Social cognitive theory follows an agentic perspective toward human psychosocial functioning, highlighting the human agency fundamental role in individuals’ self-development, adaptation, and change over time (Bandura, 2001). Human agency is defined as the human capability to intentionally influence one’s functioning and the course of life through one’s actions (Bandura, 2006b).

According to Bandura (2006b), this construct includes four core properties – intentionality, forethought, self-reactiveness, and self-reflectiveness. Nevertheless, in 2018, Bandura embedded the intentionality property in the agency construct (e.g., stating that “to be an agent is to intentionally produce certain effects by one’s actions”, p. 130) and described the human agency as subsuming three core properties (i.e., forethought, self-reactiveness, and self-reflectiveness). We followed Bandura’s (2018) understanding of human agency in this work. In forethought, individuals extend their agency in time, setting goals, designing action plans consistent with their goals, and anticipating the potentially achievable outcomes to foster their motivation (Bandura, 2006b, 2018). For example, individuals can proactively take steps to mitigate CC by anticipating the potential long-term consequences of their actions on the environment and future generations. In self-reactiveness, individuals activate their self-regulation processes (e.g., monitoring their CC mitigation behaviors and efforts, making adjustments in their plans while performing it, attributing themselves rewards or sanctions) to execute the pre-established action plans and ensure control over their behavior against their standards (Bandura, 2001, 2018). Finally, in self-reflectiveness, individuals reflect upon their behavior and evaluate their functioning (i.e., personal efficacy, values, and the meaning of their purposes, for example, CC mitigation). Through this metacognitive exercise, individuals examine their thoughts and actions and consider possible corrective adjustments (e.g., alternative strategies and pro-environmental behaviors) favoring the reach of their pursuits (Bandura, 2006b). In sum, according to Bandura’s theory of human agency, individuals are characterized as being simultaneously forethinkers, self-regulators, and self-examiners, which allows them to adopt an agentic approach in their lives (Bandura, 2018).

Moreover, these three core properties can be displayed through three modes of agency – individual, proxy, and collective (Bandura, 2006b, 2018). The individual mode includes actions and activities over which individuals exercise direct control (e.g., daily individual CC mitigation behaviors such as reducing consumption, unplugging electronic devices not in use, using public transportation, and adopting a vegetarian diet; Koskela and Paloniemi, 2023). However, actions addressing a complex environmental problem, such as CC relying exclusively on the individuals’ control, are limited. For example, individuals do not directly write the laws of the country or set the policies and practices followed in institutional contexts, such as in their workplaces or, in the case of adolescents, in their schools or home, where adults usually lead decisions.

All considered, contextual factors may favor or hinder individual agency. For example, using public instead of private transportation is only possible if these are available in the community; also, adopting a vegetarian diet may be difficult to sustain if vegetarian options are not part of the menu of restaurants, stores, and local markets (Koskela and Paloniemi, 2023). To overcome these contextual barriers, individuals can exercise their socially mediated proxy agency by “influencing others who have the resources, knowledge, and means to act on their behalf to obtain the outcomes they desire” (Bandura, 2018, p.131). In other words, individuals can act through better-equipped or positioned individuals to propose solutions favoring pro-environmental behaviors (e.g., reaching environmental institutions or politicians whose actions directly impact society). Furthermore, specific goals and purposes can only be achieved through group effort (e.g., activists’ actions to draw attention to the triggers of CC, which are rooted in human activity such as industrialized activities and low-efficient energy use, responsible for high emissions of greenhouse gasses). When working together with others and sharing common goals, individuals are likely to follow a multiagent model of agency termed collective agency (Bandura, 2018). In sum, while acting intentionally, individuals can exercise their agency: (i) on their own, (ii) through influencing people with knowledge and resources to act on their behalf when it is not in their power to do something, and (iii) merging their efforts with group efforts to achieve common major goals. This way, people could follow an ecological approach, increasing the impact of their direct and indirect actions on diverse situations as CC mitigation.

