Factor structure of the anxiety sensitivity index-3 in a sample of older adults

Abstract

Anxiety sensitivity is a transdiagnostic risk factor for internalizing psychopathology in children and adults. Several factor analytic studies have examined the factor structure of anxiety sensitivity in children, adolescents, and adults. However, to date, no studies have specifically examined the factor structure of the Anxiety Sensitivity Index-3 in older adults. The purpose of the current study was to compare the fit of a correlated three-factor, higher-order, and bifactor model of the Anxiety Sensitivity Index-3 (ASI-3) in a sample of older adults. Participants were 856 older adults aged 60–94 recruited through Qualtrics Panels. Confirmatory factor analyses demonstrated good model fit for the bifactor model and the higher-order model and adequate fit for the correlated three-factor model. These results suggests that the ASI-3 is a valid tool assessing anxiety sensitivity in older adults. Limitations and future directions are discussed.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10433-022-00736-9.

Introduction

Anxiety sensitivity (AS) is the fear of anxiety-related sensations (Reiss and McNally 1985) and is considered a risk factor for internalizing disorders and symptoms (Naragon-Gainey 2010; Olatunji and Wolitzky-Taylor 2009) such as panic disorder and posttraumatic stress disorder (Olatunji et al. 2009). Importantly, AS is considered a theoretically and empirically distinct construct from negative affect (McNally 2002) and trait anxiety (McWilliams and Cox 2001). For example, AS is significantly associated with PTSD symptoms in adults even after accounting for negative affect (Fetzner et al. 2012) and a significant predictor of panic disorder even after accounting for trait anxiety (Jurin and Biglbauer 2018).

Several self-report measures have been developed to assess AS, with the most frequently used measure being the Anxiety Sensitivity Index (ASI; Reiss et al. 1986). Although the ASI was originally conceptualized as a unidimensional construct, subsequent work has found support for a multidimensional structure. However, factor analytic studies with the ASI have reported inconsistent multidimensional factor solutions (e.g., Blais et al. 2001; Peterson and Heilbronner 1987; Reiss et al. 1985; Telch et al. 1989). In addition, it has been suggested that the subscales of the ASI may not have adequate reliability and that some of the items (e.g., “It scares me when I am nauseous”) do not target specific dimensions (Taylor et al. 2007). In response to the lack of consistency across studies and issues with reliability and content validity of the ASI, Taylor and Cox (1998) developed the ASI-R, a 36-item measure of AS, and the 60-item Anxiety Sensitivity Profile to address concerns for the content relevancy and representativeness of items on the ASI.

Subsequent studies with the ASI-R, however, also reported inconsistent factor solutions (e.g., Peterson and Reiss 1992; Zinbarg et al. 1997), which led to the development of the Anxiety Sensitivity Index-3 (ASI-3; Taylor et al. 2007). The ASI-3 was subject to an extensive and scientifically rigorous validation process, including testing across four separate samples consisting of clinical and non-clinical subjects (Taylor et al. 2007) and is a three-factor measure of AS consisting of somatic/physical concerns (i.e., fear of physical sensations), cognitive concerns (i.e., fear of cognitive dyscontrol), and social concerns (i.e., fear of publicly observable symptoms). The three-factor solution of the ASI-3 has been replicated in both clinical (e.g., Kemper et al. 2012) and non-clinical (e.g., Osman et al. 2010) samples and provides a meaningful way to examine facets of AS across internalizing disorders given that the subscales have demonstrated greater associations with specific disorders. For example, studies have shown that individuals with panic disorder have more somatic concerns than patients with other anxiety disorders (e.g., Kemper et al. 2012), while the fear of publicly observable anxiety symptoms (i.e., social concerns) is more strongly related to social anxiety disorder (McWilliams et al. 2007; Zinbarg et al. 1997). In addition, more recent studies have shown that cognitive concerns is uniquely associated with overall PTSD symptom severity (Raines et al. 2017) and depression symptoms (Saulnier et al. 2018).

Although previous investigations of the psychometric properties of the ASI-3 confirmed the initial correlated three-factor structure, several studies have found support of a bifactor structure for the ASI-3 (e.g., Jardin et al. 2018; Osman et al. 2010). The bifactor model separates a general factor (i.e., anxiety sensitivity) that is orthogonal with a set of subfactors (i.e., physical concerns, cognitive concerns, and social concerns). When tested against the original correlated three-factor model, the bifactor model has resulted in the best model fit for the ASI-3 (e.g., Ebesutani et al. 2014), providing additional support for the hierarchical structure of AS. However, despite research leading to a better understanding of the factor structure of the ASI-3, several questions about its psychometric performance across different demographic groups still remain. For example, there have been few studies that examine the ASI-3 across gender (e.g., Allan et al. 2017; Ebesutani et al. 2014), and even fewer studies that examine the factor structure of the ASI-3 across different racial/ethnic identities, sexual identities, and age groups (Jardin et al. 2018). It is important to note that while one study found evidence for measurement invariance of the ASI-3 across young adults (ages 18–20) and adults (ages 21 +) (Jardin et al. 2018), no study to date has examined the factor structure of the ASI-3 in older adults, and it is unclear if the proposed three-factor solution replicates in this population.

