In 7182 Soweto women, the standard sleep questionnaire did not hold together as one score

The Pittsburgh Sleep Quality Index is the most widely used self-report sleep instrument in research, and its global score — a single number, with 5 as the conventional cut-point — is what most studies report. A psychometric evaluation in 7,182 South African women finds that the single number does not describe what the questionnaire is actually measuring in that population [s1].

The test

Data came from women enrolled in the Bukhali randomised controlled trial in Soweto, part of the Healthy Life Trajectories Initiative, with sociodemographic information and PSQI responses collected through interviewer-administered surveys [s1].

Internal consistency was assessed with Cronbach's alpha, McDonald's omega and item-level correlations, and confirmatory factor analysis evaluated the original one-factor model alongside established two- and three-factor multidimensional models, with fit examined using RMSEA, CFI and TLI [s1].

Internal consistency was modest: α = .57, ω = .57, with item-total correlations ranging from .33 to .65 [s1].

The one-factor model — the structure implied by reporting a global score — demonstrated poor fit: χ²(14) = 1677.70, CFI = .65, TLI = .48, RMSEA = .13 [s1]. Fit was good for the two-factor model (χ²(13) = 291.46; CFI = .94; TLI = .91; RMSEA = .05) and for the three-factor model (χ²(11) = 285.21; CFI = .94; TLI = .89; RMSEA = .06) [s1].

The authors conclude that the PSQI showed a multidimensional structure in this population, that the two-factor model showed slightly better fit and is recommended for use here, and that interpreting PSQI components rather than the global score offers more meaningful insight — while highlighting the need for context-specific psychometric validation in diverse settings [s1].

What that means in practice

A CFI of .65 and RMSEA of .13 for the one-factor model [s1] is not a marginal result. It says the component scores in this sample do not behave as indicators of one underlying construct. Adding them up still produces a number; that number just does not mean "overall sleep quality" in the way the instrument's design assumes.

The internal consistency figure points the same way. An alpha of .57 [s1] would be treated as unacceptable for a unidimensional scale in most measurement contexts.

None of this makes the PSQI useless in Soweto. Two- and three-factor models fit well [s1], which means the items carry real, structured information — just organised differently than a global score presumes.

The descriptive findings are their own story

Most women, 57.4%, reported good sleep quality by the conventional threshold of PSQI ≤ 5 [s1]. Average sleep was 7.8 hours, characterised by prolonged onset latency and high fragmentation, which the authors read as continuity disruptions rather than opportunity shortage driving poor sleep [s1].

That distinction matters for anyone designing an intervention. Sleep duration is close to what guidelines describe as adequate; the problem is getting to sleep and staying asleep. Programmes built around extending sleep opportunity would be aimed at the wrong target.

The socioeconomic pattern also runs against expectation. Poor sleep quality was associated with higher household socioeconomic status measured by assets score, higher education, and single relationship status [s1]. In much of the high-income literature the gradient runs the other way. The study is cross-sectional on these associations and offers no causal account, but it is the kind of inversion that only appears when an instrument is applied in a setting it was not developed in.

Why this lands on a literature that uses global scores

To see what is at stake, consider what the regional evidence base looks like. A systematic review and meta-analysis published in May pooled nine studies with a combined sample of 3,147 people living with HIV/AIDS in Ethiopia and estimated a pooled prevalence of poor sleep quality of 52.12% (95% CI 43.58–60.66) [s2]. Symptoms of anxiety and depression, CD4 count below 200 cells/mm³ and living alone were significant factors [s2].

That pooled prevalence is a global-score prevalence. It is derived by counting people above a threshold on exactly the summed measure whose one-factor structure fit poorly in the Soweto cohort [s1]. If the global score behaves differently across African settings — and the Soweto paper is the evidence that it can — then the confidence interval around a pooled prevalence understates the real uncertainty, because it captures sampling variation and not measurement variation.

This is not a criticism of the Ethiopian meta-analysis, which did what the available primary studies allowed. It is a description of a dependency the field has not been in a position to check.

Limits of the validation itself

The sample is specific: young women in Soweto enrolled in a randomised trial [s1]. Trial participants differ from the general population, and a validation in one urban South African cohort does not establish how the PSQI behaves in men, in older adults, or in rural settings elsewhere on the continent. The paper's own recommendation is scoped accordingly — the two-factor model is recommended for use in this population [s1].

Data were collected by interviewer-administered survey [s1], which changes response patterns relative to self-completion and is itself part of what was validated.

And factor structure is not accuracy. Nothing here compares PSQI responses against actigraphy or polysomnography, so the study establishes how the items relate to one another, not whether they correspond to measured sleep.

What follows

The practical recommendation the authors make is small and immediately actionable for researchers: report and interpret PSQI components or factors, not only the global score [s1]. The larger implication is the one their conclusion states directly — context-specific psychometric validation in diverse settings is a prerequisite, not an optional extra [s1].

For a region where the sleep evidence base is thin to begin with, that is an awkward finding arriving at an awkward time. It suggests that some of what has been measured needs re-reading before more of it is accumulated.

Sources

• Psychometric evaluation of the Pittsburgh Sleep Quality Index among South African women participating in the Bukhali trial: Healthy Life Trajectories Initiative, Journal of Clinical Sleep Medicine, published online 3 August 2026
• Poor quality of sleep among people living with HIV/AIDS attending at health care setting: a systematic review and meta-analysis from Ethiopia, AIDS Care, published online 4 May 2026