DASS-Y — Simorq Persian translation v1: adaptation and clinical validation

Protocol version 1.0 — repository preregistration, 2026-09-16. The Git commit that first adds this file is the registration record. An external registry identifier may be added later without changing this plan. Outcome analysis is not authorised before registration. Amendments must be dated and must leave this version legible; an amendments.md file is created only when an amendment exists.

1. Identity and exact form

• Instrument: Depression Anxiety Stress Scales for Youth (DASS-Y), Persian self-report, code DASSY, variant youth_self.
• Exact form: simorq-translation-v1, adopted 2026-08-26: 21 items, four ordered response categories (0–3), past-week recall, digital administration, Depression / Anxiety / Stress keys fixed in the dossier. The form and scoring key are immutable for this protocol.
• Intended population in the repository: ages 7–18. Developer material says ages 8–17. Ages 7 and 18 are therefore boundary extensions that require evidence here; they are not justified by the developer study.
• Rights: the base instrument and this unbranded translation are public domain under the DASS owners' terms. This is a third-party instrument, so Simorq's authored-instrument evidence stages do not apply. Its current norm class is and remains none unless a later human evidence decision changes it.

2. Scientific rationale

The published Iranian Persian DASS-Y was clinically reviewed against the English source. Simorq changed 12 of 21 items: 1, 5, 6, 9, 11, 12, 13, 14, 17, 18, 19 and 21. Items 9, 18 and 21 carried the clearest intensity or construct defects; other changes corrected tense, calque, register and collapsed Stress facets. The item-by-item record is in the translation file and is not repeated here.

The owner review completed form lock but did not complete an ISPOR/Beaton adaptation process: there was no independent blind back-translation and no target-user cognitive interviewing. Those omissions matter before coefficient estimation. Content validity—relevance, comprehensiveness and comprehensibility—is therefore the first question.

Later DASS-Y work also makes a single structural assumption unsafe. The developer selected three correlated factors; subsequent studies have found meaningful ESEM and bifactor-ESEM solutions and high factor correlations. These findings justify prespecified secondary models, not retrospective model shopping.

3. Intended use and target claims

The intended use is dimensional description and repeated measurement of Depression, Anxiety and Stress symptoms in Persian-speaking young people in care. This study does not make DASS-Y diagnostic and does not treat a severity label as a diagnosis.

If all relevant phases succeed, Paper 1 may claim that this exact Persian form is comprehensible across the sampled ages, that its prespecified subscale structure is supported in a multi-site Simorq clinical cohort, that subscale scores have adequate internal consistency for group description, and—only for groups that meet invariance requirements—that measurement is comparable. The test–retest substudy may estimate reliability and measurement error in stable respondents.

Paper 1 cannot provide population norms, T-scores, percentiles, diagnostic accuracy, a minimal important change, or validated severity bands. It cannot validate another Persian wording or another administration mode.

4. Existing evidence and non-transfer

• Szabo and Lovibond developed the English DASS-Y in an Australian community sample and supported Depression, Anxiety and Stress scores. Their sample was not a Persian reference population.
• Shabani et al. studied the prior Persian wording in 1,277 Iranian school students. Its reliability, distribution, factor results and use of developer bands are comparative background only. Twelve changed items break form identity. The study also used a school convenience/purposive frame and did not publish Iranian percentile norms or test–retest evidence.
• More recent Serbian, Chinese and Polish studies make ESEM/bifactor alternatives, developmental invariance and longitudinal invariance relevant, but their coefficients and thresholds do not transfer across language, wording or population.
• The DASS developers describe severity bands as percentile-based descriptive ranges from the original community sample, not diagnostic thresholds. A 2026 Indonesian diagnostic-accuracy study illustrates the different design a screening cutoff requires: an external clinical criterion. Neither result transfers to this Persian form.
• The dossier retains old Persian reliability/distribution values and English cutoffs so their provenance is visible; each is marked unusable for release.

