Multidimensional assessment of anxiety through the State-Trait Inventory for Cognitive and Somatic Anxiety (STICSA): From dimensionality to response prediction across emotional contexts

Filipa Barros; Cláudia Figueiredo; Susana Brás; João M. Carvalho; Sandra C. Soares

doi:10.1371/journal.pone.0262960

Abstract

The assessment of mal-adaptive anxiety is crucial, considering the associated personal, economic, and societal burden. The State-Trait Inventory for Cognitive and Somatic Anxiety (STICSA) is a self-report instrument developed to provide multidimensional anxiety assessment in four dimensions: trait-cognitive, trait-somatic, state-cognitive and state-somatic. This research aimed to extend STICSA’s psychometric studies through the assessment of its dimensionality, reliability, measurement invariance and nomological validity in the Portuguese population. Additionally, the predictive validity of STICSA-Trait was also evaluated, through the analysis of the relationship between self-reported trait anxiety and both the subjective and the psychophysiological response across distinct emotional situations. Similarly to previous studies, results supported both a four-factor and two separated bi-factor structures. Measurement invariance across sex groups was also supported, and good nomological validity was observed. Moreover, STICSA trait-cognitive dimension was associated with differences in self-reported arousal between groups of high/low anxiety, whereas STICSA trait-somatic dimension was related to differences in both the subjective and psychophysiological response. Together, these results support STICSA as a useful instrument for a broader anxiety assessment, crucial for an informed diagnosis and practice.

Citation: Barros F, Figueiredo C, Brás S, Carvalho JM, Soares SC (2022) Multidimensional assessment of anxiety through the State-Trait Inventory for Cognitive and Somatic Anxiety (STICSA): From dimensionality to response prediction across emotional contexts. PLoS ONE 17(1): e0262960. https://doi.org/10.1371/journal.pone.0262960

Editor: Claudio Imperatori, European University of Rome, ITALY

Received: September 2, 2021; Accepted: January 7, 2022; Published: January 25, 2022

Copyright: © 2022 Barros et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the paper and its Supporting Information files.

Funding: This research was supported by National Funds through the Portuguese Foundation for Science and Technology (FCT – Fundação para a Ciência e a Tecnologia), within the William James Center for Research (UIDB/04810/2020); the CINTESIS R&D Unit (UID/IC/4255/2020); the IEETA/UA R&D unit (UIDB/00127/2020); the TEMA R&D Unit (TEMA/DEM/UA); a Doctoral Grant with the ref. SFRH/BD/118244/2016, granted to Filipa Barros; and a Doctoral grant with the ref. SFRH/BD/136815/2018, granted to João M. Carvalho. Also, this work was funded by national funds, European Regional Development Fund, FSE through COMPETE2020, through FCT, in the scope of the framework contract foreseen in the numbers 4, 5 and 6 of the article 23, of the Decree-Law 57/2016, of August 29, changed by Law 57/2017, of July 19. It was also developed under the support of the Research Program “CeNTER - Community-led Territorial Innovation” (CEN-TRO-01-0145-FEDER-000002), funded by Programa Operacional Regional do Centro (CENTRO 2020), through the ERDF and PT2020. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Anxiety is an emotional response following the anticipation of a threatening stimulus, either real or perceived [1]. It encompasses a complex and multidimensional phenomenon associated with muscular tension, subjective distress, enhanced vigilance concerning future threat and avoidance behaviors, which prepare the body for action and adaptive behavior [1, 2]. Nevertheless, anxiety is often associated with great distress and poor health outcomes when persistent and/or significantly intense [1, 3]. In fact, anxiety disorders are one of the most observed groups of psychopathologies, with an estimated prevalence of 7.3% globally (4.8–10.9%; see [4]), and are often associated with other mental disorders [5], and health problems, such as cardiovascular diseases [6]. Furthermore, anxiety symptomatology is characterized by great inter and intra-individual heterogeneity, along with variable symptom trajectories over time [7]. Altogether, these factors complexify the assessment, diagnosis and treatment of non-adaptive anxiety, catalyzing the associated societal, personal and health care costs and burden [3, 8].

