Psychological well-being in Europe after the outbreak of war in Ukraine

The study investigates the psychological impact of the February 24, 2022 Russian invasion of Ukraine on individuals living outside Ukraine, focusing on European countries where the consequences and public attention were likely most direct. The authors aim to characterize how well-being evolved in the weeks surrounding the outbreak of war, identify individual differences in responses (e.g., personality and sociodemographics), and examine how daily well-being corresponded to the daily salience of the war on social media. The work addresses the broader question of how major collective events affect momentary well-being across nations and the extent to which recovery patterns vary between individuals.

Prior research on major life events often relies on cross-sectional data collected after an event, which can introduce retrospective biases and does not permit pre-post comparisons or dynamic modeling of anticipation and recovery. Panel studies, although longitudinal, frequently have low temporal resolution (e.g., annual assessments) and are typically country-specific, limiting the ability to capture acute effects at the time of the event and to compare across nations. These constraints pose challenges for studying collective events like war outbreaks. The present study leverages high-frequency, international experience-sampling data from the CoCo project to address these limitations, enabling pre-event baselines, day-level trajectories, and cross-country analyses.

Design and data source: Longitudinal experience-sampling method (ESM) from the global Coping with Corona (CoCo) project. Participants completed a pre-survey (personality and sociodemographics), four randomly timed brief ESM surveys per day for four weeks (momentary affect), and a post-survey. Ethical approval: University of Münster IRB (2020–54-MB); informed consent obtained; preregistered (osf.io/3uqtf). Time window and sample: Primary analyses restricted to a symmetric 63-day window around the invasion (January 24 to March 27, 2022). European subsample: N = 1,341 participants, 44,894 ESM reports; mean age 25.7; 1,079 women, 252 men, 10 other. Additional analyses: narrower (±7 days) and broader (Jan 1–Apr 19, 2022) windows; global sample (N = 1,735). Inclusion/exclusion: Excluded non-European participants for main analyses; excluded participants indicating non-conscientious responding, too-fast pre-survey responders (<2 s/item), and those with state data on <2 days. Measures: - Individual well-being: Daily aggregate of positive affect (happy, enthusiastic, relaxed) and reverse-scored negative affect (angry, afraid, sad), items on 1–6 scale; daily aggregation across ESM prompts; z-standardized across all measurements. - Personality: Big Five Inventory-2 (60 items); meta-traits Stability (Agreeableness, Conscientiousness, reverse Neuroticism) and Plasticity (Extraversion, Openness); pre-survey data used. - Sociodemographics: Age, gender (female/male used for analyses), subjective social status (ladder 1–10), political orientation (1 left – 10 right). - Salience of the war: Global daily count of tweets containing “Ukraine” via Twitter Researcher API v2; counts z-standardized across days; UTC day boundaries. Statistical analysis: Multilevel models with daily observations (Level 1) nested in individuals (Level 2). Eight candidate change-trajectory models specified using Level-1 change parameters: time, level (pre/post baseline shift), pre-event linear slope, post-event linear slope. Variables rescaled to −1–1 ranges for interpretability over the 63-day window. Random-intercept and random-slope models compared via AIC; best-fitting model was 2d (independent pre- and post-event linear trends plus a level shift on invasion day). Stability and sociodemographics were tested as Level-2 predictors of random intercepts and slopes. Salience models included total tweets or partitioned into within-person and between-person components; lagged models used previous-day tweets. Frequentist ML estimation used for model comparisons; Bayesian brms models with uninformative priors used as robustness checks (four chains, 20,000 iterations, delta 0.99), yielding convergent, similar results. Robustness: Country controls (dummy-coded for countries with ≥10 participants), exclusion-by-country sensitivity checks, global sample comparisons (Europe vs non-Europe), alternate time windows, affect subcomponents (positive/negative affect), Plasticity and Big Five domains, societal well-being outcomes, and lagged salience effects. Note: Technical error on March 14, 2022 reduced submissions that day (excluded from Fig. 4 but included in analyses).