No DASS-Y psychometric outcome analysis was found in repository records or issues during protocol preparation. The data model can retain exact version, age, sex, informant, setting, clinic, item responses, item timing and pseudonymous repeat linkage. Staging/production credentials were not available in this protocol-writing session, so current row and site counts were not verified. Before the first export, a data custodian must run the count-only research-recruitment query and archive the date, total, sites and availability of required fields without running research-strata or viewing responses. Rows completed before this registration, if any, may enter Phase B only. The Phase C confirmatory cohort starts after registration. Any outcome already seen elsewhere makes that outcome exploratory and must be disclosed before export.

5. Questions in evidential order

1. Do the Persian instructions, anchors and items express the intended content and work for 7–9, 10–13 and 14–18 year olds?
2. Do first clinical administrations reveal response-option, missingness, timing, floor/ceiling or item-rest problems, especially in the 12 corrected items?
3. Does the preregistered three-correlated-factor model adequately represent responses, and are Depression, Anxiety and Stress subscale scores internally consistent?
4. Does measurement operate comparably across age bands and sex to the degree needed for the intended comparisons? Does current evidence support anything beyond descriptive use of TOTAL?
5. In stable respondents, what are test–retest reliability and measurement error? Later, is the form responsive to clinically important change?
6. In separate future designs, can any severity interpretation or population reference values be justified?

6. Phase A — translation and content validity

6.1 Translation audit

Two Persian–English bilingual reviewers who did not author v1 independently back-translate all 21 items, instructions and anchors while blinded to the English wording. At least one has child mental-health expertise and neither is given the item defect history before translating. A Persian child clinician, a measurement specialist and a Persian-language specialist compare source, back-translations, published Persian wording and v1 for semantic, idiomatic, experiential and conceptual equivalence. They record per-unit relevance and comprehensibility findings; they do not decide by majority vote or by literal similarity alone. Items 9, 18 and 21 receive explicit intensity/construct adjudication; all 12 changed items remain in scope.

This retrospectively supplies an independent check that was historically missed. It does not recreate a prospective dual-forward-translation process, developer review or harmonisation across languages, and the article will say so. Developer contact is not required and is outside programme scope.

6.2 Target-user cognitive interviews

Start with 18 Persian-speaking respondents, six in each age band 7–9, 10–13 and 14–18. Recruit purposively across sex, clinical/community context, literacy and metropolitan/non-metropolitan experience. The initial sample includes at least two 7-year-olds and two 18-year-olds; a parent may assist only as they would in the recorded administration mode, and assistance is documented.

Use individual think-aloud/retrospective probing after the respondent first answers the unaltered v1 form. For every item ask for a paraphrase, recall period, response choice rationale and unclear terms. Probe the intended facet for every changed item, then ask whether important experiences within the three named constructs seem missing or needlessly duplicated. Without teaching the answer, test particularly:

• item 9: whether به‌شدت ترسیده بودم conveys terrified/high-intensity fear and remains distinct from items 15 and 20;
• item 18: whether خیلی زود اعصابم خرد می‌شد is self-irritability rather than hurt caused by another person;
• item 21: whether زندگی خیلی بد است is devaluation of life rather than fear, transient bad events or suicidality;
• formal شما versus تو: naturalness, comprehension, interpersonal tone and whether either changes the perceived administrator or answer;
• زیاد درست بود versus تا حد زیادی درست بود: category order, separation from adjacent anchors and the respondent's explanation of each degree.

Alternatives are shown only after the current wording has been understood, to avoid priming. Interviewers code construct-level error, intensity error, lexical/register difficulty, anchor-order error, assistance and new issues. Two coders reconcile a framework matrix by item and age band; quotations are de-identified.

The initial 18 is a starting design, not a saturation claim. Add batches of three in every affected age band. Stop only when all prespecified units have been probed, no critical construct error remains, and a complete additional batch in each band introduces no new material problem and correctly paraphrases items 9, 18 and 21 and the anchor ordering. Expected information-power range is 18–30; continue beyond 30 if the rule is not met. Older adolescents cannot saturate the younger-child question.

