Gustavo Marcos-Escobar
Marcos E. Gómez
Wenceslao Unanue- Business School, Universidad Adolfo Ibáñez, Santiago, Chile
Drawing on recent advances in Beck’s cognitive model—which traditionally conceptualizes anxiety and depression as correlated but does not explicitly address their temporal ordering—this study tests whether anxiety operates as a sequential mediator linking sustained stress to depressive symptoms in a non-clinical adult population. Prior longitudinal and mediation studies have examined associations among stress, anxiety, and depression, but differences in design, population, and analytical focus limit their applicability to non-clinical adult contexts. We extend this literature with data obtained from 805 adults in the Metropolitan Region of Santiago, Chile, followed across three waves at two-month intervals during the COVID-19 pandemic, although no pre-pandemic baseline was available. Accordingly, findings should be interpreted as evidence of the stress–anxiety– depression sequence within pandemic conditions. A cross-lagged panel model (CLPM) was used to estimate the temporal relations among stress, anxiety, and depression. This statistical method accounts for the stability of each construct across repeated measurements while estimating directional relationships over time. Results confirmed a significant partial mediation: perceived stress at T1 predicted higher anxiety at T2, which in turn predicted increased depressive symptoms at T3 (standardized indirect effect β = 0.049, 95% CI [0.016, 0.091], p = 0.009). To the best of our knowledge, this is the first longitudinal study conducted with a non-clinical adult sample from the Metropolitan Region of Chile that validates this mediation sequence. The findings advance Beck’s model by demonstrating anxiety’s role as a sequential mediator, contribute methodologically through the use of a three-wave CLPM to test temporal precedence, and support preventive interventions targeting early detection of subclinical anxiety to disrupt trajectories toward depression. Together, the results update the cognitive model and inform both clinical practice and public health strategies in emotionally demanding contexts.
1 The cognitive framework of anxiety and depression: theoretical convergences, divergences, and sequential hypotheses
This study tests a sequential mediation pathway in which perceived stress increases anxiety over time, and this heightened anxiety, in turn, predicts later depressive symptoms. Grounded in Beck’s (1963, 1967, 1976) cognitive theory of depression, which posits that the activation of dysfunctional cognitive schemas can give rise to various emotional disorders, this study addresses a conceptual gap in the classic model. While Beck’s framework describes how such schemas underpin both anxiety and depression, it does not explicitly formalize anxiety as a sequential mediator linking stress to depression. Building on recent empirical and theoretical advances (Higa-McMillan et al., 2014; Kok et al., 2016; Anyan et al., 2020; Vrshek-Schallhorn et al., 2015), we propose and test a sequential mediational hypothesis: perceived stress activates dysfunctional schemas, anxiety emerges as an early and transitory manifestation, and depression follows as the consolidated outcome when these patterns remain unregulated. This reconceptualization positions anxiety as a sequential mediator—rather than a parallel or co-occurring condition—extending the explanatory power of the cognitive framework.
2 The cognitive framework of anxiety and depression: theoretical convergences and divergences
According to Beck’s cognitive theory, depressive episodes stem from the activation of ingrained schemas organized around the “negative cognitive triad,” pessimistic beliefs about oneself, the world, and the future, which distort information processing and reinforce feelings of hopelessness, worthlessness, and failure (Beck and Alford, 2009). While the classic model emphasized how past experiences shape such schemas, Beck et al. (2024) extend this framework by underscoring that these maladaptive cognitive structures also generate future-oriented expectations of threat and hopelessness, which is central to our conceptualization of anxiety as a sequential mediator. Automatic negative thoughts, including overgeneralizations and catastrophizing, sustain this maladaptive cycle. Depressive symptoms are assessed here with the Patient Health Questionnaire-9 (PHQ-9), a validated measure of symptom frequency over the past 2 weeks (Kroenke et al., 2001).
Within the same architecture, anxiety is linked to threat-oriented schemas (Beck and Clark, 1997; Beck and Emery, 2005), characterized by hypervigilance, overestimation of danger, and underestimation of coping resources. In this model, anxiety is conceptualized as a subclinical emotional state arising early in the stress response. It is measured using the GAD-7 scale, which captures key symptoms of generalized anxiety over the past 2 weeks (Spitzer et al., 2006).
Table 1 summarizes the comparative dimensions of anxiety- and depression-related schemas within Beck’s cognitive framework. Whereas anxiety is primarily linked to threat-oriented schemas that emphasize anticipated danger, vulnerability, and heightened physiological arousal, depression is associated with loss- or defeat-oriented schemas that focus on failure, hopelessness, and withdrawal. The table illustrates the distinct schema profiles of anxiety and depression in Beck’s cognitive framework, highlighting differences in thematic focus, core beliefs, biases, triggers, and emotional correlates.
Table 1. Comparative dimensions of anxiety- and depression-related schemas in Beck’s cognitive framework.
Perceived stress—defined as the appraisal that external demands exceed coping resources (Cohen et al., 1983)—can activate both depression- and anxiety-related schemas, either in parallel or sequentially, producing complex affective trajectories. In the sequential mediator framework proposed here, stress serves as the initiating condition: it activates maladaptive schemas, triggers anxiety as an early manifestation, and, if unregulated, culminates in depressive symptomatology. This theoretical model is empirically tested through a three-wave longitudinal design, evaluating the temporal and functional validity of anxiety as a sequential mediator linking perceived stress to depression.
3 From stress to depression: the articulating function of anxiety
Beck’s (1963, 1967, 1976) cognitive theory of depression proposes that stressful experiences can activate dysfunctional schemas (rigid, negative beliefs about the self, the world, and the future) that distort information processing and foster hopelessness, worthlessness, and failure (Beck and Alford, 2009; Beck et al., 2024). These schemas provide a common cognitive substrate for multiple emotional disorders, establishing a theoretical foundation for both direct and indirect pathways linking stress to affective outcomes.
3.1 Stress to anxiety
Beck et al. (1985a), Beck and Clark (1997) and Clark and Beck (2011), extended this framework to anxiety, positing that stressors appraised as threats can directly activate danger-oriented schemas. This activation manifests in cognitive processes such as threat-oriented attentional biases, overestimation of danger, underestimation of coping resources, and repetitive negative thinking (e.g., worry, rumination) (Michl et al., 2013). Empirical findings corroborate this pathway, showing that perceived stress heightens emotional reactivity and undermines regulatory capacity, thereby increasing vulnerability to anxiety (Higa-McMillan et al., 2014).
From anxiety to depression. Although Beck’s cognitive theory did not formalize anxiety as a transitional stage in the progression toward depression, it proposed that dysfunctional schemas provide a common substrate for both disorders (Beck, 1976; Beck and Alford, 2009). Building on this framework, sustained anxiety can be understood as a mechanism that, by amplifying processes such as worry and rumination, reactivates and strengthens maladaptive schemas. Over time, this recursive cycle depletes regulatory resources, narrows attentional focus toward threat cues, and diminishes engagement in rewarding activities, thereby increasing vulnerability to depression. Longitudinal studies provide partial support for this pathway, showing that trait anxiety predicts subsequent depressive symptoms in specific populations, such as trauma-exposed samples (Kok et al., 2016). However, empirical evidence remains limited in scope, particularly in non-clinical adult populations, underscoring a theoretical gap that this study aims to address.
3.2 Stress directly leads to depression
Consistent with Beck’s (1976) original formulation, stress may also precipitate depressive episodes without intermediary processes by directly activating schemas of loss and hopelessness. Longitudinal and epidemiological research corroborates this direct etiological role, showing that chronic or severe stress predicts depressive outcomes even after controlling for genetic predispositions and prior episodes (Kendler et al., 2007; Vrshek-Schallhorn et al., 2015).
In sum, three pathways with theoretical and/or empirical support emerge from this framework: (a) stress exerting a direct effect on depression, (b) stress eliciting anxiety, and (c) anxiety functioning as an intermediary mechanism linking stress to depression. Collectively, these pathways provide the conceptual rationale for the analytic strategy adopted in this study. Specifically, a three-wave cross-lagged panel model (CLPM) with bootstrapped confidence intervals was employed, enabling the simultaneous estimation of direct and indirect effects while accounting for temporal stability and reciprocal influences across measurement points (Hamaker et al., 2015; MacKinnon et al., 2004).
4 Conceptual gaps and contributions
While Beck (1976), Beck et al., (1985b) cognitive model identifies maladaptive schemas as a shared substrate for anxiety and depression, it does not specify a temporal sequence between them. Much subsequent literature has emphasized co-occurrence. For example, the tripartite model (Clark and Watson, 1991) links both syndromes via high negative affect, differing mainly in physiological hyperarousal (anxiety) or low positive affect (depression). Similarly, Brown et al. (1998) and transdiagnostic models view them as concurrent expressions of shared vulnerabilities, and epidemiological studies (Kendler et al., 2007) report strong bidirectional associations without temporal ordering. Although prior longitudinal studies have examined associations among stress, anxiety, and depression, important gaps remain. For example, Kok et al. (2016) demonstrated that trait anxiety can mediate the link between stress and depression. However, their clinical trauma-exposed sample limits the generalizability of findings to non-clinical adults. Similarly, Anyan et al. (2018) and Anyan et al. (2020) provided evidence consistent with a sequential process in adolescent populations. Nevertheless, their results cannot be directly extrapolated to adults outside clinical settings. Finally, Jacobson and Newman (2017) examined longitudinal associations between anxiety and depression in adults but did not position anxiety as a structured sequential mediator within a stress–depression mediation framework.
