Using Network Analysis to examine connections between Acceptance and Commitment Therapy (ACT) processes, internalizing symptoms, and well-being in a sample of undergraduates

Abstract

Background:

Acceptance and Commitment Therapy (ACT) has been shown to be effective in treating internalizing symptoms. Understanding which ACT processes are most closely linked to certain symptoms may help develop targeted treatments. Network analysis an approach to gain insight into the interconnection between processes and the downstream benefits of targeting a particular process. However, limited work to date has explored networks involving ACT processes specifically.

Methods:

Undergraduate students (N = 447; 76.5% female; 89.5% White/Non-Hispanic) completed online questionnaires. The ACT processes assessed included experiential avoidance (AAQ-II), openness, awareness, and engagement (CompACT), and tacting ability (TOF), and internalizing symptoms/well-being (IDAS-II). Zero-order and partial correlation networks were examined as well as resulting communities.

Results:

In the association network, Dysphoria and experiential avoidance, and suicidality (in the concentration network only) were central nodes. In community analyses, experiential avoidance had the strongest influence in the association network, whereas well-being had the strongest influence in the concentration network. Auto-detected communities were also evaluated.

Limitations:

The present study was cross-sectional and included a largely White, female, undergraduate sample. This limits generalizability to more diverse, clinical, or general community, populations. Potential concerns about data are also noted including low reliability on the TOF and two skewed domains on the IDAS-II which may impact stability of centrality metrics.

Conclusions:

Well-being, dysphoria, and suicidality may be important process-based treatment targets. Further work is needed with diverse samples and using longitudinal designs to examine within person change of the associations between ACT processes and internalizing symptoms.

Broadly, Acceptance and Commitment Therapy (ACT) is an acceptance- and mindfulness-based therapy designed to help individuals identify and move toward personal values, even in the presence of internal struggles, by increasing psychological flexibility (Hayes et al., 1999). Improving psychological flexibility in ACT is based upon six core processes targeted in treatment to help clients live meaningful lives (acceptance, defusion, attention to the present moment, self-as-context, values, and committed action; Hayes et al., 2011). In brief, acceptance is the ability to experience internal events, such as thoughts or feelings, without trying to get rid of them and defusion is the ability to create space and distance from internal events (together, are thought of as openness). Attention to the present moment is the ability to notice what shows up in the moment, such as thoughts, feelings, and physical sensations, and self-as-context is the ability to step back and observe oneself as a person who can think, feel, and act (together, are thought of as awareness). Finally, values are the people, things, and qualities of action each individual person chooses are important to them, and committed action involves the behaviors one can take to move towards those values (together, are thought of as engagement). Together, these six processes are conceptualized as the hexaflex and used by clinicians to take a holistic view of a client’s strengths and weakness. Although decreasing symptoms of distress and psychopathology (e.g., anxiety or depression) is not a main goal of ACT, this often occurs as a by-product of helping clients experience emotions in workable ways and move toward values (Gloster, 2020). Through ongoing assessment of client’s strengths and relative growth areas in each of the ACT process domains, clinicians can develop targeted interventions. There is a large body of evidence showing ACT is an effective treatment for internalizing concerns (Bai et al., 2020; Gloster et al 2020; Bluett et al., 2014). However, understanding which ACT processes are most strongly linked to certain symptoms or well-being domains may help further clarify how clinicians can target ACT processes in treatment to best meet clients’ needs.

Network analysis is one such approach to gain insight into the interconnection between processes and the downstream benefits of targeting a process. Briefly, all networks contain at least two components: nodes and edges. In intervention research, nodes typically represent symptoms or disorders whereas edges represent the connection between them, such as correlations (Borsboom et al., 2013; McNally, 2021). Networks allow for computation of centrality metrics to examine which nodes are most central to the overall network. That is, which nodes are most closely connected to other nodes in the network. These metrics are particularly important in intervention research as it may yield important information for clinicians to help determine which symptoms to target that will have the greatest downstream effect.

Although network analysis has been used increasingly in recent years to explore associations in intervention and psychopathology research broadly (e.g., Borsboom, 2013; McNally, 2016, 2021), limited work has empirically investigated networks involving ACT processes specifically (e.g., Baker & Berghoff, 2021; Benfer et al., 2021). A recent theoretical study proposed the use of network analysis to understand the links within ACT processes and across psychological functioning and behavior change (Christodoulou et al., 2019). The authors note network analysis provides a framework for understanding psychiatric disorders as systems, within a context, which aligns well with the overall ACT framework that emphasizes behavioral function in context.

