Introduction

Plastic production has grown exponentially in recent decades, and plastic products have spread extensively in daily life. According to a recent review by Villarrubia-Gomez et al. (2024), in 2022, at least 506 million tons of plastic were produced worldwide, with an estimated total of over 11,000 million tons since 1950, of which 71% were recorded in the period 2000–2022. Only 10% of the plastic waste generated globally from 1950 to 2017 was recycled; 14% was incinerated, while 76% was stored in landfills or dispersed in the environment (Geyger, 2020). Notably, if plastic is burned or buried, chemical compounds are toxic to the air and soil (Ilyas et al., 2018). These and other data clearly show how plastic contributes to environmental problems on a planetary scale, from climate change to ocean pollution and acidification to changes in freshwater. Moreover, plastic comprises thousands of chemical substances, many of which are noxious and can deteriorate ecosystems and human health (Villarrubia-Gomez et al., 2024). For example, most plastic chemical compounds are persistent in the environment and potentially hazardous to the human food chain. Therefore, plastic pollution is considered among the main environmental threats by the United Nations (United Nations Environment Programme, 2021). It cannot be limited to a waste management problem; it would be an integrative part of policies on climate change and natural resource use. In sum, plastic problems and reduction are major concerns for governments and other stakeholders (European Commission, 2018a).

The lightness, versatility and durability of plastic, combined with low production costs, have made it possible and feasible to design so-called single-use products that currently comprise most retail packaging. The rise in single-use packaging coincides with the increase in plastic production (Geyger, 2020), and single-use plastics account for at least 40% of the total plastic produced each year (e.g., Parker, 2019). One of the most common forms of single-use packaging is plastic drink bottles, among which small bottles (500 ml) are used repetitively by many people at work, school or when they engage in sports or other leisure activities.

A national survey conducted in Italy highlighted that 90.3% of Italians drink bottled water (Censis Foundation, 2018). Italy ranks first in Europe and among the top three countries globally for the consumption of bottled mineral water (Legambiente and Altraeconomia, 2018). A Greenpeace study (2021) pointed out that in Italy alone, 10 billion plastic bottles of mineral water enter the market each year. Almost 70% of these—about 7 billion—are not recycled and risk being dispersed into the environment, including the seas, contributing significantly to the planet’s pollution. To reduce pollution caused by plastics, member states of the European Union are obliged ‘to collect 90% of single-use plastic drinks bottles by 2025’ and ‘to raise consumers’ awareness about the negative impact of littering of single-use plastics and fishing gear as well as about the available re-use systems and waste management options for all these products’ (European Commission, 2018b, p. 2).

Specific background: the use of refillable bottles

Encouraging the regular use of refillable water bottles for drinking outdoors is a promising way to reduce widespread reliance on plastic bottles and, consequently, to significantly decrease pollution due to plastic production and waste accumulation. In light of what we discussed above, it is particularly salient in the Italian context. To discourage the use of plastic bottles, many companies and institutions have distributed refillable bottles to their employees and students as gifts and have installed refilling stations (Uehara and Ynacay-Nye, 2018). The University of Padova began this endeavor in 2018 (University of Padova, 2018).

In the last few years, the number of people opting to use refillable bottles rather than disposable bottles has grown, and the market is expanding increasingly (Piemontese, 2020). In 2020, a study conducted by Euromedia Research showed that 65.1% of respondents used disposable bottles outside their home, while 34.9% preferred refillable bottles. The study also found that more women (36%) used refillable bottles compared to men (33.7%). It also found that 45.9% of users were between 18 and 24 years old, and that 43.6% had medium/high or high socio-economic status (Volpe, 2020). In Italy, to facilitate widespread use, refillable bottles were included in the 2021 Italian National Statistics Institute (ISTAT) basket, an annual list of the most frequently purchased goods and services that mirror typical consumer habits to monitor the inflation rate (ISTAT, 2021).

Despite legislative and normative efforts, single-use plastic bottles remain a common choice among consumers, likely due to their perceived convenience and ease (e.g., Fedi et al., 2021). Since improper disposal of plastic bottles damages the environment, there is a need to explore ways to motivate people to adopt more sustainable options in daily life. From this perspective, the use of refillable bottles can represent a key part of the solution to plastic pollution. To shift individual behaviors, it is necessary to examine the role of psychological factors, such as attitudes, beliefs and intentions. Understanding the psychological drivers behind behaviors can inform more effective interventions, campaigns and policies aimed at reducing plastic waste (e.g., Fedi et al., 2021).

Knowledge gap and research objectives

A number of socio-psychological studies have investigated the potential obstacles to consuming tap water and the preference for bottled water over tap water (e.g., Díez et al., 2018; Etale et al., 2018; Qian, 2018). However, there is limited knowledge about the factors contributing the most to the intention to utilize reusable containers, such as refillable bottles (e.g., Ertz et al., 2017; Fedi et al., 2021).