These limitations converge on three unresolved issues: (a) the lack of explicit operationalization of anxiety as a sequential mediator in adult non-clinical samples, (b) insufficient methodological designs to establish temporal precedence and bidirectional effects—many prior studies relied on only two measurement waves, which preclude testing temporal ordering and separating intraindividual stability from change (Maxwell and Cole, 2007; Mitchell and Maxwell, 2013)—, and (c) limited generalizability across different populations. In our framework, anxiety is operationally defined as a sequential mediator characterized by heightened threat-oriented attentional bias, overestimation of danger, underestimation of coping resources, and increased repetitive negative thinking (e.g., worry, rumination). It is measured with the Generalized Anxiety Disorder-7 (GAD-7; Spitzer et al., 2006), which assesses symptom frequency over the past two weeks, and is temporally positioned between perceived stress assessed with the Perceived Stress Scale-14 (PSS-14; Cohen et al., 1983)—and depressive symptoms—assessed with the Patient Health Questionnaire-9 (PHQ-9; Kroenke et al., 2001).
The present study addresses these gaps by explicitly testing the sequential mediation pathway from stress to depression via anxiety in a non-clinical adult population, using a three-wave cross-lagged panel model (CLPM) with bootstrapped confidence intervals. This design allows for the estimation of both direct and indirect effects while controlling for intraindividual stability and reciprocal influences over time (Hamaker et al., 2015; MacKinnon et al., 2004). By combining a temporally sensitive analytical approach with an updated cognitive framework, this work reconceptualizes anxiety as a sequential mediator and refines the explanatory scope of Beck’s model.
5 Proposed model and study design
The proposed model builds directly on the theoretical rationale developed in Sections 2 and 3. Specifically, it assumes that perceived stress activates maladaptive cognitive schemas and threat-oriented processes, which increase vulnerability to anxiety. Sustained anxiety, in turn, undermines regulatory resources, fosters repetitive negative thinking (e.g., worry, rumination), and facilitates the emergence of depressive symptoms. At the same time, consistent with Beck’s (1976) original cognitive formulation and subsequent longitudinal evidence (Kendler et al., 2007; Vrshek-Schallhorn et al., 2015), stress may also exert a direct effect on depression by activating schemas of loss and hopelessness, independent of the mediating role of anxiety. Preliminary analyses of our three-wave data supported this dual structure: perceived stress predicted later anxiety, anxiety predicted subsequent depressive symptoms, and a significant indirect effect from stress to depression through anxiety was observed, while a weaker but significant direct effect from stress to depression also emerged. This dual specification provides both theoretical coherence and empirical testability, offering logical clarity and theoretical consistency, as well as empirical testability (Ashkanasy, 2016).
To examine these mechanisms, a three-wave cross-lagged panel model (CLPM) with bootstrapped confidence intervals was employed, which enables the estimation of directional effects across time points while simultaneously controlling for autoregressive stability and reciprocal influences. By incorporating repeated measures of both mediator and outcome, this analytic approach ensures temporal ordering and reduces bias in testing indirect effects (Hamaker et al., 2015; Maxwell and Cole, 2007; Mitchell and Maxwell, 2013; MacKinnon et al., 2007). Accordingly, the following hypotheses were formulated:
H1: Perceived stress at Time 1 will predict increased anxiety at Time 2.
H2: Anxiety at T2 will predict increased depressive symptoms at T3.
H3: Perceived stress at T1 will predict increased depressive symptoms at T3.
H4: Anxiety at T2 will mediate the effect of stress at T1 on depression at T3.
Consistent with Beck’s cognitive framework, we hypothesized that perceived stress at T1 would predict increased anxiety at T2 (H1), that anxiety at T2 would predict higher depressive symptoms at T3 (H2), and that perceived stress at T1 would directly predict depressive symptoms at T3 (H3). Furthermore, we posited that anxiety at T2 would mediate the longitudinal association between stress at T1 and depression at T3 (H4). We did not specify whether this mediation would be complete or partial, as Beck’s model allows for both direct and indirect effects, and we aimed to let the data inform the relative magnitude of these pathways. In addition, the inclusion of the control variable ‘fear of COVID-19’ was theoretically justified, given that pandemic-related fears have been shown to affect stress, anxiety, and depression simultaneously (Ahorsu et al., 2020). Controlling for this variable ensured that the mediation pathway captured the unique contribution of stress and anxiety, rather than reflecting pandemic-specific confounding effects.
6 Method
6.1 Procedure
This longitudinal study consisted of three measurements taken at two-month intervals and was developed according to the guidelines of the American Psychological Association and the British Psychological Society. It was approved by the university’s ethics and research committee in Chile. Data collection was carried out with the support of Netquest (www.netquest.com), an international company specializing in the management of online panel data (OPD). This company operates an OPD in Chile, serving 122,374 people, and complies with the quality standards defined by ISO 26362. In addition, Netquest is a pioneer in Latin America, holding the ISO 20252 certification, which aims to ensure high standards in market, social, and opinion research. OPDs offer significant advantages over convenience samples or face-to-face surveys, allowing data to be collected more quickly, with less bias and greater cost efficiency, especially in longitudinal studies (Porter et al., 2019). To ensure data quality, Netquest was instructed to apply the methodological guidelines of Porter et al. (2019) and to implement rigorous incentive strategies. Ongoing communication was maintained between the company and the research team. Participants were invited to complete a questionnaire designed by the researchers. An informed consent form describing the study’s objectives was included in each wave of the study. During T1, participants agreed to take part in a three-stage longitudinal study, authorizing their participation in T2 and T3. Only those who completed T1 were eligible to participate in T2, and only those who completed both T1 and T2 were invited to participate in T3.
6.2 Sample
A representative sample was drawn from the Netquest database in Santiago, Chile, with a proportional distribution by gender and age, based on the 2017 Chilean census. This sample accurately represents the adult population of the Metropolitan Region, which constitutes 41.18% of the national total. The first wave of data collection took place in August 2021, aiming to reach at least 479 participants who completed all three measurements. Recommendations regarding SEM sample size vary. Some authors suggest a minimum of 200 participants or 10 per parameter (Weston and Gore, 2006), while others propose at least 5 per parameter (Worthington and Whittaker, 2006). Monte Carlo simulations indicate that the required sample size may range from 30 to 460 cases, depending on multiple factors. In this study, the threshold recommended by Wolf et al. (2013) was surpassed, aligning with similar studies, such as those by Burkholder and Harlow (2003).
To strengthen the validity of the study, we implemented a three-wave longitudinal design. Cross-sectional studies, although common in stress and mental health research, are limited because they cannot establish temporal precedence and are unable to disentangle reciprocal or mediational processes among variables (Cole and Maxwell, 2003). In contrast, longitudinal designs enable the examination of intraindividual change and the testing of directional relationships over time. Specifically, having three waves makes it possible to estimate indirect effects more robustly and to model reciprocal associations between stress, anxiety, and depression, which is not feasible with two-wave designs (Little, 2013). This design, therefore, provides more substantial evidence regarding the temporal dynamics of psychological distress and enhances the interpretability of mediation mechanisms.
A post-hoc power analysis was performed to verify the adequacy of the sample size. Following recommendations that discourage post-hoc power based on observed effect sizes (Gelman, 2019), we conducted an RMSEA-based sensitivity power analysis at the model-fit level (close-fit vs. not-close-fit), which does not rely on any single observed parameter. Specifically, using α = 0.05, df = 450, and a close-fit null of RMSEA_H0 = 0.05, achieved power was evaluated at the observed sample size (N = 805) with the semPower package (Moshagen and Erdfelder, 2016). The analysis indicated Power (1 – β) > 0.99, confirming that the sample size was adequate to detect the hypothesized effects. This calculation is reported solely as a complementary sensitivity check of sample adequacy, rather than as a conclusive inferential test. The inclusion criteria were being over 18 years of age, residing in the Metropolitan Region, being a native Spanish speaker, and agreeing to participate in the three waves by providing informed consent. Those who had recently participated in similar research or did not meet the requirements were excluded. No additional restrictions were applied. A total of 805 adults (M = 43.27, SD = 15.54; 51.55% women) completed T1. In T2, 599 individuals participated (mean age = 45.02, SD = 15.67; 52.09% female), and in T3, 479 individuals (mean age = 46.37, SD = 15.72; 53.24% female). The sociodemographic characteristics of period T1 are presented in Table 2.
Table 2. Sociodemographic characteristics of participants at T1 (N = 805).