One more recent study evaluated associations between symptoms of post-traumatic stress and psychological flexibility (using the Multidimensional Psychological Flexibility Inventory [MPFI]) in a sample of N = 722, majority White adults (Mage = 37) using network analysis (Benfer et al., 2021). A second study modeled psychology flexibility and inflexibility, again with the MPFI as it relates to psychopathology in an N = 137, majority White female sample (Baker & Berghoff, 2021). Benfer et al. (2021) found that of the ACT processes, experiential avoidance had the strongest influence, and lack of connectedness with values had the second strongest influence on post-traumatic stress symptoms. Interestingly, the authors note that experiential avoidance, though the most influential as a bridge symptom, was not a highly central node overall. Baker and Berghoff (2021) results showed clusters consistent with psychological flexibility and inflexibility, though with no clear associations with psychopathology. Importantly, the authors note their sample was likely underpowered at N = 137 which may have led to spurious connections or the inability to detect small effects. Although the results of these two studies are preliminary evidence of the way ACT processes interact at a system level with symptoms of psychopathology, further work is needed to explore ACT processes as they relate to internalizing symptoms more broadly. Given that ACT is a transdiagnostic approach, examining how ACT processes interact with internalizing symptoms may yield important information for clinicians and researchers in identifying intervention targets.

ACT Process Measures

Numerous self-report ACT process measures have been developed to assess and track movement in the core domains over time. One of the most widely used measures is the Acceptance and Action Questionnaires, second version (AAQ-II; Bond et al., 2011). The AAQ-II was developed to measure “experiential avoidance,” or an unwillingness to experience internal events (e.g., thoughts, emotions) and subsequent engagement in avoidance-based behaviors (Hayes et al., 1996). Experiential avoidance is conceptualized as the contrast of acceptance, which is a willingness to experience unwanted internal experiences when it is worth it, or values-based. Although widely used, more recent criticisms of the scale have suggested that the AAQ-II is less a measurement of experiential avoidance, and instead measures psychological distress or neuroticism/negative affect (Wolgast et al., 2014; Rochefort et al., 2017; Tyndall et al., 2019; Ong et al., 2020). Further, a recent review has highlighted how within ACT research literature, the AAQ-II (and the first edition of the scale, the AAQ) has been described as a measure of psychological inflexibility, psychological flexibility, experiential avoidance, and acceptance—depending on the research publication (Arch et al., 2022). Thus, it appears the AAQ-II is not used consistently across studies to measure a shared ACT process. Francis et al. (2016) developed the Comprehensive assessment of Acceptance and Commitment Therapy (CompACT), which collapses the core six processes into three core pillars: openness, awareness, and engagement (Strosahl, et al., 2012). Openness is one’s willingness to experience internal experiences (e.g., thoughts, emotions, physical sensations) to pursue chosen values. Openness is thought to be a combination of acceptance and defusion. Awareness is one’s ability to pay mindful attention to the function of behaviors and is conceptualized as a blend of attention to the present moment and self-as-context. Engagement is the ability to identify chosen values and to flexibly pursue those values over time. The values and committed action processes are thought to make up engagement. Finally, Pierce and Levin (2019) developed the Tacting of Function scale (TOF). The ability to tact function entails noticing and labeling the purpose of behavior in a given context (e.g., being able to identify whether you are consuming alcohol to avoid emotions such as anxiety or depression (avoidant function) versus consuming alcohol socially to connect with friends (valued function)). Clinicians may use these measures with clients to assess strengths and weaknesses, identify targets for improvement, and measure progress from pre- to post-treatment.

Objectives of the Present Study

As such, this aim of this study is to evaluate how specific ACT processes are associated with internalizing symptoms and well-being domains from the IDAS-II using a network analysis approach. Network analysis is often exploratory in nature, particularly using cross-sectional data and relying on correlations. However, both zero-order and partial correlation networks will be examined. This study will evaluate the six core processes of the hexaflex by way of the three pillars (openness, awareness, and engagement) suggested by Strosahl, et al., 2012. Additionally, this study will examine two related ACT processes that overlap with the six core processes: experiential avoidance and tacting ability. Given the recent findings by Benfer at al. (2021), it is expected that experiential avoidance will be the strongest and most influential ACT process across both the zero-order and partial networks. This is bolstered by the noted criticism that the AAQ-II primarily taps into negative affect, which cuts across many internalizing disorders measured by the IDAS-II. Relatedly, it is expected that of the IDAS-II domains, the dysphoria domain will also be a highly central node, given expected associations between the AAQ-II and dysphoria.

Method

Participants and Procedure

447 undergraduate students (M = 18.66 years, SD = 1.02 years) were included in the current sample. Subjects were recruited from an online study pool that provided course credit for participation. Data collection took place during the Fall semester of 2020. Eligibility criteria for the current study included: enrollment in a participating psychology course, being aged 18 to 26 years old, and ability to read English. Age was missing from the surveys of study participants who otherwise had completed the survey battery (n =79), due to an auto-selection feature wherein age 18 was auto-filled. This missing data mostly (87%) corresponded to undergraduates in their first- or second-year. Descriptive statistics of the study sample are provided in Table 1. All study procedures were approved by the University of Iowa institutional review board, and all participants provided informed consent.

Table 1.

Descriptive statistics of the sample (N = 447).

