Correcting Beliefs about Wage Catch-Up after High Inflation: Evidence from a Survey Experiment on Labour-Market and Consumption Outcomes

This study embeds a randomized information-provision experiment in the ad-hoc module of the September 2026 wave of the ECB Consumer Expectations Survey (CES), a monthly online panel survey of consumers in 11 euro-area countries (approx. 19,000 respondents per round). All respondents first report a belief about when average wages in their country caught up with the 2022-2023 rise in consumer prices. Respondents are then randomly assigned to one of four groups in a 2x2 factorial design that varies (i) whether they are shown additional information relevant to that belief, and (ii) the framing of a subsequent hypothetical scenario about future income growth. All four groups complete an identical questionnaire; only the specific screens shown differ by assignment. Follow-up outcomes are measured within the same September interview and in the following month's (October 2026) wave of the same panel, covering respondents' income and inflation expectations, wage negotiation, household consumption, and labour-market behaviour. The intervention involves no deception beyond withholding information from a subset of respondents, no incentives beyond the survey's standard participation arrangements, and poses no more than minimal risk to respondents.

2026-08-27

2026-09-28

(1) Within-subject change in the personal wage-catch-up belief between its pre-treatment (AX3010) and post-treatment (AX5030) readings, September wave.
(2) Expected own-income or average-wage growth 18 months after a hypothetical energy-price shock (AX5040_3 / AX5042_3), September wave.
(3) Indicator for any large-ticket household purchase, October wave.
(4) Realized on-the-job search and job-search-entry indicators, and the re-elicited wage-demand percentage, October wave.
(5) Re-elicited personal wage-catch-up belief (FLQ_82_2100), October wave (persistence of outcome 1).
(6) Standard 12-month wage/income-growth expectation, consolidated from Q2510/Q2520/Q2530, October wave.
(7) Horizon-conditioned expected real wage gap and implied catch-up horizon at 1, 3, and 5 years, constructed from FLQ_82_2300_1-3, October wave.

Outcome (1) and (5) are belief-change scores on an identical 5-point qualitative scale ("Perfectly" to "Not at all"), asked verbatim before and after treatment. Outcome (2) is a directly reported percentage. Outcome (6) is consolidated from a qualitative direction item plus an open-ended quantitative item per the instrument's own SignedWagExp rule. Outcome (7) is constructed as RealGap_i,h = w_i,h - pi_i,h, the entered or prefilled expected income growth at horizon h minus the respondent's own displayed inflation expectation at that horizon; the implied catch-up horizon is the first h at which RealGap_i,h >= 0, right-censored at 5 years if the gap never closes. All treatment comparisons additionally interact the treatment indicator with an informational-surprise measure, Surprise_i = s(AX5000_i) - s(Y_c(i)), the respondent's prior recoded on a common ordinal scale minus the official country-specific answer on the same scale.

(1) Descriptive distribution of prior beliefs (AX5000) relative to the official answer and the resulting informational surprise (no causal test attached).
(2) Expected own-income or average-wage growth 6 months after the hypothetical energy-price shock (AX5040_1-2 / AX5042_1-2), and a categorical 18-month price-growth forecast, September wave.
(3) Total nominal household consumption and each of 12 individual expenditure categories, October wave.
(4) Employer change (proxied by tenure under 3 months with current employer), applications sent, and realized wage-negotiation incidence and outcome, October wave.
(5) Expected real wage gap at each individual horizon (1, 3, and 5 years) taken separately, and the treatment effect on the respondent's own displayed inflation expectations (reported as a diagnostic), October wave.

Secondary outcomes follow the same construction rules as the corresponding primary outcomes (see Primary Outcomes (Explanation)) but are analyzed at a finer grain (individual horizons, individual expenditure categories) or are diagnostic/descriptive rather than confirmatory. Employer change has no direct instrument item in the October wave; it is proxied by tenure with the current employer of less than 3 months, a substitution documented and flagged as an instrument-QA item in the pre-analysis plan.

Randomized controlled trial embedded in an ongoing, nationally representative online panel survey (ECB Consumer Expectations Survey). Individual-level random assignment to one of four groups in a 2x2 factorial design (information provision x framing), stratified by country/language and recruitment method, with equal target allocation across the four cells. All groups complete the same survey instrument; only the specific screens shown differ by assignment. Outcomes are measured within the same interview and in the following month's wave of the same panel.

Not available

Simple random assignment (equal probability, no minimization or covariate-adaptive algorithm), implemented in real time by the fielding contractor's (Ipsos) survey-scripting software at the point each respondent reaches the relevant module.

Individual respondent. Randomization is stratified by 11 country/language groups (Germany, France, Italy, Spain, the Netherlands, Belgium [French- and Dutch-speaking strata], Finland, Greece, Ireland, Portugal, Austria) crossed with 2 recruitment methods (CATI, CAWI), giving 22 strata, with equal target allocation across the four treatment cells within each stratum.

No

Not applicable - randomization is at the individual respondent level, not clustered.

Approximately 19,000 respondents in the September 2026 (Wave 81) CES round across 11 euro-area countries, of whom approximately 14,000-14,500 respondents (about 76% retention, based on realized wave-to-wave CES retention rates) are expected to complete the linked October 2026 (Wave 82) round used for the cross-wave outcomes.

Approximately 4,750 respondents per cell in September (Group A: no information, own-income framing; Group B: no information, average-wage framing; Group C: information shown, own-income framing; Group D: information shown, average-wage framing), based on equal allocation of the approximately 19,000-respondent Wave 81 sample across the four cells of the 2x2 design. Approximately 3,600 respondents per cell are expected to remain for the October cross-wave outcomes, assuming 76% retention applied uniformly across arms (differential attrition across arms is a pre-specified robustness check).

All minimum detectable effects (MDEs) use the two-arm formula MDE ≈ 2.8 x SD x sqrt(2/n) for continuous outcomes (SD = sqrt(p(1-p)) for binary outcomes), at alpha = 0.05 two-sided and 80% power, using moments from prior, already-delivered CES waves (no Wave 81/82 data used). Figures below are conservative post-only upper bounds; registered ANCOVA specifications will shrink them further once baseline-endline correlations are estimated. Selected main outcomes: - Personal wage-catch-up belief (H2/H6), n ≈ 9,500/arm: MDE ≈ 0.041 SD (no directly comparable prior-wave item; reported in SD units). - Standard 12-month wage-growth expectation, employees (H7), n ≈ 5,083/arm: SD = 5.49 pp (winsorized 2%); MDE ≈ 0.31 pp. - Horizon-conditioned catch-up outcome (H8), n ≈ 9,500/arm: MDE ≈ 0.041 SD (new item; reported in SD units). - Any large-ticket purchase (H4), n = 9,112/arm: base rate 42.9%, SD = 0.499; MDE ≈ 2.1 pp. - Total household consumption (H4), n = 9,112/arm: SD = EUR 5,030 (winsorized 1%); MDE ≈ EUR 209 (≈0.042 SD). - Realized on-the-job search, all respondents (H5), n = 9,107/arm: base rate 13.87%, SD = 0.346; MDE ≈ 1.4 pp. - Own wage-growth persistence, employees, arms A vs C (H3), n = 2,542/arm: SD = 5.49 pp (winsorized); MDE ≈ 0.43 pp.