📄 psychiatry and clinical psychology

Measurement equivalence of the SRQ-20 across armed-conflict exposure, sex, and region in Colombia (ENSM 2015)

This study demonstrates that the SRQ-20 is a psychometrically sound and measurement-equivalent tool for assessing mental distress across armed-conflict exposure and regions in Colombian adults, though it exhibits minor bias when comparing sexes.

Original authors: Velez-Pardo, P., Sanchez Acosta, D., Moratto-Vasquez, N. S., Quintero-Hoyos, J. M.

Published 2026-07-27

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Velez-Pardo, P., Sanchez Acosta, D., Moratto-Vasquez, N. S., Quintero-Hoyos, J. M.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are a detective trying to solve a mystery, but instead of looking for fingerprints, you are looking for feelings. In the world of psychology, scientists use special "feeling rulers" called surveys to measure how much stress, sadness, or anxiety a person is carrying. One of the most famous rulers is the SRQ-20, a short list of 20 questions used all over the world to spot mental distress. But here is the tricky part: a ruler only works if it measures the same thing for everyone. If a ruler stretches when you hold it in the sun but shrinks when you hold it in the shade, you can't trust the measurements. In science, this is called "measurement equivalence." It asks: Does a "sad" answer mean the same amount of sadness for a man as it does for a woman? Does it mean the same for someone who has never seen a war as it does for someone who has survived one? If the ruler is broken for certain groups, we might think they are sadder just because the tool is biased, not because they actually are. This is a huge deal because if we get the numbers wrong, we might send help to the wrong people or miss the people who need it most.

Now, let's step into the story of a team of researchers in Colombia who decided to test this ruler under some very intense conditions. Colombia has a long history of armed conflict, and the researchers wanted to know: Does the SRQ-20 still work fairly for people who have been directly hurt by the war, or does the trauma change how they answer the questions? They also checked if the ruler worked differently for men and women, and across different regions of the country. They didn't just ask people to fill out the survey; they used a sophisticated mathematical toolkit called Item Response Theory (IRT). Think of this like taking the ruler apart, examining every single tick mark, and seeing if the metal expands or contracts depending on who is holding it. They compared the survey answers against a gold-standard diagnostic interview to see if the ruler actually pointed to real mental health issues.

Here is what they found, and it's a bit of a plot twist. First, they discovered that the SRQ-20 is a fantastic ruler for measuring distress in Colombia. It is essentially one solid piece of metal, measuring a single type of "mental weight." But the most exciting news is about the war. They found that the ruler works equally well for people who have been exposed to armed conflict and those who haven't. Even for adults who were directly victimized—those who saw torture, kidnapping, or violence—the ruler didn't bend or break. The fact that these people reported higher levels of distress wasn't because the questions were confusing or biased against them; it was because they were genuinely carrying more weight. The researchers were able to separate the "true sadness" from "measurement errors," proving that the tool is fair. You can compare a war survivor's score to a non-survivor's score and trust that the difference is real, not a glitch in the tool.

However, the story isn't perfect. When they looked at sex, they found a tiny crack in the ruler. The survey is mostly fair, but for three specific questions (about being easily frightened, having headaches, and crying more), women tended to say "yes" slightly more often than men, even when they were feeling the exact same amount of distress. It's like the ruler has a tiny, consistent stretch for women, adding about 0.85 points to their total score. It's a small bias, but it's real. So, if you are comparing men and women, you need to know about this tiny stretch and adjust your thinking, but it doesn't ruin the whole tool.

They also tried adding four extra questions about strange thoughts or hallucinations to make a "25-item ruler," but that didn't work well. One of those extra questions ("Are you more important than others think?") was answered by nearly half the people, even those who weren't having psychotic experiences, making it a bad question for this group. The team concluded that the original 20 questions are the best bet. Finally, they checked if the ruler could actually spot real mental health disorders. It did a pretty good job, acting like a reliable net that catches most of the cases, though it sometimes catches a few fish that aren't quite what we're looking for (false positives). The bottom line? In the complex, heavy context of Colombia's history, the SRQ-20 is a sturdy, fair tool that can help us understand the mental toll of conflict without getting tricked by the trauma itself.

Technical Summary: Measurement Equivalence of the SRQ-20 in Colombia (ENSM 2015)

Problem Statement
The Self-Reporting Questionnaire (SRQ-20) is a widely deployed screening tool for common mental distress in low- and middle-income countries (LMICs). However, its psychometric evidence base is often limited to classical reliability coefficients and assumed unidimensionality, with insufficient interrogation of item-level behavior using Item Response Theory (IRT). Crucially, the measurement equivalence of the SRQ-20 across groups differing in armed-conflict exposure, sex, and region has rarely been tested using methods that distinguish true differences in latent distress (impact) from item bias (Differential Item Functioning, or DIF). In conflict-affected settings like Colombia, assuming invariance without rigorous testing risks conflating genuine group differences in distress with measurement artifacts, leading to erroneous epidemiological estimates and resource allocation.