Characteristics
Age, mean years (± SD)	18.66 (1.02)
Gender identity, N(%)
Male	101 (22.6%)
Female	342 (76.5%)
Transgender	2 (.4%)
Other	2 (.4%)
Race, N(%)
White	366 (81.9%)
Asian	43 (9.6%)
African American or Black	10 (2.2%)
American Indian or Alaska Native	1 (.2%)
Native Hawaiian or Pacific Islander	1 (.2%)
Biracial or multiracial	25 (5.6%)
Missing	1 (.2%)
Ethnicity, N(%)
Hispanic or Latinx	45 (10.1%)
Non-Hispanic and Not Latinx	400 (89.5%)
Missing	2 (.4%)

Note. Participants were asked their “gender identity” on the survey with response options of male, female, transgender woman, transgender man, genderqueer, prefer to self-describe, and prefer not to disclose.

Measures

ACT processes.

Experiential avoidance was measured with the Acceptance and Action Questionnaire, second version (AAQ-II; Bond et al., 2011). The 7 items are rated on a 7-point Likert scale, ranging from (1) never true to (7) always true, with higher scores indicating greater avoidance. Three core processes of psychological flexibility were measured with the Comprehensive assessment of Acceptance and Commitment Therapy processes (CompACT; Frances et al., 2016). The three subscales include: (1) openness to experience, or one’s willingness to experience internal experiences (i.e., thoughts, emotions, physical sensations), without trying to get rid of or control them; (2) behavioral awareness, or one’s ability to be mindfully aware of their actions in the present moment; and (3) valued action, or one’s engagement in values-based actions. For clarity, these subscales will henceforth be referred to as openness, awareness, and engagement (respectively). The 23 items are rated on a 7-point Likert scale, ranging from 0 (strongly disagree) to 6 (strongly agree). Scores were reversed such that higher scores across all measures would indicate poorer psychological flexibility to aid in interpretability. Furthermore, an important component of awareness is the ability to notice the function or purpose of one’s actions. The ability to tact (or label) function of behavior in a context was measured with the Tacting of Function scale (TOF; Pierce & Levin, 2019). The 10 items are rated on a 7-point scale, ranging from 1 (never true) to 7 (always true). Like the CompACT, scores were reversed such that higher scores would indicate poorer tacting ability. Internal consistencies of the measures in the current sample ranged from poor to excellent (AAQ-II: α = .91; openness: α = .79; awareness: α = .90; engagement: α = .86; TOF α: = .58).

Internalizing Symptoms and Well-Being.

The Inventory of Depression and Anxiety Symptoms expanded version (IDAS-II; Watson et al., 2012) was used to assess internalizing and well-being domains over the previous two weeks. The subscales used in the current analyses include appetite gain, appetite loss, checking, claustrophobia, cleaning, dysphoria, euphoria, ill temper, insomnia, lassitude, mania, ordering, panic, social anxiety, suicidality, trauma avoidance, trauma intrusions, and well-being. The general depression composite scale was not used in the current analyses, as this has overlapping items with several subscales. The well-being subscale was reverse-scored so that higher scores indicated poorer well-being. The 99 items are rated on a 5-point scale ranging from 1 (not at all) to 5 (extremely). Internal consistency across all subscales in the current sample was adequate and ranged from .75 (euphoria) to .92 (dysphoria).

Data Analytic Plan

First, sum scores for the IDAS-II subscales and three ACT measure subscales (CompACT, TOF, and AAQ-II) were created. Next, two adjacency matrices were made to be used in network analysis, a zero-order correlation matrix and a partial correlation matrix. All non-significant correlations were set to zero (Borsboom & Cramer, 2013). The adjacency matrices were then imported into Rstudio (version 1.4.1717; R Core Team, 2019) to use for network visualization. Edges represent the weight of the correlations between the individual nodes. Wider edges indicate stronger correlations between nodes. Given the expected density of the networks, edges were also set to fade in color intensity using the standard cut-points for larger networks (more than 20 nodes) in the qgraph package. Nodes represent scores of each IDAS-II or ACT process scale domains, and node size was set to be larger for nodes with higher degree centrality (i.e., more connections to other nodes).

Local centrality metrics, such as degree centrality (the number of connections to any given node) and weighted degree (the sum of the weighted connections to a given node) were examined. Global centrality metrics, such as closeness, betweenness, and eigenvector centrality, were also computed. Closeness indicates the shortest path length between a given node and all other nodes. Betweenness measures the number of shortest paths that pass through a given node; it is a way to identify bridge nodes or highly influential nodes. Eigenvector centrality identifies “important” nodes as those with high eigenvector centrality scores are connected to other “important” nodes. Given centrality metrics are highly influenced by data variability, preliminary checks of all IDAS-II and ACT subscale scores were completed to examine for limited variability (i.e., floor and/or ceiling effects).

Clustering within the network was also examined to identify whether distinct subcomponents of nodes existed. Both predefined communities (assigning nodes to either the ACT process community or the IDAS-II community) and auto-detecting communities were evaluated (see the supplemental method section for information on auto-detecting communities). Results presented will be bridge strength (which relies on absolute values of edge weights) and bridge expected influence in terms of indirect effects (which considers negative and positive edge weights). Bridge strength is the absolute value sum of all edge weights between a given node and nodes not in that community. A high bridge strength indicates a node that is strongly connected to nodes in another community. Expected influence values show potential indirect effects on a community in which the node is not a member, stronger values are those that are farther away from zero.

