Depression is one of the most prevalent mental health conditions and a leading cause of disability globally. It is characterised by persistent feelings of sadness, hopelessness and a lack of interest or pleasure in daily activities. In more severe cases, depression can lead to thoughts of self-harm or suicide (American Psychiatric Association, 2022). Depressive disorders have a significant impact on an individual’s ability to function in various aspects of life, including personal relationships and their occupation. Moreover, recent studies suggest that the prevalence of depression is increasing (Moreno-Agostino et al., 2021; Shorey et al., 2022).

In a systematic review and meta-analysis of the literature on adolescent depression, Shorey and colleagues concluded that 34% of adolescents globally were at risk of developing clinical depression (Shorey et al., 2022). Similarly, Hu and colleagues, in their meta-analytic study on depression among older adults, reported a prevalence rate of 28.4%, while Gutiérrez and colleagues, in their systematic review, reported a prevalence rate of 21% among community samples (Gutiérrez-Rojas et al., 2020; Hu et al., 2022). Studies have also highlighted the prevalence of depression among vulnerable population groups including postpartum women (Liu et al., 2022), healthcare workers (Li et al., 2021), teachers (Ozamiz-Etxebarria et al., 2021) and college students (Wang et al., 2023). These findings underscore the substantial and pervasive burden of depressive disorders and highlight the critical need for effective screening practices. Early identification of depressive symptoms through comprehensive screening is essential for timely intervention. This can significantly improve outcomes by preventing the progression of the disorder and mitigating its impact on individuals’ overall well-being.

Although depression has traditionally been approached as a categorical condition – where individuals either meet the criteria for a diagnosis or do not – there is growing evidence to suggest that depression may be better understood as a dimensional phenomenon. This perspective posits that depressive symptoms exist on a continuum, varying in severity from mild to severe, rather than as discrete categories. Growing recognition of the dimensional nature of depression has led to the development of instruments designed to assess both the presence and severity of symptoms (Bianchi et al., 2022).

The Patient Health Questionnaire (PHQ) was specifically developed to facilitate a criteria-based diagnosis of depression, aligning with established clinical guidelines (Kroenke et al., 2001). The instrument comprises nine items that correspond to the diagnostic criteria for major depressive disorder (MDD) as outlined in the Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association, 2022). This includes both cognitive-affective symptoms, such as mood disturbances, feelings of guilt or worthlessness and difficulties in concentration, as well as somatic symptoms, including changes in sleep and appetite, fatigue and suicidal ideation (American Psychiatric Association, 2022).

Since its inception, the psychometric properties of the PHQ-9 have been examined across numerous studies, and the instrument has consistently demonstrated strong diagnostic accuracy and reliability (El-Den et al., 2018). The psychometric properties of the PHQ-9 have been assessed in women experiencing perinatal depression, patients with psychiatric disorders, people with multiple sclerosis and patients suffering from major depressive disorder (Beard et al., 2016; Patrick & Connick, 2019; Sun et al., 2020; Wang et al., 2021). These studies have confirmed that the PHQ-9 is an effective tool for identifying depressive disorders in diverse populations and settings, making it a valuable resource in both clinical practice and research. However, research into the factor structure and dimensionality of the instrument has raised questions about whether it adequately captures a single underlying construct or multiple distinct dimensions of depression (Bianchi et al., 2022; Doi et al., 2018).

Using confirmatory factor analysis (CFA), Doi and colleagues examined the factor structure of the PHQ-9 across different groups, including adults diagnosed with MDD, adults with both MDD and a co-occurring anxiety disorder and those without any psychiatric disorder (Doi et al., 2018). The authors reported that the bifactor model was supported in both the clinical and nonclinical samples. Beard and colleagues (2016) conducted a comprehensive validation of the PHQ-9 in a psychiatric sample, using exploratory and confirmatory factor analyses. Their findings suggested a two-factor solution, with one factor capturing cognitive and affective symptoms and the other reflecting somatic symptoms. Using Mokken analysis, Boothroyd and colleagues (2019) concluded that a one-factor solution was viable for the PHQ-9 in a sample of adults, while González-Blanch and colleagues (2018), using CFA, found that both the one-factor and two-factor models provided adequate fit in a primary care setting. A systematic review of studies using CFA to examine the factor structure of the PHQ-9 identified four models including a one-factor solution, a two-factor solution, a bifactor model and a three-factor model (Lamela et al., 2020). Given the ongoing debates about whether the PHQ-9 should be interpreted as measuring a single construct of depression or whether it reflects distinct but related dimensions (e.g., somatic and cognitive-affective symptoms), further research is warranted to clarify its factor structure. The current study aims to contribute to the literature by investigating the dimensionality of the PHQ-9 in a sample of South African first responders using CFA, ancillary bifactor indices, parallel analysis and Mokken analysis.