To assess dropout effects, comparisons were made between participants who completed all waves and those who dropped out after T1. Attrition rates were 25.6% between T1 (n = 805) and T2 (n = 599), and 20.0% between T2 and T3 (n = 479), corresponding to an overall retention rate of 59.5% across the three waves. No significant differences were found in gender (χ2(1) = 1.34, p = 0.247), perceived stress (t(803) = −1.33, p = 0.184), anxiety (t(803) = −1.97, p = 0.050), depression (t(803) = −1.79, p = 0.073), or fear of Covid-19 (t(803) = 0.20, p = 0.840). However, there was an age difference (t(737.92) = 6.34, p < 0.001), as participants who remained in the study tended to be older. Although this difference does not undermine the internal validity of the results, it may indicate that younger participants were more likely to discontinue participation, a pattern also documented in longitudinal survey research (Fumagalli et al., 2013), potentially limiting the generalizability of the findings to younger age groups. Even so, attrition did not compromise the validity of the study. Little's (1988) test confirmed that missing data followed a pattern of missing completely at random (MCAR), χ2(12) = 11.99, p = 0.100. Therefore, the full information maximum likelihood method (FIML) was used to handle missing data, as it provides efficient and unbiased estimates under the MCAR assumption (Newman, 2014). Nonetheless, we acknowledge that MCAR tests have limited power to detect subtle patterns of non-random missingness; thus, the reported attrition rates and demographic differences should be considered when interpreting the results. In particular, although our test supported MCAR, it is more realistic to assume that missingness in longitudinal survey research may approximate “missing at random” (MAR) rather than a strict MCAR mechanism. Under both MCAR and MAR, the use of FIML remains appropriate; however, the assumption of completely random missingness should be interpreted with caution, particularly about factors such as the type of stress or educational level.
The variables analyzed exhibited acceptable skewness and kurtosis distributions, as per Kline (2016). For perceived stress, skewness was (0.17 at T1, 0.63 at T2, 0.57 at T3) and kurtosis (−0.23 at T1, 0.42 at T2, 0.18 at T3). For anxiety, skewness was (0.50 at T1, 0.58 at T2, 0.56 at T3) and kurtosis (−0.64 at T1, −0.38 at T2, −0.41 at T3). Regarding depression, skewness values were (0.81 at T1, 1.02 at T2, 0.97 at T3) and kurtosis (0.03 at T1, 0.53 at T2, 0.48 at T3).
6.3 Measurements
Validated instruments with robust psychometric properties were used, and conceptual equivalence was ensured by the back-translation procedure (Brislin, 1970) when necessary.
6.3.1 Perceived stress
Stress was assessed using the Perceived Stress Scale (PSS-14; Cohen et al., 1983), a Spanish validated version culturally adapted for Chile (Tapia et al., 2007). More recently, the PSS has also been validated in Chilean teachers from Copiapó, demonstrating adequate reliability (α ≈ 0.83) and factorial validity (Jorquera-Gutiérrez, 2023). Additionally, a transcultural adaptation of the PSS-10 was conducted in Chilean Spanish (Lagos-Riveros et al., 2023), further supporting its cultural validity. The instrument comprises 14 items that assess the frequency with which individuals perceive their lives as unpredictable, uncontrollable, and stressful over the last month. Responses were recorded on a Likert-type scale from 0 (never) to 4 (very often). Examples of items include “In the past month, how often have you felt that you could not control the important things in your life?” and “In the past month, how often have you felt that difficulties were piling up so much that you could not overcome them?”
6.3.2 Anxiety
Anxiety was measured with the Generalized Anxiety Disorder scale (GAD-7; Spitzer et al., 2006), using the Spanish culturally adapted version validated by García-Campayo et al. (2010). In Chile, the GAD-7 has also been validated in adolescents, confirming good psychometric properties and reliability (α = 0.86) (Gaete et al., 2022). The scale comprises seven items that assess anxiety symptoms over the last 2 weeks. Responses are given on a Likert scale from 0 (not at all) to 3 (almost every day). Examples of items include “Feeling nervous, anxious, or on edge” and “Worrying too much about different things.”
6.3.3 Depression
Depression was measured using the Patient Health Questionnaire (PHQ-9; Kroenke et al., 2001), specifically the Spanish validated version applied in Chilean primary care by Baader et al. (2012), which includes the nine items plus an additional functional impairment question, and demonstrated good sensitivity, specificity, and reliability (α = 0.83–0.89). Additional Chilean studies have further supported its psychometric performance, including Saldivia et al. (2019) in primary care patients and Borghero et al. (2018) in adolescents (α = 0.86). The PHQ-9 consists of 9 items, each rated on a 4-point scale (0 = “not at all,” 1 = “several days,” 2 = “more than half the days,” 3 = “nearly every day”), with a total score range of 0–27. Examples include “Feeling depressed, down, or hopeless” and “Having little interest or pleasure in doing things.”
7 Results
7.1 Analysis plan
To test our hypotheses, we modeled all constructs as latent variables in order to minimize measurement errors (Finkel, 1995). To test our longitudinal mediation hypothesis, we employed structural equation modeling (SEM; Cole and Maxwell, 2003; Selig and Preacher, 2009), following the recommendations of Maxwell et al. (2011) and O'Laughlin et al. (2018). Figure 1 illustrates our hypothesized model. We implemented a panel design in which anxiety at T2 acts as a mediating variable for the association between perceived stress at T1 and depression at T3. All variables were controlled for their lagged autoregressive trajectories. Following the approach proposed by Jöreskog (1979), we incorporated correlated residuals for the observed indicators of each latent variable and allowed all latent variables to covary freely.
Figure 1. Hypothesized model. Longitudinal mediation model of anxiety, perceived stress, and depression. All variables are latent constructs. STR, perceived stress; ANX, anxiety; DEP, depression. Time 1 (T1), Time 2 (T2), and Time 3 (T3). Paths a, b, and c represent hypothesized mediation effects corresponding to H1 (a > 0), H2 (b > 0), and H3 (c > 0).
In our sample, the internal consistencies of the instruments were adequate. For perceived stress (PSS-14), Cronbach’s α was 0.88 at T1, T2, and T3, and the construct was modeled as a latent variable with four balanced parcels. For anxiety (GAD-7), Cronbach’s α was 0.92 at T1, 0.92 at T2, and 0.93 at T3, modeled as a latent variable using three randomly distributed parcels. For depression (PHQ-9), Cronbach’s α was 0.90 at T1, 0.91 at T2, and 0.91 at T3, modeled as a latent variable with three parcels formed by random assignment of items.
The variable fear of COVID-19 was incorporated as a control due to its well-documented associations with anxiety, depressive symptoms, and stress during the pandemic (Ahorsu et al., 2020; Alimoradi et al., 2022). Its inclusion was not intended to examine the specific impact of the pandemic per se, but rather to isolate the hypothesized stress–anxiety–depression pathway. Operationally, fear of COVID-19 was specified as an exogenous predictor at each wave, directly predicting stress, anxiety, and depression. This specification controlled for the variance attributable to the pandemic context, preventing artificial inflation of associations among the psychological variables. As a result, the longitudinal trajectories represent net relationships independent of the contextual impact of fear of COVID-19, ensuring that the observed effects reflect genuine links. Since all waves occurred during the pandemic, results should be understood within this context. Consistent with theoretical completeness criteria, we report the models that incorporate this control variable. Notably, although the model’s results and substantive conclusions remained virtually unchanged when estimated without it, the reported analyses correspond to the models with the control included. This reinforces the robustness of the findings and indicates that the fear of COVID-19 did not exert a significant influence on the tested mediation process.
Following the recommendations of Hu and Bentler (1999) and Kline (2016), model fit was evaluated using multiple indices, including the chi-square test (χ2), the root mean square error of approximation (RMSEA) with its 90% confidence interval, the comparative fit index (CFI), the Tucker–Lewis index (TLI), and the standardized root mean square residual (SRMR). Interpretation was guided by widely used cutoffs, with RMSEA values ≤ 0.06 (0.08), CFI/TLI ≥ 0.95 (0.90), and SRMR ≤ 0.08 considered indicative of good (acceptable) fit, while recognizing the sensitivity of χ2 to sample size and model complexity. Descriptive statistics are reported in Table 2, and zero-order correlations among the study variables are presented in Table 3. All structural and longitudinal mediation analyses were conducted in Mplus 8.0 (Muthén and Muthén, 2017), with indirect effects tested via bias-corrected bootstrap resampling (1,000 draws, MODEL INDIRECT). Standardized indirect effects (β) and 95% bias-corrected confidence intervals were reported, and statistical significance was established when the confidence intervals did not include zero.
Table 3. Descriptive statistics and zero-order correlations between all variables at T1, T2 and T3 (T1: N = 805; T2: N = 599; T3: N = 479).