The current study followed the literature recommendations to develop scales assessing agency (e.g., Boateng et al., 2018; Tsang et al., 2017; Yasir, 2016). Concretely, we followed two steps before building our instruments: (i) an analysis of the literature on the instruments addressing human agency and (ii) instruments on human agency in the domain of CC.

Firstly, we conducted a thorough literature review on human agency. Searches using several combinations of the terms “agency,” “human agency” and “agentic” AND “instrument,” “questionnaire,” “survey,” and “scale” were run on Scopus, Web of Science (all databases), ERIC, and Google Scholar databases. As a relevant finding of these systematic searches, we highlight the recent literature review of Cavazzoni et al. (2022), addressing the quantitative instruments used to measure children, youth, and adults’ agency across distinct contexts. The detailed analysis of the 34 included studies revealed that researchers grounded their studies on diverse theoretical frameworks for agency, which translated into the adoption of distinct definitions and instruments.

Despite distinct and focusing context-related specificities, the definitions of human agency cited in this pool of papers seem to share core characteristics, such as intentionality, choice, and control over actions (e.g., Bryan et al., 2014; Zimmerman et al., 2019). Moreover, most of the instruments used in the sampled studies were built to assess diverse forms of agency (e.g., critical, political, sexual, and moral agency). Still, instruments built originally to assess other constructs were also used to measure agency within the targeted populations (e.g., Snyder’s Children Hope Scale, Poteat et al., 2018; Adolescent Autonomy Questionnaire, Beyers et al., 2003; Racial Cohesion Questionnaire targeting adults, Bentley-Edwards, 2014). The analysis of the retrieved studies also revealed that, in some studies, other constructs were added to the realm of the agency construct, such as self-efficacy (Alkire, 2005; Hitlin and Elder, 2006), autonomy (Beyers et al., 2003), and empowerment (e.g., Alkire, 2005; Berhane et al., 2019; Pick et al., 2007; Zimmerman et al., 2019). Moreover, it was also possible to identify different operationalizations of the agency construct, for example, as a single variable (e.g., Lautamo et al., 2020; Steckermeier, 2019) or as a composite variable comprising varied dimensions (e.g., dispositional, motivational, and positional, see Vaughn et al., 2020b), domains (e.g., voice, behavioral control, decision making, see Zimmerman et al., 2019) or indicators (e.g., perceived control, sense of self-efficacy, work ethic, see Burger and Walk, 2016).

The overall results of our database searches revealed a similar trend to that reported in Cavazzoni et al.’s (2022) review. Research on agency has extended its reach to cover several populations (i.e., children, youth, and adults), contexts (e.g., educational, moral, and political), and domains (e.g., career development, e.g., Betz and Hackett, 1987; Yoon, 2011, 2019; exercise and physical activity, e.g., Blacksher and Lovasi, 2012; Shields and Brawley, 2006; education, e.g., Jääskelä et al., 2016; Luo et al., 2019; Stenalt and Lassesen, 2022; Vaughn et al., 2020a; environment, e.g., Kachel, 2012; Oliveira et al., 2015). This extended coverage of the agency construct may help explain the high variability found in the definitions and instruments used.

Interestingly, just four studies in our pool were framed on Bandura’s (2006b, 2018) Human Agency Theory: an unpublished doctoral dissertation (Yoon, 2011), one empirical paper (Code, 2020), and two theoretical papers (Koskela and Paloniemi, 2023; Yoon, 2019). For the purposes of the current research, we describe briefly the two empirical studies and discuss the contributions of the two theoretical papers for the CC domain. Yoon (2011) developed the “Assessment of Human Agency” (AHA) instrument. The AHA was purposefully designed to assess the agency of adults (i.e., employees and traditional and non-traditional university students), addressing aspects related to individual performance in an organizational context. Centered on the career domain, this instrument allows mapping individuals’ strengths and deficiencies among the four core properties of individual agency mode (Bandura, 2006b).

