COPSOQ III (Copenhagen Psychosocial Questionnaire, third version, core and dimension scales)

Structural validity Moderatewell-establishedevidence form: mixed

indirect : evidence is from Turkish, Australian, Norwegian, Swedish and (for COPSOQ-II) Australian and Polish samples; no UK-specific structural validation was retrieved, and some of the strongest structural evidence is parent-form (COPSOQ-II).

COPSOQ III is built as a set of largely independent dimension scales rather than one higher-order factor, so structural validity is assessed dimension by dimension, and the network explicitly notes that factor structure for the newly introduced dimensions was still to be fully tested at launch (Burr 2019). Confirmatory and exploratory factor analyses of national versions generally support the intended per-dimension structure. The Turkish COPSOQ-3 reported an excellent-fitting model (RMSEA 0.038, SRMR 0.053, CFI 0.98) with 19 extracted factors explaining 66.1% of variance (Sahan 2018). The Australian long-version validation, using EFA followed by CFA, reported acceptable to good fit (a four-factor higher-order EFA solution, and CFA RMSEA around 0.034 to 0.036 with CFI/TLI in the 0.89 to 0.92 range depending on model) (Rahimi 2025). For the predecessor COPSOQ-II, a set-ESEM approach that permitted only theory-consistent cross-loadings improved fit markedly over strict CFA (from CFI 0.907 to CFI 0.947 to 0.971) and reduced inflated inter-factor correlations, indicating that strict independent-clusters CFA understates fit for this multidimensional instrument (Dicke 2018). A full CFA of the COPSOQ-II 33-subscale model in a Polish sample also showed good fit (RMSEA below 0.05, SRMR below 0.08) (Baka 2022). The Norwegian COPSOQ III validation combined CFA with item response theory to characterise dimension structure and item information (Ose 2023). The main structural caveat is that fit is evaluated separately per dimension or per national selection of dimensions, not for a single global model, and that some short dimensions are less stable across countries.

Sub-grades (evidence differs by subgroup):

• {"subgroup": "Established core dimensions (demands, influence, leadership, social support, social capital)", "grade": "Moderate", "note": "Consistently recovered as intended factors across national CFAs (Sahan 2018; Rahimi 2025; Baka 2022)."}
• {"subgroup": "Newly introduced COPSOQ III dimensions (e.g. work engagement, quality of work, cyber bullying)", "grade": "Low", "note": "Network flagged their factor structure as not yet fully tested at launch (Burr 2019); evidence still accumulating."}

Convergent and discriminant validity Moderatewell-establishedevidence form: mixed

indirect : convergent evidence against ERI is from a German general-population employed sample ([Nuebling 2013](https://doi.org/10.1186/1471-2458-13-538)); the strongest discriminant analysis is parent-form (COPSOQ-II) ([Dicke 2018](https://doi.org/10.3389/fpsyg.2018.00584)); no UK sample.

Convergent and discriminant validity are supported mainly through scale intercorrelations and comparison with other established work-stress models. In the Gutenberg Health Study, COPSOQ scales and the Effort-Reward Imbalance (ERI) questionnaire showed congruent patterns across occupational groups, and COPSOQ predictor scales explained comparable or slightly greater variance than ERI in shared outcome scales (for example job satisfaction R-squared 0.51 for COPSOQ versus 0.46 for ERI; burnout 0.35 versus 0.26), supporting convergent validity against an established alternative model (Nuebling 2013). Within-instrument, developers calculate scale intercorrelations to demonstrate that dimensions are distinct (divergent) while theoretically related scales correlate (convergent) (Burr 2019; Berthelsen 2020). The set-ESEM analysis of COPSOQ-II is directly relevant to discriminant validity: strict CFA produced some inflated inter-factor correlations that fell substantially once theory-consistent cross-loadings were permitted, indicating the dimensions are discriminable but not perfectly orthogonal (Dicke 2018). A dedicated study constructed and validated a global Workplace Social Capital scale from COPSOQ III justice and trust items, supporting convergent structure among the social-capital dimensions (Berthelsen 2019). Formal multitrait-multimethod or correlations with independent external gold-standard constructs are sparser than the within-instrument evidence.

Criterion validity: reference standard Not applicable (category difference)untestedevidence form: canonical

direct : the category (diagnostic reference-standard validity) does not apply to a psychosocial-exposure measure; recorded as Not-applicable rather than Absent because it is a category mismatch, and no reference-standard study was located.

