INTRODUCTION
Insomnia, characterized by difficulty initiating or maintaining sleep despite adequate opportunity, is a prevalent condition affecting approximately 10% of the adult population, with an additional 20% experiencing occasional symptoms. It is more common among women, older adults, and individuals facing socioeconomic hardships. Chronic insomnia often pereISSN 1976-3026 OPEN ACCESS sists over time, with a 40% persistence rate over a five-year period [1]. The impact of insomnia extends beyond sleep disturbances, significantly affecting daytime functioning and overall health. Individuals with insomnia are at a higher risk for developing psychiatric conditions, particularly depression and anxiety disorders. In fact, insomnia has been identified as a predictor of the onset of depression within one to three years. Moreover, insomnia is associated with various medical conditions, including cardiovascular diseases, hypertension, chronic pain, and metabolic disorders [2]. The socioeconomic consequences of insomnia are substantial. Individuals suffering from insomnia often experience impaired cognitive performance, diminished productivity, and increased absenteeism, all of which contribute to significant losses in workplace efficiency and functioning. These outcomes are further compounded by increased healthcare utilization and frequent medical consulta-tions, reflecting the broader burden insomnia places on healthcare systems [3]. The overall economic cost of insomnia, when taking into account direct medical expenses and indirect costs such as lost productivity, is estimated to range between $35 and $107 billion annually in the United States alone [4].
Ideally, sleep should be spontaneous, just like heartbeats and breathing. The psychobiological inhibition model5 assumes that sleep is a biological process, and behavioral, cognitive, and emotional factors (such as thoughts, feelings, and environmental interactions) serve as “setting conditions” that can facilitate or inhibit sleep. In the attention-intention-effort pathway [6], people who experience difficulty sleeping direct their attention toward sleep itself, leading to the formation of a conscious sleep intention. The intention escalates into sleep effort— active behaviors aimed at triggering sleep, such as relaxation, rituals, or behaviors that target sleep. Sleep effort refers to sleep-related performance anxiety and includes both cognitive (such as obsessive thoughts like “I must sleep”) and behavioral (such as trying too hard to fall asleep) components. Good sleepers typically make no effort to sleep, while poor sleepers often engage in excessive efforts like overthinking about sleep, neglecting sleep hygiene, or using over-the-counter sleep medications. This excessive effort paradoxically increases arousal and vigilance, making sleep more elusive and contributing to the development and maintenance of insomnia. Thus, measuring and individual’s sleep effort is important for exploring the mechanisms underlying insomnia.
The Glasgow Sleep Effort Scale (GSES) [7] was developed as a 7-item self-report rating scale that can measure one’s anticipatory fears and anxieties about one’s ability to sleep, which are collectively known as “sleep effort.” Each item is scored on a 3-point Likert-like scale from 0 (not at all) to 2 (very much). Higher scores, ranging from 0 to 14, indicate greater sleep effort. The GSES has been translated into various languages such as Persian [8], Arabic [9], Turkish [10], and Korean. Although the Korean version of the GSES was already developed and validated [11], we aimed to explore the psychometric properties of the Korean GSES using modern tests under the item response theory (IRT) framework, including the Rasch and graded response models (GRM). The aim of this study was to explore the psychometric properties of the Korean version of the GSES among both the general population and individuals who report insomnia, using sleep-wake patterns and previously validated rating scales.
METHODS
Participants
We collected data in two parts: Study I used the general population, and Study II used individuals who reported insomnia. Study I was conducted via an anonymous online survey during July 17-27, 2023. It was conducted with the general population via the EMBRAIN survey platform (Seoul, Korea). We aimed to collect 300 responses based on sample size estimation by allocating 30 samples for 10 cells (sex×five age groups) based on the central limit theorem [12]. All 7,312 enrollment emails were sent to 1.7 million registered panels, and 381 participants completed the survey among 680 participants who accessed the survey. Responses completed too quickly or too slowly were removed, and for each quota, the fastest 5% of respondents based on response time were excluded to ensure data validity. Additionally, responses with an average time between questions exceeding three times the overall average were excluded. Finally, the company provided the first 300 deidentified responses to the researchers. The Institutional Review Board (IRB) of Asan Medical Center approved the study protocol (approval no.: 2023-0871).