Self-report instruments have been extensively used as a simple, brief and non-invasive method of assessing anxiety across contexts [9]. Although many of them present satisfactory psychometric properties [10, 11], most present several limitations regarding, for instance, the ability to discriminate anxiety from depression [12], and the ability to embrace multiple anxiety dimensions. To surpass some of these difficulties and to provide a multidimensional assessment of anxiety, Ree and colleagues [13] developed the State-Trait Inventory for Cognitive and Somatic Anxiety (STICSA). This instrument encompasses 42 items divided in two forms of 21-items each: the state anxiety (STICSA-State) and the trait anxiety (STICSA-Trait) form. State anxiety is conceptualized as the anxiety response experienced at the moment, i.e., in a limited period; in turn, trait anxiety corresponds to individual differences regarding anxiety as an emotional response relatively stable in time [11, 14]. STICSA also addresses the cognitive and somatic components of anxiety within each one of its forms; the cognitive dimension encompasses symptoms such as worry, intrusive thoughts and rumination (e.g., “Feel agonized over problems”), while the somatic component refers to psychophysiological activation, including palpitations or excessive sweating (e.g., “Heart beats fast”). Discerning between these four dimensions can be critical to characterize anxiety profiles and to predict people’s response in certain contexts. In fact, people can present distinct cognitive and somatic symptomatology [15].

STICSA has been supported as an adequate measure of anxiety across healthy and clinical samples, demonstrating good nomological validity [13, 16–18]. Furthermore, most psychometric studies of STICSA support the construct validity of the two-factor structure within STICSA-Trait and STICSA-State forms [13, 17–21], as well as the four-dimension structure [16, 17, 22]. Yet, past studies still evidence mixed findings and have left open questions regarding STICSA’s dimensionality (see [20]). Therefore, it is crucial to extend STICSA’s psychometric studies to other settings and cultures, as well as to confirm its adequacy and utility across contexts. Also, although STICSA provides important information about self-reported anxiety symptomatology, the relationship between the reported symptoms and other dimensions of the emotional response of the individuals in certain contexts remains poorly understood. The accordance between the multiple components of the emotional response (e.g., subjective and psychophysiological responses) is often low [23], with each component providing unique and complementary information [24]. Nevertheless, in the absence of more objective measures and considering the frequent time and financial constraints observed across contexts, self-report is often one of the few and more reliable tools to assess anxiety. Therefore, studying how scores in the four dimensions of STICSA are able to predict individuals’ emotional response beyond the subjective domain is crucial to obtain more accurate information about people’s anxiety profile, health and well-being.

Considering this mindset, we propose two main aims for our research. First, we sought to extend the validation studies of STICSA, while adapting this instrument for the Portuguese population. Second, we aim to assess the relationship between STICSA-Trait scores and the individuals’ subjective and psychophysiological responses to emotional stimulation. Since trait anxiety reflects relatively stable differences between people in the proneness to experience symptoms of anxiety [14], it may be especially useful to predict an individual’s emotional response across contexts. With these aims, we intend not only to support the construct validity of STICSA, but also to maximize and integrate information about different dimensions of anxiety across emotional contexts. Moreover, this research seeks to provide a better comprehension of how distinct dimensions of self-reported trait anxiety predicts an individual’s emotional response in different emotional contexts, including how these scores relate to a biomarker of autonomic dysfunction and psychopathology, namely Heart Rate Variability (HRV). This can be very critical knowledge, not only considering research settings but also clinical practice, given that self-report is often the most accessible and cost-effective tool to support assessment, diagnosis and monitoring of anxiety.