Results

First, item-level missing data were examined. If <20% of items were missing from a scale (for the AAQ-II and TOF), or subscale (for the CompACT and IDAS-II), items were imputed with mean person imputation. If more than 20% of items were missing, the score was excluded from the current analysis (Hawthorne et al., 2005). Across scales and subscales, 0.51–1.74% items were missing and 0.06–0.29% of items were imputed.

Overall, there was a large degree of variability across model variables (see Table S1). Skewness values were within normal limits, except for suicidality and claustrophobia, with these domains showing low endorsement (i.e., floor effects). The means and the standard deviations for the suicidality and claustrophobia subscales are in line with previously established national norms (Nelson et al., 2018). As such, these variables were not transformed. However, to best contextualize the scores on the IDAS-II and levels of internalizing symptoms in the present sample, raw scores were converted to percentiles based on national norms (see Table S2). Overall, results indicated there were relatively high levels of appetite loss (M percentile = 72.15, SD = 18.64); however, this still falls within the normal range. The next highest internalizing domain was social anxiety (M percentile = 66.23, SD = 26.49). Notably, data were collected during COVID-19, during which social isolation and distancing protocols were enacted. Conversely, the lowest domain in the sample was suicidality, which was also still within the normal range (M percentile = 44.83, SD = 27.03). The remaining scales fell between approximately 47 and 64 percentiles, all within the normal range.

Association Network

The zero-order correlation network had 23 nodes and 239 total edges (see Figure 1, Panel a). The density of the network was .94, meaning 94% of all possible connections appeared in the network, indicating this is a strongly connected network. Five centrality metrics were examined (see Table 2). Ten nodes had the highest degree centralities, and notably, all ten were subscales of the IDAS-II (Dysphoria, Lassitude, Insomnia, Suicidality, Ill Temper, Panic, Social Anxiety, Claustrophobia, Trauma Intrusions, and Trauma Avoidance). Two nodes had the lowest degree centralities (Well-Being and Euphoria). For both weighted degree and eigenvector centrality, the most central node was Dysphoria, whereas the least central node was Well-Being. For closeness, the same ten nodes as listed above had the highest closeness value, whereas Well-Being and Euphoria were the most distant nodes. Lastly, for betweenness, the same ten nodes had the highest betweenness values, whereas ordering had the lowest.

Figure 1.

Panel A depicts the zero-order correlation network and Panel B depicts the partial correlation network; nodes are scaled such that larger nodes indicate higher centrality and are scaled within each graph separately given the reduced connections in the partial network. Blue = IDAS; Light purple = ACT processes; 1 = Experiential Avoidance; 2 = Openness; 3 = Awareness; 4 = Engagement; 5 = Tacting Ability; 6 = Well-Being; 7 = Dysphoria; 8 = Lassitude; 9 = Insomnia; 10 = Suicidality; 11 = Appetite Loss; 12 = Appetite Gain; 13 = Ill Temper; 14 = Mania; 15 = Euphoria; 16 = Panic; 17 = Social Anxiety; 18 = Claustrophobia; 19 = Trauma Intrusions; 20 = Trauma Avoidance; 21 = Checking; 22 = Ordering; 23 = Cleaning.

Table 2.

Network Centrality. Metrics

Node	Association					Concentration
	Degree	EgnVct	Weighted Degree	Betweenness	Closeness	Degree	EgnVct	Weighted Degree	Betweenness	Closeness
Awareness	42	0.74	9.02	0.53	0.96	12	0.23	0.75	18.23	0.54
Experiential Avoidance	42	0.82	9.94	0.53	0.96	10	0.51	0.62	6.15	0.54
Openness	42	0.73	8.75	0.53	0.96	16	0.49	0.95	37.97	0.59
Tacting Ability	40	0.35	4.38	0.35	0.92	10	0	0.58	11.48	0.49
Engagement	38	0.45	5.58	0.23	0.88	8	0	0.58	11.11	0.47
Appetite Gain	40	0.44	5.26	0.48	0.92	8	0.21	0.22	5.25	0.46
Appetite Loss	42	0.7	8.26	0.72	0.96	10	0.32	0.34	6.71	0.5
Checking	40	0.7	8.34	0.34	0.92	10	0.64	0.73	12.1	0.52
Claustrophobia	44	0.73	8.89	0.83	1	12	0.64	0.96	17.34	0.55
Cleaning	42	0.62	7.6	0.53	0.96	12	0.59	0.83	17.33	0.55
Dysphoria	44	1	12.3	0.83	1	20	1	1.96	64.86	0.65
Euphoria	34	0.35	4.1	0.36	0.81	14	0.69	0.81	39.84	0.59
Ill Temper	44	0.81	9.87	0.83	1	8	0.26	0.51	8.44	0.48
Insomnia	44	0.79	9.48	0.83	1	12	0.18	0.26	14.83	0.54
Lassitude	44	0.89	10.88	0.83	1	12	0.47	0.83	25.3	0.56
Mania	42	0.84	10.25	0.53	0.96	12	0.74	1.1	15.37	0.56
Ordering	38	0.65	7.81	0.23	0.88	10	0.6	0.81	10.32	0.51
Panic	44	0.9	10.88	0.83	1	12	0.67	1.01	23.54	0.56
Soc Anxiety	44	0.85	10.37	0.83	1	10	0.55	0.77	9.16	0.49
Suicidality	44	0.75	9.08	0.83	1	20	0.55	0.63	47.77	0.65
Trauma Avoidance	44	0.78	9.31	0.83	1	14	0.74	1.12	20.56	0.58
Trauma Intrusions	44	0.9	10.81	0.83	1	10	0.74	1.22	7.14	0.48
Well-Being	34	0.34	3.97	0.31	0.81	10	0	0.35	13.21	0.5