7.2 Estimates
To assess the factorial validity of the scales used, confirmatory factor analyses (CFA) were performed separately at each of the three measurement waves. In all cases, a single-factor model, in which all items loaded on a single latent factor, was compared with a theoretically grounded model composed of three distinct latent factors. At time one (T1), the single-factor model showed poor fit: χ2(35) = 1053.79, p < 0.001, CFI = 0.864, TLI = 0.825, RMSEA = 0.190, 90% CI [0.180, 0.200], SRMR = 0.056. In contrast, the three-factor model showed a substantially better fit: χ2(32) = 161.90, p < 0.001; CFI = 0.983; TLI = 0.976; RMSEA = 0.071, 90% CI [0.060, 0.082]; SRMR = 0.022. The difference between models was significant, Δχ2(3) = 891.89, p < 0.001. This procedure was replicated at T2. The one-factor model again presented an inadequate fit: χ2(35) = 716.63, p < 0.001, CFI = 0.875, TLI = 0.839, RMSEA = 0.180, 90% CI [0.169, 0.192], SRMR = 0.051. The three-factor model, by contrast, showed a markedly superior fit: χ2(32) = 102.53, p < 0.001, CFI = 0.987, TLI = 0.982, RMSEA = 0.061, 90% CI [0.048, 0.074], SRMR = 0.017. The difference between models was significant, Δχ2(3) = 614.11, p < 0.001. Similarly, at T3, the one-factor model showed an unsatisfactory fit: χ2(35) = 616.38, p < 0.001, CFI = 0.872, TLI = 0.836, RMSEA = 0.186, 90% CI [0.173, 0.199], SRMR = 0.051. The three-factor model again presented a better fit: χ2(32) = 80.97, p < 0.001, CFI = 0.989, TLI = 0.985, RMSEA = 0.057, 90% CI [0.041, 0.072], SRMR = 0.019. The difference between models was statistically significant, Δχ2(3) = 535.41, p < 0.001. Taken together, these results provide robust evidence supporting the discriminant factorial validity of the three constructs analyzed, justifying their treatment as conceptually and empirically differentiated dimensions over time.
To assess metric invariance, two versions of a nine-factor latent model across waves were compared. First, a base model with no restrictions on factor loadings (the free model) was estimated. This model presented an adequate fit: χ2(369) = 1320.06, p < 0.001, CFI = 0.952, TLI = 0.944, RMSEA = 0.057, 90% CI [0.053, 0.060], and SRMR = 0.025. Subsequently, a second model was estimated, imposing equality restrictions on all factor loadings (the restricted model). The fit of this model was also adequate and virtually identical to the previous one: χ2(383) = 1332.09, p < 0.001, CFI = 0.952, TLI = 0.946, RMSEA = 0.055, 90% CI [0.052, 0.059], SRMR = 0.026. The comparison between the two models showed a non-significant chi-square difference: Δχ2(14) = 12.03, p = 0.604. Furthermore, no change was observed in the CFI value (ΔCFI = 0.000), indicating that the restrictions imposed did not deteriorate fit quality. According to the criteria proposed by Cheung and Rensvold (2002), these results support metric invariance, which allows us to assume that the relationships between the items and the latent constructs remain stable over time. Therefore, the restrictions were maintained in the subsequent structural analyses.
7.3 Structural models and comparisons
Then, following Cole et al. (2005), we constrained all autoregressive and cross-lagged paths to be time-invariant. This decision was based on the premise that, by keeping the time intervals between measurements constant (2 months), there is insufficient theoretical justification to expect variations in the magnitude of the coefficients over time. This simplification enabled us to model the effects between T1 and T2, as well as between T2 and T3, using a single parameter for each trajectory, which favors model parsimony and improves statistical power. To validate this assumption, we compared constrained and unconstrained models using Chi-square tests (Kline, 2016). The resulting loading-path model showed a good fit to the data, χ2(455) = 891.38, p < 0.001, CFI = 0.978, TLI = 0.975, RMSEA = 0.035, 90% CI [0.031, 0.038], SRMR = 0.045. This model was compared with the loading-only model, whose fit was χ2(450) = 870.19, p < 0.001, CFI = 0.979, TLI = 0.975, RMSEA = 0.034, 90% CI [0.031, 0.037], SRMR = 0.043. To evaluate the imposed restrictions, a chi-square difference test was performed between the loading model (χ2 = 870.19, df = 450) and the loading-path model (χ2 = 891.38, df = 455). The result was Δχ2(5) = 21.19, with a p-value ≈ 0.0011. Although this difference was statistically significant, it should be interpreted with caution, given the sample size (n = 805) and the stability of the overall fit indices (ΔCFI = 0.001). This suggests that the constraints do not substantially deteriorate the fit, supporting the choice of the more parsimonious constrained model. The structural paths of the mediation model are illustrated in Figure 2 and summarized in Table 4. As for the autoregressive effects, and given that time invariance was assumed, only the estimates corresponding to the interval T1 to T2 are reported in the text. These were statistically significant for the three latent variables: perceived stress (β = 0.84, 95% CI [0.80, 0.87], p < 0.001), anxiety (β = 0.62, 95% CI [0.50, 0.73], p < 0.001), and depression (β = 0.50, 95% CI [0.37, 0.62], p < 0.001), indicating moderate-to-high temporal stability. As for the cross-lagged effects, and consistent with the assumption of time invariance, only the estimates corresponding to the interval T1 to T2 are reported. Stress at T1 significantly predicted higher levels of anxiety at T2 (β = 0.17, 95% CI [0.06, 0.28], p = 0.002), while anxiety at T1 predicted subsequent increases in depression at T2 (β = 0.27, 95% CI [0.17, 0.38], p < 0.001). Factor loadings ranged from 0.75 to 0.95 (all with p < 0.001), and R2 values ranged from 0.54 to 0.74 (all with p < 0.001). Taken together, these findings provide empirical support for H1, as stress at T1 significantly predicted anxiety at T2, and for H2, given that anxiety at T2 was a robust predictor of depression at T3. H3 was also supported, as stress at T1 directly predicted depressive symptoms at T3.
Figure 2. Longitudinal mediation model of anxiety, perceived stress and depression, controlling for fear of COVID-19. All variables are latent constructs. STR, perceived stress; ANX, anxiety; DEP, depression. Time 1 (T1), Time 2 (T2), and Time 3 (T3). Solid lines represent statistically significant paths. Covariances among latent variables at each wave and factor loadings of the latent constructs were omitted for simplicity; all factor loadings ranged between 0.75 and 0.95 (p < 0.001). Model fit indices: χ2(450) = 870.19, p < 0.001, CFI = 0.979, TLI = 0.975, RMSEA = 0.034, 90% CI [0.031, 0.037], SRMR = 0.043. Significance levels: ***p < 0.001, **p < 0.01, *p < 0.05.
Table 4. Standardized path coefficients in the longitudinal mediation model linking stress, anxiety, and depression.
7.4 Mediation analysis
The results also support the central hypothesis of the study, which postulates that anxiety acts as a mediating variable in the longitudinal relationship between perceived stress and depression (see Figure 3 and Table 4). To assess this mediation, a longitudinal model with three temporal measurements was specified, in which perceived stress at T1 predicts anxiety at T2 (trajectory a), in turn, predicts depression symptoms at T3 (trajectory b). We also included the direct trajectory of stress at T1 on depression at T3 (trajectory c’), which allowed us to distinguish between full and partial mediation. Perceived stress at T1 significantly predicted anxiety at T2, β = 0.17, 95% CI [0.062, 0.282], p = 0.002, and anxiety at T2 significantly predicted depression at T3, β = 0.29, 95% CI [0.172, 0.393], p < 0.001. The direct effect of perceived stress on depression was also significant, although of smaller magnitude, β = 0.08, 95% CI [0.009, 0.167], p = 0.031, indicating that the mediation is partial rather than full. The indirect effect of stress on depression, mediated by anxiety, was significant, with a bias-corrected 95% bootstrap confidence interval (1,000 iterations): indirect effect = 0.049, 95% CI [0.016, 0.091], p = 0.009. These findings empirically support the hypothesis that perceived stress exerts both direct and indirect effects on subsequent mental health, and that anxiety plays a key role as an emotional mechanism in this process.
Figure 3. Longitudinal mediation model linking perceived stress (T1) to depression (T3) through anxiety (T2). All variables are latent constructs. STR, perceived stress; ANX, anxiety; DEP, depression. T1, Time 1; T2, Time 2; T3, Time 3. Coefficients are standardized. Solid lines represent statistically significant paths. Indirect and total effects were estimated in Mplus with 95% bootstrap confidence intervals (CIs) based on 1,000 iterations. ***p < 0.001; **p < 0.01; *p < 0.05.
To further evaluate the robustness of these findings, we examined whether the final sample size (n = 479 completers) provided sufficient statistical power to detect the hypothesized indirect effect. We conducted a post-hoc Monte Carlo simulation with 200,000 replications based on the standardized path estimates from the fitted model (a = 0.172, SE = 0.055; b = 0.286, SE = 0.057). Simulations were implemented in Python using the NumPy library to generate random draws from normal distributions for each path, compute the product a × b, and evaluate its sampling distribution. The proportion of simulated indirect effects with two-tailed p < 0.05 provided the estimated power, which was 0.86, exceeding the conventional 0.80 threshold (Selig and Preacher, 2009). This analysis supports the adequacy of the sample size and reinforces the validity of the mediation results. Accordingly, H4 was supported, as anxiety at T2 significantly mediated the longitudinal association between stress at T1 and depression at T3.