Study II was performed via an anonymous online survey during June 18-24, 2025 among the general population, via the EMBRAIN survey platform, and we aimed to collect responses of participants complaining of insomnia. In this survey, participants were screened by asking if they had “difficulty falling asleep or maintaining sleep in the past three months” and the question “Do you go to bed between 9:00 PM and 1:00 AM every day?” We aimed to collect 600 responses based on sample size estimation by allocating 50 samples for 12 cells (sex×six age groups) based on the central limit theorem [12]. All 7,558 enrollment emails were sent to 1.8 million registered panels, 1,882 surveys were accessed, and 640 participants completed the survey. Responses completed too quickly or too slowly were removed, and for each quota, the fastest 5% of respondents based on response time were excluded to ensure data validity. Additionally, responses with an average time between questions exceeding three times the overall average were also excluded. Finally, the company provided the first 600 deidentified responses to the researchers. The IRB of Asan Medical Center approved the study protocol (approval no.: 2025-0607).
In both Studies I and II, we developed a survey form that included questions about participants’ age, sex, marital status, and past psychiatric history. Participants were excluded at the beginning of the two surveys if they reported working shifts. Only those who selected “yes” to the question indicating agreement to participate in this study were permitted to continue.
Measures
GSES
The GSES [7,13] is a brief, 7-item self-report questionnaire developed by Broomfield and Espie [7] in 2005 to assess the degree of cognitive and emotional effort individuals invest in trying to sleep—a key factor in psychophysiological insomnia. Each item is rated on a 3-point Likert scale, evaluating aspects such as preoccupation with sleep, anxiety about sleep, and perceived control over sleep. Higher total scores indicate greater sleep effort, reflecting maladaptive sleep-related cognitions and behaviors. The Korean version of the GSES has previously been developed [11].
Insomnia Severity Index
The Insomnia Severity Index (ISI) [14] is a brief, reliable, and validated self-report questionnaire developed to assess the severity of insomnia and its impact on daily functioning. It consists of seven items that evaluate key aspects of insomnia. Each item is rated on a 5-point Likert scale, with total scores ranging from 0 to 28. Based on the total score, insomnia is categorized as follows: 0-7 (no clinically significant insomnia), 8-14 (subthreshold insomnia), 15-21 (moderate severity), and 22-28 (severe). The Korean version of the ISI was validated [15], and the Cronbach’s alpha values were 0.820 and 0.838 in Studies I and II.
Dysfunctional Beliefs and Attitudes about Sleep-6
The Dysfunctional Beliefs and Attitudes about Sleep-6 (DBAS-6) [16] is a shortened, data-driven version of the original 16-item DBAS scale [17], developed using machine learning to improve clinical efficiency without losing accuracy. Researchers applied factor analysis and machine-learning techniques to select six key items that best represent the full scale, achieving a high predictive accuracy (R²=0.90). The DBAS-6 effectively captures core dysfunctional sleep beliefs and has strong reliability across diverse populations, making it a practical tool for assessing sleep-related cognitive distortions in both research and clinical settings. Cronbach’s alphas were 0.839 and 0.769 in Studies I and II.
Patient Health Questionnaire-9
The Patient Health Questionnaire-9 (PHQ-9) [18] is a rating scale that measures depression severity. Each item is rated on a 4-point Likert scale, with total scores ranging from 0 to 27. Severity of depression is categorized as follows: 0-4 (minimal), 5-9 (mild), 10-14 (moderate), 15-19 (moderately severe), and 20-27 (severe). The Korean version of the PHQ-9 has been validated [19]. Cronbach’s alphas were 0.886 and 0.894 in Studies I and II.
Assessing sleep indices and discrepancy between desired time in bed and desired total sleep time
We obtained participants’ sleep-wake patterns to assess their habitual sleep behaviors. Specifically, participants were asked about their sleep-related timing variables, including usual bedtime, the time they typically fall asleep, and the time they get out of bed in the morning. Using these responses, we calculated duration variables, including sleep onset latency (SOL), defined as the interval between bedtime and sleep onset, and time in bed (TIB), calculated as the duration between bedtime and wake-up time.
Additionally, we assessed the discrepancy between desired time in bed and desired total sleep time (DBST) index [20], representing the discrepancy between participants’ desired TIB and their desired total sleep time (TST). To determine this, participants were asked how many hours they wished to sleep daily and their preferred sleep window. The DBST index was then calculated as the difference between the desired TIB and the desired TST.
Statistical analysis
Prior to statistical analysis, we prepared the dataset by excluding participants who reported bedtimes before 9:00 PM or after 1:00 AM to minimize the inclusion of individuals with circadian rhythm sleep-wake disorders, such as delayed or advanced types. After these exclusions, we analyzed 208 participants from Study I and 477 from Study II.
