In 2013 the Australian Act Belong Commit team published a self-help Guide to Keeping Mentally Healthy that included a questionnaire to assess how much individuals acted (4 items), belonged (5 items) and committed (6 items) [32]. A statewide survey in 2012 had established that each of these three measures (and their combined total) correlated moderately but statistically significantly with respondents’ scores on the Warwick Edinburgh Mental Well-being Scale (WEMWBS).

In 2016, the Danish ABC research team endeavoured to translate and qualitatively explore the Guide and the measurement tool [33]. The translated measurement tool was also included in a larger representative Danish survey to statistically validate the scale in a Danish context. However, the tool was developed with the primary purpose of enlightening the general public about what the domains of A, B and C cover, and hence was not consistent in separating behaviours from emotional states. The response options included both binary and continuous options leaving it unsuitable for scientific validation methods. Hence the Danish group embarked on developing a new measurement tool to capture mental health promoting behaviours suitable for research purposes. This interdisciplinary research group—comprising experts in public health, philosophy, social sciences, and health promotion, with expertise in mental health promotion and the ABC—led the development process, with several members co-authoring this study.

Our primary sample consisted of data from the national cross-sectional Realdania survey, including a random representative sample of Danish men and women aged 18 years and above drawn from the Danish Civil Registration System. Statistics Denmark sent an electronic letter to all sampled individuals in the period 23 May 2023–30 October 2023 with information about the study and an invitation to participate. In total, 346,426 persons were invited to participate, of whom 132,468 agreed to participate (a 38.2% response rate). Of these 132,468 participants, 7189 (5.4%) were excluded because they did not answer the questions needed for this analysis, and 6699 (5.1%) were excluded because they answered the survey in a language other than Danish. Hence, the final sample size was 119,221 participants. The age range of the final study sample was 18–103 years old. 6 individuals were in the age range of 100–103 years old. All participants, including those aged 100–103 years, completed the same digital questionnaire, with no formal use of proxy respondents.

Validation of the ABC-MHPB scale examined: (1) factor structure assessing content validity, goodness-of-fit through confirmatory factor analysis (CFA), as well as measurement invariance testing (also through CFA); (2) internal consistency by Cronbach’s alpha; and (3) relations to relevant measures. Psychometric analyses were conducted using Stata.

We hypothesized a first-order three-factor model, Act (A: items a–c), Belong (B: items d–f), and Commit (C: items g–j)—in which each observed item was specified to load on its respective latent factor, with unique measurement error terms for each item, and the three latent factors freely allowed to correlate. This specification accounts for the shared variance among items within each factor and separates true score from error variance. Although this was our main hypothesis of the scale structure, we also tested a one-factor model. CFA was estimated using Stata’s SEM (version 18; StataCorp, College Station, TX, USA) with Satorra-Bentler corrections for non-normality, i.e., vce(sbentler). As recommended by Hoyle and Panter [46], we used several fit indices including the Root Mean Square Error of Approximation (RMSEA), the standardized root mean square residual (SRMR), the Comparative Fit Index (CFI), and the Tucker-Lewis Index (TLI). Values of 0.95 or greater for the CFI and TLI were considered to reflect good model fit. RMSEA and SRMR values of 0.06 or less were considered to indicate good fit, although values up to 0.08 were considered acceptable [47]. A factor loading of at least 0.30 is considered the minimum criterion for an item [48]. We assessed measurement invariance across sex (women vs. men), age groups (18–24 years of age vs. 25–80+), and education (primary or unknown vs. youth-tertiary education), examining differences in Alternative Fit Indexes. We considered a model invariant when the respective constraint produced at most −0.01 change in CFI and TLI, paired with changes of up to 0.015 in RMSEA, and 0.030 (for metric invariance) or 0.015 (for scalar or residual invariance) in SRMR [49]. If full measurement invariance did not hold, we used partial invariance testing, which means we allowed the specific item(s) that differed between groups to vary freely while keeping the rest of the model the same. This approach keeps the overall structure intact but makes small adjustments, so that the model fits fairly across groups [50].

