Participants and data collection
A convenience sample of 420 participants living in a general community setting was obtained through recruitment on various social media platforms, such as [Discord, Facebook, Instagram, LinkedIn, Pinterest, and Twitter/X], as well as instant messaging services, including [LINE, Telegram, Viber, and WhatsApp]. The inclusion criteria were age 18 years or older and the ability to read and comprehend Arabic. Participants completed an online questionnaire via Google Forms, which included basic demographic questions (age, sex, height, weight, and marital status), the Arabic JSS, the Arabic version of the Pittsburgh Sleep Quality Index (PSQI) [
38], and the Arabic version of the Athens Insomnia Scale (AIS) [
39]. A subsample of the participants (
n = 147) completed the questionnaire again at two- and four-week intervals for test-retest reliability. To ensure the test-retest reliability of our measures, participants were requested to provide their email addresses during the initial survey administration. The inclusion of email addresses allowed us to recontact participants for the retest portion of the study, facilitating the assessment of the stability and consistency of the measures over time. To minimize the unnecessary burden on the entire sample, we made a deliberate decision to include only one third of the original sample (35% of 420 participants) for the test-retest reliability assessment. To select participants for the retest validity assessment, we employed a simple random sampling technique. Specifically, we utilized a random starting point and selected every third participant from the initial sample. This sampling approach was chosen to ensure the representation of a diverse and unbiased subset of participants in the retest phase. The email addresses provided by participants served as a crucial tool in matching the initial test responses with the corresponding retest responses, enabling us to establish the test-retest reliability of the measures. Furthermore, this matching process played a pivotal role in preventing the inclusion of duplicate data, ensuring the integrity, consistency, and accuracy of the obtained results. The PSQI [
40] and AIS [
41,
42] were included to assess convergent validity. The PSQI is a questionnaire that measures sleep quality and disturbances over a one-month period [
40]. It consists of 19 self-report questions and five questions rated by the participant’s bed partner or roommate [
40]. These items constitute seven components that are routinely examined in clinical sleep assessments, including subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleeping medications, and daytime dysfunction [
40]. Total scores range from 0 to 21, with higher scores indicating worse sleep quality [
40]. The Arabic version of the PSQI has shown acceptable internal consistency (⍺ = 0.65). The AIS is an 8-item questionnaire that assesses sleep within the past month according to the International Classification of Disease-10 (ICD-10) criteria for insomnia [
42]. Total scores ranged from 0 (absence of any sleep-related problem) to 24 (a severe degree of insomnia) [
43]. The scale has also demonstrated reliability as a screening tool for insomnia [
41], and the Arabic version has shown good psychometric properties (⍺ = 0.83) [
38].
Statistical analyses
Preliminary analyses included descriptive statistics (mean, standard deviation, skewness, and kurtosis) for all the measures. Skewness and kurtosis were used to determine the normality of the data, with values between − 2 and + 2 being set as cutoff points [
47]. Values within this range are considered acceptable levels of skewness and kurtosis for assuming a normal distribution. Sex, age, and marital status differences on the Arabic JSS were assessed using independent sample t tests. Age was divided into two groups, those younger than 35 years and those older than 35 years, based on the median age.