The present study adopted the theoretical framework and methodological grid of the theory of planned behavior (TPB; Ajzen, 1991) to determine the psychosocial factors that may predict the intention and behavior of using refillable bottles for drinking away from home. The TPB model was extended, with self-identity as an additional predictor of intention and past behavior as a precursor of all model constructs (see Fig. 1). This study was based on a two-wave (Time 1 and Time 2, one month later) prospective design. The first aim was to explain and predict the intention to use a refillable bottle at Time 1 (T1) and the corresponding self-reported behavior at Time 2 (T2). The second aim was to investigate the invariance of the processes leading to intention and behavior in two groups: university students and adults from the general population who were not enrolled in the university when the study was conducted. In the TPB literature, two types of research samples are typically used—students or non-student adults from the general population. Although these samples differ in a specific condition (i.e., student vs. non-student status), they may exhibit varying degrees of overlap for other variables, such as age. Some meta-analyses on health and pro-environmental behaviors used the student/non-student sample characteristic as a moderator of the relations within the TPB models (e.g., McEachan et al., 2011; 2016, Udall et al., 2021). Differences between such groups are of interest for both theoretical and practical reasons. From a theoretical perspective, it is valuable to know whether they translate into a better fit for the TPB in students compared to adult samples; from a practical standpoint, the findings can inform the design of interventions to change certain behaviors in various populations. The status of being a student vs. a non-student should denote differences in interests, lifestyle, daily routines and personal and family responsibilities that we argue could be related to the target behavior of the present study. As mentioned above, in recent years, various initiatives have been introduced in educational institutions to encourage the use of refillable bottles.

Fig. 1
figure 1

The hypothesized TPB-based relation model.

Full size image

Some studies have shown that groups of different ages presented dissimilarities regarding attitudes toward environmental issues and sustainable behaviors (e.g., Ágoston et al., 2024). For example, in an international survey, Hickman et al. (2021) found that among individuals aged 16–25 years, 59% are extremely worried or very worried about climate change and 83% believe that people have failed to take care of the planet. Doan et al. (2025) found that Generation Z (in which most university students belong) engaged in sustainable activities in everyday life more frequently than Generation Y (or millennials). In Italy, 45.9% of individuals aged 18–24 years, 37% of those aged 25–44 and 25.9% of those over 65 years prefer using refillable bottles for drinking when away from home (Volpe, 2020). In addition, a refillable bottle can serve as an accessory, such as a customizable smartphone cover. Consequently, these items are not only fashionable but also hold greater appeal for younger people than for older adults.

Different status conditions (i.e., student/non-student) and lifestyles might influence the relationships between the constructs of the hypothesized TPB model. Previous studies have rarely examined the role of sample or context characteristics as moderators in the TPB model (e.g., Canova et al., 2020a; 2025; Kekäläinen et al., 2022; Pasi et al., 2021), revealing only a few differences in the process of intention formation. However, this approach can elucidate whether intervention strategies to encourage specific behaviors need to be tailored to specific population subgroups or whether similar strategies can be effective across them.

Theoretical background and hypotheses

The theory of planned behavior

The TPB, one of the most established social cognition models, posits that a specific behavioral intention (e.g., intention to use a refillable bottle for drinking away from home) is the most proximal antecedent of the actual behavior. It indicates the extent to which individuals are motivated to enact a behavior and plan its execution. Intention is a function of three correlated constructs—attitude toward the behavior (i.e., its positive or negative evaluation), subjective norm (i.e., the belief that important others would approve or disapprove of the behavior) and perceived behavioral control (PBC, i.e., the belief in one’s ability to enact the behavior). PBC, since it may reflect actual control, can directly predict behavior. Drawing from the assumptions of TPB, we put forward the following hypotheses (Fig. 1):

H1. Intention is positively related to the future behavior of using a refillable bottle.

H2. Each of the original TPB constructs – attitude (H2a), subjective norm (H2b), and PBC (H2c) – is positively related to behavioral intention to use a refillable bottle.

H3. PBC is positively related to the future behavior of using a refillable bottle.

The TPB has been successful in predicting a wide array of pro-environmental behaviors (e.g., de Leeuw et al., 2015; Klöckner, 2013), but only a few studies have tested its efficacy in predicting the reduction of plastic consumption (e.g., Aruta, 2022; Nguyen, 2024; Nguyen et al., 2025; Oludoye and Supakata, 2024; Raimondo et al., 2022; Truelove et al., 2023; Wang et al., 2024). Among these, only a few have considered reusable products (e.g., Ertz et al., 2017) or have specifically focused on refillable bottles (e.g., Fedi et al., 2021). Overall, the results suggest that TPB adequately predicts intentions and behavior in this field. However, these studies, with a few exceptions (e.g., Truelove et al., 2023), did not adopt a prospective design—with two or more waves of data collection—and were limited to intention or considered past or current behavior as a proxy for future behavior, but this is likely to inflate the correspondence between intention and behavior (Hausenblas et al., 2008).

