The Self-Reporting Questionnaire-20 (SRQ-20) is widely used to screen for common mental distress in low- and middle-income countries. Despite frequent use in conflict-affected settings, its measurement equivalence across exposure to armed conflict, sex, and geographic region is not often tested with methods that distinguish true group differences from item bias. This analysis used nationally representative survey data from the 2015 Colombian National Mental Health Survey (ENSM 2015) to evaluate dimensionality, item functioning, equivalence across key groups, and criterion validity.
The analytic sample comprised 10,865 adults drawn from the ENSM 2015 public-use microdata. The authors assessed dimensionality and fitted a two-parameter logistic (2PL) item response model to the 20 SRQ items. Measurement equivalence across armed-conflict exposure, sex, and region was evaluated using multiple-group models with purified anchoring and freely estimated group means. This approach separates true distress differences from item bias, enabling tests of whether observed score differences reflect real variation in distress or differential item functioning.
Item-level functioning was judged against an equivalence band on signed expected-score differences (SIDS) set at ±0.10. Test-level differential test functioning (DTF) was reported alongside severity-graded and design-weighted sensitivity analyses. Criterion validity used design-weighted receiver operating characteristic (ROC) analyses comparing SRQ-20 scores to 12-month diagnoses from the Composite International Diagnostic Interview (CIDI).
All analyses used the de-identified ENSM 2015 microdata and incorporated survey design weights. The authors made their de-identified analytic dataset and reproducible analysis code openly available as reported in the source.
Dimensionality and model fit
The SRQ-20 was essentially unidimensional. A one-factor confirmatory fit index (CFI) equalled 0.945, improving to 0.971 when four content-redundant item pairs were modelled. Explained common variance for the dominant factor was 0.71, supporting a primarily single latent distress dimension.
Item response modelling and reliability
A 2PL model fit the data well. Conditional reliability at the clinically relevant cut-points was high, reported at approximately 0.93–0.94, indicating robust measurement precision where screening decisions are typically made.
Measurement equivalence across armed-conflict exposure and region
When true distress differences were separated from item bias, the SRQ-20 showed equivalence across armed-conflict exposure and across region. All item-level |SIDS| values for conflict exposure and region were below the ±0.10 equivalence threshold, with a reported maximum SIDS of 0.03 for conflict exposure. Net DTF for conflict exposure was approximately 0.1 points. Equivalence held even among adults who had been directly victimised, indicating the scale functions comparably for exposed and non-exposed respondents in this national sample.
Measurement equivalence by sex
Sex exhibited partial invariance. Three SRQ items demonstrated non-equivalent functioning between men and women, producing an overall DTF of approximately 0.85 points. The authors note this represents a small, quantifiable bias, meaning raw-total score comparisons by sex should be interpreted with caution or adjusted for the detected differential functioning.
Criterion validity
Design-weighted ROC analyses against 12-month CIDI diagnoses yielded area under the curve (AUC) values of 0.88 for major depression and 0.84 for any disorder, supporting the SRQ-20's criterion validity in identifying common mental disorders within this population.
In this large, nationally representative Colombian sample, the 20-item SRQ-20 demonstrates strong unidimensionality, good 2PL model fit, and high conditional reliability at screening cut-points. After separating true distress differences from item bias, the SRQ-20 is equivalent across armed-conflict exposure (including direct victimisation) and region, supporting valid comparisons between exposed and non-exposed groups on this measure. Sex comparisons of raw totals show partial non-equivalence (three items) with a modest DTF (~0.85 points), indicating a small but measurable bias.
The authors recommend the 20-item SRQ as appropriate for use in Colombian adults for population screening and comparisons by conflict exposure and region, while advising caution or adjustment when comparing raw totals by sex.
The study is a secondary analysis of openly available ENSM 2015 public-use microdata published by the Colombian Ministry of Health and Social Protection and Colciencias. The de-identified analytic dataset derived from these microdata and all reproducible analysis code are openly available at the authors' reported GitHub repository. The preprint states that necessary ethical approvals and participant consent procedures associated with the ENSM 2015 were followed; the analysis used de-identified public-use data.
For researchers and program evaluators working in Colombia or similar contexts, the SRQ-20 20-item form can be used to compare distress levels between groups defined by armed-conflict exposure and by region with confidence that item bias is minimal. However, where sex comparisons are central, analysts should consider the reported partial invariance (three items) and the modest DTF when interpreting raw-score differences or apply methods that adjust for differential item functioning. The scale's good criterion AUCs against 12-month CIDI diagnoses further support its utility as a screening instrument in population surveys.