8 Discussion
Several studies have shown that chronic stress constitutes a robust antecedent of psychological distress, particularly when it is internalized through dysfunctional patterns of emotional and cognitive processing. Consistent with this line of research, our results reveal that anxiety acts as a significant mediating mechanism in the relationship between perceived stress and depression. Specifically, the longitudinal structural model confirmed that higher levels of perceived stress at T1 predict increases in anxiety at T2, which in turn predict higher levels of depression at T3. This finding provides empirical evidence that clarifies a functional sequence proposed theoretically by Beck (1976), Beck et al. (2024) and supported by recent work (Anyan et al., 2017, 2020). However, this sequence has been insufficiently validated using longitudinal designs to date.
8.1 Theoretical implications
One of the main theoretical contributions of this study is the empirical validation of a functional sequence originally suggested by Beck’s cognitive model (Beck, 1976; Beck et al., 2024) in which anxiety operates as a sequential mediator linking stress to depression. Previous studies have provided support for this pathway (Anyan et al., 2017, 2018, 2020). However, their reliance on two-wave designs limited the ability to establish temporal precedence or to disentangle intraindividual stability from change. By employing a three-wave longitudinal approach with CLPM models, the present study addresses this gap. It provides more robust evidence for the mediational role of anxiety in non-clinical adult populations. In this framework, stress activates dysfunctional schemas that elicit anxious responses, which, if sustained, evolve into depressive symptomatology. Although this progression was clinically acknowledged in earlier versions of the cognitive model, the role of anxiety as a formal mediating mechanism had not been empirically validated with sufficient rigor. In terms of magnitude, the effects observed in this study were moderate. The path from stress to anxiety (β = 0.17) and from anxiety to depression (β = 0.29) are comparable to those reported in community and student samples, such as Anyan et al., 2017, β = 0.19 and β = 0.25) and Jacobson and Newman (2017, β = 0.15–0.23 and β = 0.18–0.24). By contrast, Kok et al. (2016) reported substantially stronger effects when examining the mediating role of trait anxiety in cardiac surgery patients (β = 0.34 and β = 0.74), with full mediation of the stress–depression link. These differences can be explained both by the clinical and trauma-exposed nature of their sample and by the use of trait anxiety, a stable personality factor strongly associated with depression, rather than anxiety assessed at a specific time point as in the present study. This comparison underscores the contribution of our work: it demonstrates that Beck’s functional sequence also holds in non-clinical adults under conditions of sustained stress, albeit with more moderate effect sizes, thereby supporting its robustness and generalizability beyond clinical contexts.
First, methodologically, our work strengthens the internal validity of this mediational sequence. Using CLPM modeling, we were able to establish temporal ordering while controlling for intraindividual stability, meeting the classic criteria of mediation (Baron and Kenny, 1986). This refinement not only confirms anxiety’s role within a cognitive–affective trajectory but also represents a concrete contribution to the field by advancing longitudinal methods suited to testing cognitive–emotional processes.
Second, this work contributes to the refinement of the classical cognitive model by clarifying the specific role of anxiety within the stress-depression sequence. Whereas Beck’s original model emphasizes the activation of dysfunctional schemas as an antecedent of depressive states, our findings show that anxiety is not merely a co-occurring symptom, but a mediating mechanism that cognitively transforms the perception of threat into depressive symptomatology. This refinement allows us to describe the model as a temporal sequence with distinct affective phases.
Third, the study contributes to the literature on temporal affective mediation by providing evidence that anxiety acts as a transient emotional state with a prospective effect, which has implications for theories of emotional regulation and the maintenance of negative affect. Thus, it opens the possibility of incorporating this mediation into broader frameworks that integrate dimensions of time, intensity, and emotional reactivity, moving towards a more dynamic understanding of the psychopathological course.
At the same time, it is important to situate these findings within broader theoretical debates. While Beck’s (1976), Beck et al., (1985b) cognitive model identifies maladaptive schemas as a shared substrate for anxiety and depression, it does not specify a temporal sequence. Subsequent frameworks have emphasized co-occurrence rather than sequential ordering. For example, the tripartite model (Clark and Watson, 1991) links anxiety and depression through high negative affect, differing mainly in physiological hyperarousal or low positive affect. Transdiagnostic approaches (Brown et al., 1998) similarly conceptualize them as concurrent expressions of shared vulnerabilities, and epidemiological studies (Kendler et al., 2007) report strong bidirectional associations without temporal ordering. Our findings challenge these perspectives by providing temporal evidence that anxiety prospectively mediates the stress–depression pathway, thereby imposing sequential structure on a relationship often treated as simultaneous or reciprocal.
Unlike the tripartite model (Clark and Watson, 1991), which conceptualizes anxiety and depression as concurrent outcomes of heightened negative affect differentiated only by physiological arousal or diminished positive affect, our findings demonstrate temporal precedence, thereby challenging the assumption of synchronicity and highlighting anxiety’s prospective role in shaping depressive outcomes. Similarly, transdiagnostic approaches (Brown et al., 1998) and epidemiological studies (Kendler et al., 2007) emphasize strong bidirectional associations without specifying temporal ordering; however, the present evidence suggests that stress-related anxiety exerts a directional effect on subsequent depression rather than emerging simultaneously. Beyond these models, rumination theory (Nolen-Hoeksema et al., 2008) and metacognitive frameworks (Wells, 2009) propose cognitive mechanisms, repetitive negative thinking, and dysfunctional metacognitive beliefs that may interact with anxiety in shaping vulnerability to depression. Importantly, empirical findings support this proposition. For instance, Hong (2007) demonstrated that worry (anxious repetitive thought) predicted subsequent increases in both anxiety and depression, whereas rumination showed a more specific association with depressive symptoms. Similarly, Dickson et al. (2012) reported that perseverative worry exacerbates negative affect and facilitates the transition from anxious arousal to depressive mood. These findings illustrate how stress-induced anxiety may trigger worry processes, which in turn reinforce rumination and deepen depressive trajectories. Integrating such evidence strengthens the interpretation that anxiety not only functions as an independent mediator but may also operate sequentially with cognitive vulnerabilities, thereby extending Beck’s framework toward a more dynamic model of risk.
Although our study did not directly test these mechanisms, future research should clarify whether anxiety interacts with repetitive thought processes in parallel or as part of a broader sequential trajectory within Beck’s framework.
Ultimately, a key strength of this study lies in its three-wave longitudinal design, which allows for the evaluation of temporal ordering and indirect effects with greater rigor than cross-sectional or two-wave approaches (Cole and Maxwell, 2003; Little, 2013).
8.2 Practical implications
Notably, the study was conducted during the COVID-19 pandemic, a period characterized by unprecedented health, social, and economic stressors. Although our design does not allow us to compare stress levels before and during the pandemic, this context nonetheless provides a meaningful backdrop for interpreting the mediational sequence. Specifically, sustained stressors present in crisis contexts may intensify the transition from anxiety to depression. Framing the findings within this context underscores their real-world significance while recognizing that the results should be interpreted as evidence of the stress, anxiety, and depression pathway under pandemic conditions. Importantly, fear of COVID-19 was explicitly modeled as an exogenous predictor of stress, anxiety, and depression at each wave, ensuring that the reported associations reflect genuine psychological mechanisms rather than pandemic-driven confounding.
From an applied perspective, this study offers relevant implications for clinical settings, community strategies, and the design of public mental health policies. Our findings suggest that anxiety not only coexists with depression but also functions as an early indicator with potential mediating properties. Therefore, preventive strategies should prioritize early detection and clinical management of subclinical anxiety to interrupt its progression to more severe depressive conditions. This orientation aligns with evidence from randomized controlled trials showing that early anxiety symptoms in at-risk individuals can predict the later onset of depressive episodes. This orientation is consistent with the evidence provided by controlled trials that have demonstrated the efficacy of early interventions in at-risk populations (Garber et al., 2009).
In this framework, it is crucial to implement psychoeducational and therapeutic interventions adapted to the social and emotional context in which stress manifests. Cognitive-behavioral programs that consider situational factors have shown promising results, especially when they incorporate training in emotional self-regulation, symptom monitoring, or the provision of containment spaces in clinical and community settings. Nevertheless, implementing such strategies may face significant barriers. In the Chilean context, inequities in access to specialized services, disparities between urban and peri-urban areas, and sociocultural norms that influence help-seeking behaviors represent challenges that limit the scalability and effectiveness of interventions. Acknowledging these barriers highlights the importance of tailoring preventive strategies to the sociocultural and geographical characteristics of the populations they are directed towards, thereby ensuring ecological validity and enhancing their real-world applicability. Specifically, interventions that target anxiety in its early phases may function as critical “disruption points” in the sequential pathway, reducing the likelihood that stress exposure crystallizes into depressive symptomatology.