Code (2020) developed the Agency for Learning Questionnaire (AFLQ) to assess multidimensional aspects of undergraduate students’ agency in learning. This author grouped the items (extracted from other instruments) according to the four agency core properties (Bandura, 2001, 2006b): intentionality, forethought, self-reactiveness [self-regulation], and self-reflectiveness [self-efficacy] (see Code, 2020, pp. 2–3). Notwithstanding, the selected items do not represent all aspects included in Bandura’s definitions of the core properties. Both instruments (i.e., AHA instrument, Yoon, 2011, and AFLQ, Code, 2020) are exclusively centered on the individual agency mode.

The two theoretical papers focused on exploring and providing relevant inputs on human resource development (Yoon, 2019) and conceptualizing human agency for sustainability transformations (Koskela and Paloniemi, 2023). For example, the article of Koskela and Paloniemi (2023) contributes to bridging Bandura’s theory with the CC domain, which was particularly relevant for our purposes.

Secondly, we narrowed our focus while searching for literature addressing instruments on human agency in the domain of CC or environment. The searches combined “climate change,” “human agency” and “measures,” “scales” or “instruments.” The sample of studies found was not grounded on Bandura’s theoretical framework and followed a qualitative approach (e.g., Blanchet-Cohen, 2008; Muller and Wood, 2021; Oliveira et al., 2015; Toivonen, 2022; Trott, 2020). In sum, based on the results of the searches performed, no instrument theoretically grounded in Bandura’s theory has been developed to assess adolescents’ three modes of agency toward CC.

Adolescence is a pivotal developmental phase marked by the development of the capacity to think beyond concrete phenomena – abstract thinking (e.g., Byrnes, 2006), and an emerging connection and interest in global society problems (as CC) in a more independent way (Holden, 2007). Throughout this developmental period, parallel to the development of diverse skills (e.g., emotional intelligence, critical thinking, socialization, self-regulation; Zimmer-Gembeck and Collins, 2006), adolescents acquire a growing understanding of the causes and consequences of broader sustainability and environmental issues while gradually experiencing autonomy in decision-making and behavior (Ojala and Lakew, 2017). The capacity to think abstractly and critically about certain phenomena allows them to make choices that fit their understanding of their world and their role as citizens (Zimmer-Gembeck and Collins, 2006). As citizens of today and of the future, adolescents can either exacerbate CC through unsustainable lifestyles or contribute to its solution by adopting climate-friendly habits, influencing their parents and peers, and advocating for change through political protest (O’Brien et al., 2018; Metzger et al., 2010). However, despite facing the same developmental changes, not all adolescents are in the same stage of development (e.g., middle and high school students), and neither exhibit the same level of knowledge, concern, and willingness to act in similar areas as in the case of CC. Literature indicates that although adolescents’ CC knowledge tends to increase with age, their willingness to act on this problem tends to decrease (Lee et al., 2020). On the other hand, some authors (e.g., Olsson and Gericke, 2016; Otto et al., 2019) describe a fluctuation in environmental concern across adolescence that tends to be reestablished during emerging adulthood, called “adolescence dip.” For these reasons, adolescence could be a critical period since it could influence climate engagement in adulthood (Nash et al., 2020).

To address CC, it is necessary to mitigate its triggers rooted in human activity responsible for high greenhouse gas emissions (Swim et al., 2011). As Sanson et al. (2019) alert, this is even more relevant for adolescents, given that younger generations will suffer the most consequences of CC. However, not all adolescents are equally interested and committed to combat CC. Importantly, despite the increasing development of several skills throughout adolescence (e.g., Byrnes, 2006; Zimmer-Gembeck and Collins, 2006), empirical data also indicate that this period can also be marked by a decrease in CC action (e.g., Lee et al., 2020; Otto et al., 2019). These data from developmental and environmental psychology literature underscores the need to study adolescents in different stages of development, in the case of the current study, early and middle adolescence, that approximately correspond to middle and high school levels.

Bandura’s theory of human agency provides insights into the CC problem by emphasizing the role of the individual, proxy, and collective agency in initiating and sustaining action and behavior change (Bandura, 2018). These agency modes are particularly relevant when dealing with CC, where solutions must be built at individual and collective levels and by influencing people with knowledge, resources, or decisive power (Bandura and Cherry, 2020; Ojala, 2012; Sanson et al., 2019).