First, an exploratory factor analysis was conducted using the Study I and II samples. Before conducting factor analysis, the normality assumption for all items was checked based on skewness and kurtosis within ±2. The adequacy of the sampling and the suitability of the data for factor analysis were evaluated using the Kaiser-Meyer-Olkin (KMO) measure and Bartlett’s test of sphericity. A parallel analysis was performed to determine the number of factors to retain by comparing the eigenvalues from the actual data with those generated from randomly simulated datasets of the same size, using the 95th percentile of the simulated eigenvalue distribution as the retention threshold. Confirmatory factor analysis (CFA) was conducted for the Study I and II samples, utilizing the diagonally weighted least squares estimation method to evaluate model fit. Good model fit was defined as a standardized rootmean-square residual (SRMR) value of ≤0.05, a root-meansquare error of approximation (RMSEA) value of ≤0.10, and comparative fit index (CFI) and Tucker-Lewis index (TLI) values of ≥0.90.
Second, under the IRT framework, we used the Rasch model and GRM to measure the item-level characteristics of the GSES. This was done to provide a rigorous evaluation of item fitness with the underlying latent trait. In the Rasch model, the infit and the outfit mean square (MnSq) values were calculated to measure the matching of actual item responses to the expected responses. Infit and outfit MnSq values closer to 1 reflect the item’s good alignment to the underlying latent construct, with values ranging between 0.5 and 1.7 indicating acceptable model fit [21]. Furthermore, to assess the scale’s discriminating power at different ability ranges of the latent trait and to check the consistency in item difficulty estimates, person and item reliabilities, along with their separation indices, were used. As indicative of the measurement precision and consistency, the reliability coefficients should be greater than 0.80 and the separation indices should range from 1.5 to 3.0, providing information about the robustness of the scale [22].
Along with the Rasch model, we assessed the item-level properties of the GSES using the GRM. Firstly, GRM fits were assessed through the IRT assumptions, including unidimensionality, monotonicity, and local independence. A Loevinger’s H value greater than 0.5 reflects the scale’s strong unidimensionality. For local independence, a Yen’s Q3 residual coefficient of >0.2 suggests local dependence between scale items. The monotonicity of scale items was determined through a crit value less than 40 [23]. In the GRM, we assessed the item discrimination (α) and difficulty parameters (b) of the GSES. Discrimination values of 0.65-1.34, 1.35-1.69, and >1.70 indicate moderate, high, and very high ability (respectively) of scale items to effectively discriminate among individuals across the latent construct [24]. Item difficulty parameters indicate individual item responses to each category with 50% probabilities across varying ability ranges of the latent trait. We also checked item information curves (IICs) and the test information function (TIF) of the GSES to ensure adequate information coverage across the spectrum of the underlying latent trait.
Third, internal consistency reliability was assessed using McDonald’s omega for both Study I and II samples. Convergent validity was evaluated by calculating Pearson’s correlation coefficients between the GSES with sleep indices, and rating scales such as the ISI, PHQ-9, and DBAS-6, within each study sample. All statistical analyses were performed using the JASP software, version 0.14.1.0 (JASP team; https://jasp-stats.org/). The R package ltm (version 1.2.0; https://www.r-project.org/) was applied to determine the discrimination and difficulty parameters in the GRM model. The R package mokken (version 3.1.2; https://cran.r-project.org/web/packages/mokken/index.html) was used for unidimensionality and monotonicity, and the mirt (version 1.4.2; https://cran.r-project.org/web/packages/mirt/mirt.pdf) was used to determine the Q3 coefficients.
RESULTS
We collected responses from 300 general population participants in Study I, and 600 responses from participants who reported insomnia for at least 3 months in Study II. After excluding participants who did not complete the survey and reported their bedtime being before 9:00 PM or after 1:00 AM, a total of 208 from Study I and 477 from Study II were finally analyzed. Demographic and clinical characteristics of each study are described in Table 1.
Factor analysis
We first explored the factor structure of the Korean version of the GSES using samples from Studies I (Table 2) and II (Table 3). Before factor analysis, we checked whether all items of the GSES were normally distributed based on skewness and kurtosis within ±2 for all items. The sampling was adequate and the data was suitable for factor analysis based on the KMO measures (0.865, Study I; 0.862, Study II) and Bartlett’s test (p<0.001, both Studies I and II). Parallel analysis suggested a single-factor model of the GSES, with actual eigenvalues (3.89 in Study I and 3.34 in Study II) greater than the simulated 95th percentile (1.28 in Study I and 1.17 in Study II). CFA was conducted using samples from Studies I and II. CFA for the single-factor model of the GSES showed a good fit among samples of Study I (CFI=0.986, TLI=0.980, RMSEA=0.066, SRMR=0.043) and Study II (CFI=0.990, TLI=0.986, RMSEA=0.081, SRMR=0.057).
