Measuring psychological flexibility to promote wellbeing: validation of the Spanish Psy-Flex in a healthy sample

Full reference: Crego, A., Yela-Gómez, S., Yela, J. R., Ruiz, F. J., & Gloster, A. T. (2026). Measuring psychological flexibility to promote wellbeing: validation of the Spanish Psy-Flex in a healthy sample. Frontiers in Psychology, 17, 1808313. https://doi.org/10.3389/fpsyg.2026.1808313

Study type: Adaptation and validation of an instrument (classical and network psychometric analysis of the Spanish Psy-Flex).

Background and objectives

Psychological flexibility—the ability to be in contact with the present moment, remain open to internal and external experience, and engage in value-consistent action—is a core process of Acceptance and Commitment Therapy (ACT) and a key transdiagnostic mechanism for wellbeing. Its counterpart, psychological inflexibility (experiential avoidance, cognitive fusion, etc.), has been linked to a broad range of psychological problems. The authors note important methodological limitations in measuring this construct: the terms flexibility and inflexibility are often used interchangeably, and the most common instrument, the Acceptance and Action Questionnaire-II (AAQ-II), was originally designed to measure inflexibility, with debated discriminant validity (possible overlap with neuroticism and negative affect).

In this context, the Psy-Flex (Gloster et al., 2021) emerges as a brief, theoretically grounded, context-sensitive scale composed of six items, each referring to one of the six processes of the ACT Hexaflex (defusion, acceptance, present moment/mindfulness, self-as-context, values/presence of meaning, and committed action). Unlike other measures, its items assess the presence of each flexibility skill rather than its absence. The original scale showed a unifactorial structure, excellent reliability, and evidence of convergent, divergent, and incremental validity; subsequent adaptations (Korean, Chinese, Portuguese, and Hebrew) confirmed the single-factor structure and good psychometric properties. The instrument has already been translated into 20 languages.

There are two Spanish translations of the Psy-Flex: that of Ruiz et al. (2021), not yet empirically analyzed, and that of Navarrete et al. (2025), studied in people with chronic pain. The present study aims to analyze the psychometric properties of the Ruiz et al. (2021) translation in a sample of healthy individuals. Specifically, it examines its factor structure, internal consistency, and temporal stability (test-retest), as well as its construct validity (divergent, discriminant, and convergent) and criterion validity (correlations with measures of flexibility, inflexibility, wellbeing, and mental health), and conducts a network analysis of the six items to explore the interplay and relative importance of the Hexaflex processes.

Method

Participants

The sample comprised 170 healthy individuals, predominantly female (75.9%). Mean age was 42.81 years (SD = 15.51; range 18-65). Educational level was distributed as follows: 1.8% basic education; 13.5% professional training or secondary education; 55.3% undergraduate university studies; and 29.4% postgraduate studies. Most participants were employed (61.2%), followed by students (15.9%), unemployed (14.7%), retirees (7.6%), and 0.6% in other categories. Participants were recruited in Salamanca, Spain, in November-December 2022 through local media advertisements and social media. Data come from the MSC-Health project (ClinicalTrials.gov NCT05695586); the pre-intervention measurement scores (collected online in January-February 2023) were used primarily. For temporal stability, five measurement points were used: pre-intervention (T1), post-intervention (T2), and 3-month (T3), 6-month (T4), and 12-month (T5) follow-ups; the longitudinal N (listwise) was 119, with participants who received mindfulness-based interventions (n = 87) or formed the control group (n = 32).

Instrument under study

The Psy-Flex (Gloster et al., 2021; Spanish translation by Ruiz et al., 2021) consists of six items rated on a 5-point Likert scale, from 1 (Never) to 5 (Very often), where higher scores indicate greater psychological flexibility. In the present study, its Cronbach's alpha was 0.839.