6.3 Phase A decision

Retain v1 only if target users understand the intended constructs, response period and ordered anchors without systematic age-dependent drift. Formality or awkwardness alone is not a reason to revise. Any semantic change to an item, instructions, register or anchor creates a new form version; v1 data do not validate that version. The new version repeats affected cognitive work and starts Phase B anew. Typographic corrections are versioned and documented even when judged non-semantic.

7. Phase B — clinical pilot

Analyse the first 150 eligible first fills, targeting 50 per age band, from at least three clinics. Pre-registration rows, if any, are ordered by answered_on then pooled_at and may be included here only if outcomes were unseen. At n=150, observing zero occurrences puts the simple 95% upper bound near 2% for a form-wide catastrophic event; 50 per band provides a deliberately rough developmental signal, not definitive item ranking.

Report response-category frequencies, missingness, item and whole-form time, floor/ceiling, polyserial corrected item-rest correlations within the assigned subscale, subscale distributions and all corresponding age-band views. Contrast the 12 corrected items with the nine unchanged items descriptively; items 9, 18 and 21 are displayed individually. Extremely short/long completion, a rare category or a weak item-rest coefficient is a flag for transcript/form review, not an automatic exclusion or edit. All-equal responses remain valid unless a documented technical/test record proves otherwise.

The pilot may stop a defective form. It cannot establish structure, reliability, cutoffs or norms. If v1 is unchanged and the analysis plan is not altered in response to its outcomes, post-registration pilot rows may remain in the Phase C total with a sensitivity analysis excluding them. Any outcome-informed analytical change is an amendment and moves the primary test to a fresh holdout. Any wording change creates a new form and v1 rows are not pooled with it.

8. Phase C — clinical validation

8.1 Sample

Recruit consecutive eligible first fills during ordinary care after registration. Target 600 complete first fills, with at least 200 in each age band, at least 240 in each of the two sex groups used for the planned comparison, neither group below 40% within a band, and at least 30 participants aged 7 and 30 aged 18. Recruit from at least three clinics; no clinic should provide more than 60%. Participants with another/unknown sex remain in overall analyses and are described but are not silently assigned for a two-group test.

The binding analysis is three-group age invariance, not a persons-per-item rule. Around 200 per age group gives useful precision for ordinal thresholds and moderate loadings and permits detection of practically material loading differences; approximately 300 per sex group is stronger. Six hundred is also ample for the 21-indicator primary CFA. It is not automatically sufficient: if an age band or response category is sparse, the affected invariance step is refused rather than collapsed post hoc. Age-by-sex invariance and single-year factor models are not planned.

8.2 Primary structural model

Fit the fixed developer key as three correlated factors to the 21 ordered four-category items using an ordinal probit model with WLSMV (or an exactly equivalent diagonally weighted least-squares estimator), robust standard errors and polychoric associations. No correlated residual is freed and no item is reassigned. Report robust chi-square, CFI, TLI, RMSEA with 90% CI, SRMR, loadings with CIs, thresholds, residuals and factor correlations.

Support requires a proper solution; all target loadings positive and statistically distinguishable from zero; no standardised loading below .30; factor-correlation upper CIs below .95; CFI and TLI at least .95, RMSEA no greater than .06 with upper 90% CI no greater than .08, and SRMR no greater than .08. These are preregistered decision gates, not universal laws. A small miss on one global index with otherwise coherent parameters is reported as mixed, not relabelled success. An improper solution, a factor correlation indistinguishable from one, a loading below .30, or material failure of two or more global gates refuses the three-subscale structural claim. Modification indices never change the confirmatory model.