IRT analysis using the Rasch model and GRM
To assess the item-level psychometric properties of the GSES, the Rasch rating scale model was employed under the IRT approach. For both Study I and II samples, the scale items were well-aligned with the Rasch model, indicating good fit statistics. In line with this, the infit and outfit MnSq values were within the commonly accepted ranges (0.5-1.7), with infit and outfit MnSq values ranging from 0.62-1.29 for Study I and 0.72- 1.25 for Study II (Table 4). Moreover, the item reliabilities (0.96 and 0.86 for Study I and II samples, respectively) and separation indices (4.63 and 2.46, respectively) in Table 4 demonstrated excellent measurement precision and higher differentiation among item difficulty levels. Similarly, the person reliabilities (0.77 and 0.82 for Studies I and II, respectively) and separation indices (1.84 and 2.13, respectively) (Table 4) indicated good precision and robustness of the GSES in discriminating individuals across varying levels of the latent construct.
Examination of the model fits for IRT assumptions under the GRM revealed that the GSES had strong unidimensionality (H=0.63 and 0.48 for Studies I and II, respectively) and no local dependence among scale items, with Q3 coefficients ranging from -0.36 to 0.12 for Study I and -0.33 to 0.04 for Study II. In addition, no violations in monotonicity of scale items were found with crit values of 0 for all items in both samples. The discrimination parameters in Table 5 show that the α coefficients ranged from 1.65-4.11 for Study I and from 0.94-2.26 for Study II, indicating that the scale items had moderate-to-very high discriminating power in effectively differentiating individuals across the latent trait. The difficulty parameters (b coefficients) in Table 5 and the TIFs in Figures 1B and 2B indicate that the scale provided more information about the participants across a moderate spectrum of the ability ranges, with theta levels of -0.75 to +2.0 for Study I and -2 to +2 for Study II, indicating more endorsement in some item response categories for Study I to acquire information at wider ability ranges. Moreover, an inspection of the IICs in Figures 1A and 2A indicated that items 3-7 for Study I and items 2-7 for Study II provide a good level of information with distinct abilities about the underlying latent trait. However, items 1 and 2 for Study I and item 2 for Study II provided a low-to-moderate level of information, suggesting the use of under-reported response categories for future validation studies with more diverse samples.
Reliability and evidence based on relations to other variables
Reliability of internal consistency was measured with Mc-Donald’s omega, which showed that the GSES is reliable among the general population (Study I: 0.874) and individuals who reported insomnia (Study II: 0.815). Convergent validity of the GSES with sleep indices and rating scales scores in each study is described in Table 4. In Study I, the GSES total score was significantly correlated with wake-up time (r=0.15, p< 0.05), SOL (r=0.16, p<0.05), TIB (r=0.15, p<0.05), TIB/day (r=-0.24, p<0.01), and DBST index (r=0.20, p<0.01) (Table 6). Furthermore, the GSES was significantly correlated with the ISI (r=0.60, p<0.01), PHQ-9 (r=0.40, p<0.01), and DBAS-6 (r=0.70, p<0.01). In Study II, the GSES total score was significantly correlated with age (r=-0.17, p<0.01), sleep onset time (r=0.20, p<0.01), SOL (r=0.15, p<0.01), desired bedtime (r=-0.09, p<0.05), desired wake-up time (r=0.16, p<0.01), desired TIB (r=0.22, p<0.01), and DBST index (r=0.16. p<0.01). Furthermore, the GSES was significantly correlated with the ISI (r=0.49, p<0.01), PHQ-9 (r=0.43, p<0.01), and DBAS-6 (r=0.61, p<0.01).
DISCUSSION
This study demonstrated that the GSES is a reliable and valid a rating scale for measuring an individual’s preoccupation with sleep, or sleep effort, within the general population, including those reporting insomnia. The GSES demonstrated a significant correlation with prolonged SOL, the DBST index, and rating scales such as the ISI, PHQ-9, and DBAS-6 in both groups.
This study shows good reliability of internal consistency among the general population and those reporting insomnia. As a single-factor model, good model fits were observed among both groups. This indicates that the GSES can be used to measure one’s preoccupation with sleep among the general population complaining of insomnia, as well as the broader general population.