Other outcome measures

The following were used for validation: the SCS (Self-Compassion Scale, 26 items; α = 0.921); the MAAS (Mindful Attention Awareness Scale, 15 items; α = 0.897); the Presence of Meaning subscale of the MLQ (Meaning in Life Questionnaire, 5 items; α = 0.865); the CFQ (Cognitive Fusion Questionnaire, 7 items; α = 0.920); the AAQ-II (Acceptance and Action Questionnaire-II, 7 items; α = 0.876); the Activation subscale of the BADS (Behavioral Activation for Depression Scale, 7 items; α = 0.860); the HADS (Hospital Anxiety and Depression Scales, 14 items; anxiety α = 0.831, depression α = 0.793); the PSS (Perceived Stress Scale, 14 items; α = 0.892); the PANAS (10 items per dimension; positive affect α = 0.906, negative affect α = 0.898); the SHS (Subjective Happiness Scale, 4 items; α = 0.853); and the SWLS (Satisfaction with Life Scale, 5 items; α = 0.887).

Data analysis

Factor structure was examined via Confirmatory Factor Analysis (CFA) with AMOS 24, evaluating fit with Hu and Bentler's (1999) criteria: χ²/df ≤ 3; GFI, NFI, TLI, and CFI ≥ 0.90; SRMR ≤ 0.08; RMSEA ≤ 0.10. An RMSEA-based power analysis was conducted (MacCallum et al., 1996). Reliability was assessed with Cronbach's alpha (N = 170), item-total correlations (adequate if r > 0.30), alpha-if-item-deleted, and composite reliability (CR, threshold 0.70). Temporal stability was estimated with the intraclass correlation coefficient (ICC; two-way random effects, consistency type) on the total scores across the five points. Construct validity included divergent validity (t-test by sex, correlation with age; SPSS 22), convergent validity (AVE, threshold 0.50, with Fornell and Larcker's [1981] flexible criterion), and discriminant validity (Fornell-Larcker criterion: square root of the AVE greater than the correlations). Criterion validity was analyzed with Pearson correlations interpreted following Cohen (1988). Finally, a regularized partial correlation network (EBICglasso, γ = 0.5) was estimated with the R packages bootnet and qgraph, assessing centrality via standardized node strength and robustness via bootstrapping (1000 iterations): non-parametric for the 95% CIs of edge weights and case-dropping for the centrality stability (CS) coefficient (acceptable > 0.25, ideal > 0.50).

Results

Factor structure (CFA). A unifactorial structure was tested, allowing—following the modification indices and previous studies (Gloster et al., 2021; Navarrete et al., 2025)—the correlation between the errors of items 5 (values) and 6 (committed action). The one-factor solution showed a reasonable fit: χ² = 20.435, df = 8, p = 0.009; χ²/df = 2.554; GFI = 0.961; NFI = 0.952; TLI = 0.943; CFI = 0.970; SRMR = 0.046; RMSEA = 0.096 (90% CI 0.045-0.148), PCLOSE = 0.065. The model's post-hoc power (df = 8, N = 170) was 0.75.

Reliability. Internal consistency was good (α = 0.839; 95% CI 0.812-0.866). Alpha always decreased when any item was removed, except for "committed action," whose removal would slightly improve it to 0.847. Corrected item-total correlations were adequate in all cases: mindfulness r = 0.564, acceptance r = 0.622, defusion r = 0.730, self r = 0.724, values r = 0.652, and committed action r = 0.403. Temporal stability over one year was high (ICC = 0.887; 95% CI 0.852-0.916). In the experimental conditions, the overall ICC was 0.880, with a lower value during the 8-week training period (ICC = 0.666, when scores increased) and a higher value in the subsequent periods of stability (ICC = 0.902). Composite reliability was CR = 0.83.

Construct validity. Divergent validity was confirmed: there were no significant differences between men (M = 3.33, SD = 0.81) and women (M = 3.08, SD = 0.81; t = -1.732, df = 169, p = 0.085), nor an association with age (r = -0.014, p = 0.853). The one-factor model yielded an AVE of 0.47, slightly below the ideal criterion of 0.50 but acceptable given that the CR exceeded 0.60 (Fornell and Larcker, 1981). The square root of the AVE (0.685) exceeded all correlations of the Psy-Flex with the other constructs, supporting discriminant validity.