Although the validity of the TPB and its predictive power have been widely recognized, the original TPB model has been extended by exploring the effects of other constructs. In the relational model hypothesized in this study (Fig. 1), two constructs were added – past behavior and self-identity.

The role of past behavior in the theory of planned behavior

Considerable empirical research and theoretical considerations have focused on the role of past behavior in social cognition theories (e.g., Conner et al., 1999; Hennessy et al., 2010; Ouellette and Wood, 1998). Hagger et al. (2018) summarized the effects of past behavior as follows. As a predictor of future behavior alongside TPB constructs, past behavior usually increases the amount of variance in intentions and behaviors accounted for by the model. It directly predicts future behavior and the other TPB variables, including intention, and attenuates the effects of other constructs on intentions and behaviors. The direct effects of past behavior on future behavior may represent unconscious processes, implicit cognitions or behavioral scripts that represent sets of information developed over time and affect behavior beyond an individual’s awareness (Hagger et al., 2018). This interpretation of the effects of past behavior is consistent with dual-process theories, which assume that the execution of behavior is a function of two processes—a reasoned, planned process represented by the constructs in social cognition theories, such as TPB, and an implicit, non-conscious process represented by past behavior (Hagger, 2016). One non-conscious process modeled by past behavior is habit (Ouellette and Wood, 1998), which can be defined ‘as an action or behavioral tendency that is enacted spontaneously, with little conscious awareness or reflection, in response to a set of associated conditions or contextual cues’ (Hagger et al., 2018, p. 86). When the target behavior is expected to be performed frequently (e.g., using a refillable bottle for drinking away from home in this study), there is a higher likelihood of habit formation. Thus, past behavior will probably exert greater direct effects on future behavior than intentions.

Hagger et al. (2018) conducted a meta-analysis of the reasoned action approach applied to health behaviors. They extended a previous meta-analysis by McEachan et al. (2016) and considered past behavior to be a precursor of model constructs. They found that past behavior not only predicted prospective health behavior both directly and indirectly through the TPB constructs but also attenuated the effects of intention and social cognition variables on health behaviors.

In the current study, following Hagger et al. (2018) and Hennessy et al. (2010), we included past behavior as a precursor of TPB model constructs (Fig. 1) to gauge the extent to which non-conscious processes control the behavior under examination and the extent to which TPB constructs account for the effects of past behavior. This approach also allowed for an evaluation of the sufficiency of the TPB model in explaining behavior. Based on these theoretical considerations and previous results, the following hypotheses were proposed:

H4. Past behavior is positively related to the future behavior of using a refillable bottle.

H5. Past behavior is positively related to intention.

H6. Past behavior is positively related to attitudes (H6a), subjective norm (H6b) and PBC (H6c).

Self-identity

Another predictor often included in the TPB model is self-identity, defined as ‘the extent to which I consider myself to be the kind of person who performs the behavior of interest’ (Ajzen, 2020, p. 317). The inclusion of self-identity was originally based on identity theory (Stryker, 1987). From a social cognition perspective, self-identity can be defined as a category label to which individuals self-associate, guiding their cognition and behavior (Randers and Thøgersen, 2023a; 2023b; Reed et al., 2012). Through self-labeling, people categorize themselves according to their personal characteristics (e.g., personality, appearance, behavior), the roles they fulfill (e.g., sister, mother) or their group memberships. Personal characteristics, roles and memberships can induce specific actions to support the validation of one’s self-concept (Stets and Burke, 2000).

Ajzen (2020) and Fishbein and Ajzen (2010) stated that additional predictors should be included in the TPB with caution. These further constructs should be conceptually independent of the theory’s existing predictors and behavior-specific, conforming to the principle of compatibility. It should be possible to define and measure these constructs in terms of the target, action, context and time (TACT), which define the behavioral criterion.

Fishbein and Ajzen (2010) argued that self-identity could overlap with past behavior, as people could possibly deduce self-identities by taking into consideration past behaviors. Nevertheless, a meta-analytic study by Rise et al. (2010) reported that self-identity explained an additional 6% of the variance in intentions independently from both past behavior and other TPB constructs. Different studies, including those that controlled for past behavior (e.g., Canova et al., 2020a; Carfora et al. 2017a; 2017b), confirmed that self-identity is a valuable and non-redundant addition within the TPB. Moreover, individuals may consider consuming or using specific products as a strategy for self-presentation, using such behaviors to express the kind of person they are and their identity to others (Goffman, 1959). For example, self-identity as a recycler (e.g., Nigbur et al., 2010; White and Hyde, 2012), green consumer (e.g., Carfora et al., 2019; Dean et al., 2012; Sparks and Shepard, 1992) or meat eater (Carfora et al., 2017a) can predict a person’s intention to recycle, purchase organic food or reduce meat intake, respectively.