• Best-fitting trajectory (Model 2d): No significant anticipation trend pre-invasion (b = 0.005, 95% CI [−0.097, 0.107], p = 0.923). Significant abrupt decline in well-being on the invasion day (b = −0.200, 95% CI [−0.278, −0.122], p < 0.001), approximately 0.2 SD. Post-event linear trend small positive (b = 0.089, 95% CI [−0.006, 0.184], p = 0.066), suggesting slow recovery; supplementary broader/narrower windows showed significant positive post-event slopes. - Cross-regional comparison: During the period around the outbreak, participants from European countries reported lower well-being than those from non-European countries (b = −0.245, 95% CI [−0.329, −0.162], p < 0.001). - Individual differences: Stability (meta-trait) associated with higher overall well-being (b = 0.245, 95% CI [0.180, 0.311], p < 0.001). Stability did not significantly moderate pre-event changes or the invasion-day drop but significantly moderated post-event recovery (b = 0.161, 95% CI [0.066, 0.255], p = 0.001): higher Stability predicted faster recovery; low-Stability individuals showed little recovery in the first month. Sociodemographics (age, gender, subjective social status, political orientation) did not significantly explain variability in change parameters. - Salience of war on social media: Greater global counts of “Ukraine” tweets were associated with lower daily well-being (b = −0.070, 95% CI [−0.096, −0.044], p < 0.001). Decomposed effects: within-person (b = −0.065, 95% CI [−0.092, −0.037], p < 0.001) and between-person (b = −0.116, 95% CI [−0.207, −0.024], p = 0.013) were both significant. Lagged analyses (previous day’s tweets) showed comparable associations. No credible evidence that personality or sociodemographics moderated salience effects. - Effect size context: The 0.2 SD population-level dip is small compared to personal tragic life events (e.g., bereavement/disability ~0.6–0.7 SD) but larger than many prior population-level event effects (e.g., no significant effects during 2020 COVID-19 lockdowns; Fukushima ~0.12 SD in affected areas).

The analyses show an acute, population-level decline in well-being coinciding with the invasion day, followed by slow recovery. The uniformity of the initial drop across personality and sociodemographic groups suggests a strong situational shock. Recovery trajectories, however, varied by personality: individuals high in Stability (lower threat sensitivity, less rumination, greater emotional regulation) recovered more quickly, whereas low-Stability individuals showed minimal recovery over the first month. Daily well-being tracked the salience of the war on social media: both people sampled on high-salience days and individuals on their own higher-salience days reported lower well-being, consistent with a reciprocal dynamic in which distress and exposure to war-related content reinforce each other. Results align with contemporary revisions of set-point models of well-being, indicating temporary adaptation on average but meaningful interindividual differences. Exploratory post-hoc analyses suggested that lower well-being related primarily to dysfunctional indicators (e.g., worries, avoidant coping), with limited evidence for translation into prosocial behaviors like donations or protests. Potentially scalable supports (e.g., group-based, tele/online interventions, and self-help like expressive writing) could target vulnerable subgroups, though such implications were not directly tested here and should be tailored to clinical severity. Country-specific differences may exist, but unbalanced data preclude firm conclusions.

This study provides high-temporal-resolution, preregistered evidence that the outbreak of war in Ukraine corresponded with an immediate decline in well-being among European participants, followed by slow, personality-dependent recovery. It highlights the psychological dimension of collective crises and identifies trait Stability as a key correlate of resilience in recovery. Future research should (a) obtain more balanced, representative, and cross-nationally harmonized samples; (b) include key populations most directly affected (e.g., Ukrainian and Russian residents) to assess larger, context-specific impacts; (c) integrate causal designs and richer exposure measures beyond social media salience; and (d) delineate mechanisms and interventions that foster adaptive recovery while considering individual differences.

• Causality: Although temporal patterns align with the invasion and salience measures, the observational design limits strong causal inference. - Sample representativeness: Skew toward younger, more female, and more educated participants limits generalizability. - Cross-country balance: Unequal numbers of participants and uneven timing of reports across countries complicate country-specific inferences; exploratory analyses suggested some country-specific differences. - Exclusions of key populations: Ukrainian and Russian participants were not included, where larger effects are likely. - Technical issue: A data collection error on March 14, 2022, reduced submissions that day (omitted from figure but retained in analyses).