We obtained cross-sectional data from a sample of 1,292 teachers. The data were collected via the administration of paper-and-pencil questionnaires directly to the participants. This approach facilitated direct, face-to-face interactions, which enhanced the response rates and the completeness of the surveys. The questionnaire survey was conducted between October 16 and November 25, 2023, approximately 1 month after the start of the school year. The research project gathered participants with an average age of 45.22 ± 5.99 years from college and secondary schools in different regions of Tunisia. The sample was split into two groups: an exploratory sample consisting of 544 participants, with 52.60% being female, and a confirmatory sample consisting of 748 participants, with 50.50% being female.

An Arabic version of the Teacher Stress Inventory (TSI) was used to assess teachers’ occupational stress. The questionnaire consists of 49 items conceptualized into 10 subscales, grouped into two main categories: Stress Sources and Stress Manifestations. Stress Sources include time management (which evaluates the teacher’s ability to manage time for both professional and personal needs), work-related stressors (including workload associated with teaching or administration, class size, and professional responsibilities), professional distress (stress factors such as income, career development, promotion opportunities, and recognition), discipline and motivation (which determines whether a teacher feels capable of managing the classroom while motivating students), and professional investment (which assesses the teacher’s commitment and beliefs regarding their work). Stress Manifestations include emotional manifestations (teachers’ emotional responses to stress, such as anxiety, frustration, anger, sadness, etc.), fatigue manifestations (teachers’ physical responses to stress, such as sleep disturbances, headaches, fatigue, etc.), cardiovascular manifestations (teachers’ cardiovascular responses to stress, such as elevated blood pressure and increased heart rate), gastrointestinal manifestations (teachers’ gastrointestinal responses to stress, such as stomach pain, loss of appetite, cramps, etc.), and behavioral manifestations (strategies teachers adopt to cope with stress, such as self-medication, use of prescribed medications, alcohol consumption, etc.). Response modalities are measured on a 5-point Likert scale, with five statements ranging from 1 ‘Strongly Disagree’ to 5 ‘Strongly Agree.’ To simplify the response options, a brief description was provided: ‘Strongly Agree’ corresponded to ‘Extremely Perceptible,’ ‘Agree’ corresponded to ‘Very Perceptible,’ ‘Neutral’ corresponded to ‘Moderately Perceptible,’ ‘Disagree’ corresponded to ‘Barely Perceptible,’ and ‘Strongly Disagree’ corresponded to ‘Not Perceptible.’

The normality of the data was demonstrated by skewness values ranging from −0.249 to 0.723 and kurtosis values ranging from −1.116 to −0.279 (see Table 1). Skewness values in the range of +2 to −2 are considered acceptable for a normal distribution (Elliott and Woodward, 2007; Tabachnick and Fidell, 2013), indicating that the data distribution is normal.

To examine the factor structure of the scale, we conducted an orthogonal varimax exploratory factor analysis (EFA) using Kaiser’s varimax criterion with principal component analysis (PCA) on our 49-item scale (Everitt and Dunn, 1991; Kaiser, 1958; Rummel, 1988; Bollen, 2014). Items were retained if their factor loadings were 0.40 or greater (Pett et al., 2003). Sampling adequacy was assessed using the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s sphericity test (Bartlett, 1950). A KMO value greater than 0.6 and a significant p-value for Bartlett’s test is recommended (Worthington and Whittaker, 2006).

A first-order confirmatory factor analysis (CFA) with maximum likelihood estimation was conducted to verify the factor structure across ten dimensions of the TSI-49. Several indices were analyzed to assess the fit of the model to the collected data. We used the goodness of fit index (GFI) (Hooper et al., 2008; Bollen, 2014; Marsh et al., 1988; Schermelleh-Engel et al., 2003) and its adjusted index (AGFI), both of which should be 0.90 and higher for a good fit (Muenjohn and Armstrong, 2008). In addition, the root mean square error of approximation (RMSEA) should be ≤0.05 for a good fit (MacCallum et al., 1996). The standardized root mean square residual (SRMR) should be ≤0.10 for an acceptable fit (Hu and Bentler, 1999). We also checked the comparative fit index (CFI) (Watchravesringkan et al., 2008), the non-normed fit index (NNFI), the normed fit index (NFI), the parsimony comparative fit index (PCFI), and the parsimony NFI (PNFI). Values for CFI, NFI, TLI, PCFI, and PNFI should be ≥0.95, and NFI should be ≥0.90 (Schermelleh-Engel et al., 2003). A significant χ² value compared to the degrees of freedom indicates that the proposed model does not fit the observed data well, while a non-significant χ² value indicates that the model fits adequately (Satorra and Bentler, 1994). The ratio of χ² to degrees of freedom (χ²/df) was used to compare model fit indices. A ratio of χ²/df < 3 indicates an acceptable model fit (De Carvalho and Chima, 2014). However, chi-square (χ²) is affected by sample size and model complexity, with more complex models being more sensitive to χ² (Kelloway, 1995).