Note. EgnVct = Eigenvector centrality; the first five nodes are the ACT processes and the rest are the IDAS subscales

For the predefined community analysis (see Table 3), in the IDAS-II community, dysphoria had the strongest bridge strength and expected influence value when considering indirect effects, whereas euphoria was weakest for both. For the ACT community, experiential avoidance had strongest bridge strength and expected influence value when considering indirect effects, whereas tacting of function had the weakest for both.

Table 3.

Predefined Community Analysis Metrics

	Association Network		Concentration Network
	Bridge Strength	Expected Influence (2-step)	Bridge Strength	Expected Influence (2-step)
Awareness	7.09	72.5	0.49	0.45
Experiential Avoidance	8.06	80.85	0.71	0.98
Openness	7.04	71.82	0.72	0.89
Tacting Ability	3.17	33.77	0.21	0.28
Engagement	4.13	42.95	0.53	0.22
Appetite Gain	0.94	9.88	0	0.02
Appetite Loss	1.66	16.79	0	0.06
Checking	1.28	15.31	0.13	−0.1
Claustrophobia	1.34	16.46	0.16	−0.13
Cleaning	1.05	13.43	0.29	−0.09
Dysphoria	2.72	23.94	0.37	0.51
Euphoria	0	6.51	0	−0.08
Ill Temper	2	19.04	0.1	0.15
Insomnia	1.89	18.87	0	0.11
Lassitude	2.27	21.33	0.11	0.33
Mania	1.76	18.99	0.14	0.25
Ordering	0.94	13.92	0	0
Panic	2.01	20.97	0.1	0.17
Social Anxiety	2.13	20.14	0	0.07
Suicidality	1.78	17.76	0.13	−0.16
Trauma Avoidance	1.76	17.72	0.38	0.56
Trauma Intrusions	2.07	20.91	0.11	0.28
Well-Being	1.89	9.93	0.63	0.89

Note. ACT measures, above the line, are in community one and the IDAS measures, below the line, are in community two. For Expected influence, a score closer to zero is weaker

Community analysis based upon auto-detected communities (see Table 4) indicated that there were two communities in the network. The first was made up of appetite gain, mania, euphoria, claustrophobia, trauma avoidance, checking, ordering, and cleaning. The remaining IDAS-II scales and all ACT process measures were in the second community. In the first community, euphoria had the lowest bridge strength and expected influence while mania had the highest. For bridge strength in the second community, well-being had the lowest value and trauma intrusions had the highest. For expected influence for indirect effects, well-being again had the lowest value whereas dysphoria had the strongest value.

Table 4.

Auto-detected Community Analysis Metrics

		Association			Concentration
	Symptom	Community Membership	Bridge Strength	EI (2-step)	Community Membership	Bridge Strength	EI (2-step)
IDAS	Appetite Gain	1	3.3	35.65	2	0.59	0.05
IDAS	Appetite Loss	2	2.19	26.45	3	0.39	0.03
ACT	Awareness	2	2.33	27.25	1	0.39	0.17
IDAS	Checking	1	5	56.17	2	0.28	0.18
IDAS	Claustrophobia	1	5.82	59.81	2	0.62	0.51
IDAS	Cleaning	1	4.27	49.77	2	0.29	0.1
IDAS	Dysphoria	2	3.57	37.91	3	0.89	1.63
IDAS	Euphoria	1	2.14	26.45	2	0.56	0.1
ACT	Experiential Avoidance	2	2.67	30.73	1	0.43	0.52
IDAS	Ill Temper	2	3.19	31.25	1	0.21	0.43
IDAS	Insomnia	2	2.76	30.04	3	0.22	−0.23
IDAS	Lassitude	2	3.31	34.1	3	0.45	0.63
IDAS	Mania	1	6.75	69.14	2	0.25	0.59
ACT	Openness	2	2.52	27.37	1	0.37	0.43
IDAS	Ordering	1	4.27	51.78	2	0.21	0.06
IDAS	Panic	2	3.54	34.69	2	0.42	0.93
IDAS	Social Anxiety	2	3.37	32.69	3	0.43	0.9
IDAS	Suicidality	2	2.74	28.75	2	1.04	0.41
ACT	Tacting Ability	2	0.81	11.95	1	0	−0.02
IDAS	Trauma Avoidance	1	6.32	63.97	1	0.58	0.95
IDAS	Trauma Intrusions	2	3.58	34.75	1	0.53	1
ACT	Engagement	2	0.73	14.92	1	0.23	−0.16
IDAS	Well-Being	2	0.55	9.91	1	0.29	−0.54

Note. EI = expected influence; association network computed with zero-order correlations, concentration network computed with partial correlations. For Expected influence, a score closer to zero is weaker.