However, to the best of our knowledge, literature lacks an instrument (i) assessing the three agency modes proposed by Bandura, (ii) targeting adolescents, or (iii) focusing on the CC mitigation domain. To fill this three-fold gap, the current research aims to develop and validate new theoretically based scales for the Portuguese population, assessing adolescents’ individual, proxy, and collective modes of agency toward CC. Following research guidelines for developing and validating instruments (e.g., Boateng et al., 2018; Tsang et al., 2017), we conducted the current study in two phases: (i) development process of the AGENcy Toward Climate Change Scales (AGENTC2-S), and (ii) AGENTC2-Scales testing and validation.

These new scales for assessing the three agency modes are grounded in the human agency theory by Bandura (2018); therefore, we hypothesized that each scale has a multidimensional and multilevel structure (H1). Specifically, the first level encompasses three factors, i.e., properties (i.e., forethought, self-reactiveness, and self-reflectiveness), loading on a second-order factorial level, the overall agency mode (i.e., individual, proxy, or collective, see Figure 1 – Model 3a-c). Secondly, we hypothesized the invariance of this model across sex (H2) and grade level (H3), i.e., the model measures the same construct for girls and boys, as well as for adolescents in different grade levels (7th–12th).

Concerning reliability, we hypothesized that individual (H4), proxy (H5), and collective (H6) agency mode scales show high internal consistency at the first and second waves of data collection (Cronbach’s alpha ≥ 0.70). We also hypothesized high correlations between both data collection waves for each agency mode scale, indicating test–retest reliability (H7–H9).

To confirm convergent validity, we tested whether agency modes were positively and statistically correlated to theoretically related variables. For example, prior research has shown positive relationships between pro-environmental or CC mitigation behaviors and CC self-efficacy (e.g., Busch et al., 2019), nature connectedness (Krettenauer et al., 2020), and CC concern (e.g., Lawson et al., 2019). Hence, we hypothesized that adolescents’ agency modes toward CC are positively related to CC mitigation behaviors (H10), self-efficacy to combat CC (H11), nature connectedness (H12), and CC concern (H13).

Lastly, literature on learning agency, self-regulation, and pro-environmental or CC mitigation behaviors shows similar results. Girls tend to report more agentic and self-regulation learning behaviors (e.g., Martins et al., 2024; Motie et al., 2012; Zimmerman and Martinez-Pons, 1990) and pro-environmental or CC mitigation behaviors (e.g., Fielding and Head, 2012; Zeeshan et al., 2021) than boys. In addition, throughout adolescence, pro-environmental or CC mitigation behaviors tend to decline (e.g., Krettenauer, 2017; Negev et al., 2008). All considered, we hypothesized that girls report higher agency toward CC than boys (H14), and agency scores decline throughout schooling (H15).

To build questionnaires to assess adolescents’ modes of agency toward CC, first, we developed an item pool based on the AHA instrument (Yoon, 2011) that is grounded on Bandura’s conceptualization of agency. These efforts were preceded by an extensive literature review on the human agency construct (see section “Literature review: Instruments assessing agency”). In the following step, to improve item formulation and calculate content validity, we asked experts on human agency to evaluate the quality of a pool of items (i.e., the AGENTC2-S items) selected to measure the three modes of agency. Finally, we interviewed adolescents from middle and high school levels to learn their understanding of the AGENTC2-S items. All these processes are detailed below.

After concluding the developmental process, we tested and validated the AGENTC2-S. Specifically, as we aimed to test a clear, theoretically driven factor structure of the construct of human agency applied to the CC domain, we assessed the scale’s factor structure using a confirmatory factor analysis (CFA). We also assessed the measurement invariance, reliability, and convergent validity of the scales.