Methodology
The study utilized data from the 2015 Colombian National Mental Health Survey (ENSM), a nationally representative, multistage probability survey of 10,865 adults. The analysis employed a modern psychometric toolkit:

Dimensionality and Model Fit: Tetrachoric correlations and weighted least squares (WLSMV) estimators were used for Confirmatory Factor Analysis (CFA). Dimensionality was assessed via Horn's parallel analysis, bifactor indices (Explained Common Variance, $\omega$ -hierarchical), and models accounting for local dependence (testlet parameterization).
IRT Calibration: A two-parameter logistic (2PL) model was fitted to estimate item discrimination ( $a$ ) and difficulty ( $b$ ). Model fit was evaluated using the $M_2^*$ statistic, RMSEA, and item-level $S-X^2$ statistics.
Measurement Equivalence Testing: A hierarchy of multiple-group 2PL models (configural, metric, scalar) was tested. To address the limitation of non-significant DIF results in large samples, the authors employed equivalence testing on item-level effect sizes (Signed Expected-Score Differences, SIDS) against a pre-specified band of $|SIDS| < 0.10$.
Separation of Impact and Bias: A "purified anchoring" procedure was used. An initial configural model identified invariant items to serve as anchors. The focal group's latent mean and variance were freely estimated to separate true group differences in distress (impact) from item bias.
Validity and Sensitivity: Criterion validity was assessed via design-weighted ROC analysis against 12-month diagnoses from a reduced Composite International Diagnostic Interview (CIDI). Sensitivity analyses included severity-graded comparisons (direct victimization vs. non-exposed) and design-weighted re-estimation.

Key Results

Dimensionality: The SRQ-20 is essentially unidimensional. While parallel analysis suggested multiple factors, bifactor indices indicated a dominant general distress factor (ECV = .71, $\omega$ -hierarchical = .79). Apparent multidimensionality was largely attributed to four content-redundant item pairs (e.g., digestion/stomach, tiredness items), which were successfully modeled via correlated residuals without invoking a second substantive dimension.
IRT Parameters: The 2PL model provided a superior fit compared to a one-parameter model. Items showed high discrimination (ranging from 1.10 to 2.79), with affective items (anhedonia, sadness) discriminating better than somatic ones. The scale is highly informative at screening thresholds ( $\theta \approx 1.1–1.5$ ), where conditional reliability is high (.93–.94), though marginal reliability is lower (.65) due to the concentration of information above the population mean.
Measurement Equivalence:
- Armed-Conflict Exposure: The SRQ-20 demonstrated full equivalence between exposed and non-exposed adults. Once the higher distress level of the exposed group (impact = +0.29 SD) was modeled, all item SIDS fell within the equivalence band (max $|SIDS| = .033$). Net Differential Test Functioning (DTF) was negligible (+0.12 points). This equivalence held even in a severity-graded analysis of directly victimized adults.
- Region: Full equivalence was found across five macro-regions (max $|SIDS| = .046$).
- Sex: Partial invariance was observed. While scalar invariance held by $\Delta CFI$ standards, equivalence testing revealed three items with bias (easily frightened, headaches, crying more) where women endorsed symptoms more than men at equal latent distress. This resulted in a net DTF of +0.85 points, indicating a systematic bias in raw total scores favoring higher scores for women.
Criterion Validity: Design-weighted AUC was .877 for major depression and .836 for any disorder. Optimal cut-points (e.g., $\ge 5$ ) showed good specificity but modest sensitivity, reflecting the trade-off inherent in screening. The low Positive Predictive Value (PPV) was attributed to low prevalence rather than poor screen quality.
25-Item Extension: The addition of four psychotic-experience items did not form a coherent subscale (Cronbach's $\alpha = .45$ ), largely driven by a poorly discriminating "grandiosity" item. The 20-item form is recommended.

Significance and Claims
The paper claims that the SRQ-20 is a valid, reliable, and essentially unidimensional instrument for measuring common mental distress in the Colombian adult population. Its primary contribution is establishing that the instrument functions equivalently across groups with different levels of armed-conflict exposure and across regions, supporting valid comparisons between exposed and non-exposed individuals within this specific population.

The study emphasizes a methodological distinction: by separating true distress differences from item bias, it demonstrates that the SRQ-20 measures the same construct on the same metric for conflict-exposed and non-exposed individuals, despite the exposed group having higher levels of distress. This validates the use of the SRQ-20 for within-population comparisons in conflict settings, a practice often assumed but rarely validated at a national scale.

However, the authors are modest regarding the scope of these claims. They explicitly state that the results support comparisons within the Colombian population and do not license cross-national comparisons or comparisons between conflict and non-conflict settings globally. Furthermore, they caution that raw total score comparisons by sex are biased and should be adjusted (approximately +0.85 points for women) or interpreted on the latent metric. Finally, the study recommends against using the extended 25-item form for psychosis screening in this context, advocating for the standard 20-item version.

Technical Summary: Measurement Equivalence of the SRQ-20 in Colombia (ENSM 2015)

More like this