What nursing stress subscales do and do not measure: general-factor dominance and item applicability in a 57-item German nursing stress instrument—a psychometric evaluation study

Background Nursing Stress Scale family instruments are intended to distinguish workplace stress domains, but validation studies have produced varying subscale structures. Few have examined whether a dominant general factor contributes to this variation or how ‘not applicable’ responses in the Expanded Nursing Stress Scale Discrimination domain affect measurement. Objectives To evaluate validity evidence for use of the German-language Stress in Nursing-57 (STERN-57), examine whether observed subscale scores add information beyond conditional stress severity, and test whether ‘not applicable’ responses are interchangeable with score 1 or reflect differences in which stressors apply across respondents. Design Cross-sectional psychometric evaluation study. Setting Single-centre online survey recruited through an acute-care university hospital in [city blinded], Germany. Participants 440 nursing-service and midwifery respondents (429 and 11, respectively). Methods We compared four ordinal confirmatory factor models: one-factor, correlated nine-factor, partial higher-order and bifactor. ‘Not applicable’ responses were not imputed; pairwise-available estimation yielded effective item-pair sample sizes of 162–412. We estimated general and subscale-specific reliable variance, explored whether observed subscales added information beyond conditional stress severity, reported applicability coverage separately, and compared ‘not applicable’ directly with score 1. Exploratory and confirmatory analyses used the same sample. Results The bifactor model showed the best relative fit (comparative fit index = .993; root mean square error of approximation = .058), although residual fit remained poor (standardised root mean square residual = .108) and the one-factor model fitted poorly. A general factor accounted for 91% of reliable total-score variance (omega hierarchical = .910; explained common variance = .923). Recoding ‘not applicable’ as score 1 reduced mean conditional severity from 2.44 to 2.20 (mean change = −0.23, 95% confidence interval −0.26 to −0.21) and changed 35.8% of scores by at least 0.25 points. Incremental-validity tests were underpowered and inconclusive. Conclusions The German-language STERN-57 is provisionally supported for conditional use in comparable nursing and midwifery samples, not as a measure validated for the German nursing population as a whole. Conditional severity describes stress among applicable situations and should be reported with applicability coverage; ‘not applicable’ and score 1 should remain distinct. Most subscales are not supported as separate inferential outcomes. Comparisons across occupations, settings, groups or time require independent evidence of score comparability. Parent study registration: DRKS00028779 (Deutsches Register Klinischer Studien); registered 2022-05-19; first participant recruited 2022-06-01.