8.3 Prespecified secondary structure

Fit (1) a three-factor ESEM with target rotation and (2) a bifactor-ESEM with a general distress factor plus Depression, Anxiety and Stress specific factors, using the same ordinal estimator. They exist because recent DASS-Y evidence shows that forced-zero cross-loadings can inflate factor correlations and that a general factor may coexist with specific factors. They do not replace the primary model because they fit better.

Report target and cross-loadings, factor correlations, factor determinacy, explained common variance, percentage uncontaminated correlations, omega hierarchical for the general factor and omega hierarchical-subscale for specific factors. A higher-order model is not tested: with only three first-order factors it adds little independently testable information. A one-factor model is not promoted as a candidate.

TOTAL remains a descriptive arithmetic composite unless the bifactor-ESEM is proper and replicable, general-factor determinacy is at least .90, omega hierarchical is at least .80, and explained common variance is at least .60 without erasing meaningful specific factors. Failing any gate leaves TOTAL without a clinically distinct interpretation. It may never substitute for the three subscales merely because it is more reliable.

8.4 Internal consistency

Estimate categorical model-based omega with 95% bootstrap CIs for each supported subscale. A lower CI of at least .70 is the minimum for group-level interpretation; individual change requires separate reliability and error evidence. Report coefficient alpha with CI only for comparability with prior papers, explicitly noting its stronger assumptions. No omega/alpha is interpreted before structural support. TOTAL reliability is reported only under the TOTAL rule above.

8.5 Construct validity and clustering

There is no validated, rights-cleared Persian comparator spanning ages 7–18 in the repository. Paper 1 therefore makes no confirmatory convergent-validity or criterion-validity claim. Associations with routinely co-administered PHQ-9, GAD-7, SIM-YI, diagnosis or functioning may be described only as exploratory, with their timing, age limits and evidence limitations; they cannot rescue a failed structure. Clinic heterogeneity is described and assessed by leave-one-clinic-out sensitivity. With fewer than 20 clinics, no unstable multilevel factor model will be fitted.

9. Cross-group validity and invariance

Comparisons that matter now are the repository's age bands and recorded sex. Other administration, region and clinic comparisons are descriptive until an intended use and adequate sample exist.

For the supported primary factor model, use the ordered-categorical sequence and identification of Wu and Estabrook: configural, equal thresholds, equal loadings with thresholds, then equality sufficient for latent-mean comparison under the chosen parameterisation; residual equality is reported as a stricter optional step. Test age and sex separately. Judge changes using robust scaled tests, parameter differences and approximate guides of ΔCFI no worse than -.010, ΔRMSEA no greater than .015, and ΔSRMR no greater than .030 for loading constraints or .010 for the latent-mean level. The guides do not override large item-level noninvariance.

If full invariance fails, examine modification indices once to locate—not erase—the problem. A prespecified partial-invariance model may free the smallest defensible set, must retain at least two invariant items per factor, and is labelled partial. Item-level ordinal DIF by age and sex, with false discovery-rate control and an effect-size estimate, is exploratory because n=600 is not powered for every item-by-group contrast. A flag prompts content review; it never deletes an item automatically.

Without the relevant latent-mean level of invariance, no observed or latent age/sex mean comparison is interpreted and no group-conditioned reference value is allowed. Invariance is a measurement property, not a new Simorq evidence stage.

10. Test–retest reliability and measurement error

Nest a separate substudy of 120 participants to obtain at least 100 analysable stable pairs after attrition. Sample all three age bands across at least three clinics. Repeat the unchanged form after 8–14 days, before a major planned treatment change where feasible. At retest collect youth global change and document acute events/treatment changes. The primary stable subset reports no meaningful change and no documented major intervention or crisis; all pairs are shown in sensitivity analysis.

For each structurally supported subscale estimate two-way absolute-agreement single-measure ICC with 95% CI, mean change with CI and Bland–Altman limits. At n=100 a true ICC around .80 is estimated with roughly ±.07 precision, which is adequate for a first reliability study but not a universal individual decision rule. Estimate agreement SEM from the test–retest error component and SDC_individual = 1.96 × sqrt(2) × SEM; report uncertainty. The generic alpha-based SEM formula retained elsewhere in the repository is not used for longitudinal error here: internal consistency is not temporal agreement.