Study 1: Psychometric study of STICSA

Previous psychometric studies of STICSA assessed and supported factorial validity of the state-trait and cognitive-somatic distinction by computing full scale confirmatory factorial models and observing, for instance, a four-factor correlated model [16, 17, 22], or a hierarchical model with a global anxiety factor supported in the four factors (state-cognitive, state-somatic, trait-cognitive and trait-somatic anxiety; [17]). Dimensionality was also established by estimating separate models, namely a two-factor model for state [13, 17–20] and trait forms [13, 17–21]. Furthermore, studies of measurement invariance across gender and age groups [18, 19], as well as a multi-method approach with an informant within friendship dyads of trait anxiety [21], gave further support for the quality of the STICSA measurement. Previous studies have also revealed good convergent and nomological validity [13, 16–19, 22]. Nevertheless, STICSA evidenced medium to high correlations with depression measures (r = .42 - .67), especially when considering the correlation with cognitive anxiety (r > .48; [16, 18, 22]), despite of being better at discriminating anxiety from depression in comparison with, for instance, the State-Trait Anxiety Inventory (STAI; [11]), an instrument broadly used to measure anxiety in its state and trait dimensions (e.g., [16]). STICSA has been proven to be, therefore, a suitable instrument to measure anxiety in general and clinical populations, but it is necessary to extend its psychometric studies to better characterize the instrument’s factor structure, validity, and its adequacy across cultures and settings. In order to assess STICSA’s psychometric properties in the Portuguese population, its dimensionality, reliability, measurement invariance and nomological validity were assessed in Study 1.

Materials and methods

Participants.

The inclusion criteria included: 1) Age ≥18 years old; 2) Portuguese nationality; 3) Currently residing in Portugal. A final sample of 1153 Portuguese adults (753 females; 65.3%) from different regions of Portugal was collected. Age ranged from 18 to 78 years old (M = 29.47; SD = 13.70). Most of the participants were higher education students (n = 658; 57.2%) or active employed individuals (n = 338; 29.4%). Most of the participants were single (n = 825; 71.7%) and did not have children (n = 872; 75.7%). Furthermore, 13.4% of the participants (n = 154) reported having a psychological/psychiatric problem, and 6.2% (n = 71) reported being monitored/receiving treatment regarding those problems. For further information about sample’s characteristics please refer to Table 1. All participants were informed about the voluntary nature of their participation and the anonymity of collected data. The study was approved by the Ethics and Deontology Committee of the University of Aveiro (ref. 08/2018) and was performed according to the guidelines of the Declaration of Helsinki and the American Psychological Association.

Download:

Table 1. Sample’s demographic information.

https://doi.org/10.1371/journal.pone.0262960.t001

Instruments.

In a first step, the 42 items of STICSA were translated to the Portuguese language by the research team, followed by a retroversion conducted by an English native speaker. The two versions were then discussed to obtain a preliminary version of the scale. This version was tested in a group of individuals to assess clarity and to identify possible misinterpretations associated with the items wording and instructions, as well as to evaluate the average response time of the overall protocol. After collecting their feedback, the research team made minor adjustments, finalizing the process of translation and adaptation of the 42 items (21 items per subscale).

Considering the procedure followed by one of the original psychometric studies of STICSA [16], STAI [11], the Depression, Anxiety and Stress Scales–Depression scale (DASS-D; [25]) and the Positive and Negative Affect Schedule (PANAS; [26]) were selected to evaluate nomological validity. These instruments have been widely used to measure anxiety, depressive symptomatology, as well as positive and negative affect, respectively, aspects that were on the basis of STICSA’s development. In fact, STICSA was developed considering a state/trait distinction similar to STAI, and it sought to surpass one of STAI’s greatest limitations, related to its high associations with depression and low positive affect [27, 28]. The protocol was then firstly constituted by a sociodemographic questionnaire, with relevant information for the sample’s characterization, followed by STICSA (42 items), STAI (40 items), PANAS (20 items) and DASS-D (7 items).

State-Trait Inventory for Cognitive and Somatic Anxiety (STICSA). The STICSA [13] is a self-report instrument that measures anxiety considering four dimensions: state, trait, cognitive and somatic anxiety. It is divided in two forms, each one with the same 21 items: one evaluating anxiety in its state dimension (“how do you feel right now”; STICSA-State) and the other evaluating its trait dimension (“how often this is true for you”; STICSA-Trait). Both forms evaluate cognitive and somatic components of anxiety, with items being rated in a 4-point scale that ranges from 1 (Not at all) to 4 (Very much; see [16]).