A second network collapsing the three CompACT subscale nodes into one total score node (for a network with 21 nodes total) was also evaluated and detailed results are presented in the supplemental section. Results were largely the same in this network, with both the AAQ-II and CompACT total scores nodes being identified as central nodes of the ACT process measures, depending on which centrality metric is prioritized, and Dysphoria being more central overall.

Concentration Network

The network using partial correlations had 23 nodes and 68 total edges (see Figure 1, Panel b). The density of the network was .27, meaning 27% of all possible connections appeared in the network, indicating this is a sparsely connected network. The same five centrality metrics were examined in the partial network (see Table 2). Dysphoria and suicidality had the highest degree centrality while appetite gain, engagement, and ill temper had the lowest. For weighted degree and eigenvector centrality, dysphoria had the highest value. Appetite gain was the least central based upon weighted degree whereas engagement, well-being, and tacting of function were the least central, based upon eigenvector centrality. Both dysphoria and suicidality had the highest closeness values, and appetite gain was the most distant. Lastly, dysphoria had the highest betweenness value, whereas appetite gain was the lowest.

For the predefined community analysis, (see Table 3) in the IDAS-II community, well-being had the strongest bridge strength whereas insomnia, appetite loss, appetite gain, euphoria, social anxiety, and ordering all had bridge strengths of zero (not connected to the ACT community). For expected influence when considering indirect effects, well-being again had the strongest value while appetite gain had the weakest value. Of note, suicidality had the strongest negative value. For bridge strength in the ACT community, openness had the strongest value, and tacting of function had the weakest value. For expected influence when considering indirect effects, experiential avoidance was the strongest whereas engagement was the weakest.

Community analysis based upon auto-detected communities (see Table 4) indicated there were three communities in the network. The first was made up of the five ACT process scales, well-being, ill temper, traumatic intrusions, and traumatic avoidance. The second community was comprised of suicidality, appetite gain, mania, euphoria, panic, claustrophobia, checking, ordering, and cleaning. The third community was comprised of dysphoria, lassitude, insomnia, appetite loss, and social anxiety. In the first community, tacting of function had a bridge strength of zero, indicating it was not connected to nodes in any other community. Traumatic avoidance had the highest bridge strength. Tacting of function also had the weakest expected influence, whereas traumatic intrusions had the strongest; and although not the strongest overall, well-being had the strongest negative expected influence value. In community two, ordering had the weakest bridge strength, and suicidality had the strongest bridge strength. As for expected influence, appetite gain was weakest, whereas panic was strongest. Lastly, in community three, insomnia had the lowest bridge strength while dysphoria had the highest. For expected influence, appetite loss had the weakest value, and dysphoria had the strongest value; insomnia had the strongest negative expected influence value.

Again, a second network collapsing the three CompACT subscale nodes into one total score node was also evaluated results are presented in the supplemental section. Results differed most within the auto-detected community analyses. The CompACT and AAQ nodes were again most central of the ACT process measure nodes, depending on the metric prioritized, although dysphoria remained most central overall.

Discussion

Generally, results indicated the levels of psychopathology present are consistent with that of a community recruited national sample of adults. As such, this sample is well suited to evaluate associations among internalizing domains and ACT processes, particularly given a core tenet of ACT and larger movement in the field is moving away from symptom-focused approaches and toward a transdiagnostic approaches that prioritize human functioning (Hofman & Hayes, 2019). Understanding processes that improve human functioning regardless of symptom presentation, even in community samples, is an important research direction. Within the present undergraduate sample, it was expected that experiential avoidance would be a central node across both the association and concentration network, and Dysphoria would be a central node when it comes to the IDAS-II domains.

In sum, this was supported as dysphoria and experiential avoidance had the highest eigenvector centralities indicating they were the two nodes most strongly connected to other important nodes in the network. However, when looking at betweenness values in the concentration network specifically, Dysphoria and Suicidality had the highest values. As noted, betweenness is one way to identify bridge nodes or hubs, nodes that link many other nodes in the network. Both dysphoria and suicidality may be important symptoms or areas to target via treatment. Suicidality being a more central node was an unexpected finding; however, in the context of a college student sample with average levels of psychopathology overall (given percentile scores on the IDAS-II), this result makes sense. This is especially true given college students are at higher risk for experiencing both psychopathology generally and suicidality specifically (Mortier et al., 2018). Additionally, suicidality may also be considered a transdiagnostic symptom in that it is a diagnostic criterion for several psychiatric disorders (e.g., depression, borderline personality disorder) and is considered a consequence of many more; further, there are genetic, environmental, and socio-cultural risk factors for suicidality (Fehling & Selby, 2021). All of this may contribute to why suicidality appeared as a more central node across networks.