In addition, the results present an opportunity for the development of public policies aimed at preventing affective disorders from a population-based approach. Educational institutions, primary health services, and employers could implement structured screening programs that identify risk profiles and activate early referral mechanisms to support individuals at risk. Screening programs that systematically detect subclinical anxiety could serve as a feasible, cost-effective strategy to intervene at the sequential stage before depression emerges. In this regard, our findings highlight the value of community- and workplace-based programs that strengthen resilience and stress-management skills in emotionally demanding contexts, as a concrete preventive strategy directly aligned with the mediating role of anxiety identified in this study. In parallel, the development of digital platforms and mobile applications capable of monitoring distress patterns and delivering automated interventions could substantially broaden the preventive reach of traditional clinical strategies. Such tools are particularly valuable for underserved populations or individuals facing barriers to timely access, as they can provide scalable, low-cost, and context-sensitive support that complements face-to-face interventions.
8.3 Limitations and future lines of research
The longitudinal design employed in this study represents a significant methodological advance, as it enables the evaluation of temporal relationships between variables, the estimation of prospective mediations, and the control of intraindividual stability. However, this type of design does not allow for the modeling of specific aspects of affective change, such as nonlinear trajectories, cumulative effects, or feedback between constructs. In this sense, latent growth models can be a complementary tool for more accurately capturing intraindividual variability and exploring how mediating effects change over time (Curran et al., 2010; McArdle, 2009).
An additional limitation is that we did not account for potential confounding variables such as stressful life events, clinical comorbidities, or personality traits, which may influence the longitudinal trajectories. Future research could incorporate additional measures or high-resolution methodologies, such as ecological momentary assessment, to better capture these influences.
Also, while this study employed a structural equation model with lagged trajectories (CLPM), which allows for estimating prospective relationships while controlling for intra-individual stability effects, there are more sophisticated models such as the Random Intercept Cross-Lagged Panel Model (RI-CLPM), which allows for explicitly differentiating between intra- and inter-individual processes in longitudinal analysis (Hamaker et al., 2015; Orth et al., 2021). However, the reliable implementation of the RI-CLPM requires at least four measurements to ensure parameter identification and stability, as well as adequate statistical power, conditions that exceed the three-wave design used in this study. In this sense, the choice of the CLPM responds to reasonable methodological criteria that are theoretically consistent with the proposed objectives. Future research with denser time series could benefit from the use of the RI-CLPM to model more accurately the affective dynamics involved in the transition from stress to depression, with anxiety as a sequential mediator.
Although our study was not initially designed with a priori power calculation, a post-hoc Monte Carlo analysis (200,000 replications) based on the observed path estimates indicated a statistical power of 0.86 for detecting the hypothesized indirect effect, suggesting that the sample size was adequate for the mediation model.
Beyond methodological considerations, the sociocultural context of data collection constitutes an additional limitation. Cultural characteristics such as prevailing social values, interpersonal relationship patterns, and broader economic conditions may shape how stress, anxiety, and depression unfold and interact. In this study, the sample consisted exclusively of adults residing in the Metropolitan Region of Chile, which restricts the generalizability of the findings to other regions or cultural contexts. This geographic concentration reflects characteristics specific to Santiago, such as higher urbanization, socioeconomic inequality, and the impact of the COVID-19 pandemic. Chile also exhibits a low individualism score, reflecting a more collectivist orientation that may influence coping styles, emotional expression, and the interplay among stress, anxiety, and depression. These contextual features could partly explain variations in the strength or form of the observed pathways. Future research should therefore incorporate more diverse samples across regions, socioeconomic groups, and demographics, as well as cross-cultural comparisons to test whether the mediational sequence identified here generalizes to societies with different cultural orientations and socioeconomic conditions. Additionally, it should examine potential moderators, such as gender, predominant stressors, or educational level. Another relevant limitation relates to the exclusive use of self-reported measures. Although validated instruments were used and possible biases were controlled, this approach may still be susceptible to common method errors and social desirability bias (Podsakoff et al., 2003). Future research should therefore incorporate additional sources of information, with particular emphasis on structured clinical interviews, which provide a more objective diagnostic assessment. Complementary sources, such as third-party reports or physiological indicators, would further strengthen the convergent validity of the findings.
The model analyzed focused exclusively on anxiety as a mediator of the relationship between stress and depressive symptoms, based on a hypothesis derived from Beck’s cognitive model. This decision is informed by methodological and theoretical criteria that favor the use of parsimonious and clearly defined models when analyzing longitudinal data with three waves (Hayes, 2013). Nevertheless, we acknowledge that other psychological processes—such as rumination, attributional styles, and self-efficacy—constitute plausible alternative pathways in the progression from stress to depression. These mechanisms were not included in the present model due to design restrictions; however, their examination through multivariate analyses and longitudinal designs with greater temporal density represents a crucial direction for future research.
Although the present study is explicitly grounded in Beck’s cognitive model, its findings suggest the possibility of exploring complementary conceptual frameworks in future research. In particular, approaches such as rumination theory (Nolen-Hoeksema et al., 2008) and dysfunctional metacognition (Wells, 2009) offer relevant theoretical hypotheses for understanding how anticipatory anxiety might amplify or maintain dysfunctional affective trajectories. However, since these dimensions were neither operationalized nor included in the analytical model proposed here, their mention is limited to an orienting theoretical opening, without constituting a conceptual integration within the present work.
Thus, although the proposed model captures a plausible mediational route within the cognitive framework, it does not pretend to exhaust the multiple possible pathways in the affective progression from stress to depression. Its value lies in offering a parsimonious and empirically tested theoretical sequence, which can serve as a starting point for more comprehensive and comparative models in future research. Finally, although the results obtained align with Beck’s cognitive model, they also open up the possibility of examining similar affective trajectories in other clinical conditions. This projection suggests an exploratory direction towards more integrative models, in which anxiety could play an articulating role within complex emotional sequences. In particular, the hypothesis of a shared matrix of cognitive and emotional vulnerability, proposed by transdiagnostic models such as those of Ingram and Luxton (2005) or the rumination theory of Nolen-Hoeksema et al. (2008), constitutes a promising avenue for future research. However, its empirical validation would require more extensive and multivariate designs, which are beyond the scope of the present study.
Additionally, although attrition analyses suggested no systematic bias in stress, anxiety, depression, or fear of COVID-19, the fact that participants who remained were on average older indicates that the generalizability of our findings may be more limited for younger adults. While this age-related attrition does not compromise the internal validity of the mediation model, it highlights the need for caution when extrapolating results to younger populations.
9 Conclusion
The findings demonstrate that anxiety plays a mediating role in the temporal sequence connecting perceived stress with the onset of depressive symptoms. Using a three-wave longitudinal design and a structural model with lagged paths, we identified an emotional sequence in which stress precedes increases in anxiety, which, in turn, predicts higher levels of depressive symptoms. These findings update the classical cognitive model of depression by positioning anxiety as a sequential emotional component that links threat perception with the development of depressive states. This perspective offers a more dynamic understanding of affective disorders by integrating theories of emotion regulation and cognitive vulnerability. In applied terms, the results support the need for preventive strategies focused on subclinical anxiety. Early detection and treatment of these symptoms could interrupt their progression into fully developed depressive conditions. It also highlights the potential for developing clinical, community, or digital tools to monitor and manage these emotional trajectories. Overall, the results contribute to a more accurate understanding of affective trajectories by integrating cognitive and emotional components in a sequential model with potential applications in preventive interventions. In sum, the study offers both a theoretical and practical contribution by proposing a longitudinal mediational model that improves the understanding of psychopathological trajectories by clarifying how stress evolves into depression via anxiety, thus consolidating an affective sequence of high clinical and preventive value.
Author’s note
This work is part of Gustavo Marcos Escobar’s doctoral research conducted within the Ph.D. program in Management at Business School, Universidad Adolfo Ibáñez.
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Ethics statement
The studies involving humans were approved by Comité Ético de Investigación de la Universidad Adolfo Ibáñez. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.
Author contributions
GM-E: Resources, Formal analysis, Writing – review & editing, Funding acquisition, Validation, Writing – original draft, Project administration, Visualization, Methodology, Conceptualization, Data curation, Investigation, Software, Supervision. MG: Resources, Writing – original draft, Investigation, Writing – review & editing, Software, Funding acquisition, Visualization, Formal analysis, Data curation, Validation, Conceptualization, Project administration, Methodology, Supervision. WU: Resources, Data curation, Writing – review & editing, Project administration, Conceptualization, Validation, Investigation, Formal analysis, Writing – original draft, Methodology, Software, Funding acquisition, Visualization, Supervision.
Funding
The author(s) declare that financial support was received for the research and/or publication of this article. WU was awarded by a grant from the Agencia Nacional de Investigación y Desarrollo (ANID; Chilean Government) through the Fondecyt Regular 2023 (Fondo de Investigación Científica y Tecnológica), Project #1230626.