Tables 1–3 show correlation coefficient data between the items of each agency mode. Table 4 presents the fit results of the factor models of the three agency modes (individual, proxy, and collective), and Table 5 presents the Standardized Regression Weights. Data show that the unifactor model has a limited and poorer fit than the other two models. All statistics show that models 2 and 3 present a better fit, including the AIC and BIC scores (individual agency: ΔAIC = 180.36, p < 0.001; ΔBIC = 165.31, p < 0.001; proxy agency: ΔAIC = 224.68, p < 0.001; ΔBIC = 209.63, p < 0.001; collective agency: ΔAIC = 677.30, p < 0.001; ΔBIC = 662.25, p < 0.001), which show an optimal and similar fit for the three agency modes. Therefore, models 2 and 3 are equally suitable to represent the factor structure of the scales for each agency mode.

To study the invariance by sex and grade level, we used the two-factorial level model as the reference model because model 3 is more complete and presents a similar fit to model 2. The invariance by sex and grade level was examined for each agency mode (individual, proxy, and collective agency).

The sex invariance analysis of individual agency indicates that the fit of the base model is good [χ²(102) = 289.66, p < 0.05; TLI = 0.970; CFI = 0.977; RMR = 0.030; RMSEA = 0.041]. Acknowledging this finding, invariance was analyzed considering the (i) factorial weights (measurement weights) (ΔCFI = 0.000, ΔRMSEA = 0.002), (ii) structural weights (ΔCFI = 0.000, ΔRMSEA = 0.000), (iii) Structural model (structural covariances) (ΔCFI = 0.000, ΔRMSEA = 0.000), (iv) structural residuals (ΔCFI = 0.000, ΔRMSEA = 0.001), and (v) residuals of the measurements (measurement residuals) (ΔCFI = 0.006, ΔRMSEA = 0.002). Considering the sex invariance of proxy agency, data indicate that the base model fit is good [χ²(102) = 277.27, p < 0.05; TLI = 0.975; CFI = 0.981; RMR = 0.026; RMSEA = 0.039]. Invariance was analyzed considering the (i) factorial weights (measurement weights) (ΔCFI = 0.000, ΔRMSEA = 0.001), (ii) structural weights (ΔCFI = 0.000, ΔRMSEA = 0.001), (iii) Covariances of the structural model (structural covariances) (ΔCFI = 0.000, ΔRMSEA = 0.000), (iv) structural residuals (ΔCFI = 0.001, ΔRMSEA = 0.001), and (v) residuals of the measurements (measurement residuals) (ΔCFI = 0.001, ΔRMSEA = 0.001). Finally, we analyzed the sex invariance of collective agency. Data indicate that the fit of the base model is good [χ²(102) = 379.99, p < 0.05; TLI = 0.968; CFI = 0.976; RMR = 0.033; RMSEA = 0.050]. Consistently, invariance was studied while considering the (i) factorial weights (measurement weights) (ΔCFI = 0.001, ΔRMSEA = 0.002), (ii) structural weights (ΔCFI = 0.000, ΔRMSEA = 0.001), (iii) Covariances of the structural model (structural covariances) (ΔCFI = 0.000, ΔRMSEA = 0.000), (iv) structural residuals (ΔCFI = 0.001, ΔRMSEA = 0.000), and (v) measurements (measurement residuals) (ΔCFI = 0.001, ΔRMSEA = 0.001). Current results indicate that considering the five dimensions assessed (i.e., measurement weights, structural weights, covariances of the structural model, structural residuals, and measurement residuals) and the three agency modes (i.e., individual, proxy, and collective) data are invariant across sex.

