Insomnia is a sleep disorder that is highly prevalent in the general population but it often remains underdiagnosed and untreated (Morin et al., 2023). It is characterized by difficulty initiating and/or maintaining sleep despite opportunities for sleep and is subjectively experienced as adversely affecting daytime functioning. Insomnia can be either acute, lasting for 3 days to a week, or it can be chronic and persist for at least 3 months (Vargas et al., 2020). Insomnia has consistently been associated with adverse physical health outcomes and is a risk factor for hypertension and cardiovascular disease owing to the elevation of cortisol levels arising from lack of sleep (Dean et al., 2023). Insomnia commonly co-occurs with mental health disorders such as anxiety and mood disorders. This has been ascribed to a bi-directional relationship between insomnia and these conditions. For instance, both generalized anxiety disorder and major depressive disorder are associated with persistent worrying, rumination about potential negative outcomes and frequent negative thoughts about oneself, the world and the future. These cognitive processes trigger autonomic arousal and emotional distress, which can impact sleep initiation and maintenance. Furthermore, it can lead to attentional bias toward internal and external cues that signal that sleep is being disrupted leading to an escalation in anxiety, which culminates in sleep deficits (Harvey, 2002; Tang et al., 2023).

Insomnia, in turn, can produce distress and adversely impact interpersonal and occupational functioning, which can aggravate psychological symptoms. Owing to the co-occurrence of insomnia and mental health disorders, interventions have typically prioritized treating the primary mental health condition, based on the assumption that the insomnia will remit once the underlying disorder is addressed. However, insomnia has been found to precede comorbid mental health disorders, aggravate the symptoms associated with them and persist despite effective treatment of the co-occurring condition (Morin et al., 2023). This has prompted the recognition of insomnia as an independent disorder, necessitating direct treatment and it has driven the development of instruments to accurately assess, diagnose and monitor sleep disturbances.

The Pittsburgh Sleep Quality Index (PSQI) and the Epworth Sleepiness Scale (ESS), for example, are frequently used instruments to assess sleep problems (Buysse et al., 2008; Salihu et al., 2022). However, these tools offer only general assessments of factors such as daytime sleepiness, sleep quality, and disturbances, and are not specifically designed to measure insomnia. The Insomnia Severity Index (ISI) is a prominent instrument for measuring insomnia severity and perceptions of the condition (Bastien et al., 2001). The ISI captures the diagnostic criteria for insomnia outlined in the Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association, 2022). The ISI has been translated and validated into many languages including Korean (Cho et al., 2014), Portuguese (Clemente et al., 2017), Moroccan (Oneib et al., 2022), Spanish (Fernandez-Mendoza et al., 2012), Hausa (Salihu et al., 2022), and Persian (Sadeghniiat-Haghighi et al., 2014). These studies have confirmed that the ISI is a reliable instrument. The factor structure of the ISI has been examined in several studies, predominantly using Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA) and have provided mixed and inconsistent results.

The original validation study examining the psychometric properties of the ISI was conducted with a clinical population and employed EFA, which yielded a three-factor solution (Morin, 1993). This was supported by a study that used CFA in a non-clinical Spanish-speaking population (Fernandez-Mendoza et al., 2012). Subsequent research identified three distinct types of one-factor models and five different types of both two-factor and three-factor models. For instance, Gerber and colleagues validated the German version of the ISI among three independent samples and reported the scale was unidimensional (Gerber et al., 2016). Similarly, Dragioti and colleagues, analyzed the Swedish version of the ISI and found that a one factor solution had the best fit. Support for the two-factor solution has emerged from a range of studies including a validation study of the ISI in a sample of Korean university students (Lee and Kim, 2023) and research among Arabic patients diagnosed with chronic conditions (Al Maqbali et al., 2022). In contrast, the three-factor solution has been supported in research undertaken with Italian clinical patients diagnosed with insomnia (Castronovo et al., 2016) and Chinese undergraduate students (Lin et al., 2018). A systematic review and meta-analytic study on the structural validity of the ISI identified thirteen distinct models across the literature (Dilshad Manzar et al., 2021). The wide range of multifactorial models suggests that insomnia is composed of heterogeneous latent constructs that emerge from a common set of symptoms. This variability implies that individuals may experience and interpret insomnia symptoms in different ways, leading to diverse underlying factor structures. It also highlights the complexity of insomnia as a clinical phenomenon, where shared symptoms such as difficulty falling asleep, staying asleep, or experiencing non-restorative sleep, may cluster differently depending on the population studied, the context, or methodological approaches used.