Thus, in the case of behavioral choices that show other people the ethical commitment of individuals and their attention to sustainability issues (e.g., the preference for refillable vs. single-use plastic bottles), self-identity as a person who uses refillable bottles can play a role in determining intention and behavior. Therefore, the following hypotheses were proposed (Fig. 1):

H7. Past behavior is positively associated with self-identity.

H8. Self-identity is positively related to the intention to use a refillable bottle.

Regarding the invariance of the processes leading to intention and behavior in the two groups (university students and non-student adults), and based on prior research findings (Canova et al., 2020a; 2025), we anticipated that the hypothesized paths in the model represent generalized associations that apply across groups, with no expected differences.

Summary and research contribution

This study focused on psychosocial factors that predict the behavior of using refillable bottles and adopted an extended TPB model where past behavior was added as a precursor to the original TPB constructs and self-identity as an additional determinant of intention. Our research represents one of the few studies in the literature that investigates the use of these specific containers instead of non-eco-friendly single-use plastic bottles. A two-wave prospective online survey design was employed and the relation models were tested via structural equation modeling. Through multi-group analysis the processes leading to intention formation and behavior were compared between two groups: university students and non-student adults. Understanding potential differences between these two groups has both theoretical and practical implications for designing interventions for changing behavior targeted at specific subgroups of the population.

Method

Procedure and participants

The total sample size was established using Soper’s (2024) calculator for structural equation models. With 22 observed and 7 latent variables (see the Data analysis section and Fig. 1), the minimum sample size required to achieve a power of 0.80, with a 0.05 probability level and 0.20 effect size (between small, f = 0.10, and medium, f = 0.30), is 425 respondents. Thus, we aimed to recruit ~1000 participants to achieve more than the required power, accounting for the expected attrition across the two time points of the research and the project to test the invariance of the hypothesized processes in the two groups (university students vs. non-student adults).

Two waves of data collection were organized—Time 1 (T1) and Time 2 (T2), with a time lag of one month. The questionnaires were designed and distributed via the Qualtrics platform from March to June 2023. At T1, participants filled out a questionnaire including the measures of the variables constituting the extended TPB model (i.e., past behavior, attitude towards the behavior, subjective norm, PBC, self-identity, and intention) as well as socio-demographic variables. One month later, at T2, self-reported behavioral measures were collected, asking the participants if they had used a refillable bottle for drinking away from home during the last month, before the second wave of the research.

The study was based on convenience sampling. Participants were recruited thanks to the collaboration of undergraduate students attending psychology lectures in different degree courses at two Italian universities (Padua University and the Salesian University Institute of Venice). The students received the questionnaire at their institutional email addresses. They were asked to personally fill it out and send the link to at least two non-student adults, not from the same family, to avoid the potential resemblance of daily habits, who also completed the same questionnaire. To be eligible for participation in this study, people were required to be of legal age (i.e., over 18 years old). Along with the link to the questionnaire, all participants received an instruction letter in which they were made familiar with the aim of the study, the estimated duration of the task (about 15 min) and the possibility of withholding their consent to participate at any time. They were also assured that participation was voluntary (without any form of compensation) and that all answers would remain confidential and be treated as a group. The instruction letter also stated that the research had two phases and that they would receive another very short questionnaire a month later. The questionnaires were paired using self-generated alphanumeric codes. The study was conducted according to the guidelines laid down in the Declaration of Helsinki. The ethics committee of the institution financially supporting the research also approved the procedures involving human subjects. Participants indicated their informed consent by selecting the ‘I accept’ button on the first page of the questionnaire. Qualtrics allowed each participant to use the link to the questionnaire only once.

Of the 1548 individuals who accessed the questionnaire link at T1, 1316 completed the first questionnaire (85% response rate). Among them, 835 also completed the second questionnaire at T2 (53.9% final response rate). Only those who completed both questionnaires constituted the final sample for the data analysis.

Table 1 summarizes the socio-demographic composition of the total sample and of student and non-student adult subsamples. The final overall sample comprised 835 participants, with a prevalence of women (73.7%), particularly in the student group (87.4%). The majority of the participants lived in the northeast of Italy (87.5%). As expected, the mean age between the student and non-student adult groups was significantly different (t832 = 32.67, p < 0.001), even if a modest degree of overlap between the two groups existed. More importantly, in line with the recruiting criteria, the occupational status and level of education of the two groups differed.

Table 1 Survey sample characteristics.
Full size table

Attrition analysis conducted via the chi-square test, t-test and multivariate analysis of variance (MANOVA) revealed only a difference between the 835 participants included in the final sample and the 481 participants who did not fill in the second questionnaire. In the final sample vs. ‘dropouts’, women were likelier than men to answer the second questionnaire (73.7% vs. 66.1%) (χ²2 = 9.04, p < 0.02). No difference was found in either age (t1304 = –0.61, ns) or TPB constructs (principal multivariate effect – F7,1304 = 1.78, p < 0.09, η²par = 0.01).