The internal consistency of the Arabic JSS was evaluated using Cronbach’s alpha [
48] and McDonald’s omega [
49] coefficients. Values of 0.70 or higher were considered satisfactory [
50]. To evaluate the test-retest reliability of the JSS, we calculated intraclass correlation coefficients (ICCs) between JSS scores at two time points (i.e., test vs. retest at 2 weeks AND test vs. retest at 4 weeks). To assess construct validity, confirmatory factor analysis (CFA) using the maximum likelihood extraction technique [
51] was conducted to test the unidimensional factor structure of the Arabic JSS. Model fit was assessed using the comparative fit index (CFI), Tucker‒Lewis index (TLI), and root mean square error of approximation (RMSEA). CFI and TLI values above 0.90 and an RMSEA less than 0.08 indicated acceptable model fit [
52]. To evaluate the generalizability and comparability of our findings, we conducted a series of measurement invariance tests using multigroup CFA across sex, age, and marital status. Configural, metric, scalar, and residual (strict) invariance were examined by systematically imposing equality constraints and evaluating changes in fit indices used in global CFA (i.e., CFI, TLI, RMSEA, and SRMS) [
53]. A CFI difference (ΔCFI) of less than 0.01 and an RMSEA difference (ΔRMSEA) of less than 0.015 indicated no significant decrease in fit between models [
53]. A significant chi-square difference test indicated that invariance could not be assumed [
53]. For configural invariance, we specified the same CFA model across groups with no equality constraints imposed [
53]. This model had an adequate fit with conventional criteria (CFI > 0.90, RMSEA < 0.08) [
53]. Metric invariance was supported if ΔCFI and ΔRMSEA were less than the cutoff, and the chi-square difference test was nonsignificant after constraining factor loadings to be equal [
53]. Scalar invariance was supported after additionally constraining intercepts equal, with ΔCFI < 0.01, ΔRMSEA < 0.015, and a nonsignificant chi-square difference test [
53]. Residual invariance was supported after constraining residual variances equal, with minimal changes in fit [
53].
To examine convergent validity, Pearson’s correlation coefficients were calculated between total scores on the Arabic JSS and total scores on the PSQI and AIS. Strong positive correlations were expected based on these instruments’ assessments of similar sleep constructs.
A network analysis was conducted on the data from the 4-item JSS questionnaire. The JSS includes questions on trouble falling asleep (JSS-1), waking up during the night (JSS-2), having trouble staying asleep (JSS-3), and not getting enough rest from sleep (JSS-4) [
54]. The network consisted of 4 nodes, one for each JSS item. Edges were defined between nodes based on Pearson correlation coefficients between all pairs of JSS items, with edges retained between nodes with correlations greater than 0.3 [
54].
Several common network analysis metrics were calculated to characterize the centrality and interconnectedness of nodes. The nodal centrality measures included betweenness centrality, closeness centrality, node strength, expected influence, Barrat’s measure, Onnela’s measure, weighted symmetrical uncertainty (WSa), and Zhang’s centrality [
54]. A high betweenness centrality indicates that the node lies on many of the shortest paths between other nodes, while a high closeness centrality means that the node can reach others quickly [
54]. The node strength sums the edge weights connected to the node. The expected influence measures the total strength of a node’s neighbors. Barrat’s measure of the node’s weighted degree [
54]. The Onnela measure incorporates the intensity and number of links, while the Zhang measure considers indirect as well as direct links [
54]. Clustering coefficients were also calculated per node to quantify the interconnectedness between a node and its neighbors [
54].
Constructing a network model enables the visualization and quantification of complex associations within the scale’s structure. This provides additional insight compared to traditional techniques such as factor analysis, which focus solely on relationships between items and latent variables. Specifically, network analysis can identify highly interconnected core items, detect clustering, and, through analysis of connections between items, highlight any redundancies or weak associations. Ultimately, network analysis complements conventional psychometric assessments by generating a more nuanced understanding of the underlying relationships and connections between scale items themselves.
Based on guidelines for network analysis in psychometric studies, we applied an edge weight cutoff of 0.25 [
55,
56]. This means that only connections with a partial correlation greater than 0.25 were depicted as edges in the network graph [
56]. Applying this cutoff filters out weaker connections and provides a more parsimonious visualization focused on the most relevant associations between items [
56]. The specific cutoff value was selected based on prior recommendations for network modeling of psychometric scales to balance detail with interpretability [
56].
All analyses were performed using the lavaan package (version 0.6–17) in R (version 4.3.2 (eye holes)) and were released on 2023-10-31. Network metrics were computed using R statistical software and the qgraph package (version 1.9.8). Visualization of the network structure was performed with the same software. The qgraph implements graphical modeling techniques to estimate network connections between items and generates graphical displays of these connections [
55]. After estimating a network model, we used qgraph to visualize nodes (scale items) and edges (partial correlation coefficients between items) [
55]. The Fruchterman-Reingold algorithm was applied to determine node placement, with strongly connected nodes placed closer together [
55]. Network graphs were rendered with nodes color-coded by subscale membership and weighted edges representing the strength of association between items [
55]. This approach allowed clear visualization of the overall network structure, clusters of related items, and central nodes [
55].