Acknowledgments
GM-E was supported by funds from the Fondecyt Regular 2023 (Project #1230626) awarded to WU.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The authors declare that Gen AI was used in the creation of this manuscript. The author(s) verify and take full responsibility for the use of generative AI in the preparation of this manuscript. Generative AI was used exclusively for editorial support, including English language editing (grammar, fluency, and academic tone), as well as verification of consistency between in-text citations and the reference list. Specifically, OpenAI's ChatGPT (model GPT-4o, June 2024 version) was employed. The AI was not used to generate content, formulate hypotheses, analyze data, interpret results, or write any part of the manuscript. All outputs generated by the AI were critically reviewed and manually verified by the authors to ensure factual accuracy, conceptual integrity, and compliance with academic standards. The initial and final prompts used during this process are provided in Supplementary materials.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2025.1668518/full#supplementary-material
References
Ahorsu, D. K., Lin, C. Y., Imani, V., Saffari, M., Griffiths, M. D., and Pakpour, A. H. (2020). The fear of COVID-19 scale: development and initial validation. Int. J. Ment. Health Addict. 20, 1537–1545. doi: 10.1007/s11469-020-00270-8
Alimoradi, Z., Broström, A., Tsang, H. W. H., Griffiths, M. D., Haghayegh, S., Ohayon, M. M., et al. (2022). Fear of COVID-19 and its association with mental health-related factors: a systematic review and meta-analysis. BJPsych Open 8:e73. doi: 10.1192/bjo.2022.26
Anyan, F., Bizumic, B., and Hjemdal, O. (2018). Specificity in mediated pathways by anxiety symptoms linking adolescent stress profiles to depressive symptoms: results of a moderated mediation approach. J. Affect. Disord. 228, 109–117. doi: 10.1016/j.jad.2017.12.011
Anyan, F., Ingvaldsen, S. H., and Hjemdal, O. (2020). Interpersonal stress, anxiety and depressive symptoms: results from a moderated mediation analysis with resilience. Stress. Health 36, 241–252. doi: 10.1002/smi.2926
Anyan, F., Worsley, L., and Hjemdal, O. (2017). Anxiety symptoms mediate the relationship between exposure to stressful negative life events and depressive symptoms: a conditional process modelling of the protective effects of resilience. Asian J. Psychiatr. 29, 41–48. doi: 10.1016/j.ajp.2017.04.019
Ashkanasy, N. M. (2016). What matters for excellence in theory and theorizing. Organ. Psychol. Rev. 6, 219–224. doi: 10.1177/2041386616648171
Baader, T., Molina, J. L., Venezian, S., Rojas, C., Farías, R., Fierro-Freixenet, C., et al. (2012). Validación y utilidad de la encuesta PHQ-9 (Patient Health Questionnaire) en el diagnóstico de depresión en pacientes usuarios de atención primaria en Chile. Rev. Chil. Neuropsiquiatr. 50, 10–22. doi: 10.4067/S0717-92272012000100002
Baron, R. M., and Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J. Pers. Soc. Psychol. 51, 1173–1182. doi: 10.1037/0022-3514.51.6.1173
Beck, A. T. (1963). Thinking and depression: I. Idiosyncratic content and cognitive distortions. Arch. Gen. Psychiatry 9, 324–333. doi: 10.1001/archpsyc.1963.01720160014002
Beck, A. T. (1967). Depression: clinical, experimental, and theoretical aspects. New York, NY: Harper & Row.
Beck, A. T., and Alford, B. A. (2009). Depression: causes and treatment. 2nd Edn. Philadelphia, PA: University of Pennsylvania Press.
Beck, A. T., and Clark, D. A. (1997). An information processing model of anxiety: automatic and strategic processes. Behav. Res. Ther. 35, 49–58. doi: 10.1016/S0005-7967(96)00069-1
Beck, A. T., and Emery, G. (2005). Anxiety disorders and phobias: A cognitive perspective. 20th anniversary Edn. New York, NY: Basic Books.
Beck, A. T., Emery, G., and Greenberg, R. L. (1985a). Anxiety disorders and phobias: A cognitive perspective. New York, NY: Basic Books.
Beck, A. T., Rush, A. J., Shaw, B. F., and Emery, G. (1985b). Cognitive therapy of depression. New York, NY: Guilford Press.
Beck, A., Rush, A. J., Shaw, B. F., Emery, G., De Rubeis, R. J., and Hollon, S. D. (2024). Cognitive therapy of depression. 2nd Edn. New York, NY: The Guilford Press.
Borghero, F., Martínez, V., Zitko, P., Vöhringer, P. A., Cavada, G., and Rojas, G. (2018). Tamizaje de episodio depresivo en adolescentes: Validación del instrumento PHQ-9. Rev. Med. Chile 146, 479–486. doi: 10.4067/s0034-98872018000400479
Brislin, R. W. (1970). Back-translation for cross-cultural research. J. Cross-Cult. Psychol. 1, 185–216. doi: 10.1177/135910457000100301
Brown, T. A., Chorpita, B. F., and Barlow, D. H. (1998). Structural relationships among dimensions of the DSM-IV anxiety and mood disorders and dimensions of negative affect, positive affect, and autonomic arousal. J. Abnorm. Psychol. 107, 179–192. doi: 10.1037/0021-843X.107.2.179
Burkholder, G. J., and Harlow, L. L. (2003). An illustration of a longitudinal cross-lagged design for larger structural equation models. Struct. Equ. Model. 10, 465–486. doi: 10.1207/S15328007SEM1003_8
Cheung, G. W., and Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Struct. Equ. Modeling 9, 233–255. doi: 10.1207/S15328007SEM0902_5
Clark, D. A., and Beck, A. T. (2011). Cognitive therapy of anxiety disorders: Science and practice : Guilford Press.
Clark, L. A., and Watson, D. (1991). Tripartite model of anxiety and depression: psychometric evidence and taxonomic implications. J. Abnorm. Psychol. 100, 316–336. doi: 10.1037/0021-843X.100.3.316
Cohen, S., Kamarck, T., and Mermelstein, R. (1983). A global measure of perceived stress. J. Health Soc. Behav. 24, 385–396. doi: 10.2307/2136404
Cole, D. A., Martin, N. C., and Powers, B. (2005). A competency-based model of child depressive symptoms in the context of peer and academic functioning: a longitudinal study of cognitive vulnerability-stress interaction. J. Abnorm. Psychol. 114, 1–15. doi: 10.1037/0021-843X.114.1.1
Cole, D. A., and Maxwell, S. E. (2003). Testing mediational models with longitudinal data: questions and tips in the use of structural equation modeling. J. Abnorm. Psychol. 112, 558–577. doi: 10.1037/0021-843X.112.4.558
Curran, P. J., Obeidat, K., and Losardo, D. (2010). Twelve frequently asked questions about growth curve modeling. J. Cogn. Dev. 11, 121–136. doi: 10.1080/15248371003699969
Dickson, K. S., Ciesla, J. A., and Reilly, L. C. (2012). Rumination, worry, cognitive avoidance, and behavioral avoidance: examination of temporal effects. Behav. Ther. 43, 629–640. doi: 10.1016/j.beth.2011.11.002
Fumagalli, L., Laurie, H., and Lynn, P. (2013). Experiments with methods to reduce attrition in longitudinal surveys. Journal of the Royal Statistical Society: Series A (Statistics in Society), 176, 499–519.
Gaete, J., Martínez, V., Fritsch, R., Rojas-Barahona, C. A., Montgomery, A., Araya, R., et al. (2022). Validation of the Spanish version of the GAD-7 in Chilean adolescents. Rev. Med. Chile 150, 1252–1260. doi: 10.4067/S0034-98872022000901252
Garber, J., Clarke, G. N., Weersing, V. R., Beardslee, W. R., Brent, D. A., Gladstone, T. R., et al. (2009). Prevention of depression in at-risk adolescents: a randomized controlled trial. JAMA 301, 2215–2224. doi: 10.1001/jama.2009.788
García-Campayo, J., Zamorano, E., Ruiz, M. A., Pérez-Páramo, M., López-Gómez, V., Rejas, J., et al. (2010). Cultural adaptation into Spanish of the generalized anxiety disorder-7 (GAD-7) scale as a screening tool. Health Qual. Life Outcomes 8, 1–11. doi: 10.1186/1477-7525-8-8
Gelman, A. (2019). Don’t calculate post-hoc power using observed estimate of effect size. Annals of Surgery, 269, e9–e10.
Hamaker, E. L., Kuiper, R. M., and Grasman, R. P. P. P. P. (2015). A critique of the cross-lagged panel model. Psychol. Methods 20, 102–116. doi: 10.1037/a0038889
Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York, NY: The Guilford Press.
Higa-McMillan, C. K., Francis, S. E., and Chorpita, B. F. (2014). “Anxiety disorders” in Child psychopathology. eds. E. J. Mash and R. A. Barkley. 3rd ed (New York, NY: The Guilford Press), 345–395.
Hong, R. Y. (2007). Worry and rumination: differential associations with anxious and depressive symptoms and coping behavior. Behav. Res. Ther. 45, 277–290. doi: 10.1016/j.brat.2006.03.006
Hu, L., and Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct. Equ. Model. 6, 1–55. doi: 10.1080/10705519909540118
Ingram, R. E., and Luxton, D. D. (2005). “Vulnerability-stress models” in Development of psychopathology: A vulnerability-stress perspective. eds. B. L. Hankin and J. R. Z. Abela (Thousand Oaks, CA: Sage Publications), 32–46.