Measures

The measures of the TPB constructs complied with the guidelines for correctly designing a TPB questionnaire and were based on the principle of compatibility between TACT (Ajzen, 2020; Fishbein and Ajzen, 2010). The items, already used and validated in previous studies in the Italian context (Table 2), were contextualized to the target behavior (i.e., to use a refillable bottle for drinking away from home). Two experts reviewed the adapted versions of the measures, and the items were then slightly modified based on their suggestions. For all TPB items (except past behavior and behavior at T2), the time reference was ‘next month’. At the beginning of the questionnaire, the target behavior was described as follows: ‘To use a refillable bottle for drinking away from home, that is, a bottle that can be refilled several times and can hold any type of beverage, such as water or other soft drinks. Think, for example, of metal, hard plastic or glass bottles. Disposable (or single-use) plastic bottles are excluded, even if they are refilled multiple times.’

Table 2 Measures and factor loadings.
Full size table

In the first wave (T1), participants were asked to report their past behavior, attitude toward the behavior, subjective norm, PBC, self-identity, intention and demographic information (following the same order for all participants). The demographics were gender (response categories—woman, man, I prefer not to answer), age, geographic area of residence, education and employment status (only for non-student adults). In the second wave (T2), participants reported whether they used a refillable bottle during the month between T1 and T2. For this purpose, the first two items that measured past behavior at T1 were administered again. All items were scored on a five-point Likert-type scale; only the semantic differential adopted a seven-point scale (Table 2).

Data analysis

Analyses were conducted using MPLUS 8.5 and SPSS 28. All responses to the items measuring the variables included in the model were mandatory; thus, there were no missing values. First, confirmatory factor analysis (CFA) using the maximum likelihood method was performed on the whole sample (n = 835). The measurement model included seven latent factors and twenty-two indicators. Goodness-of-fit was evaluated by means of the following indices: χ², comparative fit index (CFI), Tucker–Lewis index (TLI), root mean squared error of approximation (RMSEA), and standardized root mean square residual (SRMR). A satisfactory model is indicated by nonsignificant χ2, RMSEA ≤0.06, CFI ≥0.95, TLI ≥0.95 and SRMR ≤0.08 (Hu and Bentler, 1999).

Convergent validity was assessed by calculating the average variance extracted (AVE) for each factor: AVE ≥ 50 was considered adequate. Next, discriminant validity was evaluated by comparing the square root of AVE values with correlations among factors: when the square root of AVE for each latent variable is higher than the correlations with other latent variables, discriminant validity is considered adequate (Fornell and Larcker, 1981). Composite reliabilities (CR) and Cronbach’s alpha coefficients were computed to estimate reliability. The hypothesized model of relations (Fig. 1) was tested via structural equation modeling. The indirect effects were deemed statistically significant if the bootstrapped 95% confidence interval (CI) did not include 0.

Finally, a multi-group procedure was applied to test whether all the parameters of the model were invariant across students and non-student adults. The following consequential hypotheses were tested—configural invariance, which requires an equal number of factors and the same pattern of factor–item relations in both groups; metric invariance, which denotes the invariance of factor loadings across groups; and structural invariance, which reflects the invariance of regression coefficients between exogenous and endogenous latent variables. Model comparisons were based on chi-square difference (i.e., Δχ²) and ΔCFI. Descriptive statistics were computed, and the differences between the mean scores of students and non-student adults were examined using MANOVA.

Results

Measurement model

CFA indicated that the seven-factor model showed satisfactory goodness-of-fit indices in the overall sample: χ2188 = 707.87, p ≅ 0.00, RMSEA = 0.06, 90% CI [0.05, 0.06], CFI = 0.96, TLI = 0.95, SRMR = 0.04. All standardized factor loadings were significant (Table 2), as were the correlations among the latent factors (Table 3).

Table 3 Correlations among latent factors, reliability and AVE coefficients.
Full size table

For each construct, the AVE was higher than the suggested value of 0.50 (Fornell and Larcker, 1981). Furthermore, the square root of AVE values was higher than the correlations between each pair of constructs, except for the PBC–intention pair (Table 3). However, even in this case, the 95% CI obtained by considering two standard errors above and below the coefficient did not include a perfect correlation (i.e., 1.00) (Bagozzi et al., 1994). Overall, these results supported adequate convergent and discriminant validity. Composite reliability values and Cronbach’s coefficients were satisfactory for all measures (Table 3).