Under the IRT approach, item-level properties of the GSES assessed with the Rasch model and the GRM demonstrated good fit statistics, with infit and outfit MnSq values and itemand person-levels separation indices with corresponding reliability coefficients within the commonly accepted ranges. Moreover, the discrimination and difficulty parameters of the scale items clearly demonstrated the scale’s ability to distinguish between individuals across varying levels of the latent trait. The scale items were highly informative for items 3-7 and low-to-moderately informative for items 1 and 2, indicating different levels of information across the latent trait. Overall, the fit statistics derived from the Rasch model and GRM indicate excellent measurement precision and psychometric robustness of the Korean version of the GSES with strong item differentiation. The GSES was significantly associated with long sleep latency and the DBST index in both groups. Although a causal relationship cannot be established in this study, individuals experiencing prolonged SOL may exhibit increased sleep effort or preoccupation with sleep. Among the Study I sample, the GSES total score was significantly associated with late wake-up time, prolonged TIB, and prolonged TIB/day. Extended TIB or TIB/day may decrease sleep efficiency, which plausibly correlates with the GSES.
The DBST index was significantly associated with the GSES in both studies. The DBST index is a concept that reflects the paradoxical sleep behavior of patients with insomnia. These individuals may desperately desire to achieve a minimum amount of sleep while unconsciously adopting maladaptive sleep patterns, such as going to bed early and waking up late, to obtain more rest. We previously reported a significant correlation between the ISI and DBST index across various studies [20,25,26]. However, the relationship between the DBST index and sleep-related cognition scales has been a subject of debate. The DBST index was not significantly associated with the DBAS-1627 among samples of clinical insomnia, Cancer-Related Dysfunctional Beliefs about Sleep among cancer patients [26], or Dysfunctional Beliefs about Sleep-2 items among the general population [25]. However, the DBST index was associated with the GSES among the older adult population [28], with Cancer-related Dysfunctional Beliefs and Attitudes about Sleep-14 among cancer patients [29], and with some subscales of Positive and Negative Sleep Appraisal Measure [30]. The discrepancy remains challenging to interpret. One possible explanation is that the relationship between the DBST index and sleep-related cognition scales may be weaker when circadian rhythm is influenced. Among shift workers, the DBST index was not associated with the GSES even with the ISI [31,32]. The DBST index demonstrated limited validity among shift workers. The DBST index reflects the discrepancy between desired TIB and desired TST. Consequently, a higher DBST index may result from a prolonged desired TIB combined with a shorter desired TST. However, shift workers may exhibit a stronger desire for additional sleep compared to individuals with insomnia, who often express a more urgent and desperate need for even minimal sleep. Further study is needed to confirm this.
Nonetheless, individuals with insomnia frequently attempt to go to bed early or spend excessive TIB to compensate for their difficulty falling asleep. This behavior reflects a dysfunctional preoccupation with sleep and an intensified effort to induce sleep, which paradoxically may prolong SOL. According to the two-process model of sleep regulation, sufficient wakefulness is necessary to build the homeostatic drive for sleep, and an excessive focus on attempting to sleep can disrupt this process. Based on this hypothesis, the DBST index shares a conceptual similarity with the GSES.
This study has some limitations. First, this survey was conducted online, which may introduce bias, despite the existence of several studies developing assessment scales using online methods [33]. In addition, anonymous surveys may lead to bias, as participants may respond to the questions indifferently. Approximately 7,000-8,000 emails were sent to recruit participants for each study, resulting in 300-600 enrolled samples. Accounting for potential biases introduced by non-respondents is important, as they may have been unable to complete the survey due to differences in health, motivation, or access to the Internet. Second, we examined the psychometric properties of the Korean version of the GSES within the general population, rather than among clinical samples of people with insomnia, whose responses might be gathered in a clinical setting. To mitigate this limitation, we collected responses from those among the general population complaining of insomnia (Study II). The limited generalizability of the data to patients with clinically diagnosed insomnia should be acknowledged. Third, subjective measures were employed to evaluate participants’ sleep-wake patterns instead of objective methods, such as polysomnography or actigraphy. This approach may introduce potential biases in the assessment.
In conclusion, the Korean version of the GSES is a reliable and valid instrument for assessing individuals’ anticipatory fears and anxieties regarding their ability to sleep. Its concise format may enhance its utility in evaluating patients’ sleep effort behaviors to achieve improved sleep outcomes.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Print
