AI versus Human Sources of Earnings Risk and Perceived Economic Well-Being

This randomized online survey experiment examines whether the source of otherwise identical prospective earnings risk affects subjective ex-ante economic well-being. U.S. working adults recruited through Prolific are randomly assigned to imagine a 50% chance that, within the next year, some of their current paid-work tasks will be transferred either to an artificial intelligence (AI) system or to another person. If the transfer occurs, total earnings from work fall by 40% and remain at that lower level with no set end date. The objective probability, earnings consequence, timing, task content, and all other scenario features are held constant across conditions. The risk remains unresolved when outcomes are measured.

2026-08-24

2026-09-21

Financial satisfaction (0–10 scale), measured after the assigned hypothetical earnings-risk scenario while the risk remains unresolved.

Financial satisfaction is measured using a single 0–10 item, with higher values indicating greater satisfaction with the participant’s financial situation under the hypothetical unresolved earnings risk. This is the sole primary confirmatory outcome for the study. The adapted five-item CFPB financial well-being index is analyzed separately as convergent/supportive evidence and is not a second primary confirmatory outcome.

Adapted five-item CFPB financial well-being index; perceived economic position; financial threat; perceived fairness; redistribution and policy preferences; source appraisals (controllability, recoverability, breadth, objectivity, and attribution to own skills/abilities); and measures of how respondents mentally process the unresolved earnings risk.

The adapted CFPB financial well-being index consists of five items scored from 0 to 4 and summed to form a 0–20 index, with higher values indicating greater financial well-being. It provides convergent/supportive evidence for the primary hypothesis. Perceived economic position is measured using a 1–9 economic ladder and is analyzed as a related secondary outcome.

Financial threat is measured as the mean of five items, each scored from 1 to 5, with higher values indicating greater perceived financial threat. Fairness is measured using a researcher-developed single-item global fairness judgment. Financial threat and fairness are the two secondary confirmatory outcomes.

Redistribution and policy-preference outcomes, the remaining source appraisals, and risk-processing measures are exploratory. Source appraisals include perceived controllability, recoverability, breadth of impact, objectivity, and attribution to the participant’s own skills or abilities. Risk processing is measured using a categorical item describing how respondents mainly interpreted or imagined the unresolved risk and two 1–5 imagining-intensity items. The latter are also used to construct Downside Focus (downside imagining minus no-loss imagining) and Overall Imagining (the sum of the two imagining ratings).

This is a two-arm, between-subject randomized online survey experiment with U.S. working adults recruited through Prolific. Participants are individually randomized with approximately equal probability to the AI-source or Human-source earnings-risk condition. Pretreatment/background measures are collected before treatment assignment, and outcome measures are collected after participants receive their assigned hypothetical scenario. The design compares responses across the two treatment conditions while holding the objective earnings-risk characteristics constant.

Not available

Individual-level computer randomization using the Qualtrics Survey Flow randomizer, with approximately equal assignment probability to the AI-source and Human-source conditions. No stratification or covariate-adaptive randomization is used.

Individual survey participant.

No

645 individual participants (nonclustered design; the individual is the unit of randomization and observation).

645 approved, completed Main Study participants.

645 participants total, with approximately equal random assignment: approximately 322–323 participants in the AI-source condition and 322–323 participants in the Human-source condition.

The primary outcome is the 0–10 financial-satisfaction measure. For a two-sample, two-sided comparison at alpha = .05, with 645 analyzable participants (approximately 322–323 per condition), power is about 88.7% for d = 0.25, 96.7% for d = 0.30, and 71.8% for d = 0.20. The corresponding minimum detectable standardized effect size at 80% power is approximately d = 0.22. These calculations do not imply adequate power for very small effects or for exploratory interaction/moderator analyses. Pilot effect estimates are not used as the sole basis for the confirmatory sample size because the pilots were small and designed primarily for instrument development.