Model testing and invariance

In the total sample, the goodness-of-fit indices of the hypothesized model were satisfactory: χ2191 = 708.54, p ≅ 0.00, RMSEA = 0.06, 90% CI [0.05, 0.06], CFI = 0.96, TLI = 0.95, SRMR = 0.04. The model explained 73% of the intention variance and 68% of the future behavior of using refillable bottles. Past behavior was significantly associated with TPB core constructs and self-identity, supporting H6a, H6b, H6c and H7 (Table 4). Past behavior accounted for the following percentages of TPB constructs and self-identity—attitude (29%), subjective norm (7.6%), PBC (47%) and self-identity (36%). Attitude, PBC and self-identity were associated with intention, which predicted behavior, thus confirming H2a, H2c, H8 and H1. In addition, past behavior was directly related to the intention and future behavior of using refillable bottles, as hypothesized (H5 and H4, respectively). Contrary to H2b and H3, we found no statistically significant effect of subjective norm on intention, and PBC did not predict behavior at T2.

Table 4 Path coefficients of the model and the test of structural invariance.
Full size table

An examination of the indirect effects showed that past behavior was associated with intention through attitude (β = 0.13, p < 0.001, 95% CI [0.08, 0.19]), PBC (β = 0.19, p < 0.001, 95% CI [0.10, 0.30]) and self-identity (β = 0.11, p < 0.004, 95% CI [0.04, 0.19]). It was also associated with future behavior at T2 through intention (β = 0.05, p < 0.02, 95% CI [0.01, 0.09]). The indirect effects of past behavior accounted for a modest proportion (20%) of the total effect of past behavior on future behavior at T2 (total effect =0.81, direct effect = 0.65, total indirect effects = 0.16).

The multi-group procedure was applied by contrasting students (n = 429) with non-student adults (n = 406). First, we estimated the model independently for each group. Goodness-of-fit statistics of the model were satisfactory in both groups; in the student sample, it was χ2191 = 426.80, p ≅ 0.00, RMSEA = 0.05, 90% CI [0.05, 0.06], CFI = 0.96, TLI = 0.96, SRMR = 0.04; and in the adult sample, it was χ2191 = 491.17, p ≅ 0.00, RMSEA = 0.06, 90% CI [0.06, 0.07], CFI = 0.96, TLI = 0.95, SRMR = 0.05. Table 4 presents the path coefficients for both samples.

The configural invariance model was supported (χ2382 = 917.97, p ≅ 0.00, RMSEA = 0.06, 90% CI [0.05, 0.06], CFI = 0.96, TLI = 0.95, SRMR = 0.04). Comparing the metric invariance model (χ2397 = 933.72, p ≅ 0.00, RMSEA = 0.06, 90% CI [0.05, 0.06], CFI = 0.96, TLI = 0.95, SRMR = 0.05) with the configural model showed that Δχ2 was not significant (Δχ215 = 15.76, ns) and ΔCFI = 0.00. This sustained factor loading invariance across the two groups. In the last model (structural invariance), the regression coefficients between the latent variables were constrained to be equal. A comparison of this model (χ2409 = 961.74, p ≅ 0.00, RMSEA = 0.06, 90% CI [0.05, 0.06], CFI = 0.96, TLI = 0.95, SRMR = 0.06) with the metric model revealed that the Chi-square difference test was statistically significant (Δχ212 = 28.02, p < 0.01) while ΔCFI was equal to 0.00. These results indicate some differences in the structural model weights between the two groups. Therefore, we proceeded with the investigation by constraining one path at a time to identify the specific differences between student and non-student adults. The results showed that invariance did not hold for only three paths. In particular, the effect of self-identity on intention was stronger in the non-student adults than in the student group (Table 4).

Differences between university students and non-student adults

Table 5 presents the means and standard deviations of all constructs obtained in the total, student and adult samples. Overall, the respondents intended to use a refillable bottle for drinking away from home, and the evaluation of this behavior was very positive. Participants stated that they perceived themselves as having enough control (PBC) over their behavior and feeling moderate social pressure from significant others. They also reported a good level of self-identity as a person who uses a refillable bottle as well as a good level of past and future behavior.

Table 5 Descriptive statistics and differences between students and non-student adults.
Full size table

We used MANOVA to examine the mean differences between students and non-student adults. The main multivariate effect was significant (F7,827 = 12.29, p < 0.001, η²par = 0.09). Non-student adults scored lower than students on past and future behavior, intention, attitude and PBC. Students used refillable bottles more frequently than adults, declared to have a higher intention to use them for drinking away from home and had a more favorable attitude toward this behavior. Finally, as opposed to the adults, the students perceived the behavior as easier to enact and under their control.

Discussion

Staying anchored to the TPB theoretical and measurement prescriptions and capitalizing on the extant literature, the present two-wave prospective research targeted refillable bottle use. It also proposed an extended TPB model in which past behavior was added as a precursor of the original TPB constructs and self-identity as an additional determinant of intention. Moreover, the processes leading to intention formation and behavior were compared between two groups, namely, university students and non-student adults, who were assumed to have differences in terms of both their daily routines and lifestyles and in the attention, appreciation and use of refillable bottles.