In the association network, the predefined community analysis (i.e., ACT process community and IDAS-II community) revealed that experiential avoidance and dysphoria had the largest direct and indirect influence on the other community. This indicates that overall, experiential avoidance has the largest influence on internalizing domains, and dysphoria has the largest influence on ACT processes. In the concentration network, however, both experiential avoidance and openness had nearly identical direct influences, while experiential avoidance had the highest indirect influence. As for the IDAS-II domains, it was well-being that had the highest direct and indirect influence. Interestingly, once all connections were accounted for, well-being had the greatest influence on ACT processes. This finding is consistent with previous literature, which has shown psychological flexibility to associate with well-being (Brassell et al., 2016; Marshall & Brockman, 2016; Wersebe et al., 2018). In fact, a recently published meta-analysis that examined the efficacy of ACT interventions on improving well-being (Stenhoff et al., 2020). Overall, the findings of the review indicate that ACT is helpful in improving well-being in both non-clinical and clinical populations.

It is important to mention the influence of experiential avoidance in both the concentration and association networks, As stated previously, experiential avoidance, as measured by the AAQ-II, has been criticized as being a measure of psychological distress or negative affect (Wolgast et al., 2014; Rochefort et al., 2017; Tyndall et al., 2019; Ong et al., 2020). This may be why experiential avoidance demonstrated high influence in the networks and should be considered when interpreting the results of the current study.

For the auto-detected community analysis, two communities were found in the association network. The first community was made up of the bipolar scales (mania and euphoria), five anxiety scales (claustrophobia, traumatic avoidance, checking, ordering, cleaning), and one symptom of major depression: appetite gain. The second community was made up of the collapsed ACT processes per the hexaflex (experiential avoidance, openness, awareness, engagement, tacting ability), well-being, symptoms of major depression (appetite loss, ill temper, lassitude, insomnia, suicidality) and anxiety symptom domains (panic, social anxiety, and traumatic intrusion). These findings indicate that the ACT processes may be more helpful in the treatment of major depressive disorder and/or panic, social anxiety, and intrusive thinking than bipolar disorder or obsessive/compulsive-spectrum disorders. However, another possibility may be that the auto-detected communities identified scales based on initial development and validation. The original IDAS scale contained 11 subscales (dysphoria, social anxiety, lassitude, ill temper, traumatic intrusions, panic, suicidality, insomnia, appetite loss, appetite gain, and well-being; Watson et al., 2007), all of which were included in the second community, except for appetite gain. When the expanded version of the IDAS (the IDAS-II) was developed, seven new scales were created (traumatic avoidance, checking, ordering, cleaning, claustrophobia, mania, and euphoria; Watson et al., 2012), which make up the scales of the first community. Given the auto-detected community clusters, and the clusters of the scales when developed, it may be difficult to discern if the patterns in the nodes reflect true associations between ACT processes and psychopathology or represent artificially inflated connections resulting from the IDAS development.

Regarding the auto-detected community analysis in the concentration network, three communities were found. The first was made up of all the ACT processes and well-being, as well as ill temper, traumatic intrusions, and traumatic avoidance. The second community (suicidality, appetite gain, mania, euphoria, panic, claustrophobia, checking, ordering, and cleaning) seems to relate to activated/arousal-based symptoms. Interestingly, this is in line with the proposed Hierarchical Taxonomy of Psychopathology (HiTOP) system, where symptoms of mania are thought to associate with obsessive-compulsive symptoms like checking, ordering, and cleaning (Kotov et al., 2017, 2021). Finally, the third community (dysphoria, lassitude, insomnia, appetite loss, and social anxiety), appears to group together low arousal or low positive emotion-based symptoms. In contrast to the association network, the concertation network appears to be less affected by method effects and measurement artifacts and appears to show a clearer communities of symptoms related to avoidance and trauma memory, high arousal, and low arousal.