Distributions were examined for normality and any floor and ceiling effects. We also examined individual items and total scores for floor and ceiling effects. Instruments exhibit floor or ceiling effects if more than 15% of respondents record the lowest or highest score [51]. Next, internal consistency (reliability) was evaluated using Cronbach’s alpha, with a conventional threshold of 0.70 considered to reflect good reliability (39, 41). For short scales with few items (e.g., three-item subscales), lower alpha values are expected due to the limited number of items, and values around 0.50–0.60 can still be considered reasonable. Discriminant validity was assessed by calculating Pearson correlations between the ABC-MHPB scale and level of education. Convergent validity was assessed by calculating Pearson correlations between the ABC-MHPB scale and mental well-being, life satisfaction, health satisfaction, and loneliness. We hypothesized that ABC-MHPB average scores would be positively correlated with mental well-being, life satisfaction, health satisfaction and education, and negatively correlated with loneliness [24,26,52]. The size of correlation coefficients was based on Cohen’s rule of thumb, i.e., weak: r = 0.1; moderate = 0.3; strong = 0.5 [53].

Finally, we conducted simple linear regressions with mental well-being and loneliness, and we also proposed an alternative practical operationalization reflecting high versus low ABC engagement, where associations were investigated with the categorical three-level mental well-being scale (multinomial logistic regression) and the binary loneliness scale (logistic regression).

Sample characteristics are shown in Table 1. The mean age of the study sample was 52.1 years (SD = 19.1), and 54.9% were females.

Table 2 shows intercorrelations between the 10 ABC-MHPB items.

CFA (see Table 3) was conducted with all items, both as a single-factor and three-factor model. None of these produced good model fit results. Examination of factor loadings revealed that item h (items are listed in Supplementary Table S1) had a factor loading of less than 0.30. Hence, this item was omitted from a further CFA (see Table 3) conducted on the remaining 9 items, which produced a good model fit for the three-factor solution only: χ² (24) = 5196.1; CFI = 0.96; TLI = 0.95; SRMR = 0.048; RMSEA = 0.026. The same table shows the factor loadings and factor correlations for the 9-item three-factor model. The factor correlations suggest that Commit (C) overlaps more with Act (A) and Belong (B), than Act (A) and Belong (B) overlap with each other.

Given the 9-item model was the best fit, representing a three-factor model of mental health promoting behaviours, we further studied measurement invariance across different groups. Measurement invariance (Table 4) was supported for education. For sex, partial scalar invariance was achieved by freeing items c and e, and for age, partial scalar invariance was achieved by freeing item d. After these adjustments, differences in alternative fit indexes (ΔCFI, ΔTLI, ΔRMSEA, ΔSRMR) remained below our cut-off points.

Five individual items (a, b, c, d and g) in the 9-item ABC-MHPB scale showed ceiling effects, and floor effects were observed for three individual items (e, h and j) (Supplementary Figs. S1 and S2). Neither floor nor ceiling effects were observed for the Belong (B) and Commit (C) subscales. The Act (A) subscale displayed a tendency towards ceiling effects, although it did not qualify as having a ceiling effect according to our criterion (due to the fact that averaging scores for three items produces additional values relative to the number of response options for each item). The final 9-item ABC-MHPB scale is shown in Table 5 and illustrated in Fig. 1.

Regarding distribution, the Act (A) subscale displayed a left-skewed distribution, while the Belong and Commit subscales displayed normal distributions. The mean (SD) for each of the three subscales were as follows: Act (A) 5.4 (1.1); Belong (B) 4.3 (1.4); Commit (C) 3.3 (1.4). In terms of internal consistency (reliability) and discriminant validity (see Table 6), each factor in the ABC-MHPB scale had alphas around 0.5–0.6, and positive weak correlations with education, positive moderate correlations with mental well-being, life satisfaction, and health satisfaction, and negative moderate correlations with loneliness. All correlations were statistically significant.