The hypothesized model explained substantial percentages of variance in intention and future behavior. PBC and attitude emerged as significant predictors of intention, followed by self-identity. Intention predicted the future behavior of using a refillable bottle. Therefore, individuals are more motivated to engage in target behavior when they have a positive attitude toward performing it in the future, believe that doing so is under their control and consider the behavior itself to be coherent with their self-identity. Attitude and PBC were significantly associated with intention in almost all studies on plastic reduction (e.g., Aruta, 2022; Nguyen, 2024; Oludoye and Supakata, 2024; Raimondo et al., 2022; Truelove et al., 2023; Wang et al., 2024). Contrary to hypothesis H2b, and in line with Fedi et al. (2021), the subjective norm turned out to be linearly independent of intention. This finding is not new in the TPB literature. The subjective norm is often the weaker determinant of intention (Armitage and Conner, 2001), suggesting a lesser role for the perceived influence of significant others in different types of behavior, including those related to reducing the use of plastic or plastic waste (e.g., Brown et al., 2020; Canova et al., 2022; Carfora et al., 2017a; Nguyen, 2024; Truelove et al., 2023; Wang et al., 2024). PBC does not directly predict future behavior. Studies on healthy eating (e.g., Canova et al., 2020a) and the reduction of plastic drinking bottles (e.g., Raimondo et al., 2022) have obtained the same result. This non-significant relationship with behavior seems unproblematic. While such a direct effect is allowed by the TPB, Ajzen (1991) clarified that this relationship will not appear if the behavior is perceived to be under complete volitional control.

Self-identity was significantly associated with intention and, as hypothesized (H8), acted as a predictor of intention over and above TPB constructs and past behavior. This result is in line with previous findings (e.g., Canova et al., 2020a; Carfora et al., 2016; 2017a; 2017b; Rise et al., 2010) and supports the importance of individuals’ self-perceptions about environmental issues (Capasso et al., 2025).

Past behavior was significantly associated with attitude, subjective norms, PBC and self-identity. It also directly predicted intention and behavior, and, as in previous studies where the target behavior was performed repeatedly (e.g., Canova et al., 2020a; Phipps et al., 2020), exerted the largest effect on the target behavior. Past behavior was also indirectly related to intention and behavior. The indirect effects may reflect making similar decisions in the past or the influence of past experiences in forming beliefs regarding the future performance of the behavior (Ajzen, 2002). As in previous studies (e.g., Brown et al., 2018; Hagger et al., 2018), the indirect effects of past behavior through social cognition variables accounted for a relatively small proportion of the total effect of past behavior on behavior. However, although the size of the effects of the TPB constructs could be reduced with the inclusion of past behavior, they were not fully eliminated. These results indicate the pervasive effects of past behavior in social cognition models, but they are also consistent with the perspective of dual models of information processing and decision-making (Hagger et al., 2018). When the frequency of behavior increases, individuals are likelier to form habits, and the role of the reasoned decision-making process is diminished.

Findings from the multi-group analysis showed that the hypothesized extended TPB model fitted the data well in both student and non-student groups, further confirming configural and metric invariance. However, structural invariance did not receive full support. A stronger effect of self-identity on intention was found in the adult sample than in the student group. Overall, the results indicate that the process leading to the enactment of the behavior is habit-based for both groups. Contrary to expectations, for adults, the factors related to the presentation of the self as a person who uses refillable bottles—and is therefore concerned with environmental issues—seem more important.

With regard to the mean differences, students resorted to refillable bottles more frequently than adults did. This result is in line with the limited data available on the use of this specific container (Volpe, 2020). Their attitude toward using a refillable bottle for drinking away from home was more favorable, and they declared a higher intention to use the bottles. Finally, compared to the adults, the students perceived the behavior as easier to enact and more under their control. This aligns with evidence that younger people are more active and committed to fighting climate change and promoting eco-friendly ways of life, starting with their habits (Doan et al., 2025; Hickman et al., 2021).

We argue that our study can contribute significantly to the literature on pro-environmental behaviors and has strengths worth mentioning. This is one of the very few studies that considers the specific pro-environmental behavior of refillable bottle use. Measures of TPB constructs adhered to the principle of compatibility (Ajzen, 2020), and the study adopted a prospective design with two-wave data collection. Lastly, this study evaluated the role of a sample characteristic (students vs. non-student adults) as a possible moderator of the relationships within the model.