11. Later longitudinal questions

Responsiveness is not forced into Paper 1. A later cohort needs at least 200 eligible pairs (target 250 allowing attrition), a prespecified clinically meaningful interval, unchanged form/informant, treatment/context, and an independent youth-centred global-change or functional anchor; aim for at least 60 anchor-improved and 60 anchor-unchanged participants. Test longitudinal threshold/loading invariance before interpreting change, then test directional hypotheses and report change-score distributions by anchor group.

Reliable change may be estimated only from the test–retest error above and for the same version, informant and population. SEM is not minimal important change; reliable change is not clinical importance; percentage change is not assumed valid. A minimal important change is attempted only if the anchor is credible, its association with change is at least .50, and at least 100 participants each are available in prespecified minimally improved and stable groups. Otherwise the result is an explicit refusal. No LLM computes any authoritative statistic.

12. Severity and cutoff strategy

Paper 1 cannot justify cutoffs_only. Its clinical distribution is neither a population percentile frame nor an external criterion.

1. Reuse developer bands: possible only if a representative Persian reference arm supports cross-cultural comparability and shows that the developer raw boundaries reproduce their intended percentile locations with acceptable uncertainty. They remain descriptive severity ranges, not diagnoses.
2. Adapt bands: requires the same reference study and a prespecified reason to preserve percentile meaning while changing raw boundaries. This is more likely to create a distribution readout than cutoffs_only.
3. Derive a screening cutoff: a distinct intended use and study. Use consecutive participants from the target clinical/screening setting, an assessor blinded to DASS-Y, and a prespecified structured diagnostic interview or other defensible criterion. Depression, anxiety and stress need construct-matched criteria; Stress has no simple diagnostic gold standard. Precision, not a round total, sets n: for sensitivity or specificity near .80 with a 95% half-width of .07, about 125 criterion-positive and 125 criterion-negative participants are needed, followed by independent or held-out validation. The 2026 Indonesian total-score cutoff answers a different language/population/composite-disorder question and does not transfer.
4. Decline bands: required whenever the reference or criterion design is absent, incomparable or imprecise. This is the default.

13. Population/reference strategy

Reference interpretation is optional. For clinical monitoring, raw subscales plus defensible measurement error and responsiveness may be more useful than a population rank. T-scores and percentiles do not express clinical importance. Before recruitment, Simorq must state the decision the reference value would change; otherwise this arm is not run.

If useful, recruit a separate school/household probability or calibrated multi-stage sample of Persian-speaking 7–18 year olds not selected for care. Cover all single years of age; balance sex; sample multiple provinces and metropolitan, smaller-city and rural settings; measure socioeconomic/school type and nonresponse. Clinical/help-seeking fills are never pooled with it.

Planning target is 1,200 (about 100 per single year), with a minimum of 960 only if prespecified bootstrap precision is met, neither sex below 40% in any year, and no major region/socioeconomic gap. Fit continuous age distributions with GAMLSS or an equivalently validated regression-norming model; include sex only if invariance and incremental model fit support it. The stopping target is 2,000-bootstrap 95% CI half-width no greater than 0.15 SD for the modelled 90th percentile at every prespecified age/sex profile. If 1,200 does not meet it, continue recruitment or refuse tail percentiles rather than publish unstable cells. Validate smoothness/fit out of sample or by bootstrap; report raw means, SDs and quantiles with uncertainty before considering T-scores.

The Shabani school sample may inform expected variance, age coverage and recruitment feasibility. Its convenience/purposive frame and old wording mean it cannot be pooled, used as the Simorq reference, or treated as a prior that overrides the new data.

