Losses spiral faster than gains. The feeling of being replaceable is what does the damage.

At a glance

1 · Run the spiral

The paper estimates how a person’s own year-to-year shifts in a psychological resource and in distress feed each other, after removing stable differences between people (a random-intercept cross-lagged panel model). The coefficients below are the published ones. Choose a pair and a disturbance size; the chart traces what those coefficients imply for the years that follow a single one-time change, with everything else at its expected level.

2 · Which way out of distress?

About one employee in six sat in the Distressed configuration at any wave (high neuroticism with preserved agreeableness). Over a decade, most moved. The question the paper asks is not whether they left but where they went: toward the Resilient profile (recovery) or toward the Reserved profile (withdrawal with lower engagement). Each resource below was tested in its own model among employees leaving Distressed.

3 · What the study found, in order

• The four profiles travel. Distressed, Antagonistic, Reserved, and Resilient, first recovered in U.S. midlife adults on the MIDI inventory, re-emerged on the IPIP-50 in a Dutch working-age panel (Tucker congruence .96 after Procrustes rotation). The Antagonistic and Resilient anchors replicated inside independently recruited entry cohorts; Distressed is a partial shape match recoverable only in the pooled model.
• Perceived replaceability and exhaustion carried forward five years. Displacement risk in 2020 predicted lower job satisfaction (β = −.11) and life satisfaction (−.11) in 2025. Exhaustion predicted all three criteria (−.22, −.31, and −.20 on positive affect). Algorithmic management reached only life satisfaction; technology self-efficacy ran the other way (βs ≈ .10). Ten of 16 paths were significant and all survived false-discovery control.
• Distressed membership did not amplify the stressors. The a priori moderation hypothesis was not supported. Exploratory slopes were steepest among Antagonistic members, which is a description, not a rescue.
• Resources steered the exit. Optimism (OR 1.78), self-esteem (1.62), and work meaning (1.77) per SD predicted recovery over disengagement and held up under soft-weighted, pseudo-class, and Firth sensitivities, and under direct classification-error correction in a latent annual Markov model. Workplace support was frame-sensitive; PsyCap was underpowered (n = 44).
• Losses outran gains within persons. Lagged distress depleted self-esteem (−.14) more than lagged self-esteem relieved distress (−.08), and the asymmetry survived a likelihood-ratio test. The optimism–negative affect pair showed reciprocal paths but no reliable asymmetry.
• Appraisal, not occupation title. Perceived displacement related to lower work meaning, support, and job satisfaction and higher distress. Occupation-level AI exposure (Felten AIOE) was null on all but one small path, and Frey–Osborne automatability failed a pre-trend falsification check, pointing to occupational selection rather than disruption.
• Costs surfaced before anyone left. Distressed members reported about 65 more lost work hours a year than Resilient members and twice the quit-intent rate, yet realized exits by 2025 did not differ and Distressed members reported more self-rated effort. Euro translations were partly compositional under a double/debiased machine-learning check.