Reflecting the multidimensional structure identified in the CFA, the ABC-MHPB scale should be applied as three distinct subscales (Act, Belong, and Commit), with subscale means serving as the primary scoring method, as this approach most accurately reflects the underlying model. This represents the standard scoring approach. In what follows, we also propose an alternative practical operationalization to aid interpretation.

Finally, linear regressions using the continuous subscale means (rounded to whole numbers) were estimates with continuous outcomes on mental well-being and loneliness. The results consistently showed that Belong (B) was more strongly associated with the outcomes than Act (A) and Commit (C). With regard to loneliness, associations for the two were similar, whereas for mental well-being, Act (A) was more strongly associated with the outcome than Commit (C). Further, a practical operationalization of the ABC-MHPB scale was proposed, based on weekly activity (corresponding to at least 1–2 times per week) on each Act (A), Belong (B), and Commit (C) subscale mean (subscale means were rounded to whole numbers). Weekly activity was chosen because the most frequent responses to the majority of the items were weekly activity and because prior research has indicated that mental health benefits are experienced particularly when engaging in ABC-related behaviours on a minimum weekly basis [54,55]. At least weekly engagement at the domain-level was defined as ‘high engagement’ and below weekly engagement at the domain-level was defined as ‘low engagement’. The subscale mean represents an estimate of the domain itself: the individual items function as indicators of that broader behavioural tendency, so the mean score reflects the domain-level engagement rather than just the simple average of the specific activities.

A four-point variable was generated with the following operationalization: 0 (low engagement in A and B and C); 1 (high engagement in A, B, or C); 2 (high engagement in A and B, A and C, or B and C); 3 (high engagement in A, B and C). The distribution was as follows: 0-low engagement in A, B, and C: 12,803 (10.7%); 1-high engagement in either A, B, or C: 45,771 (38.4%); 2-high engagement in AB, AC, or BC: 43,093 (36.2%); 3-high engagement in A, B, and C: 17,554 (14.7%).

Subsequently, a multinomial logistic regression model was performed, with the three-level mental well-being variable as the outcome (relative to the moderate mental well-being group), and a logistic regression model with the binary loneliness variable as the outcome. For each, the predictor was the Act Belong Commit (ABC) variable, with high engagement in either A, B or C as the reference category. The models were adjusted for sex (binary), age (continuous), country of origin (categorical), education (categorical), employment status (binary), living arrangements (categorical), region (categorical), and health satisfaction (continuous).

Table 7 shows the results of the regressions. People who had low ABC engagement were twice as likely to have low mental well-being, and almost 50% more likely to be lonely, as compared to people who had high engagement in either A, B or C. People who had low ABC engagement were also about 30% less likely to have high mental well-being. Regarding people who engaged in more ABC domains than the reference group: People who had high engagement in two out of three ABCs were about 60% less likely to have low mental well-being, and about 50% less likely to be lonely. They were also about 75% more likely to have high mental well-being. Finally, people who had high engagement in both A, B, and C were about 70% less likely to have low mental well-being and 70% less likely to be lonely, and they were more than two and a half times (OR = 2.68) more likely to have high mental well-being.

To our knowledge, no validated instruments exist that specifically target mental health promoting behaviours as a distinct construct. Some existing measures include items that may indirectly reflect such behaviours, such as items on physical activity, social engagement, or purpose in life, but these are typically embedded within broader well-being or lifestyle scales and are not structured around a coherent theoretical framework. Moreover, these instruments often focus on outcomes or attitudes rather than on behaviours that can be actively promoted to support mental health. The ABC-MHPB scale addresses this gap by offering a brief, theory-driven measure that captures concrete behaviours aligned with the Act (A), Belong (B), and Commit (C) domains. The ABC-MHPB scale may serve as a useful tool for monitoring population-level engagement in mental health promoting behaviours, evaluating the impact of interventions, and exploring behavioural pathways to mental health outcomes. However, compared to broader well-being measures, the ABC-MHPB scale may not capture more nuanced psychological or emotional dimensions. It focuses on frequency of behaviour and not on motivational or contextual factors.