COPSOQ III is an exposure measure of psychosocial working conditions, not a screener for a health condition, so validity against a diagnostic or clinical reference standard is not a design goal and is largely not evaluated. No study retrieved in the latest review pass compared COPSOQ III dimension scores against a diagnostic reference standard (for example a structured clinical interview for depression) with sensitivity or specificity. The embedded strain scales (self-rated health, burnout, stress) are self-report and are treated as outcomes correlated with exposures rather than validated against a clinical criterion. This is an appropriate absence for an exposure instrument, but it means the field is genuinely untested rather than merely weak.

Criterion validity: organisational Very lowthinevidence form: mixed

indirect : the only quantified associations are cross-sectional and against self-reported outcomes in a German general-population sample ([Nuebling 2013](https://doi.org/10.1186/1471-2458-13-538)); objective-outcome prediction is unstudied and no UK data exist.

Evidence that COPSOQ III scores predict objective organisational or health outcomes (register-based sickness absence, staff turnover, diagnosed conditions, performance) is a recognised gap that the developers themselves flag as future work. The Swedish validation explicitly states that evaluating predictive criterion validity against register data on absence, staff turnover and performance remains to be done in longitudinal multilevel designs (Berthelsen 2020), and the Norwegian study likewise lists predictive validity as not yet established for their version (Ose 2023). The available criterion-type evidence is cross-sectional and against self-reported outcomes: in the Gutenberg Health Study COPSOQ psychosocial scales explained meaningful variance in self-reported job satisfaction (R-squared 0.51), burnout (0.35), satisfaction with life (0.18) and general health (0.11), with theoretically sensible predictors (for example meaning of work and sense of community for job satisfaction; work-privacy conflict for burnout) (Nuebling 2013). This supports concurrent association with self-reported strain outcomes but not prediction of objective work outcomes. Genuine organisational criterion validity against registers or hard outcomes is essentially Absent in the retrieved literature.

Internal consistency Highwell-establishedevidence form: mixed

indirect for the UK specifically (evidence from Sweden, Germany, Australia, Turkey, Greece, China and the seven-country international sample; no UK sample), but the property is directly and repeatedly measured on the fielded COPSOQ III versions.

Internal consistency is the most extensively documented property and is generally acceptable to good for the multi-item dimensions, with well-characterised weak spots at the short two-item and emotion-related scales. Across the seven-country international middle version, most of the 23 tested scales reached Cronbach alpha above 0.70; three fell below: Commitment to the Workplace (two items, mean alpha 0.64, 95% CI 0.61 to 0.67), Demands for Hiding Emotions (three items, mean alpha 0.66, 0.58 to 0.73) and Control over Working Time, with additional country-specific shortfalls for Predictability, Meaning of Work and Job Insecurity (all two-item scales, alpha around 0.62 to 0.66 in France and Turkey) (Burr 2019). The German middle version reported that 20 of 25 multi-item scales exceeded alpha 0.70 and 13 reached 0.80 or higher; a small number of two-item scales (for example Degrees of Freedom, reduced to two items, alpha 0.53) were weak (Lincke 2021). The Australian long version found all 31 three-or-more-item scales acceptable except Demands for Hiding Emotions (0.66), and among two-item scales used Spearman-Brown coefficients, with Variation of Work unacceptably low (0.24) while Meaning of Work was acceptable (0.78) (Rahimi 2025). The Turkish version reported 23 dimensions above 0.70 with Control over Working Time (0.54) and Predictability (0.66) below (Sahan 2018). The Greek long version found 22 of 40 scales with alpha above 0.70 (Kotsakis 2025), and the Chinese long version reported an overall alpha of 0.92 with per-dimension values from 0.60 to 0.92 (Huang 2025). The consistent pattern is that longer scales are reliable and the recurring weak scales are the two-item and hiding-emotions dimensions, which is a structural feature of the compact design rather than a country-specific defect.

Sub-grades (evidence differs by subgroup):

• {"subgroup": "Multi-item dimensions (three or more items)", "grade": "High", "note": "Alpha consistently above 0.70, frequently above 0.80 (Burr 2019; Lincke 2021; Rahimi 2025)."}
• {"subgroup": "Two-item and hiding-emotions scales", "grade": "Low", "note": "Recurring alpha/Spearman-Brown shortfalls (e.g. Commitment 0.64, Hiding Emotions 0.66, Variation of Work 0.24 in Australia) (Burr 2019; Rahimi 2025)."}

Test-retest reliability Lowthin

indirect : no test-retest study of the fielded COPSOQ III was located; the adequate ICC evidence is from the Danish parent COPSOQ over about three weeks ([Thorsen 2010](https://doi.org/10.1177/1403494809349859)), and the only longitudinal COPSOQ-II retest used a 12-month interval that conflates true change with instability ([Baka 2022](https://doi.org/10.1371/journal.pone.0262266)).