State-Trait Anxiety Inventory (STAI—Form Y). The STAI—Form Y [11] is a psychological assessment instrument that measures anxiety. This scale encompasses two self-report scales of 20 items each: the state anxiety scale (STAI-Y1), which assesses how the individual feels at the moment; and the trait anxiety scale (STAI-Y2), assessing how the individual generally feels. Participants are asked to respond to each item using a four-point scale that ranges from 1 (Nothing at all) to 4 (Very much so) in the state scale, and from 1 (Almost never) to 4 (Almost always) in the trait form. This instrument was adapted and validated for the Portuguese population by Santos and Silva [29] and further psychometric studies supported its adequacy as a measure of anxiety [30].

Positive and Negative Affect Schedule (PANAS). The PANAS [26] is a self-report instrument that measures positive and negative affect. This scale encompasses 10 items evaluating positive affect and 10 items evaluating negative affect. Each item corresponds to an emotion and the individual is asked to indicate how much he/she felt that emotion in the last weeks, using a scale ranging from 1 (Very slightly or not at all) to 5 (Extremely). PANAS was adapted for the Portuguese population by Galinha and Pais-Ribeiro [31] confirming the original factor structure of the instrument, and its adequate psychometric properties.

Depression, Anxiety and Stress Scales–Depression scale (DASS-D). The DASS-21 is a self-report instrument comprising three scales evaluating anxiety, depression and stress, with 21 items distributed equally in each dimension [25]. Response scale encompasses a four-point scale that ranges from 0 (Did not apply to me at all) to 3 (Applied to me very much or most of the time), considering the frequency or severity of the negative emotional symptoms experienced in the last week [25]. This instrument was adapted for the Portuguese population by Pais-Ribeiro and colleagues [32], revealing adequate psychometric properties for the three subscales in an adult sample. In the current study, only the depression subscale was used.

Procedure.

The participants’ recruitment and data collection followed a mixed-mode survey procedure [33], namely: 1) data collection in paper, mostly carried out in classroom context (targeting higher education students) and with older people of the general population; and 2) data collection through a digital platform (targeting the general population). In both modes, participants were informed about the study mainly through e-mail, class visits, flyer distribution, and social networks. All participants received detailed information about the study and gave their informed consent before their participation (by consenting and proceeding with the protocol in the digital platform, or by providing written informed consent when data collection was performed in paper). There was no compensation for their participation in the study. The protocol took, on average, between 15 and 20 minutes to complete.

Analytic procedures.

The Mplus 8 version [34] was used to conduct Confirmatory Factorial Analysis (CFA), both in one group and multi-group (measurement invariance). The IBM SPSS Statistics (version 23) was employed for descriptive and reliability analyses. Additionally, R package’s Amelia II [35] was used to handle missing data. The missing data analysis was performed considering a sample higher than 1000, through a multiple imputation procedure. The used algorithm first created a bootstrapped version of the original data, through which estimated the sufficient statistics by Expected Maximization (EM) on a bootstrapped sample, and then imputed the missing values of the original data using the estimated sufficient statistics. This procedure was carried out by producing five datasets that were aggregated afterwards, assuming data as ordinal; in fact, three of the four measurement instruments with imputed missing data had 4-point answer scales, and the remaining instrument had a 5-point scale that ought to be considered ordinal as well. The imputation procedures were performed separately for each instrument. All participants presenting more than 10% of missing values in an instrument were excluded from the final sample [36]. The maximum imputed by variable was 1.2%, and the overall imputation encompassed less than 0.2% of total responses in the dataset.

The CFA were computed using a categorical robust Weighted Least Square Estimator (WLSMV) developed by Muthén and Muthén [34]. This estimator has been suggested as the one that best performs when testing categorical data [37, 38]. The assessment of model fit adjustment was based on robust chi-square statistic, as well as in the following fit indexes: Comparative Fit Index (CFI), Tucker and Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Residual (SRMR). Given the sample size of the present study, the interpretation of chi-square statistic was less helpful, since a non-significant result would not be expected [39]. Therefore, the model assessment assumed the following cut-off points: 1) CFI and TLI values higher than .95; and 2) RMSEA and SRMR values equal or less than .07 [38, 40].