First responders are routinely exposed to a wide range of stressful and traumatic events during the course of their work. This increases their risk of adverse mental health outcomes. The most common trauma-related disorders encountered among first responders include post-traumatic stress disorder, anxiety and depression. A clearer understanding of the factor structure of the PHQ in this population group may aid in early identification of symptoms of psychological distress and promote targeted interventions.

Data were analysed using IBM^® Statistical Package for Social Sciences (SPSS) for Windows, version 29, IBM Amos for Windows, version 28 and R, version 4.3.1 (R Development Core Team, 2020) and the package Mokken, version 3.1.2 (Van der Ark, 2012).

The reliability of PHQ-9 can be considered satisfactory, as both Cronbach’s alpha and McDonald’s omega were 0.89. The interitem correlations, descriptive statistics, item-total correlations (ITC) and factor loadings are presented in Table 1.

Table 1 indicates that the interitem correlations (0.23 to 0.64) and the average interitem correlation (0.48) were within the recommended range, thus indicating that the items reflect the same content domain and there were no redundant items. The ITC correlations ranged between 0.57 and 0.70 and were all above 0.50. Similarly, the factor loadings ranged between 0.66 and 0.78 and were all above 0.55. The ITC and factor loadings confirm that all items contribute to the measurement of the latent variable. In general, all the indices in Table 1 support the internal consistency of the PHQ-9.

Kaiser–Meyer–Olkin was greater than 0.50 (0.89), and Bartlett’s test was significant (p < 0.001), thus confirming the suitability of the data for factor analysis. The results of the EFA are reported in Table 2.

Table 2 shows that the EFA resulted in two factors, similar to previous factor analysis, and they were labelled similarly. However, it is also noticeable that four items cross-loaded with loadings above 0.32 (Tabachnick et al., 2013). With the exception of the factor loading of the item ‘fatigue’ on the cognitive-affective factor, all the factor loadings and cross-loadings were statistically significant.

The three models of the PHQ-9 that were examined with CFA, namely the one-factor, the correlated two-factor and the bifactor models, are presented in Figure 1. The CFA fit indices are reported in Table 3.

Table 3 indicates that all three models fit the data to an acceptable degree (GFI, TLI, CFI ≥ 0.95, RMSEA ≤ 0.08), and AIC indicated that the one-factor model was marginally the best model. Figure 1 shows that the factor loadings for the one-factor model all exceeded 0.50 and ranged between 0.55 and 0.78. Similarly, the factor loadings for the correlated two-factor model ranged between 0.64 and 0.77. However, the two factors were strongly associated (0.79), suggesting that a two-factor structure is redundant. Although the fit indices for the bifactor model showed an acceptable fit, the loadings for the two subscales were problematic. For the somatic subscale, three of the loadings were below 0.50, and for the cognitive-affective subscale, all of the loadings were nonsignificant and two were negative.

The results of the ancillary bifactor analysis overwhelmingly indicated that the PHQ-9 is essentially unidimensional:

The unidimensionality of the PHQ-9 was also confirmed by parallel analysis. In this regard, a principal component analysis (PCA) of the current dataset identified only one eigenvalue (4.83) that was greater than the 95th percentile (1.29) of a range of simulated eigenvalues. The second eigenvalue in the current dataset (1.16) was lower than the 95th percentile (1.20) of the simulated eigenvalues.

The results of the Mokken analysis are reported in Table 4.

Table 4 indicates that all of the PHQ-9 items loaded on a single scale, as indicated by the AISP value of 1. The H-coefficient of the unidimensional PHQ-9 and MS_rho were 0.52 and 0.90, respectively, reflecting a very strong and reliable scale. The H_i values of the individual items were all above 0.30 and ranged between 0.47 and 0.55. There were no violations of monotonicity, as confirmed by the Crit value as well as #vio and #zsig.

There is ongoing debate regarding the dimensionality of the PHQ-9, with studies reporting various factor structures, ranging from unidimensional models to more complex multifactor solutions (Lamela et al., 2020). Some studies have identified a two-factor model, typically distinguishing between cognitive-affective and somatic symptoms, while others have proposed three- or four-factor models (Beard et al., 2016; Bianchi et al., 2022). The variability in findings highlights the need for further research of the underlying structure of the PHQ-9 from a variety of perspectives. Against this backdrop, the current study used exploratory and confirmatory factor analyses as well as ancillary bifactor indices, parallel analysis and Mokken analysis to examine the dimensionality of the PHQ-9 in a sample of South African first responders. There were several important findings.