14. Measures, recruitment and provenance

Phase A uses the exact form, interview guide and optional alternatives only. Phases B/C use routine clinical v1 fills; no research-only product feature is required. The test–retest and later longitudinal studies add a change anchor under ethics approval. A diagnostic or reference arm receives its own later protocol and consent materials.

Eligibility: Persian-speaking self-report respondents aged 7–18; exact form version and digital mode; research consent/assent as applicable; valid age, informant and completion timestamp. Exclude staff/test/preview records, technical corruption, wrong form/version/mode, duplicate same-person first fills (retain earliest; timestamp tie by pooled_at) and absent research permission. Do not exclude on score, completion time, clinic, diagnosis or response pattern. Report recruitment flow and all exclusions.

15. Missing data

Report item-level and record-level missingness by age band, sex and clinic. The dossier authorises no prorating, so a subscale score and its reliability require all seven items; the primary CFA uses complete 21-item vectors. If fewer than 90% of otherwise eligible rows are complete, the primary result is reported but the structural claim is refused pending a missing-data study; pairwise ordinal or multiple-imputation analyses may be shown only as exploratory sensitivity. Timing fields may be missing without excluding a response from non-timing analyses. No value is imputed for product scoring.

16. Confirmatory and exploratory work

Confirmatory: Phase A questions/stopping rule; Phase B displays; primary CFA and gates; omega; age/sex invariance sequence; TOTAL rule; test–retest model. Prespecified secondary: ESEM and bifactor-ESEM for structural explanation. Exploratory: DIF, routine-comparator associations, clinic sensitivity, post-hoc residual/local misfit displays and any analysis proposed after seeing outcomes. Exploratory findings generate a new protocol; they do not rewrite this one. All code, software versions, random seeds, extraction identifier and exact template hash are archived under the shared analysis standard.

17. Decision and refusal rules

• A statistic flags; a human decides. Every evidence decision cites the dated report and retains contrary findings.
• Phase A failure changes the form version or narrows the supported age range. Quantitative evidence never repairs misunderstood wording.
• Pilot failure pauses recruitment; no automatic item deletion occurs.
• Failed/mixed CFA leaves subscale structural validity unsupported. Omega does not rescue it.
• Failed invariance prohibits the affected group comparison/stratification; it does not create an evidence stage.
• TOTAL is clinical only under its stricter bifactor rule; otherwise it is an explicitly descriptive sum.
• Paper 1 does not promote the norm class. Bands/reference readouts require the separate evidence above and a human dossier edit.
• A precise refusal—no cutoff, no norm, no MIC, no comparison—is a valid study result.

18. Ethics, consent and data provenance

Obtain ethics review before recruitment/export. Use developmentally suitable assent plus guardian consent or a documented lawful waiver for secondary use of consented clinical data. Care, safety escalation and access are independent of research participation; DASS-Y is not a suicide assessment. Minimise and pseudonymise clinical data, retain the clinic/setting partition, control access to free text and publish only disclosure-safe aggregates. Record assistance and mode. Community participants must be told whether anonymous submissions cannot later be withdrawn. Report conflicts plainly: Simorq selected the translation, operates the platform and conducts the study; seek independent psychometric review/co-authorship before submission.

19. Publication plan

Paper 1 — Persian adaptation and clinical measurement properties. One article, with Phase A as a clearly separated study and Phases B/C plus the test–retest substudy as the clinical validation study. It reports the defect and 12-item correction history, historical omissions, target-user content validity, pilot behaviour, structure, internal consistency, supported invariance and measurement error. It explicitly refuses population norms and unsupported bands. The translation work is not a separate paper: it is scientifically necessary but too interdependent with validation to justify administrative splitting.

Paper 2 — Persian population reference values, only if useful. The separately recruited reference arm answers population distribution, continuous age/sex modelling and whether the developer percentile bands retain meaning.

Possible later paper — longitudinal performance. Only if a sufficiently large repeated clinical cohort with a credible anchor supports a distinct responsiveness/reliable-change/MIC question. A diagnostic cutoff, if ever an intended use, is a separate criterion study rather than an appendix to Paper 1.

