TABS 2026 CRP Results

Three-Tier Selection Methodology

The CRP dataset selects N=200 responses from Prolific Accepted participants using a three-tier quality-based strategy. Higher tiers are automatically included; remaining slots are filled by quality-ranked responses from lower tiers.

• Tier 1 - Conservative Clean (auto-include): All responses passing every quality gate: 3/3 IRI attention checks, duration ≥ 540s, reCAPTCHA ≥ 0.5, no straightlining, no auth flags (N=79)
• Tier 2 - Flexible Clean surplus (auto-include): Responses passing basic quality (all 3 IRIs + duration ≥ 480s) that did not qualify for Tier 1 (N=37)
• Tier 3 - Quality-ranked fill: Remaining Prolific Accepted responses ranked by a 100-point composite quality score, selected until N=200 is reached

Quality Scoring

Each response receives a composite quality score derived from six independent indicators, scaled to a 100-point range. The six indicators are: IRI attention check performance (35 pts), survey completion duration (20 pts), reCAPTCHA bot-detection score (15 pts), Prolific authentication flags (15 pts), full straightlining detection (8 pts), and partial straightlining / within-person response variance (7 pts).

All N=200 responses are included in the CRP dataset. The quality score determines Tier 3 selection priority and supports sensitivity analysis across sample definitions (Conservative Clean vs. full CRP sample).

De-Identification Protocol

The public dataset is de-identified following the NIST 5-step Expert Determination protocol. All direct identifiers (Prolific PIDs, IP addresses, geolocation data) are removed prior to release. Indirect identifiers (organization size, industry sector) are retained at categorical granularity to support meaningful analysis without enabling re-identification.

The de-identification process is documented in the analysis scripts and can be independently verified. See the Open Data & Reproducibility page for full details.