The current study focused on South African first responders and examined the psychometric properties of the ISI from three different psychometric perspectives: Classical Test Theory (CTT), Rasch analysis and Mokken Scale Analysis (MSA). First responders are frontline workers responsible for emergency medical care. This group of professionals include emergency medical service personnel (e.g., paramedics and ambulance personnel), firefighters and law enforcement officials. Their primary responsibility is to provide rapid and often life-saving assistance in situations that range from medical crises and accidents to natural disasters, violent incidents, and other critical emergencies. Due to the nature of their work, first responders are frequently exposed to high-stress environments, traumatic events, and unpredictable scenarios (Arjmand et al., 2024).

Existing research has confirmed that insomnia is a highly prevalent condition among first responders. In their meta-analytical study on sleep disorders among first responders, Huang and colleagues reported a prevalence rate of 28% for insomnia (Huang et al., 2022). These sleep disturbances have been attributed to the demands of their occupation including irregular work shifts, frequent exposure to potentially traumatic events and the physical and emotional toll of responding to crisis situations. Recent evidence also supports fear of sleep as a central mechanism underlying insomnia among first responders and develops in response to certain post-traumatic stress disorder (PTSD) symptoms, including intrusive re-experiencing of trauma through nightmares (Reffi et al., 2023; Lebeaut et al., 2022). These fears drive maladaptive coping behaviors including avoidance of sleep or reluctance to sleep at night (Reffi et al., 2023; Werner et al., 2021). Chronic insomnia is associated with fatigue, mood disturbances, difficulties with alertness and concentration, cognitive impairment and reduced quality of life (Bard et al., 2023). Given the prevalence of insomnia among first responders and its serious implications for both physical and mental health, it remains essential to ensure that the tools used to assess this condition are psychometrically sound and contextually appropriate. Furthermore, existing research has highlighted a wide range of multifactorial models for the ISI, underscoring variability in how the instrument performs across different populations and settings (Dilshad Manzar et al., 2021). These inconsistencies point to the importance of examining the underlying structure and item functioning of the ISI in specific contexts rather than assuming universal applicability.

The ISI has been used in the South African context but the reporting of the psychometric properties of the scale has been limited. Bentley et al. (2025), for example, used the ISI to examine sleep deterioration in a retrospective study of the general population during the COVID-19 pandemic. The ISI has also been used to assess a sleep intervention among adolescents (Rossouw et al., 2024), sleep difficulties among youth (De Doncker and Mclean, 2022) and the impact of sleep, physical activity and sedentary behavior on symptoms of depression and anxiety before and during the COVID-19 pandemic (Lewis et al., 2021). The ISI has also been used to investigate fear of not being safe during sleep and its association with sleep quality among adults (Correia et al., 2024). However, the psychometric properties of the ISI were not reported. Phaswana-Mafuya et al. (2019) used two items of the ISI to measure insomnia symptoms among pregnant women and reported satisfactory reliability (α = 0. 82).

By employing three distinct psychometric perspectives, the current study aims to provide a more comprehensive and nuanced understanding of the instrument's performance. While CTT assesses the reliability and validity of the scale at a global level and provides information on internal consistency and factor structure, Rasch analysis provides insights into the functioning of individual items (Meijer et al., 1990). MSA is a nonparametric item response theory (IRT) method that further evaluates an instrument's scalability and monotonicity, which are important for determining whether items consistently reflect increasing levels of insomnia severity (Franco et al., 2022). Taken together, these analyses have the potential to provide robust evidence on the reliability, dimensionality, and construct validity of the ISI among South African first responders.