20. What this protocol cannot justify

It cannot make v1 diagnostic; validate the published Shabani wording; carry over old coefficients, distributions or bands; claim national representation from clinics; infer responsiveness from test–retest reliability; call SEM an MIC; impose a universal RCI or percentage-change rule; compare groups without invariance; or claim ages 7/18 if those boundary participants are absent or do not understand the form.

21. References

• Beaton DE, Bombardier C, Guillemin F, Ferraz MB. Guidelines for the process of cross-cultural adaptation of self-report measures. Spine. 2000;25: 3186–3191. doi:10.1097/00007632-200012150-00014.
• Cao C-H, et al. Evaluating the psychometric properties of the Chinese DASS-Y and DASS-21. Child Adolesc Psychiatry Ment Health. 2023;17:106. doi:10.1186/s13034-023-00655-2.
• Chen FF. Sensitivity of goodness of fit indexes to lack of measurement invariance. Struct Equ Modeling. 2007;14:464–504. doi:10.1080/10705510701301834.
• Dunn TJ, Baguley T, Brunsden V. From alpha to omega. Br J Psychol. 2014;105:399–412. doi:10.1111/bjop.12046.
• Jovanović V. The DASS-Y: evidence of validity and cross-national and gender measurement invariance. Eur Child Adolesc Psychiatry. 2025;34:1317–1327. doi:10.1007/s00787-024-02561-0.
• Jovanović V. A re-examination of the structure of the DASS-Y: a move beyond the standard model. Eur J Psychol Assess. 2026. doi:10.1027/1015-5759/a000950.
• Kurnianto A, et al. Diagnostic accuracy of the Indonesian DASS-Y for screening internalizing disorders among adolescents. Psychol Res Behav Manag. 2026;19:617191. doi:10.2147/PRBM.S617191.
• Malterud K, Siersma VD, Guassora AD. Sample size in qualitative interview studies: guided by information power. Qual Health Res. 2016;26:1753–1760. doi:10.1177/1049732315617444.
• Mokkink LB, et al. The COSMIN study reached international consensus on taxonomy, terminology and definitions of measurement properties. J Clin Epidemiol. 2010;63:737–745. doi:10.1016/j.jclinepi.2010.02.006.
• Rodriguez A, Reise SP, Haviland MG. Evaluating bifactor models. Psychol Methods. 2016;21:137–150. doi:10.1037/met0000045.
• Shabani MJ, Gharraee B, Zahedi Tajrishi K. Exploring the psychometric properties of the Persian DASS-Y. Front Psychol. 2025;15:1452878. doi:10.3389/fpsyg.2024.1452878.
• Śliwerski A, Koszałkowska K, Socha I. Validation of the DASS-Y in Poland. PLOS ONE. 2025;20:e0323835. doi:10.1371/journal.pone.0323835.
• Szabo M, Lovibond PF. Development and psychometric properties of the DASS-Y. Front Psychol. 2022;13:766890. doi:10.3389/fpsyg.2022.766890.
• Terwee CB, et al. COSMIN methodology for evaluating the content validity of patient-reported outcome measures. Qual Life Res. 2018;27:1159–1170. doi:10.1007/s11136-018-1829-0.
• Wang X, et al. Does the DASS-Y remain consistent across time and diverse school-aged youth? Int J Ment Health Promot. 2026;28. doi:10.32604/ijmhp.2026.075149.
• Wild D, et al. Principles of good practice for the translation and cultural adaptation process for PRO measures. Value Health. 2005;8:94–104. doi:10.1111/j.1524-4733.2005.04054.x.
• Wu H, Estabrook R. Identification of CFA models of different levels of invariance for ordered categorical outcomes. Psychometrika. 2016;81:1014–1045. doi:10.1007/s11336-016-9506-0.

Developer pages checked 2026-09-16: DASS-Y form and scoring, severity bands, and DASS FAQ.