab

Transparency

Methodology

This page describes how experiments are conducted, what data is stored, and the limitations that affect interpretation. We believe transparency strengthens trust.

Current Sample Size

5,080

anonymous responses across all experiments

By participating, you consent to anonymous data collection.

No personal data is stored. This means no names, email addresses, IP addresses in a form that could identify you, device fingerprints, or behavioral tracking. Participation is entirely voluntary and you may close your browser at any time without submitting data.

We store only: (1) the experiment slug, (2) your response value, (3) the experimental variant you were assigned, and (4) a randomly generated session ID that is created fresh each browser session and cannot be linked to any person.

No cookies are used. No analytics providers are embedded. This site operates without any third-party tracking.


Data Collection

Responses are submitted via a standard HTTPS POST request to our API. Each request is rate-limited to 30 responses per 10-minute window per IP address to prevent data flooding.

The database schema for each response:

{
  experimentSlug: string,   // e.g. "anchoring"
  variant:        string?,  // e.g. "high-anchor"
  choiceValue:    string,   // the user's response
  numericValue:   float?,   // numeric form where applicable
  sessionId:      string,   // random, ephemeral, non-identifying
  createdAt:      datetime
}

Variant assignment is performed client-side using Math.random(). This produces roughly equal group sizes over time but does not guarantee exact balance in small samples.


Limitations

Self-selection bias

Participants choose to visit this platform, likely because they have prior interest in psychology or behavioral economics. This is not a random sample of any population. Results may underestimate biases (curious, reflective individuals) or overestimate them (users deliberately testing their limits).

Non-random sample

The participant pool is an internet convenience sample — skewed toward people with access to computers, English fluency, and higher education. Findings cannot be generalized to the global population.

Social desirability & demand characteristics

Knowing the experiment is about cognitive biases may influence responses. Participants aware of the anchoring effect, for example, may attempt to compensate, potentially attenuating or reversing the effect.

No experimental control

Unlike laboratory studies, we cannot control the environment in which participants respond: distractions, time pressure, mood, and prior task engagement all vary freely. This introduces noise that reduces statistical power.

Sample size variability

Small experiments (N < 30 per condition) should be interpreted with caution. Aggregate statistics may be unstable and confidence intervals wide. We display raw counts to help users assess reliability.

Client-side randomization

Condition assignment uses JavaScript's Math.random(), which is pseudorandom and seeded by the browser. In theory, a determined user could manipulate their condition assignment. In practice this is unlikely to affect aggregate results meaningfully.


Replication Fidelity

Each experiment is based on a canonical paradigm from the behavioral economics literature. Where possible, we use the original stimuli or close adaptations. Deviations are noted below:

ExperimentOriginalAdaptation
AnchoringSpinning wheel + estimation (T&K 1974)Random number display; same UN Africa question
FramingAsian Disease Problem (T&K 1981)Verbatim replication
Loss AversionMixed gambles (K&T 1979)3 rounds with varying stake sizes
OverconfidenceCalibration curves (L,F&P 1982)Short 5-question quiz; prediction slider