The inter-item correlations of the ISI as well as the CTT, Rasch and Mokken indices are reported in Table 1.

Table 1 shows that the inter-item correlations ranged between 0.33 and 0.72 and all were statistically significant (p < 0.001). The average inter-item correlation was 0.50. The item-total correlations ranged between 0.59 and 0.73 and were all statistically significant (p < 0.001). KMO was greater than 0.50 (0.86) and Bartlett's test was significant, indicating that the data was suitable for factor analysis. EFA (principal components with varimax rotation) extracted one factor which accounted for 57.68% of the variance. The factor loadings ranged between 0.71 and 0.82 and were all statistically significant (p < 0.001).

With respect to the Rasch indices, infit and outfit MnSq values were at an acceptable level and ranged between 0.78 and 1.33 for infit MnSq and between 0.83 and 1.25 for outfit MnSq. The items of the ISI demonstrated measurement invariance across gender (DIF = −0.23 to 0.29) and across type of first responder (DIF = −0.31 to 0.46).

With regards to the Mokken indices, Table 1 also shows that AISP found that all of the items of the ISI loaded on one scale as denoted by the value 1 for all items. Furthermore, the Hi coefficients for the individual items of the ISI were all above 0.30 and ranged between 0.49 and 0.59. There were no violations of monotonicity as #vi and Crit values were 0 for all items of the ISI.

The three models of the factor structure of the ISI that were examined with CFA are shown in Figure 1, together with standardized factor loadings. Figures 1A–C depict the one-factor, correlated two-factor, and bifactor models, respectively.

The fit indices associated with the three models in Figure 1 are reported in Table 2.

The fit indices in Table 2 indicated that all three models fit the data to an acceptable degree (relative χ² < 3, CFI > 0.95, SRMR < 0.08, RMSEA ≤ 0.06, PCLOSE non-significant). However, the model comparison index, AIC, indicated that both the bifactor model and the one-factor model (AIC difference was 2) were better fits than the correlated two-factor model (AIC difference between the bifactor model and the correlated two-factor model was 5.7). In addition, the two factors were substantially correlated (r = 0.77, p < 0.001). Also, while the bifactor model demonstrated an acceptable fit, Figure 1C shows that the regression coefficients of the severity subscale were very problematic since two of the items were very low (“Difficulty falling asleep” and “Difficulty staying asleep”) and two were negative (“Early morning awakenings” and “Dissatisfaction with sleep”). This is in stark contrast to the factor loadings associated with these four items on the same subscales in the correlated two-factor model where the factor loadings were greater than 0.60. Both factor loadings that were very low were statistically significant (i.e., 0.23 and 0.13). This is to be expected given a sample size of n = 429, but it is way below the accepted threshold of 0.50 for meaningful factor loadings.

Since CFA found that the bifactor structure fitted the data to an acceptable degree, ancillary bi-factor indices were calculated to determine whether the specific factors (two subscales) explained a sufficient amount of variance over and above that explained by the general factor (total scale). In this regard the bifactor indices indicated that the general factor, insomnia, explained 73% of the variance in the items (ECV) while the specific factors explained 27% of the variance (ECV_severity = 8.3%, ECV_impact = 18.7%). OmegaH for the general factor was 0.82, reflecting a very reliable general factor, while for the specific factors it was 0.01 and 0.39, respectively. Further, a PUC value of 0.57 together with an ECV of 0.73 and omegaH of 0.82 provides strong evidence that the ISI should be regarded as essentially unidimensional. Parallel analysis also extracted only one component with an eigenvalue (4.04) that was greater than the eigenvalue of the 95th percentile (1.25) of simulated eigenvalues. The eigenvalue of the second component (0.93) was less than the 95th percentile (1.15) of simulated eigenvalues.

The CTT, Rasch and Mokken indices at scale level for the unidimensional ISI are reported in Table 3.