Test-retest reliability for COPSOQ III specifically is essentially Absent and is repeatedly named as outstanding future work. The COPSOQ III network states that test-retest of the newly introduced dimensions was still to come (Burr 2019), and the Norwegian, Greek and Chinese validations all explicitly did not assess test-retest (Ose 2023; Kotsakis 2025; Huang 2025); the German validation deliberately excluded a formal test-retest for practical reasons (Lincke 2021). The two structured findings recorded here are both parent-form: Thorsen and Bjorner's dedicated Danish test-retest study of the original COPSOQ found ICCs of 0.70 to 0.89 (one scale, mutual trust, 0.64) over a median 22-day interval, which is good (Thorsen 2010); the Polish COPSOQ-II longitudinal study reported low retest correlations (r 0.15 to 0.34) but over a 12-month interval that the authors note is far longer than the recommended few-weeks-to-months window, so it indexes real exposure change as much as instrument instability (Baka 2022). Direct short-interval COPSOQ III test-retest evidence has not been established.

Measurement invariance Very lowthinevidence form: mixed

indirect : the only explicit longitudinal invariance evidence is parent-form (COPSOQ-II, Australian principals) ([Dicke 2018](https://doi.org/10.3389/fpsyg.2018.00584)); cross-country comparability for COPSOQ III is argued by design and similarity of solutions, not by pooled invariance testing; no UK data.

Formal measurement invariance testing (configural, metric, scalar) for COPSOQ III is thin, and the international network was candid that it could not directly test differential item functioning across countries at launch because of data-protection constraints on pooling item-level data, while observing that scale properties differed somewhat across working populations (Burr 2019). The clearest invariance-relevant evidence is parent-form: the COPSOQ-II set-ESEM study of Australian school principals tested longitudinal factor structure across two time points and reported good, stable fit (for example CFI 0.95, TLI 0.94, RMSEA 0.02 at Time 1), supporting configural stability over time (Dicke 2018). Comparability across the many national COPSOQ III versions is handled largely by design (a common obligatory core, standardised translation procedures) and demonstrated indirectly through similar factor solutions and reliability patterns across countries (Burr 2019; Rahimi 2025; Kotsakis 2025) rather than by pooled multi-group invariance models. Group-level aggregation properties (ICC(1)/ICC(2) across occupations and workplaces) were examined in Sweden to justify comparing group mean scores, which is a related but distinct property from measurement invariance (Berthelsen 2020). Full multi-group scalar invariance across sex, age, occupation and language for the fielded COPSOQ III has not been established in the retrieved literature.

Sub-grades (evidence differs by subgroup):

• {"subgroup": "Longitudinal (over time)", "grade": "Low", "note": "Configural stability supported for parent COPSOQ-II via set-ESEM (Dicke 2018)."}
• {"subgroup": "Across countries/languages", "grade": "Very low", "note": "Not formally tested at item level; DIF analysis was precluded by data-protection constraints (Burr 2019)."}
• {"subgroup": "Across sex/age/occupation", "grade": "Absent", "note": "No formal multi-group invariance study located for COPSOQ III."}

Responsiveness and MIC Absent (a finding about the literature)untestedevidence form: canonical

indirect : the only relevant material is a proposed 5 to 10 point minimum important difference convention from the Swedish benchmarking work ([Berthelsen 2020](https://doi.org/10.3390/ijerph17093179)); no formal responsiveness study and no UK data.

Responsiveness (sensitivity to change over time or after intervention) and a minimal important change threshold have not been established for COPSOQ III in the retrieved literature. The Swedish validation lists responsiveness alongside test-retest and predictive validity as properties still to be evaluated in future longitudinal designs (Berthelsen 2020). As an interpretation aid rather than a formal responsiveness statistic, the Swedish work proposes that a 5 to 10 point difference on the 0 to 100 scale metric can be treated as a minimum important difference, and reports Cohen's d effect sizes for group contrasts (Berthelsen 2020), but this is a benchmarking convention, not an anchor-based or distribution-based minimal-important-change study. No study retrieved in the latest review pass estimated responsiveness against an external change anchor.