While the ABC-MHPB scale offers a theory-driven approach to measuring and monitoring mental health promoting behaviours, we acknowledge that it does not represent an exhaustive list of all behaviours that may support mental health. The scale is grounded in the Act Belong Commit (ABC) framework, which emphasizes various forms of active engagement. However, other important behaviours such as adequate sleep or healthy eating also play significant roles in promoting mental health but fall outside the scope of this scale.

Several limitations should be considered. One limitation concerns the scale’s face validity, which could have been explored more thoroughly. As part of the pilot testing, 20 qualitative interviews were conducted by trained interviewers at Statistics Denmark to explore respondents’ understanding of the questionnaire. While the interviews contributed to assessing face validity, they focused on the overall questionnaire rather than specifically on the items developed for this scale, which may have limited the depth of insight into those particular items. Future research could benefit from a more extensive cognitive testing phase to further strengthen the scale’s face validity. The response rate of the survey was relatively low (38.2%). While this is not uncommon in large-scale population surveys, it raises the possibility of selection bias. In line with this, the current validation was based on a single, albeit large, sample of an adult population in Denmark. If individuals with lower levels of mental well-being were less likely to participate, and were also less likely to engage in Act, Belong, and Commit behaviours, the sample may underrepresent individuals with lower well-being and lower behavioural engagement. This could result in overestimation of average engagement levels and average well-being in the population. In addition, restricted variability due to underrepresentation of individuals at the lower end of the well-being distribution could attenuate observed associations between behavioural engagement and well-being. However, if respondents were disproportionately health-conscious individuals with relatively high engagement but varying well-being, associations could also be overestimated. Thus, mean levels and effect sizes should be interpreted with this in mind. Future research should aim to improve participation rates and assess potential non-response bias, for example by comparing respondents and non-respondents on available demographic characteristics or through registry linkage where feasible. Longitudinal studies would also help clarify the directionality of associations between engagement in mental health promoting behaviours and subsequent changes in mental well-being. Additionally, studies specifically targeting populations with lower levels of well-being may provide important insights into whether the observed associations differ in more vulnerable groups. Finally, the current study was cross-sectional in nature, and the scale has not yet been tested for its sensitivity to change over time, and the regression results are subject to the possibility of reverse causality or a third variable driving associations.

Future research should investigate the applicability and psychometric properties of the ABC-MHPB scale in other populations, including adolescents and individuals with specific mental health vulnerabilities. Replicating the validation process in more diverse and international samples will be essential to assess the scale’s generalizability across cultural and socio-demographic contexts. To determine the scale’s suitability for use in intervention studies, such as pre-post evaluations or longitudinal designs, future studies should examine its test-retest reliability and sensitivity to change. Establishing these properties will clarify whether the scale can effectively detect changes in mental health promoting behaviours over time. Furthermore, prospective and intervention studies are warranted to explore associations between ABC-MHPB scores and mental health outcomes.

Practically, the ABC-MHPB scale offers a standardized method for comparing mental health promotion initiatives and interventions. Within organisations applying the ABC framework, it may support the identification of interventions that most effectively foster behavioural engagement. If adopted internationally, the scale could facilitate cross-national comparisons. The scale has been forward-backward translated into English to enable the inclusion in this paper but has not been tested and validated in other contexts or languages than Danish. If the scale is being used internationally, thorough translations and validations of the scale are needed (see Availability of Data and Materials section for how to register use of the scale).

Conceptually, the ABC-MHPB scale advances the field by defining and operationalizing mental health promoting behaviours as a distinct and measurable construct. The development and validation of this scale confirms the theory that three distinct, yet related behavioural factors belong under the same overall multidimensional ABC-behavioural construct. This emergent construct enables systematic investigation of such behaviours in empirical research and contributes to the broader theoretical understanding of mental health promoting behaviours.