We analyzed the grade level invariance of individual agency; data indicate that the fit of the base model is good [χ²(102) = 298.19, p < 0.05; TLI = 0.970; CFI = 0.977; RMR = 0.030; RMSEA = 0.042]. Grounded on this finding, invariance was analyzed considering the (i) factor weights (measurement weights) (ΔCFI = 0.001, ΔRMSEA = 0.002), (ii) structural weights – that is, the relationship between the factor general (mode) and the three first-order factors (proprieties) – (ΔCFI = 0.000, ΔRMSEA = 0.000), the (iii) Covariances of the structural model (ΔCFI = 0.000, ΔRMSEA = 0.001), (iv) structural residuals (ΔCFI = 0.000, ΔRMSEA = 0.001), and the (v) residuals of the measurements (measurement residuals) (ΔCFI = 0.002, ΔRMSEA = 0.001). Regarding the grade level invariance of proxy agency, data also indicate that the fit of the base model is good [χ²(102) = 256.63, p < 0.05; TLI = 0.979; CFI = 0.983; RMR = 0.024; RMSEA = 0.037]. Therefore, we analyzed invariance regarding the (i) factorial weights (measurement weights) (ΔCFI = 0.001, ΔRMSEA = 0.002), (ii) structural weights (ΔCFI = 0.000, ΔRMSEA = 0.000), (iii) Covariances of the structural model (structural covariances) (ΔCFI = 0.000, ΔRMSEA = 0.000), (iv) structural residuals (ΔCFI = 0.000, ΔRMSEA = 0.001), and (v) residuals of the measurements (measurement residuals) (ΔCFI = 0.006, ΔRMSEA = 0.004). Finally, considering the grade level invariance of collective agency, data indicate that the base model fit is good [χ²(102) = 372.10, p < 0.05; TLI = 0.969; CFI = 0.976; RMR = 0.034; RMSEA = 0.049]. Regarding the individual and proxy agency modes, invariance was analyzed considering the (i) factorial weights (measurement weights) (ΔCFI = 0.000, ΔRMSEA = 0.002), (ii) structural weights (ΔCFI = 0.000, ΔRMSEA = 0.000), (iii) Covariances of the structural model (structural covariances) (ΔCFI = 0.000, ΔRMSEA = 0.000), (iv) structural residuals (ΔCFI = 0.000, ΔRMSEA = 0.001), and (v) residuals of the measurements (measurement residuals) (ΔCFI = 0.008, ΔRMSEA = 0.004).

Current results indicate that data are invariant across grade levels when considering the five dimensions assessed (i.e., measurement weights, structural weights, covariances of the structural model, structural residuals, and measurement residuals) and the three agency modes (i.e., individual, proxy, and collective).

The reliability of the three agency modes is very good: individual agency (αT1 = 0.94; ωT1 = 0.95; αT2 = 0.95; ωT2 = 0.95); proxy agency (αT1 = 0.94; ωT1 = 0.94; αT2 = 0.96; ωT2 = 0.96) and collective agency (αT1 = 0.95; ωT1 = 0.95; αT2 = 0.97; ωT2 = 0.97). In addition, Pearson’s correlation coefficients between T1 and T2 for individual (0.768), proxy (0.738), and collective (0.767) agency modes are high.

As Table 6 shows, the bold correlations for the three agency modes indicate convergent validity. Specifically, the three modes of agency are positively and significantly correlated with the four related variables (i.e., individual mitigation behaviors, self-efficacy, nature connectedness, and CC concern). Data indicate that the higher the agency modes scores (individual, proxy, and collective), the higher the individual mitigation behaviors, self-efficacy, nature connectedness, and CC concern, and vice versa.

The differences in the three modes of agency regarding sex and grade level were analyzed. Table 7 shows data for the means and standard deviations. Regarding the variable sex, we also found statistically significant differences [λ_wilks = 0.997; F(3,1,109) = 8.790; p < 0.001; η² = 0.023]. Girls scored higher than boys in the three agency modes: individual [F(1,111) = 25.26; p < 0.001; η² = 0.022], proxy [F(1,111) = 16.09; p < 0.001; η² = 0.014] and collective [F(1,111) = 15.08; p < 0.001; η² = 0.013]. Regarding the variable grade level, we found statistically significant differences [λ_wilks = 0.957; F(15,3,053) = 3.283; p < 0.001; η² = 0.015] there is generally a tendency showing that the agency modes decrease as the grade level grows: individual [F(5,1,108) = 2.97; p < 0.05; η² = 0.013], proxy [F(5,1,108) = 4.68; p < 0.001; η² = 0.021] and collective [F(5,1,108) = 4.41; p < 0.001; η² = 0.020]. However, considering sex and grade level, the effect sizes of the differences found are small.