A relevant dimension in psychometric validation studies is the equivalence of the measurement scale across different groups, i.e., to provide evidence that the same items are composing the same constructs in different groups of individuals, a condition that supports the accuracy of the group statistics [38, 41]. In this line, several studies suggest that anxiety disorders are more prevalent in women than in men (e.g., [42]). Therefore, sex invariance was tested in our sample, considering both state and trait models separately. Measurement invariance was assessed by testing several constrained models. First, it was assessed for each group separately, following three increasingly constrained models to compare the two groups regarding configural, metric, and scalar invariance [38, 41]. Model fit decision was assessed considering the parameters previously mentioned, and the comparison between the constrained models was done considering the difference in robust chi-square (chi-square difftest), being this option the most recommended when using the WLSMV estimator [43].

Reliability analysis was supported in McDonald’s Omega, which accounts for the effective results from the CFA analysis conducted [44, 45]. Cronbach alpha was also computed, as it is the most common internal consistency coefficient with Likert-type response scales, allowing the comparison between different studies [46]. Lastly, convergent and concurrent validity were observed through Pearson correlation [47]. The interpretation of the correlation was supported considering Cohen´s suggestion: small effect [.10- .30[, medium effect [.30 - .50[, and large effect: [.50–1] [48].

Results

Dimensionality analysis.

According to the procedures used to develop STICSA, as well as to the constructs it is meant to reflect, the dimensionality analysis can approach the full model or divide the analysis by trait and state forms. The full models are expected to express an extensive degree of multicollinearity since the items of state and trait have the same wording. Nevertheless, the dimensionality analysis assumed and explored both options. The CFA of the full model of STICSA scale (42 items) was assessed through five concurrent models:

(MF1) a unidimensional model with all items loading in one factor;
(MF2) a four correlated factor model with the state-cognitive, state-somatic, trait-cognitive and trait-somatic dimensions;
(MF3) a four-factor model including state-cognitive, state-somatic, trait-cognitive and trait-somatic factors, as well as a second-order anxiety factor;
(MF4) a four-factor model including state-cognitive, state-somatic, trait-cognitive and trait-somatic factors, as well as second-order state and trait factors;
(MF5) a four-factor model including state-cognitive, state-somatic, trait-cognitive and trait-somatic factors, as well as a second-order state and trait factors and a third-order anxiety factor.

The MF4 and MF5 models were tested in the present study since they reflect the most complex models underlying the instrument’s structure, considering the somatic and cognitive dimensions within higher-order trait and state dimensions (MF4), as well as a third higher-order anxiety factor (MF5). Considering that all items of state and trait forms are exactly the same, with differences only in the scales’ instructions [16], the error terms were allowed to correlate between the mirror items.

All five models showed acceptable to good fit indexes. The decision about model adequacy was based on the results of the selected fit indexes (CFI, TLI, RMSEA and SRMR), since the results for the chi-square test were statistically significant, as expected due to the sample size [39]. Results showed that the four-factor model with second-order state and trait factors (MF4), as well as the three-order factor model (MF5) presented good adjustment. The model with four-correlated factors (MF2) was the one with the best overall adjustment within the full models (Table 2).

Download:

Table 2. Goodness of fit statistics for all concurrent models (n = 1153).

https://doi.org/10.1371/journal.pone.0262960.t002

The three models with good fit indexes were theoretically sound, with these results supporting the factorial structure proposed in the development of STICSA. However, these models presented very high correlations between the symmetric latent variables in state and trait forms. For instance, the MF2 model showed a correlation of .817 between the two dimensions of somatic anxiety and .916 between the two dimensions of cognitive anxiety. In general, the trait and state dimensions were highly associated in all models, suggesting that the respondents did not completely separate these two forms of anxiety.

Considering these findings, as well as the complexity of the models when all the mirror items were allowed to have their error terms correlated, separated models for STICSA-State and STICSA-Trait were tested. This option enables further assessment of the hypothesized models and their validity to be conducted in simpler models, respecting the assumption of local independence [49], as well as assuring the parsimony of the models by, for instance, not increasing the parameters to be estimated [39]. Different concurrent models were compared for each form. Three models were computed both for state and trait forms: (MS1 and MT1) a unidimensional model with all items loading in one factor; (MS2 and MT2) a two-factor model including somatic and cognitive factors within state or trait anxiety; and (MS3 and MT3) a two-factor model for somatic and cognitive dimensions explaining a second-order factor for state or trait anxiety. Once more, the MS3 and MT3 models were not tested in previous studies but were tested in the present study since they reflect a more complex model of the instrument’s structure. Results of the CFA performed for both trait and state models showed that the best overall adjustment was observed for the models with two factors (somatic and cognitive anxiety; MS2 and MT2). The results from the selected fit indexes for the unidimensional and the second-order factor model were below .95 for CFI and TLI and above .07 for SRMR and RMESEA, being these the suggested cut-off points for the models to be considered acceptable (Table 2).