COVID-19

Finally, it is important to consider the findings of this study in the context of the COVID-19 pandemic. Data collection took place during the Fall semester of 2020. Participants in the study were enrolled in an undergraduate psychology course, which included attending an online lecture and in-person discussion section. This study took place before COVID-19 vaccinations and non-hospital-based treatments (e.g., Paxlovid) were available. There is evidence that college students in the U.S. experienced increases in depressive and anxiety symptoms and stress during the Spring and Fall 2020 semesters (Charles et al., 2021; Roche et al., 2022). Interestingly, however, all IDAS-II subscale T-scores in the current study were within one standard deviation of the normative mean. The symptoms reported may have differed during the period of lockdowns during COVID-19. Moreover, there is some literature that indicates that college students may be more likely to endorse symptoms of stress, in comparison to depression or anxiety symptoms (Ramón-Arbués et al., 2020). Some of the contextual stressors that college students experienced may have resulted in reports of higher perceived stress but may not have presented as clinical or diagnostic symptoms. It is also important to consider that the COVID-19 pandemic has been referred to as a collective traumatic event (Sanchez-Gomez, et al., 2021), which may have had an impact on the trauma subscales of the IDAS-II (traumatic intrusions and traumatic avoidance). Unfortunately, however, the IDAS-II does not require participants to specify which event or memory respondents are considering when answering. Thus, it cannot be determined whether respondents had the COVID-19 pandemic in mind when responding to the items that measure traumatic intrusions (e.g., “I had nightmares that reminded me of something bad that happened” and “I got upset thinking about something bad that happened”) and traumatic avoidance (e.g., “I tried to ignore upsetting memories” and “I avoided talking about bad experiences from my past”).

Limitations

Despite the strengths of the study, there are several limitations to note. First, including partial correlations for the concentration network yields important information about the present sample, however partial correlation networks have received criticisms as well. Importantly, these networks are often not stable over time or across samples and thus often may be unable to be replicated (Forbes et al., 2017a, 2017b). Second, the skewed data on the IDAS-II (i.e., floor effects for suicidality and claustrophobia), although in line with national community samples and what’s expected, can impact the computation of centrality metrics and make them less stable. As such, the centrality metric results related to suicidality and claustrophobia should be interpreted with caution. Next, the reliability of the TOF was low. This may also contribute to unstable model estimates and may indicate the need to refine this scale in future research; however, this was out of the scope of the present study. Lastly, an important limitation of the present sample that should be noted is that it is non-clinical, majority White (82%) and cisgender, and results may not generalize to people of color, gender diverse individuals or other marginalized identities, or more broad community and clinical samples. It will be important moving forward to evaluate these questions within clinical and treatment seeking samples, and diverse samples including diversity based on race, socioeconomic status, gender identity, and other minoritized identities.

Clinical Implications

Although not a controlled trial or intervention study, findings from the current study suggest the potential importance of using a process-based approach for individuals experiencing internalizing symptoms. It has been suggested that network analysis may help clinicians generate accurate case conceptualizations and select appropriate strategies for treatment (Hoffman, Curtiss, Hayes, 2020). To implement this approach, clinical science needs to move away from symptom-based, disease models, and move toward process-based models (Hofmann & Hayes, 2019). This is important for bridging across treatment modalities, cultures, and contexts. Findings from this study, although not at the individual idiographic level, do suggest that individuals experiencing symptoms of dysphoria and suicidality may benefit from targeting experiential avoidance, openness, and awareness. As such, treatment may involve implementing strategies to increase one’s willingness to experience these internal experiences (via acceptance and defusion), and to be aware of the function of their behaviors following these experiences (via present moment awareness and self-as-context). Findings from the current study also indicate that to promote well-being, treatments should focus on increasing values-based behaviors. Though the ACT process measures used to examine engagement in the current study focus mostly on behaviors associated with values, a natural extension in clinical practice would include increasing values identification as well (i.e., who and what is important, qualities of action). However, these hypotheses must be tested through individual level idiographic models within treatment-providing contexts.

Conclusions

The current study employed a network analysis approach to gain insight into the interconnection between ACT processes (experiential avoidance, openness, awareness, engagement, and tacting of function) and internalizing symptoms among 447 undergraduates. Both zero-order and partial correlations were examined. Findings indicate that well-being had the greatest influence on the ACT processes, which is consistent with previous literature. Findings also revealed Dysphoria and Suicidality to have the highest values when looking at betweenness, meaning that they were the two nodes with the most influence on other nodes in the network. Although treatment effects could not be evaluated in this study, these results suggest that these two internalizing symptoms may be particularly important when it comes to treatment as interventions aimed at reducing dysphoria and suicidality may also have downstream effects and reduce other internalizing symptoms as well. Within the process-based therapy literature, Hayes et al. (2019) specifically note the importance of using network analyses to analyze process-focused data at an idiographic level, which would be an important extension of the present work. Though not available in the present dataset, Hayes et al. (2019) note the importance of dynamic network models and including longitudinal data to understand how processes unfold overtime, influence one another, and may influence symptom or treatment trajectories. This will be an important step for future research, particularly the idea of collecting repeated process-based measures from patients to create individual-level dynamic network models. This would allow for an empirical test of the change in processes over time, within and across individuals, regardless of the presenting concerns or treatment modalities used, in line with the goals of extended evolutionary meta-model (EEMM; Hayes et al., 2020). Additionally, future studies should continue to examine the association of ACT processes and internalizing symptoms, particularly within samples whose participants are actively engaged in therapy, clinically referred, and have more diverse identities. These considerations are imperative to understanding the generalizability of the present findings.

Data Accessibility Statement:

In congruence with the IRB approved protocol, we will share de-identified and aggregated data with researchers upon reasonable request.