The assessment of AS in older adults is particularly important given that previous investigations have found that AS significantly predicts panic symptoms (Deer and Calamari 1998) and hypochondriacal concerns (i.e., illness anxiety; Bravo and Silverman 2001) among older adults. In addition, more recent research has shown that AS predicts both subjective cognition and everyday functioning among older adults (Divers et al. 2021). The purpose of the current study was to examine the factor structure of the ASI-3 in an older adult sample using a correlated three-factor model, higher-order, and a bifactor model. Given the limited research on the ASI-3 in older adult populations, the present study is exploratory in nature, and no assumptions about model fit were made with regard to the best fitting model.

Materials and method

Participants

Participants were 856 older adults aged 60–94 years (M = 67.62, SD = 5.79). The current sample represents a combined sample from three separate studies on older adults. Data from a subset of these participants have been reported previously (Divers et al. 2021). Participants were predominantly female (74.4%) and White (90.4%). Additional demographic information is given in Table 1.

Table 1.

Demographic characteristics of combined sample

	n	%
Age
60–69	573	66.9%
70–79	250	29.2%
80–89	31	3.6%
90 +	2	0.2%
Race/Ethnicity
White	774	90.4%
Black	50	5.8%
Asian	13	1.5%
Native American	1	0.1%
Multiracial	12	1.4%
Other	2	0.2%
Unknown	4	0.4%
Gender
Female	637	74.4%
Male	217	25.4%
Other	2	0.2%
Education
Less than high school	17	2.0%
Some high school	4	0.5%
High school graduate/GED	204	23.8%
Some college/specialized training	267	31.2%
College/University graduate	264	30.8%
Master’s degree	83	9.7%
Doctorate/Professional degree	13	1.5%
Not Reported	4	0.5%

Measures

The Anxiety Sensitivity Index-3 (ASI-3; Taylor et al. 2007) is an 18-item measure that assesses the degree of fear related to negative consequences of experiencing anxiety symptoms. Items are rated on a 5-point Likert-type scale from 0 (very little) to 4 (very much). Items on the ASI-3 reflect three theoretically derived facets of anxiety sensitivity such as physical concerns (e.g., “It scares me when my heart beats rapidly”), cognitive concerns (e.g., “When I feel “spacey” or spaced out, I worry that I may be mentally ill”), and social concerns (e.g., “It is important not to appear nervous”). The ASI-3 has demonstrated good internal consistency reliability for the total score and subscales (Cronbach’s α = 0.80–0.93; Wheaton et al. 2012).

Procedure

Participants were recruited through three separate data collection efforts via Qualtrics Panels, an online recruitment platform. All participants agreed to receive a personalized incentive (e.g., flyer miles) through Qualtrics Panels before beginning the studies. Participants from all three studies provided informed consent to complete the survey, and all study procedures for each data collection effort were approved by the appropriate Institutional Review Board.

Analytical Plan

All data were pre-screened to check for missing data. Four participants who did not complete the ASI-3 were excluded (n = 4). Descriptive analyses (n = 852) were conducted in SPSS before preparing the data to be exported into Mplus (Muthén and Muthén 2017) and is shown in Table 2.

Table 2.

Descriptives and reliability estimates of the ASI-3

	M	SD	Skewness	Kurtosis	Cronbach’s α	Omega/ Omega S
Physical concerns	4.67	4.96	1.35	1.53	0.88	0.89
Cognitive concerns	3.09	4.59	1.99	3.92	0.91	0.92
Social concerns	5.58	4.75	1.11	1.15	0.79	0.82
Total	13.32	12.73	1.49	2.31	0.93	0.95

Confirmatory Analyses. Model comparisons were conducted with confirmatory factor analysis (CFA) with robust maximum likelihood estimation (MLR) in Mplus (Muthen et al. 2017). The first model was a correlated three-factor model in which the three factors were allowed to correlate with each other. The second model was a higher-order model in which all subordinate factors (lower-order) load onto a superordinate factor (high-order factor) with the 18 items loading only onto their respective subordinate factors (i.e., physical concerns, cognitive concerns, social concerns). The third model was a bifactor model in which all 18 items were allowed to load onto a general factor as well as their respective factors (i.e., physical concerns, cognitive concerns, social concerns).