From a theoretical perspective, the study helps to reinforce the proposal that the TPB should consider the role of self-identity in influencing behavioral intentions, since its effect persisted even when past behavior was controlled, although some scholars argued that self-identity might act merely as a surrogate for the influence of behavioral habits (Fishbein and Ajzen, 2010). Second, the effect of past behavior on future behavior aligns with dual process theories, which hypothesize two routes to action—a reasoned one that is shaped by social cognition variables and a non-conscious, habitual one that is rooted in past behavior (Brown et al., 2018; Hagger et al., 2018). At the practical application level, the results of the study suggest that facilitating the widespread use of refillable water bottles can form part of a broader movement toward sustainability, which fits within the larger eco-surplus culture, defined by Vuong and Nguyen (2024) as a set of pro-environmental attitudes, values, beliefs, and behaviors aimed at reducing negative anthropogenic impacts on the environment—which should no longer be denied—as well as conserving and restoring nature. The shift from single-use plastic containers to reusable ones represents a significant step toward achieving the Sustainable Development Goals (United Nations, 2024), as it may help reduce everyday plastic use and pollution.

Limitations and future research

Some limitations must be recognized. First, convenience samples were recruited, with a prevalence of women, especially in the student group, which placed limits on the generalizability of the results. A possible further limitation could be the slight overlap in age between the student vs. non-student groups. Future research should evaluate the possibility of concentrating more precisely on age differences, following the line of recent studies (e.g., Doan et al., 2025) that compared different generations. In addition, differences in educational level could be treated as a moderating variable in TPB models. Furthermore, even though the study considered a prospective measure of behavior, it was correlational in design, so the direction of relationships could only be inferred from the relationships outlined in the reference theory. Experimental designs are needed to confirm the direction of causality.

The scales measuring all variables required self-reported appraisals that may suffer in terms of social desirability and retrieval inaccuracy. The reliance on self-reported behavior, which is very common in TPB studies, means that the impact on objectively measured behavior is difficult to estimate. The target behavior could also be reasonably refined. For example, future research could focus only on refillable non-plastic (e.g., aluminum) bottles to promote not only the reduction but also the complete abandonment of plastic.

Lastly, socio-demographic variables were not considered, even though some background factors, such as education and income, could also be linked with the constructs and relationships considered in this study. Some studies have highlighted that environmental concerns are positively associated with a high per capita income and education level (Baiardi and Morana, 2021; Volpe, 2020). Furthermore, women and highly educated people were found to be likelier to participate in no-plastic campaigns (Afroz et al., 2017; Heidbreder et al., 2019). Future studies should evaluate the possibility of involving people who are less educated and have a lower income per capita or, in any case, samples representative of the general population.

Practical implications

Past behavior is largely influential in determining TPB constructs and future behavior. Hence, interventions should be directed to the family and school domains, where the young are supposed to assume—from parents and peers—the most important ethical and cultural references as well as some of the behavioral routines that could be retained for a lifetime. Interventions, in addition to raising awareness of environmental issues, including topics such as environmental protection and the consequences of single-use plastics, should also seek to implement alternative routines. For example, Zorpas et al. (2017) provided stainless steel refillable bottles to primary school children as a way to reduce plastic bottle use, achieving an increase in the use of these containers and a decrease in the use of plastic ones in the next evaluation period. Hence, schools, universities, companies and other organizations should increase initiatives that involve providing water bottles to students or employees.

Moreover, intervention strategies should focus on fostering positive attitudes toward the use of eco-friendly bottles. Mass media campaigns should highlight the value of incorporating refillable water bottles into daily life and their significant contribution to environmental protection and preservation for future generations. Additionally, traditional media and social media should critique the excessive use of single-use plastic bottles, highlighting their negative consequences, such as plastic dispersion.

Especially in the adult population, promoting self-identity as a user of non-disposable items, such as refillable bottles, can be a useful way to drive intention and behavior. However, eliciting such an identity may be challenging. It can be beneficial to enhance awareness of the active role that consumers can easily assume in environmental protection and to try to create positive associations between the use of non-disposable items and the positive characteristics of the people who use them.

Finally, to increase the perception of control, specific policy measures should be implemented to create situations that facilitate both the sourcing of refillable bottles on the market and their use and to make recourse to single-use plastic bottles complicated (Ertz et al., 2017). For example, some measures may focus on less expensive and more versatile refillable bottles, providing incentives for companies to distribute eco-friendly bottles and installing user-friendly water-bottle-filling stations. By implementing these recommendations, it would be possible to stimulate the adoption of this behavior among the population, contributing to a more sustainable future and reducing the environmental impact of single-use plastics.

Conclusion

Our research endorses the utility of the TPB as a general framework for understanding and predicting the behavior of refillable bottle use, considering different samples of university students and non-student adults. On the one hand, this daily behavior is very promising for achieving the Sustainable Development Goal, connected to the urgency of decreasing pollution due to plastic production, accumulation and dispersion into the environment. On the other hand, it is still underinvestigated in the socio-psychological literature. The results of our study attested that among the core constructs of the TPB, the most important for predicting intention were attitudes and PBC. The future behavior of using a refillable bottle was predicted by intention and past behavior, which acted directly and indirectly. Self-identity was comparatively more relevant in determining intentions in the non-student adult vs. university student group, showing that this behavior can also be connected to self-presentation strategies.