In addition, a sensitivity analysis (ANOVA) was conducted to assess the effect of age, gender, and teaching experience on the scores of each dimension of the instrument.

An exploratory factor analysis (EFA) was conducted on a random sample of 544 participants, while a confirmatory factor analysis (CFA) was performed on a sample of 748 participants. EFA was conducted using the Statistical Package for Social Sciences (SPSS, version 28, IBM, Armonk, NY, USA) and JASP software (version 0.18.3.0), while CFA was analyzed using AMOS software (version 28, IBM, Armonk, NY, USA).

Table 2 presents the key characteristics of the sample population.

Table 1 displays the means, standard deviations, kurtosis, and skewness for the TSI-A items. The data were normally distributed regarding skewness and kurtosis.

The data demonstrated a normal distribution in terms of skewness and kurtosis. Furthermore, the results indicated that the TSI scale demonstrated robust temporal stability (r = 0.91 at p < 0.001; ICC = 0.90) (Table 3 and Figure 1).

The principal component analysis of the TSI version yielded a 10-factor solution, indicating that the 49 items of the scale can be aggregated to generate scores for these indices. Ten factors exhibited eigenvalues greater than one. The scree plot indicated a distinct point of inflection after the first factor (Figure 2). The initial eigenvalues for the 10 factors ranged from 1.15 (Time Management) to 10.68 (Behaviors Manifestations). The first 10 factors accounted for more than 62.37% of the total variance. In light of the eigenvalue and the proportion of variance accounted for, the 10-factor solution was deemed optimal for the subsequent analysis. The loadings of the 49 items on the 10 factors were substantial (Table 4).

In conclusion, based on the screen plot, it is appropriate to consider the first 10 principal components for further analysis, as they capture most of the variance in the data. Subsequent components contribute little additional information, as indicated by the leveling off the curve after the tenth component.

The EFA results indicated that the TSI-A showed a high degree of alignment with the expected theoretical model and demonstrated good overall internal reliability. The internal consistency of the Arabic version of the scale was evaluated using McDonald’s omega, Cronbach’s alpha, and Guttman’s λ6 values, which demonstrated strong reliability (ω TSI-A = 0.912; α TSI-A = 0.923; λ6 TSI-A = 0.954). McDonald’s omega is regarded as a more robust measure of internal reliability for the scale (Béland et al., 2017), as illustrated in Table 5. A number of statisticians (Béland et al., 2017; Cortina, 1993; Field, 2013; Kline, 2016; Taber, 2017; Taber, 2018; Vaske, 2008) have proposed that reliability coefficients above 0.60 are acceptable. Consequently, the majority of parameters demonstrated significantly high correlations with excellent values (see Table 5).

The results of the factor analysis (KMO = 0.910, Bartlett’s test of sphericity p < 0.001) supported the factor analysis and identified 10 independent main factors with values above the thresholds set by Schönbrodt and Perugini (2013), Cattell (1966) and Staw (1980). These factors accounted for 62.37% of the variance in the data. The initial principal component “Time Management” explained 21.80% of the total variance (eigenvalue = 10.68), while “Behaviors Manifestations” accounted for 62.37% (eigenvalue = 1.15). Following the recommendations set forth by Horn (1965), all dimensions with eigenvalues exceeding 1 were taken into consideration.

Concerning the EFA, the factor loadings of the 10-factor model items exhibited a range from 0.596 to 0.828, as illustrated in Table 4.

The CFA employed a range of model fit indices, by the criteria established by various researchers. Following the established critical standards, the optimal standardized Chi-Square value with the number of degrees of freedom (χ²/df) is within the range of 2 to 3 (2 ≤ χ²/df ≤ 3). This conclusion is supported by the findings of Wheaton et al. (1977) and Tabachnick and Fidell (2013). The results of the Chi-Square and normalized χ² (χ²/df) statistics indicated that the model fit was satisfactory.