The factor loadings for all items in both state and trait models were significant (p < .001). Standardized loadings for the state anxiety model ranged from .486 to .787 regarding the somatic factor and from .404 to .821 in the cognitive factor. Accordingly, on average, a square multiple correlation of .443 was observed for the somatic factor (.236≤SMC≤.620), and of .516 for the cognitive factor (.163≤SMC≤.675). Standardized loadings for the trait models ranged between .584 till .832 for somatic factor and between .447 till .876 for the cognitive factor. Square multiple correlation for somatic dimension of trait anxiety corresponded to, on average, .506 (.341≤SMC≤.692), and .607 for cognitive dimension (.200≤SMC≤.767). These results suggested that the factors were explaining, on average, at least half of the variance of the indicators (Table 3). The correlation between somatic and cognitive latent factors was of large magnitude (greater than .50; see [48]), specifically .674 for state model and .705 for trait model.

Download:

Table 3. Standardized (Unstardardized) factor loadings: State and trait models of STICSA (MS2 and MT2).

https://doi.org/10.1371/journal.pone.0262960.t003

Reliability analysis.

The reliability of the factors was assessed by computing Omega for categorical data, with results supporting a very good reliability level. In the state dimension, the Omega was .895 for the somatic factor and .912 for the cognitive factor. In the trait dimension, the index presented values of .918 for the somatic dimension and .945 for the cognitive dimension. Cronbach alpha was also computed, allowing to compare our results with other studies. Results also supported a very good level of internal consistency. The state dimensions showed alphas of .803 for the somatic factor, and .869 for the cognitive factor, with item-total correlations of at least .35 for all the items of both factors. The trait dimensions presented an alpha of .869 for the somatic factor and .857 for the cognitive factor, with strong item-total correlations of at least .45.

Measurement invariance.

Sex invariance was tested considering both state and trait models separately, aligned with the chosen strategy to further explore the psychometric models. In a first step, the models for male and female participants were tested separately, evidencing very good fit indexes. Although chi-square tests were significant, it was possible to assume the structure as equivalent in the two groups. The nested models were then tested sequentially adding restrictive constrains: (1) configurational, to assure that the factorial structure is equivalent across the sex groups; (2) metric, to test if the loadings are equivalent in the two groups; and (3) scalar, to test if the loadings and the intercepts are considered equal and, therefore, the scores on the two dimensions (somatic and cognitive) can be compared across sex groups. The results showed a very good level of overall adjustment, with CFI and TLI values above the cut-off of .95, and RMSEA lower than .07 in all the tested models. Although the individual results of the models were good, and even considering the slight improvements in the models with the increase in the constrains, the chi-square differences revealed that it was possible to assume configural and metric invariance, but not scalar invariance, in both trait and state forms (Table 4). The overall findings suggest that both male and females interpreted the underlying factorial structure of STICSA-State and STICSA-Trait, their loadings, but not intercepts, in an equivalent way.

Download:

Table 4. Fit indexes for sex invariant models (n_male = 400; n_female = 753).

https://doi.org/10.1371/journal.pone.0262960.t004

Concurrent and nomological validity.

Regarding concurrent validity, STICSA was highly associated with a measure of anxiety (STAI), specifically regarding its cognitive dimensions in both state and trait forms. The highest associations were observed between STICSA trait-cognitive dimension and STAI trait (r = .796), followed by the relationship between STICSA state-cognitive dimension and STAI state (r = .764), suggesting that STAI covers largely the cognitive aspects of anxiety. These correlation values are in line with the cut-off suggested by Kline [50] for concurrent validity (r < .75). On the other hand, the association between the somatic dimensions of STICSA and the dimensions of anxiety measured by STAI was lower (between .452 and .508), highlighting the distinction between cognitive and somatic dimensions.