Criterion validity. Psy-Flex total scores correlated significantly, with medium-to-large magnitudes, with all analyzed variables. Among the process variables, the strongest negative correlation was with the AAQ-II (r = -0.617), followed by the CFQ (r = -0.567) and the SCS (r = 0.596); it also correlated with the MAAS (r = 0.426), the MLQ-Presence (r = 0.404), and the BADS-A (r = 0.362). Among the outcome variables, the PSS (r = -0.625), negative affect (r = -0.594), anxiety (HADS, r = -0.577), and positive affect (r = 0.534) stood out, along with happiness (SHS, r = 0.494), depression (HADS, r = -0.455), and life satisfaction (SWLS, r = 0.398). At the item level, each item showed its highest correlation with the corresponding flexibility or inflexibility process scale.

Network analysis. Defusion emerged as the most central node (strength = 1.25), followed by self-as-context (0.92) and values (0.36), whereas acceptance (-0.50), mindfulness (-0.75), and especially committed action (-1.28) showed below-average centrality. The strongest connections were between self and defusion (weight = 0.52) and between acceptance and defusion (weight = 0.38). The bootstrap analysis confirmed a stable network: CS coefficients were 0.67 for edge weights and 0.52 for node strength, both above the ideal threshold of 0.50.

Discussion and conclusions

The authors conclude that the Spanish Psy-Flex (Ruiz et al., 2021) has adequate psychometric properties in a healthy population. The unifactorial structure fit the data well, consistent with the original scale and its adaptations; the correlation between the errors of items 5 and 6 is justified conceptually (committed action presupposes the clarification of values) and linguistically (both items use the descriptor "important"). Internal consistency was high and comparable to that of other versions. A notable contribution is the analysis of stability over a full year (ICC = 0.887), which suggests that the Psy-Flex is both sensitive to change following an intervention and stable in the absence of further gains.

Divergent validity (independence from sex and age), discriminant validity (Fornell-Larcker criterion), and an acceptable convergent validity—despite an AVE slightly below 0.50—were confirmed. The Psy-Flex's strongest correlation was, in the negative direction, with the AAQ-II, the most widely used measure of inflexibility, replicating the pattern of the original validation and the Brazil-Portugal cross-cultural validation. A strong association with self-compassion (SCS) was also found, a promising research avenue. The network analysis revealed a cohesive single-component structure in which defusion and self-as-context act as central nodes, while committed action occupies a peripheral position, possibly because its behavioral enactment depends on external environmental opportunities and barriers.

Regarding practical implications, the authors emphasize that the Psy-Flex is a brief instrument of great value in research and clinical practice, and that identifying core variables in the process network (flexibility, inflexibility, and self-compassion) is relevant for intervention design. As limitations, they acknowledge the relatively small sample size (N = 170) and a power of 0.75, somewhat below the 0.80 threshold (although the presence of several high-loading items > 0.70 provides anchor stability to the construct), as well as the predominance of women in the sample (though no sex differences were found).

Relevance to measurement in ACT/CBS

This study provides the contextual behavioral science community with a validated Spanish version of the Psy-Flex (Ruiz et al., 2021 translation) for healthy populations, extending the evidence on an instrument that directly measures the presence of psychological flexibility—rather than its absence inferred from inflexibility—and in a context-sensitive way, addressing classic methodological criticisms directed at measures such as the AAQ-II. By combining classical psychometrics with network analysis, it also offers a perspective on the internal architecture of the Hexaflex, positioning defusion and self-as-context as central processes. Its brevity, good properties, and theoretical grounding in ACT's multicomponent model make it a valuable tool for the clinical and research assessment of psychological flexibility processes.

---