A variety of fit indices were calculated, including the AGFI, GFI, CFI, NNFI (TLI), NFI, PNFI, PCFI, RMSEA, and SRMR. The critical values for NFI, CFI, and NNFI (TLI) should exceed 0.95, by the recommendations of Bryant and Yarnold (1995), Hu and Bentler (1999), Fan (1998), and Byrne (2014). Nevertheless, for the present scale, the AGFI and GFI values indicate 0.90, which some researchers, such as Schermelleh-Engel et al. (2003), consider to be acceptable. Furthermore, the PNFI and PCFI values exceeded the acceptable threshold of 0.50 (PNFI = 0.86; PCFI = 0.90). As indicated by Byrne (2001), the RMSEA and SRMR values should not exceed 0.05.

Overall, the 49-item model demonstrated an excellent fit to the theoretical model across all tested indices, thereby supporting the factor structure for the Tunisian teacher sample. The confirmatory factor analysis yielded a good fit for the 10-factor Arabic version of the TSI-A model (χ² = 2059.607, degrees of freedom = 1,112, p < 0.001; NFI = 0.391; CFI = TLI = IFI = 0.95; GFI = AGFI = 0.90; SRMR = 0.04; RMSEA = 0.03, 90% CI = 0.031–0.036). All of the hypothesized parameters of this model were found to be statistically significant at the p < 0.05 level.

Additionally, CFA was used to assess whether the number of factors and variables observed were consistent with the theory initially proposed. The latent variable “sources of stress” was associated with the following dimensions: time management, work-related stressors, professional distress, discipline and motivation, and professional investment. In parallel, the latent variable “Stress Manifestations” grouped emotional manifestations, fatigue manifestations, cardiovascular manifestations, and gastronomic and behavioral manifestations. To validate this structure, one- and two-factor models were tested. The two-factor model, named Model 2, obtained the best fit, confirming its relevance. This optimal fit to the theoretical model validates the two-factor structure of the IST model in the Arabic-speaking context (see Table 6 and Figure 3).

The ANOVA results, along with the sensitivity analysis, reveal no significant effect between the different dimensions of the scale and the various factors.

To assess convergent validity, we examined the magnitude and significance of the loadings of the first-order factors onto the second-order factor. The standardized factor loadings for the first-order factors ranged from 0.768 to 0.836 (p < 0.001), indicating moderate to strong relationships between individual items (I1- I49) and their corresponding latent constructs. These loadings suggest that the items effectively capture the essence of their respective constructs.

At the second-order level, each of the first-order factors demonstrated significant standardized loadings onto the overall construct, with loadings ranging from 0.768 for discipline and motivation to 0.836 for cardiovascular manifestations (all p < 0.001). Cardiovascular manifestations emerged as the strongest contributor to the second-order construct, while discipline and motivation, although weaker, still exerted a notable influence. This confirms the multidimensional nature of the model and highlights the interrelatedness of its components.

Similarly, the fit indices support the convergent validity of the second-order factor model. Both the CFI and TLI are equal to 0.95, indicating an excellent fit between the model and the observed data. The RMSEA is 0.03, with a 90% CI = 0.031–0.036, providing further evidence that the model fits the data well. Additionally, the SRMR value of 0.04 reflects a good model fit. These fit indices confirm that the first-order factors are appropriately modeled as indicators of the second-order factor.

Results for discriminant validity showed that AVE values ranged from 0.588 for time management to 0.698 for cardiovascular manifestations. The square root of the AVE values, reported along the diagonal, were as follows: 0.767 for time management, 0.782 for work-related stress factors, 0.774 for professional distress, 0.768 for discipline and motivation, 0.829 for professional investment, 0.800 for emotional manifestations, 0.802 for fatigue manifestations, 0.836 for cardiovascular manifestations, 0.832 for gastronomical manifestations, and 0.791 for behaviors manifestations. Comparing each AVE square root with the correlation coefficients between the constructs showed that the AVE square roots were higher, thus demonstrating good discriminant validity. These findings confirm that the factors evaluated in this model are consistent with their items (good convergent validity) and distinct from each other (good discriminant validity).

Despite the promising results, it is important to recognize the main limitation of the present study. The investigation was limited to Tunisian instructors, which might have limited the applicability of the results to other Arabic-speaking nations. Tunisia exhibits numerous cultural and linguistic similarities with other Arabic-speaking countries; however, it also displays distinct variations in dialects, educational systems, and cultural subtleties throughout the Arab world. These variations have the potential to affect how teachers perceive and react to stress-related elements in the TSI-A. Future research should prioritize the validation of the TSI-A with educators from diverse Arabic-speaking nations to verify its appropriateness and efficacy across different Arabic dialects and educational contexts within the broader Arabic-speaking world.