Correlations with depression (DASS-D) were lower compared to the ones with anxiety (STAI), regarding the four dimensions of STICSA. Still, the level of association between the depression measure’s score and STICSA’s state-cognitive (r = .652) and trait-cognitive (r = .634) dimensions presented a large effect (>.50; [48]). When comparing the correlation between depression and both STAI state (r = .643) and STAI trait (r = .722) scores, STICSA dimensions evidenced lower levels of association with depression in all dimensions, with the exception of state-cognitive dimension. The difference in the correlations of the four STICSA factors and the positive and negative affect (PANAS) was about twice the level of association, being the association with positive affect small (.10≤r < .30) and with negative affect moderated (.30≤r < .50) in somatic dimensions and large in cognitive dimensions (>.50; Table 5). On the other hand, STAI evidenced moderate correlations with positive affect (STAI State: r = -.468; Trait: r = -.517), and strong correlations with negative affect (STAI State: r = .694; Trait: r = .723).

Download:

Table 5. Cronbach’s α and Pearson correlation between STICSA factors and anxiety (STAI), depression (DASS-D), positive affect (PA) and negative affect (NA).

https://doi.org/10.1371/journal.pone.0262960.t005

Discussion

The present study aimed to extend STICSA’s psychometric studies by exploring its dimensionality, measurement invariance, reliability and nomological validity. Our results replicated previous findings, providing strong evidence for STICSA’s construct validity, including the distinction between the four dimensions, in non-clinical samples [13, 17, 22]. The theoretical structure was reflected in the underlying factorial structure considering either the test with the total of items and the four dimensions (full factorial model), and the models separated by state and trait dimensions. This further corroborates the psychometric properties and validity qualities of this new anxiety measure in the Portuguese population.

Several full and state/trait factor models were tested in the present study, considering the theoretical structure of STICSA. Regarding the full model with the 42 items, in the present study, the best overall adjustment was observed for the model with four correlated factors, thus replicating the previous findings provided by studies testing the full model in non-clinical samples [17, 22]. Yet, considering the high multicollinearity observed between the state-cognitive and the trait-cognitive, as well as between the state-somatic and trait-somatic dimensions, several models with the state and trait forms analyzed separately were tested. These analyses revealed that the model with the best adjustment encompassed two correlated factors (cognitive and somatic) within each form. The correlations between somatic and cognitive dimensions were higher than .65 across state and trait forms, suggesting that, despite of being independent measures, they are closely associated [13]. Similar results were also found for non-clinical samples [13, 17–20] which, altogether, corroborates this instrument’s construct validity and the distinction between the four dimensions of anxiety (trait, state, as well as cognitive and somatic anxiety). The reliability analyses conducted in the present study also support STICSA as a reliable measure of anxiety, extending the previous literature [13, 16–19, 22].

In order to be accurate in group comparisons, individuals of different groups should interpret the scale and its components in an equivalent way. The multi-group CFA performed in the present study supported the equivalence of the underlying structures between groups regarding sex (males vs. females) for STICSA’s scale interpretation, at the level of metric invariance, which adds further strength to the quality of this measure and reinforces the confidence in comparing the latent variable scores across groups. A similar result regarding measurement invariance across sex was reported by past studies [18, 19]. The intercepts (the means) differ between the groups, which reinforces the expected sex differences [42], maintaining the good characteristics of the STICSA regarding the factorial structure and its loadings, although not its intercepts.

The level of association between the dimensions of STICSA and the two subscales of STAI gave further evidence of convergent validity, particularly regarding the cognitive dimensions. The difference in the levels of association between both cognitive (>.75) and somatic (.45 - .51) dimensions and the STAI state and trait scores highlights the newness and relevance of the somatic subscales of STICSA. Our results, in line with the previous literature [17, 18], thus suggested that STAI is a more specific measure of cognitive anxiety, while STICSA opens the possibility of evaluating anxiety in a more comprehensive fashion, i.e., in a multidimensional perspective. Nevertheless, although STICSA aims to provide a better differentiation between anxiety and depression, our results do not fully support the achievement of this goal. In fact, the Portuguese version of STICSA was shown to be strongly correlated with depression, particularly its cognitive dimensions (r>.50). These results are in line with previous psychometric studies of STICSA [17, 19], are also expected in light of the tripartite model of anxiety and depression [51]. This model explains the frequent overlap and comorbidity between anxiety and depression, and highlights a component specific to anxiety (high physiological arousal), as well as a component specific to depression (low positive affect), and a third component shared by both of them, corresponding to negative affect.

