Causal Literacy

How to cite

List, John. 2026. "Causal Literacy." AEA RCT Registry. AEARCTR-0018561. https://doi.org/10.1257/rct.18561-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

John List*

* Primary investigator

Registry ID
AEARCTR-0018561
DOI
https://doi.org/10.1257/rct.18561-1.0
First registered
Last updated
Status
In development
Trial dates
2026-05-11 – 2026-05-31
Countries
Private
JEL codes
C93, D83, D91, I20
Registry record
http://www.socialscienceregistry.org/trials/18561

Abstract

Causal literacy, the ability to distinguish correlation from causation, underlies sound decision-making across health, finance, policy, and parenting. Yet no validated instrument measures this foundational skill. This study addresses this gap by using a field experiment to provide the first population-level estimates of causal literacy and its responsiveness to psychological and economic incentives. I motivate the empirical structure with a simple human capital model in which causal reasoning comprises six distinct cognitive operations, each with a corresponding failure mode: confounding, selection bias, reverse causation, proxy distortion, base rate neglect, and survivorship bias. The model generates both the structure of the assessment instrument (the Causal Literacy Assessment) and testable predictions about psychological and economic incentives.

Intervention

Field experiment testing whether financial incentives and social visibility improve causal reasoning performance.

Experimental design

RCT testing whether financial incentives and social visibility improve causal reasoning performance and reduce the application gap. 2×2 factorial design: Incentive (High $5 / Low $0.50) × Visibility (Private / Social). N=2,000 (500 per cell).

Design details

Randomization method
Simple random assignment (1:1:1:1 allocation across four cells) via Qualtrics’ built-in randomizer. Target: N=500 per cell.
Randomization unit
The randomization unit is an individual.
Clusters
2000
Observations
2000
Arms
N=500 per cell
Primary outcomes
Primary Outcomes • Causal Literacy Score (Abstract): Part D score (0-20), percentage correct on 20 scenario-based questions testing recognition of six “data traps. • Causal Literacy Score (Personal): Part D-2 score (0-20), percentage correct on 20 first-person decision scenarios. • Application Gap: (Part D %) − (Part D-2 %), measuring bias blind spot Secondary Outcomes • Scientific Reasoning Score: Part C score (0-12), vignette-based pseudoscience rejection and science trust • Numeracy Score: Part B score (0-6) • Trap-Specific Scores: Performance on each of 6 traps (Hidden Influence, Self-Selection, Sequence, Proxy, Deluge, Survivorship) • Overconfidence Measure: Residual from regressing E1/E2 self-assessment on actual Part D performance Process Measures (Mediators) • Effort (Time): Total time spent on assessment (seconds); time per question • Attention: Performance on attention checks (C13, Q31); proportion passing both • Completion Rate: Proportion completing all sections • Response Consistency: Within-trap correlation across Parts D and D-2 Real-World Outcomes Battery (Outcomes Correlates Module) • Financial Behavior Index: Investment approach (evidence-based vs. recency/social); binary loss-to-scam indicator (MLM, high-yield, crypto scam, paid course); correct identification of survivorship bias in advertised 18% return (scored item). Composite: sum of financially rational responses. • Health Behavior Index: Annual spend on non-prescribed supplements ($); use of pseudoscientific health products (homeopathy, detox, energy healing, unproven treatments); correct antibiotic completion response; correct identification of self-selection confound in supplement observational study. Composite: sum of evidence-consistent health behaviors. • Career Outcomes Index: Employment level (ordinal: entry to C-suite/self-employed); frequency of data-driven decision-making in role (ordinal); number of promotions to date; correct identification of attribution error in product-launch story. Composite: sum of career-advancement indicators. • Consumer Susceptibility Index: Influencer/celebrity endorsement effect on purchases (4-point); willingness to pay premium for “natural”/“chemical-free” labels (5-point); correct evaluation of before/after ad as weak causal evidence; reason for last brand switch. Higher scores indicate greater susceptibility to non-causal marketing cues. • Political & Parenting Causal Reasoning: Correct identification of confound in politician’s unemployment claim; correct identification of omitted variable bias in early-voting/fraud graph; spend on child educational enrichment ($); endorsement of learning-styles myth (scored); preference for value-added vs. raw test scores in school choice; correct identification of selection bias in early-childhood program graduation statistic. Social Media Module • Daily Platform Use (SM_Use): Q1 grid — minutes per day on Facebook, Instagram, TikTok, YouTube, Reddit, X/Twitter, Snapchat (7-point ordinal per platform). Aggregate: total daily social media hours (sum across platforms, midpoint-coded). Also constructed: TikTok/Instagram “short-form” sub-index. • Age at First SM Use (SM_AgeOnset): Q2 — ordinal (younger than 10 through older than 25). Recoded to approximate midpoint age for continuous analysis. Pre-adolescent onset (≤13) flagged as binary indicator for moderation analysis. • Consumption Style (SM_Passive): Q3 — binary indicator: passive scrolling (“mostly scroll without a particular goal”) vs. purposeful use. Passive = 1. • Misinformation Exposure (SM_MisInfo): Q4 — 7-point Likert (Never to Very frequently). Used as covariate and as moderator (high exposure predicted to correlate with lower causal literacy, especially on Proxy and Hidden Influence traps).

Cited by

No citation data available for this registration yet.

Keywords