Firstly, the PHQ-9 demonstrated satisfactory internal consistency, further underscoring its reliability as a measure of depressive symptoms. The high interitem correlations and ITC values confirmed that the scale items cohesively represent the same underlying construct without redundancy. This is particularly relevant in both theoretical and practical contexts, as it affirms that the PHQ-9 accurately captures a comprehensive range of depressive symptoms. The reliability of the instrument confirms that it can be used to screen for depression, track symptom changes over time and inform treatment decisions.

Secondly, the results of the EFA align with previous studies, identifying two distinct factors within the PHQ-9, which were labelled similarly to earlier research (Beard et al., 2016; Doi et al., 2018). However, the presence of cross-loadings suggests that certain items may tap into multiple dimensions of depression simultaneously, complicating the interpretation of the two-factor model. The identification of these cross-loadings is important, as it raises questions about the clear delineation of cognitive-affective and somatic symptoms within the PHQ-9, potentially indicating an overlap that needs to be considered in both research and clinical applications.

Thirdly, the CFA further explored the dimensionality of the PHQ-9 by comparing three models: a one-factor model, the correlated two-factor model and the bifactor model. The results indicated that while all the models provided an acceptable fit, the one-factor model was marginally superior. This finding supports the notion that depression, as measured by the PHQ-9, may be best understood as a unidimensional construct, where all items contribute to a single underlying factor. The correlated two-factor model, despite having acceptable fit indices, revealed a high correlation between the two factors, similar to the high correlation reported by Boothroyd and colleagues (2019). This strong association suggests that distinguishing between cognitive-affective and somatic symptoms may be redundant, as both factors appear to be closely intertwined. This finding challenges the utility of a two-factor model, reinforcing the idea that depression symptoms may not neatly divide into separate categories but rather exist along a continuum that is best captured by a single dimension. The bifactor model, although showed an acceptable overall fit, presented significant issues with the loadings on the subscales. The problematic loadings, particularly the nonsignificant and negative loadings on the cognitive-affective subscale, indicated that the bifactor model does not provide a coherent or meaningful representation of the data, further contesting the case for more complex factor structures.

Further support for the unidimensionality of the PHQ-9 came from the results of the ancillary bifactor indices and parallel analysis. The fit indices of the bifactor model confirmed a dominant general factor explaining 70% of the item variance, while the two specific factors did not add value beyond the variance extracted by the general factor. Similarly, the parallel analysis found only one eigenvalue that was greater than the 95th percentile of a range of eigenvalues that was simulated over 1000 datasets, thus confirming that one factor is sufficient to account for the factor structure of the PHQ-9.

Finally, the results of the Mokken analysis further confirmed the unidimensionality of the PHQ-9, with all items loading on a single strong and reliable scale. This reinforces the interpretation of the PHQ-9 as a measure of a single, cohesive construct of depression, rather than a collection of disparate symptoms.

In terms of theoretical implications, these findings suggest that the PHQ-9 predominantly measures a single construct of depression, consistent with the unidimensional view of the disorder. This supports the theoretical perspective that depressive symptoms, whether cognitive-affective or somatic, are manifestations of the same underlying condition, rather than distinct dimensions. In practice, the confirmation of unidimensionality has important implications for the use of the PHQ-9 in clinical and research settings. It simplifies the interpretation of scores, allowing clinicians to assess overall depression severity without needing to differentiate between symptom types. From a theoretical standpoint, these findings support the notion that depression can be reliably measured through a set of diverse yet interrelated symptoms, as operationalised in the PHQ-9. This consistency aligns with established theories of depression, which conceptualise it as a multifaceted disorder encompassing cognitive, affective and somatic components. The solid factor loadings across all items further strengthen this perspective, indicating that each item is a meaningful reflection of the latent depression construct.

The study had certain limitations. The cross-sectional design of the study limits the ability to draw causal inferences. Future studies using longitudinal designs could provide more comprehensive insights into the stability and predictive validity of the PHQ-9. Furthermore, the reliance on self-reported data may introduce response biases, such as social desirability or recall bias, which could affect the accuracy of the findings. Although the PHQ-9 is widely used and validated, the incorporation of clinician-administered assessments or corroborative data from other sources could enhance the robustness of future studies. The sample was limited to first responders, which may affect the generalisability of the findings to other populations. First responders are exposed to unique stressors and traumatic events that may influence the expression of depressive symptoms differently compared to the general population or other occupational groups. Participants also came from one province in South Africa, which also limits generalisability.

This study confirms the reliability and validity of the PHQ-9 as a unidimensional measure of depressive symptoms among first responders in South Africa. The strong psychometric properties observed, including internal consistency, factor structure and item functioning, support its use as an effective tool for assessing depression in this high-risk occupational group.