Hereupon, it is expected that anxiety and depression overlap and correlate at a certain extent, in particular considering the cognitive dimensions of STICSA, which seem to greatly reflect negative affect in comparison with the somatic dimensions which, in turn, reflect physiological arousal [16]. Furthermore, according to the tripartite model, it is not expected that a measure of anxiety encompasses items reflecting low levels of positive affect, which is exclusive to depression. Our results, once more, support both the tripartite model assumptions and STICSA as measure of anxiety, by showing that its cognitive dimensions are more associated with high negative affect (r≥.30 for somatic dimensions and r>.50 for cognitive dimensions) than with low positive affect (.10≤r < .30). Therefore, STICSA seems to better discriminate anxiety from depression in comparison with STAI, given STICSA’s weaker correlations with low positive affect, as well as the lower levels of association with the depression measure. Nevertheless, the lower associations with depression were observed for all dimensions of STICSA, except for the state-cognitive; this result is intriguing and new, and possibly suggests that the state dimension reflects negative affect in a higher extension than the trait-cognitive dimension. Importantly, these results highlight the relevance of the somatic subscales’ scores to differentiate anxiety from depression [17, 18].

Study 2: Relationship between trait anxiety and the subjective and psychophysiological emotional response towards distinct emotional situations

Given the multidimensional nature of anxiety, it is of utmost relevance to consider, in self-report validations, how the scores are associated with such dimensions, as well as to the emotional experiences under distinct situations. Ree and colleagues [13] explored how the STICSA’s cognitive and somatic dimensions of STICSA-Trait predicted the state anxiety response in a stressful event. They observed that trait-cognitive anxiety predicted a significant part of the variance in the self-reported state-cognitive and state-somatic anxiety under a cognitive stressor. Conversely, trait-somatic anxiety was associated with the self-reported state-cognitive and state-somatic anxiety in response to a somatic stressor. These results support trait-cognitive and trait-somatic anxiety as independent constructs, as well as the usefulness of STICSA to predict the anxiety response towards different types of stressors [13]. Yet, it is critical to extend this knowledge to other emotional contexts (e.g., such as under fear or happiness emotional states), which can be related to different demands, and, therefore, with distinct subjective, cognitive, behavioral, emotional and physiological responses (for reviews see [24, 52]).

On the other hand, it would be equally helpful to understand how trait-cognitive and trait-somatic anxiety are differently associated not only with the subjective, but also with the psychophysiological dimension of the emotional response. In fact, the anxiety response is multidimensional, and is often characterized by differences in the psychophysiological domain. For instance, high self-reported trait anxiety has been associated with autonomic dysfunction, specifically with a decrease in vagally-mediated indexes of HRV [53, 54]. Reduced HRV, in turn, reflects decreased flexibility and adaptability, influencing general well-being and health [55, 56]. To date, no study has explored this relationship considering self-reported trait anxiety as measured by STICSA; its differentiation between the cognitive and somatic dimensions of anxiety can yield important implications, considering that many anxiety instruments, such as STAI, possibly cover items that mainly assess the cognitive dimensions of anxiety (as supported by our results of Study 1).

Adopting a multimodal perspective to assess the predictive validity of STICSA-Trait can add to the conceptualization of trait-cognitive and trait-somatic anxiety as distinct constructs [13], which have been shown to distinctively predict behavior [57]. Also, given the inter-individual variability found in the emotional response associated with the different components of emotions [23], a multimodal approach can contribute to better understand people’s emotional responses [58]. The current study sought to explore how STICSA-Trait explains differences in psychophysiological and self-report measures. We designed an emotional induction procedure, during which the subjective and psychophysiological responses of the participants were collected. We expected to observe an overall reduced HRV in individuals with high trait anxiety in comparison with individuals with low trait anxiety [53].