Model estimation and comparison. Goodness of fit was determined using indices for root-mean-square error of approximation (RMSEA), standardized root-mean-square residual (SRMR), Chi-square, Tucker-Lewis Index (TLI), and comparative fit index (CFI). The following cutoffs were used to evaluate the fit of the models: non-significant χ² value, TLI > 0.95, SRMR < 0.08 (Hu and Bentler 1999), CFI > 0.90 (Bentler 1990), and RMSEA < 0.08 (MacCallum et al. 1996). It should be noted that χ² values are often significant with large sample sizes (Floyd and Wideman 1995). We compared the fit of the bifactor model relative to these two competing models using Akaike information criterion (AIC) and Bayesian Information Criteria (BIC), where smaller AIC and BIC values indicate better fit.

Bifactor model evaluation. Additional fit indices were calculated to assess the reliability of the general and specific factors (i.e., Omega/OmegaS, OmegaH/OmegaHS, H) (Hancock and Mueller 2001; Reise et al. 2013a, b) of the bifactor model. Omega is a model-based estimate of internal reliability of the multidimensional composite. For the general factor, all items are considered and for specific factors, only items loading on that factor are relevant to OmegaS. OmegaH reflects the percentage of systematic variance in raw total scores that can be attributed to the individual differences on the general factor, while omegaHS reflects the proportion of reliable systematic variance of a subscale score after partitioning out variability attributed to the general factor (Reise et al. 2013a, b). When omegaH is greater than 0.80, total scores can be considered essentially unidimensional. H is a measure of construct replicability which represents the correlation between a factor and an optimally weighted item composite, with high values (> 0.80) suggesting a well-defined latent variable (Hancock et al. 2001).

Results

The correlated three-factor structure provided an adequate fit for the data [χ²(132) = 540.160, p < 0.001, TLI = 0.914, CFI = 0.926, RMSEA = 0.060, SRMR = 0.048]. The three factors were correlated strongly with each other (r = 0.83–0.89), and all items are loaded strongly onto their specified factors (0.62–0.89). The higher-order model provided a good fit for the data [χ²(131) = 536.067, p < 0.001, TLI = 0.914, CFI = 0.926, RMSEA = 0.060, SRMR = 0.048]. All factors are loaded strongly onto the higher-order factor (0.84–0.90), and all items are loaded onto their respective factors (0.39–0.85). The bifactor model also provided a good fit for the data [χ²(117) = 525.434, p < 0.001, TLI = 0.903, CFI = 0.926, RMSEA = 0.064, SRMR = 0.038]. However, it should be noted that not all items loaded onto their respective factors. For example, factor loadings were low (0.15–0.27) for all items on the cognitive concerns factor. Examination of specific bifactor indices for the general factor indicated support for the general factor (Omega = 0.95, OmegaH = 0.88, H = 0.95). Examination of specific bifactor indices for the specific factors indicated some support for the physical concern (OmegaS = 0.89, OmegaHS = 0.28,  H = 0.65) and social concern (OmegaS = 0.82, OmegaHS = 0.23, H = 0.48) subscales. Little support was found for the cognitive concern subscale (OmegaS = 0.92, OmegaHS = 0.00, H = 0.14). Examination of AIC and BIC revealed that the bifactor model had the lowest AIC and BIC values compared to the competing models. Standardized factor loadings for all models are shown in Table 3, and global fit indices for all models are given in Table 4.

Table 3.

Standardized factor loadings

	Three-factor	Higher-order	Bifactor
Physical concerns
Item 3	0.73	0.73	0.45 (0.56)
Item 4	0.66	0.66	0.21 (0.64)
Item 7	0.84	0.84	0.56 (0.65)
Item 8	0.83	0.83	0.65 (0.59)
Item 12	0.82	0.82	0.48 (0.65)
Item 15	0.65	0.65	0.17 (00.65)
Cognitive concerns
Item 2	0.75	0.75	0.19 (0.75)
Item 5	0.77	0.77	0.07 (0.77)
Item 10	0.82	0.82	0.06 (0.82)
Item 14	0.81	0.81	0.13 (0.82)
Item 16	0.85	0.85	−0.27 (0.86)
Item 18	0.81	0.81	−0.15 (0.80)
Social concerns
Item 1	0.39	0.39	0.30 (.29)
Item 6	0.77	0.77	0.50 (0.63)
Item 9	0.82	0.82	0.40 (0.70)
Item 11	0.61	0.61	0.14 (0.60)
Item 13	0.75	0.75	0.39 (0.63)
Item 17	0.54	0.54	0.30 (0.44)

Loadings greater than .4 are in bold. Loadings for the general factor are in parenthesis

Table 4.

