Sleep Test Fails Its Global Score: Soweto Study Reshapes How We Measure Women’s Sleep

For more than three decades, the Pittsburgh Sleep Quality Index has been one of the most trusted tools in sleep science, a 19-item questionnaire that clinicians and researchers worldwide use to decide whether someone sleeps well or badly. But a sweeping new study from South Africa suggests that the instrument’s famous single global score may be hiding as much as it reveals, at least for young women living in under-resourced urban neighborhoods. In an analysis of 7,182 women enrolled in the Bukhali randomized controlled trial, part of the Healthy Life Trajectories Initiative, researchers found that the questionnaire simply does not behave as a unified measure of sleep quality in this population. Instead, sleep in Soweto splits into two related but distinct dimensions, a finding with consequences that stretch far beyond one South African township.

The study, published in the Journal of Clinical Sleep Medicine, set out to answer a deceptively simple question: does a questionnaire validated largely in high-income, Western settings measure sleep the same way among young, predominantly low-income women in urban South Africa? The answer, according to the research team led by Stephanie Alcock of the University of the Witwatersrand, is a qualified no. When the researchers subjected the data to confirmatory factor analysis, the statistical technique used to test whether a set of questionnaire items hangs together as a single underlying construct, the original one-factor model of the PSQI performed poorly. The model produced a Comparative Fit Index of just 0.65 and a Tucker-Lewis Index of 0.48, both far below the accepted threshold of 0.90, with a Root Mean Square Error of Approximation of 0.13, more than double the 0.06 cutoff that signals good fit.

In plain terms, the numbers mean that treating the seven components of the PSQI, which cover subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleep medication, and daytime dysfunction, as reflections of one single latent trait called sleep quality does not describe how these women actually experience their sleep. When the researchers instead tested multidimensional alternatives, the picture changed dramatically. A two-factor model, which grouped sleep duration and habitual sleep efficiency under one latent factor and the remaining components under another, achieved a Comparative Fit Index of 0.94 and an RMSEA of 0.05, comfortably within the range considered good. A three-factor model performed nearly as well, but with a caveat: the correlation between its sleep latency and sleep quality factors exceeded one, a statistical red flag indicating those two constructs could not be reliably distinguished in this sample.

The internal consistency figures tell a similar story of an instrument under strain. Cronbach’s alpha and McDonald’s omega, the two standard measures of how coherently a scale’s items hang together, both came in at 0.57, well below the conventional benchmark of 0.70, though some methodologists accept values of 0.50 or higher. Item-total correlations ranged from 0.33 for the sleep medication component to 0.65 for sleep duration, and the sleep medication item showed the weakest relationship with the rest of the scale at just 0.15. The researchers note that removing the medication component barely changed the overall reliability, a pattern likely rooted in context: prescription sleep aids are often unavailable or unaffordable in this setting, and many women may not view poor sleep as a condition that warrants treatment at all.

Perhaps the most striking finding is not statistical but epidemiological. On average, the women in the study spent 9.5 hours in bed and reported sleeping 7.8 hours, figures that would suggest ample sleep opportunity by almost any standard. Yet 42.6 percent of the sample scored above the PSQI’s global cutoff of 5, classifying them as poor sleepers, a prevalence that exceeds estimates from Germany but sits in line with figures from Spain and South Korea. The median bedtime was 22:00 and the median wake time 07:00, but median sleep onset latency stretched to 20 minutes and fragmentation scores pointed to frequent nocturnal disruptions. The data paint a picture in which the problem is not too little time for sleep but too little continuous sleep, a continuity crisis rather than an opportunity shortage.