Table 3 indicates that the CTT, Rasch and Mokken indices for the unidimensional ISI were all at an acceptable level. In the first instance, the eigenvalue of the first contrast was less than 2. Secondly, all the estimates of internal consistency were greater than 0.70 (α and ω = 0.87, CR = 0.90, MSrho = 0.89). Thirdly, AVE was greater than 0.50 and MSV and ASV were less than AVE. The scalability coefficient in Mokken analysis was greater than 0.50 (H = 0.55). The item separation reliability and index in the Rasch analysis were greater than 0.80 and 3, respectively, while the person separation reliability and index were greater than 0.80 and 2, respectively.

The descriptive statistics for and the intercorrelations between insomnia and the other variables that were included for the purpose of establishing concurrent validity, are reported in Table 4.

The indices of skewness in Table 4 ranged between 0.12 and 0.38, and those of kurtosis between −0.81 and −0.25. The values of skewness and kurtosis were thus within the acceptable range of −2 to +2. Table 4 also indicates that insomnia was significantly positively related to depression (r = 0.52, p < 0.001, large effect size), anxiety (r = 0.49, p < 0.001, medium effect size), and fatigue (r = 0.61, p < 0.001, large effect size). The positive relationships would indicate that higher levels of insomnia were associated with higher levels of depression, anxiety, and fatigue. The PHQ-9 contains one item related to sleep difficulty and to ensure that the correlation between depression and insomnia was not excessively influenced by this one item, we also examined the correlation between the two variables without this one item. The correlation remained a large effect (r = 0.50, p < 0.001 and the reliability of the PHQ-9 was not significantly affected (α = 0.88).

This study aimed to evaluate the psychometric properties of the ISI among South African first responders, a population uniquely vulnerable to sleep disturbances due to the demanding and stressful nature of their work. Drawing on CTT, Rasch analysis and MSA, the study aimed to provide a comprehensive assessment of the instrument's reliability, validity and dimensional structure within this context.

Overall, the psychometric properties of the ISI, inclusive of reliability and validity, can be regarded as highly satisfactory. In the first instance, evidence from the three different psychometric paradigms overwhelmingly favored a unidimensional interpretation of the ISI, a finding that supports the one-factor solution reported in other studies (e.g., Gerber et al., 2016; Kaufmann et al., 2019). EFA extracted one factor and CFA fit indices found that a one-factor model and a bifactor model fit the data to an acceptable degree. However, ancillary bifactor indices found that the majority of the item variance was explained by a very reliable general factor. The unidimensionality of the ISI was confirmed with parallel analysis in that only one eigenvalue in the current dataset was greater than the 95th percentile of simulated eigenvalues. Further, a PCA of the residuals in Rasch analysis confirmed a unidimensional solution and AISP in Mokken analysis showed that all of the items of the ISI loaded on a single scale. The scalability coefficient in Mokken analysis, H was greater than.50 indicating a very strong unidimensional scale.

All the estimates of the reliability of the scores of the ISI may be considered satisfactory. The CTT reliability indices, alpha, omega and CR as well as the MSA index, Ms_rho exceeded 0.80 and was therefore well above the recommended threshold of 0.70 (Devon et al., 2007).

All three psychometric paradigms provided evidence for the construct validity of the ISI. In CTT the inter-item correlations, item-total correlations and factor loadings were within the ranges recommended in the literature (Hajjar, 2018; Devon et al., 2007; Paulsen and Brckalorenz, 2017; Hair et al., 2010) thus demonstrating that all the individual items of the ISI contribute to the measurement of the latent variable, insomnia. In Rasch analysis infit and outfit MnSq values were within an acceptable range (Linacre, 2023), thus confirming that all the items of the ISI fit the Rasch model. The person separation index and reliability as well as the item separation index and reliability confirmed that the ISI can distinguish between low and high scorers on the latent variable, and that an item hierarchy exists. DIF in Rasch analysis confirmed the measurement invariance across gender and type of first responder, thus indicating that the items of the ISI measure the same construct in men and women as well as in police officers and paramedics. Lastly, in MSA the scalability coefficients for the individual items of the ISI, H_i, exceeded the threshold identified in the literature (Mokken, 2011), and these scalability coefficients, similar to item-total correlations, confirmed that all the items of the ISI contributed to the measurement of insomnia and that the latent construct of insomnia was well represented by its indicators.