Global fit indices

	BIC	AIC	χ2	RMSEA	CFI	TLI	SRMR
Model
Three-factor	35,101.358	34,830.679	540.160**	0.06	0.92	0.91	0.04
Higher-order	35,108.107	34,832.679	536.067**	0.06	0.92	0.91	0.04
Bifactor	34,943.468	34,601.557	525.434**	0.06	0.92	0.90	0.03

BIC Bayesian information criteria, AIC Akaike’s information criteria, RMSEA root-mean-square error of approximation, CFI comparative fit index, TLI Tucker-Lewis index, SRMR standardized root-mean-square residual

**p < 0.01

Discussion

The primary goal of the current study was to examine the factor structure of the ASI-3 in an older adult sample. Although previous studies have examined the psychometric properties of the ASI-3 using younger adult clinical and non-clinical samples, previous studies have not examined the factor structure of the ASI-3 in older adults. The results of the current study demonstrated good model fit for all three models. The current study therefore extends previous work on the psychometric properties of the ASI-3 and suggests that the ASI-3 is a valid tool assessing anxiety sensitivity in older adults.

Similar to previous work, the current study found that the bifactor model resulted in the best model fit for the ASI-3 (Ebesutani et al. 2014). Although previous studies have found evidence for a three-factor second order (i.e., hierarchical) and three-factor structure in clinical and non-clinical samples, these studies used younger adult samples (e.g., Kemper et al. 2012; Wheaton et al. 2012). To the authors’ knowledge, this is the first study to analyze the factor structure of the ASI-3 in older adults. Therefore, continued research examining the psychometric properties of the ASI-3 in older adults is necessary, such as in clinical samples. Although the bifactor model demonstrated the best model fit out of the three models examined, it also demonstrated the lowest specific factor loadings, particularly for the cognitive concerns factor, which is consistent with one previous study (Allan et al. 2015) and inconsistent with others (Ebesutani et al. 2014; Osman et al. 2010). However, items on the cognitive concerns factor had the highest loadings on the general factor which is consistent with all previous studies (Allan et al. 2015; Ebesutani et al. 2014; Osman et al. 2010). Although the reason for the low cognitive concerns factor loadings is not clear, relationships between subjective cognitive impairment and physical complaints in older adults have been found (Jenkins et al. 2019) which may impact the clear separation of these constructs in this population. This is further complicated by findings that changes in cognitive functioning associated with aging may contribute to anxiety symptoms (Wilkes et al. 2013). Because there is a relationship between anxiety, future cognitive decline (e.g., Gulpers et al. 2016), and subjective cognitive impairment (Divers et al. 2021), future research should also continue to examine anxiety sensitivity in older adults, particularly how anxiety sensitivity may interact with aging.

Some limitations of the current study are that our sample primarily consisted of highly educated, Caucasian, older adult women, which may limit generalizability of the findings to this specific population. In addition, as stated in the Method section, participants were recruited through three separate data collection efforts via Qualtrics Panels. Although efforts were made to remove duplicate participant data among the different databases by removing individuals with matching demographic information (i.e., age, gender, race/ethnicity, date of birth), there is a possibility of some overlap across the three samples. Because identifying information was not collected, such as names or IP addresses, it is unclear what the exact percentage of overlap may be. Overall, the results of the current study suggest that the factor structure of the ASI-3 in older adults is similar to what has previously been found in younger adults. Future studies should continue to explore the psychometric properties of the ASI-3 including divergent validity from theoretically distinct constructs (e.g., negative affect; McNally 2002). Other self-report measures of anxiety (e.g., Geriatric Anxiety Inventory; GAI) (Pachana et al. 2007) demonstrate variability in divergent validity from other self-report measures (Johnco et al. 2015) depending on whether clinical or non-clinical samples were used in the study. Therefore, replication studies using older adults, including clinical samples, is needed for continued validation of the ASI-3. Future studies can examine the psychometric properties of the ASI-3 using different methodologies to help address issues related to measurement invariance (Van De Schoot et al. 2015). For example, future studies should compare the ASI-3 across different groups (e.g., men and women, young adults and older adults) and administer the ASI-3 over multiple time points (e.g., pre and post-test, longitudinal design) to determine the stability of the ASI-3 factors over time.

Supplementary Information

Below is the link to the electronic supplementary material.

Declaration

Conflict of interest

The authors report there are no competing interests to declare.