Jacobson, N. C., and Newman, M. G. (2017). Anxiety and depression as bidirectional risk factors for one another: a meta-analysis of longitudinal studies. Psychol. Bull. 143, 1155–1200. doi: 10.1037/bul0000111
Jöreskog, K. G. (1979). “Statistical models and methods for analysis of longitudinal data” in Advances in factor analysis and structural equation models. eds. K. G. Jöreskog and D. Sörbom (Cambridge, MA: Abt Books), 129–169.
Jorquera-Gutiérrez, R. (2023). Análisis psicométrico de la Escala de Estrés Percibido (PSS-14 y PSS-10) en un grupo de docentes de Copiapó, Chile. Liberabit. Rev. Peru. Psicol. 29:e683. doi: 10.24265/liberabit.2023.v29n1.683
Kendler, K. S., Gardner, C. O., Gatz, M., and Pedersen, N. L. (2007). The sources of co-morbidity between major depression and generalized anxiety disorder in a Swedish national twin sample. Psychol. Med. 37, 453–462. doi: 10.1017/S0033291706009135
Kline, R. B. (2016). Principles and practice of structural equation modeling. 4th Edn. New York, NY: Guilford Press.
Kok, L., Sep, M. S., Veldhuijzen, D. S., Cornelisse, S., Nierich, A. P., van der Maaten, J., et al. (2016). Trait anxiety mediates the effect of stress exposure on PTSD and depression risk in cardiac surgery patients. J. Affect. Disord. 206, 216–223. doi: 10.1016/j.jad.2016.07.020
Kroenke, K., Spitzer, R. L., and Williams, J. B. W. (2001). The PHQ-9: validity of a brief depression severity measure. J. Gen. Intern. Med. 16, 606–613. doi: 10.1046/j.1525-1497.2001.016009606.x
Lagos-Riveros, G., Domínguez-Cancino, K., Luengo-Martínez, C. E., Farías-Fritz, B. L., and Guzmán-Garcés, F. (2023). Adaptación transcultural al español chileno de la Perceived Stress Scale-10 (PSS-10) en profesores: Estudio Delphi. Med. Segur. Trab. 69, 149–159. doi: 10.4321/s0465-546X2023000300003
Little, R. J. A. (1988). A test of missing completely at random for multivariate data with missing values. J. Am. Stat. Assoc. 83, 1198–1202. doi: 10.2307/2290157
MacKinnon, D. P., Fairchild, A. J., and Fritz, M. S. (2007). Mediation analysis. Annu. Rev. Psychol. 58, 593–614. doi: 10.1146/annurev.psych.58.110405.085542
MacKinnon, D. P., Lockwood, C. M., and Williams, J. (2004). Confidence limits for the indirect effect: Distribution of the product and resampling methods. Multivariate Behavioral Research, 39, 99–128. doi: 10.1207/s15327906mbr3901_4
Maxwell, S. E., and Cole, D. A. (2007). Bias in cross-sectional analyses of longitudinal mediation. Psychol. Methods 12, 23–44. doi: 10.1037/1082-989X.12.1.23
Maxwell, S. E., Cole, D. A., and Mitchell, M. A. (2011). Bias in cross-sectional analyses of longitudinal mediation: partial and complete mediation under an autoregressive model. Multivar. Behav. Res. 46, 816–841. doi: 10.1080/00273171.2011.606716
McArdle, J. J. (2009). Latent variable modeling of differences and changes with longitudinal data. Annu. Rev. Psychol. 60, 577–605. doi: 10.1146/annurev.psych.60.s110707.163612
Michl, L. C., McLaughlin, K. A., Shepherd, K., and Nolen-Hoeksema, S. (2013). Rumination as a mechanism linking stressful life events to symptoms of depression and anxiety. J. Abnorm. Psychol. 122, 339–352. doi: 10.1037/a0031994
Mitchell, M. A., and Maxwell, S. E. (2013). A comparison of the cross-sectional and sequential designs when assessing longitudinal mediation. Multivar. Behav. Res. 48, 301–339. doi: 10.1080/00273171.2013.784696
Moshagen, M., and Erdfelder, E. (2016). A new strategy for testing structural equation models. Struct. Equ. Model. Multidiscip. J. 23, 54–60. doi: 10.1080/10705511.2014.950896
Muthén, L. K., and Muthén, B. O. (2017). Mplus User's Guide. 8th Edn. Los Angeles, CA: Muthén & Muthén.
Newman, D. A. (2014). Missing data: five practical guidelines. Organ. Res. Methods 17, 372–411. doi: 10.1177/1094428114548590
Nolen-Hoeksema, S. (2000). The role of rumination in depressive disorders and mixed anxiety/depressive symptoms. Journal of Abnormal Psychology, 109:504.
Nolen-Hoeksema, S., Wisco, B. E., and Lyubomirsky, S. (2008). Rethinking rumination. Perspect. Psychol. Sci. 3, 400–424. doi: 10.1111/j.1745-6924.2008.00088.x
O'Laughlin, K. D., Martin, M. J., and Ferrer, E. (2018). Cross-sectional analysis of longitudinal mediation processes. Multivar. Behav. Res. 53, 375–402. doi: 10.1080/00273171.2018.1454822
Orth, U., Erol, R. Y., and Luciano, E. C. (2021). Development of self-esteem from age 4 to 94 years: a meta-analysis of longitudinal studies. Psychol. Bull. 147, 1045–1080. doi: 10.1037/bul0000338
Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., and Podsakoff, N. P. (2003). Common method biases in behavioral research: a critical review of the literature and recommended remedies. J. Appl. Psychol. 88, 879–903. doi: 10.1037/0021-9010.88.5.879
Porter, S. R., Whitcomb, M. E., and Weitzer, W. H. (2019). Multiple surveys of students and survey fatigue. New Dir. Inst. Res. 2004, 63–73. doi: 10.1002/ir.101
Saldivia, S., Vicente, B., Pihán, R., Valdivia, M., and Torres, S. (2019). Validación de la versión chilena del Patient Health Questionnaire-9 (PHQ-9) en atención primaria. Rev. Med. Chile 147, 53–60. doi: 10.4067/S0034-98872019000100053
Selig, J. P., and Preacher, K. J. (2009). Mediation models for longitudinal data in developmental research. Res. Hum. Dev. 6, 144–164. doi: 10.1080/15427600902911247
Spitzer, R. L., Kroenke, K., Williams, J. B. W., and Löwe, B. (2006). A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch. Intern. Med. 166, 1092–1097. doi: 10.1001/archinte.166.10.1092
Tapia, D., Cruz, C., Gallardo, I., and Dasso, M. (2007). Adaptación de la Escala de Percepción Global de Estrés (EPGE) en estudiantes adultos de escasos recursos en Santiago, Chile. Psiquiatr. Salud Ment. 24, 109–119.
Vrshek-Schallhorn, S., Stroud, C. B., Mineka, S., Hammen, C., Zinbarg, R. E., Wolitzky-Taylor, K., et al. (2015). Chronic and episodic interpersonal stress as statistically unique predictors of depression in two samples of emerging adults. J. Abnorm. Psychol. 124, 918–932. doi: 10.1037/abn0000088
Wells, A. (2009). Metacognitive therapy for anxiety and depression. New York, NY: The Guilford Press.
Weston, R., and Gore, P. A. (2006). A brief guide to structural equation modeling. Counsel. Psychol. 34, 719–751. doi: 10.1177/0011000006286345
Wolf, E. J., Harrington, K. M., Clark, S. L., and Miller, M. W. (2013). Sample size requirements for structural equation models: an evaluation of power, bias, and solution propriety. Educ. Psychol. Meas. 73, 913–934. doi: 10.1177/0013164413495237
Keywords: perceived stress, anxiety, depression, longitudinal mediation, cognitive model
Citation: Marcos-Escobar G, Gómez ME and Unanue W (2025) Understanding the transition from stress to depression: a longitudinal mediational analysis of anxiety in adults from the Metropolitan Region of Chile. Front. Psychol. 16:1668518. doi: 10.3389/fpsyg.2025.1668518
Edited by:
Michiko Sakaki, University of Tübingen, GermanyReviewed by:
Ana Teresa Martins, University of Algarve, PortugalZayra Antúnez, Austral University of Chile, Chile
Copyright © 2025 Marcos-Escobar, Gómez and Unanue. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Gustavo Marcos-Escobar, Z21hcmNvc0BhbHVtbm9zLnVhaS5jbA==; Marcos E. Gómez, bWFyY29zLmdvbWV6QHVhaS5jbA==; Wenceslao Unanue, d2VuY2VzbGFvLnVuYW51ZUB1YWkuY2w=
†ORCID: Gustavo Marcos Escobar, orcid.org/0009-0005-6755-9591
Marcos E. Gómez, orcid.org/0000-0002-0795-9244
Wenceslao Unanue, orcid.org/0000-0002-2610-8082