CTT analysis also provided evidence for other forms of construct validity. AVE was greater than 0.50 and less than CR and according to Posch et al. (2019) this serves as evidence for convergent validity. AVE was greater than MSV and ASV and since this would indicate that the latent construct had more in common with the items that contributed to its measurement than with other related variables it provides evidence for discriminant validity. Lastly the strong correlations between insomnia as measured by the ISI and depression, anxiety, and fatigue provides evidence for the concurrent validity of the ISI.

The findings of this study offer important theoretical and practical implications for the assessment of insomnia, particularly within high-risk occupational groups such as first responders. From a theoretical perspective, the results reinforce and extend the existing understanding of the latent structure of insomnia as measured by the ISI. Evidence from all three psychometric paradigms favored a strong unidimensional solution, thus supporting the conceptualization of insomnia as a unified construct rather than a multidimensional syndrome in this population. This suggests that despite symptom diversity, the experience of insomnia can be meaningfully captured along a single underlying dimension. These findings affirm that the ISI is a stable and theoretically coherent tool for assessing insomnia severity across diverse groups within the first responder population. In terms of practical implications, the findings support the use of the ISI as a screening and diagnostic tool for insomnia among South African first responders. Its unidimensional structure simplifies scoring and interpretation, making it suitable for routine clinical use, as well as for incorporation into mental health surveillance and employee wellness programs. The ISI can assist in the early identification of individuals at risk, facilitate timely referrals, and guide the implementation of evidence-based interventions to mitigate the impact of sleep disturbances on overall wellbeing and job performance. The absence of measurement bias across gender and professional roles (e.g., paramedics vs. police officers) also enhances the practical utility of the ISI, as it ensures equitable assessment across subgroups within the first responder workforce.

While the findings of this study provide strong support for the unidimensionality, reliability and validity of the ISI when used with South African first responders, several limitations should be acknowledged. The study employed a cross-sectional design, which limits the ability to draw conclusions about the temporal stability of the ISI scores. Longitudinal research is needed to assess test–retest reliability and to evaluate how insomnia symptoms and ISI scores may change over time, particularly in response to interventions or varying occupational stressors. All data, including responses to the ISI and other psychological scales, were based on self-report. While self-report is a widely accepted method in research, it is susceptible to response biases such as social desirability, recall bias, or underreporting of symptoms due to stigma, especially within occupational groups like first responders. Although the study included a diverse sample of first responders (e.g., paramedics and police officers), the findings may not be generalizable to all emergency service personnel across South Africa or to first responders in other low- and middle-income countries. Furthermore, the use of the same subject samples for EFA and CFA represents another limitation. This approach may lead to overfitting and inflated estimates of model fit. Future research is needed to address these limitations by utilizing separate samples for EFA and CFA to enhance the validity and generalizability of the findings. Finally, the survey was administered online, which limited the ability to fully validate the population. The online format, while providing broad accessibility, restricted the ability to track those recruited from specific locations. Future studies could address this limitation by using more controlled recruitment methods to enhance population validation.

This study provides comprehensive evidence for the reliability and validity of the ISI among South African first responders, using three complementary psychometric approaches, namely CTT, Rasch and Moken analyses. The ISI had a unidimensional structure across all methods, which suggests that the instrument effectively captures the core features of insomnia as a single latent construct in this high-risk population. Furthermore, the instrument demonstrated strong internal consistency, item functioning, and measurement invariance across gender and professional subgroups, indicating that it is both psychometrically robust and fair. Given the high prevalence of insomnia and its impact on the physical and psychological functioning and occupational performance of first responders, the availability of a validated, contextually appropriate instrument remains essential. The ISI emerges from this study as a valuable resource for clinicians, researchers, and occupational health professionals working with South African first responders.