This distinction matters because it reframes what poor sleep means in this population. The researchers found that longer time in bed was moderately associated with longer sleep duration, meaning that extra opportunity did translate into more sleep, but greater fragmentation was linked to shorter sleep duration. In other words, women were getting enough hours in bed, yet nocturnal interruptions, whether from noise, crowding, caregiving, or hypervigilance related to safety concerns, were eroding the restorative value of that time. The authors suggest that in urban South African townships, environmental and psychosocial stressors such as material deprivation, overcrowded households, caregiving demands, and fear-driven alertness may disrupt sleep onset, reduce efficiency, and normalize fatigue, causing women to experience disturbed sleep, daytime tiredness, and negative evaluations of their sleep as one blended phenomenon rather than separate problems.

The sociodemographic correlates add another layer of intrigue. Poor sleep quality was associated with higher household socioeconomic status measured by an asset score, higher educational attainment, and single relationship status, while age made no difference in this young cohort. The authors caution that these associations, drawn from a global cutoff score that the study itself undermines, should be interpreted carefully. Still, they speculate that educational and employment pressures, stressors accompanying low socioeconomic conditions, and reduced social support among single women may combine to shape sleep in ways that defy simple assumptions about poverty and rest. The finding that higher asset scores tracked with worse sleep is counterintuitive and underscores how context-specific sleep research must be.

The implications for measurement science are considerable. The PSQI’s factor structure has been contested for years, with studies reporting one-, two-, and three-factor solutions across populations as varied as Canadian adults, Sri Lankan patients, Portuguese older adults, Iranian students, Peruvian pregnant women, and New Zealand female workers during COVID-19 lockdowns. Even within a single multinational student study, the factor structure shifted between countries. The new South African results reinforce a growing consensus that sleep quality is not a universal, monolithic construct but a context-sensitive one, and that exporting a scoring scheme developed in Pittsburgh to a township in Johannesburg without local validation risks misclassification and misinterpretation. The authors recommend the two-factor model for this population because it is more parsimonious, fits slightly better, and avoids the discriminant validity problems that plagued the three-factor alternative.

Clinically, the message is that screening and intervention should target specific PSQI components, particularly nocturnal disruptions and the environmental factors that drive them, rather than relying on a global score that conflates distinct experiences. For young South African women, the authors argue, interventions may need to address stress-related sleep disturbance, environmental noise and crowding, and perceived sleep quality rather than simply exhorting people to spend more hours in bed. The study also positions sleep as a social determinant of health, shaped by safety, housing conditions, caregiving burdens, and financial stress, and calls for sleep health to be integrated into community health, women’s health, and public policy initiatives.

The researchers are candid about the limitations of their work. The cross-sectional design cannot capture how sleep patterns change over time, the sample of young urban women limits generalizability to men, older adults, and rural populations, and self-report measures carry the risk of reporting bias and misinterpreted items. Because no qualitative interviews accompanied the survey, the team cannot say how respondents actually understood phrases like sleep quality, and the data violated assumptions of multivariate normality, which may affect some parameter estimates. Future research, they write, should test whether the PSQI measures sleep equivalently across different groups within South Africa, deploy longitudinal designs to untangle the relationships between sleep domains and mental health, caregiving burden, and socioeconomic stress, and use qualitative methods to learn how women in Soweto define good sleep on their own terms. Until then, the study stands as a warning to global health researchers: a questionnaire is only as good as the evidence that it means the same thing everywhere it is used.

Subject of Research: Psychometric validation of the Pittsburgh Sleep Quality Index among young South African women

Article Title: Psychometric evaluation of the Pittsburgh Sleep Quality Index among South African women participating in the Bukhali trial: Healthy Life Trajectories Initiative

Article References: Psychometric evaluation of the Pittsburgh Sleep Quality Index among South African women participating in the Bukhali trial: Healthy Life Trajectories Initiative. (n.d.). https://doi.org/10.1007/s44470-026-00146-z

Image Credits: AI Generated

DOI: 10.1007/s44470-026-00146-z

Keywords: Pittsburgh Sleep Quality Index, sleep quality, psychometrics, confirmatory factor analysis, South Africa, Soweto, women's health, sleep fragmentation, Bukhali trial, Healthy Life Trajectories Initiative, public health, measurement invariance