Time spent on separating waste is never wasted:
Changing consumers’ waste separation behavior through the use of a mobile application
Kelly Kim de Wildt 12390348 Research Master’s Thesis Research Master’s programme Communication Science Graduate School of Communication University of Amsterdam Supervisor: Dr. M.H.C. Meijers 26 June 2020
Abstract
Research has shown that the rules concerning waste separation for private households are complex, hard to grasp, and calls for better knowledge provision in an easy and accessible way. One way to increase knowledge, could be using mobile applications. The current study investigates: 1) whether using a recycling app increases recycling knowledge and
subsequently actual recycling behavior; and 2) which theorie(s) between the Theory of Planned Behavior vs. Technology Acceptance Model can explain if, and why, people use a sustainable mobile app. A within-subjects experiment was conducted (N =139) in which a baseline week of recycling was compared with an intervention week, in which participants used a recycling app. Experience sampling methodology (ESM) was used to assess daily recycling behavior. While there were no significant effects of green app use on recycling knowledge, there was still evidence for the effect of recycling knowledge on recycling behavior. Furthermore, the results showed that both the TPB and the TAM were suitable to explain green app use. The findings are promising for the sustainability sector, and show that the development of apps to enhance knowledge in their customers can stimulate actual sustainable behavior. Since the field of mobile applications has huge potential for scaling, it could be a great platform tool for green interventions.
Introduction
The European Parliament approved a resolution declaring a global climate and environmental emergency on 28 November 2019 (European Parliament, 2019). To limit global environmental issues such as climate change, water, and biodiversity problems, it becomes increasingly important to encourage people to adopt more sustainable behaviors (Tate, Stewart & Daly, 2014). Performing proper waste separation is one of those behaviors, since separating waste for recycling saves natural resources, such as oil for plastics, trees for paper, and metals for devices and tinplate cans (Milieu Centraal, n.d.). In addition, saving natural resources means that there is more land that can be left in its natural state. Also, waste recycling leads to less emission of greenhouse gases compared to waste incineration (Milieu Centraal, n.d.). Lastly, recycling materials saves up on the energy of the production of new raw materials (Milieu Centraal, n.d.). That in turn, also reduces the production of greenhouse gasses. To achieve these benefits, however, it is of vital importance that people recycle their waste, and that they separate their waste in a correct manner.
According to Milieu Centraal (n.d.), it is possible to separate up to 80 percent of the municipal waste in the Netherlands. Currently, a single Dutch person produces almost 500 kilograms of waste per year. However, in 2017 only 56 percent of the Dutch waste was actually separated (Milieu Centraal, n.d.). The categories that were separated most correctly were tinplate packaging (95%), domestic chemical waste (86%), paper and cardboard (78%), glass (73%), and organic waste (60%), while the waste categories that were separated most incorrectly were textiles (31%), household appliances (49%), drink cartons (50%), and plastics (59%) (Milieu Centraal, n.d.). Since plastics and drink cartons are common everyday waste products, this shows there is potential for better and more waste separation.
It comes as no surprise that there is still room for improvement when it comes to waste separation since the directions are not always comprehensible, coherent, and logical (Árnadóttir, Kok, Van Gils & Ten Hoor, 2019), which may as such undermine people’s self-efficacy beliefs (i.e. they believe they are able to successfully execute the behavior that is required to produce the desired outcome; Bandura, 1977), an important predictor for
(recycling) behavior. For example, in the Netherlands black plastic food trays belong to the general waste, and not the plastic waste. Grease stained pizza boxes belong to the general waste instead of paper waste, since boxes smeared with fat and oils are unsuitable for recycling (Milieu Centraal, n.d.). Metal lids of glass jars, aluminum, and tinplate cans are sometimes allowed to be collected with plastics, but this differs between municipalities. For example, the Dutch municipality of Rotterdam allows them to be separated with plastics, but the municipality of Amsterdam does not.
To further illustrate the problem of ambiguous waste separation instructions: a recent study at a Dutch university cafeteria showed that most participants thought that napkins could be separated as paper waste (correct waste category is general waste), chips bags as plastic waste (correct: general waste), sandwich bags as plastic waste (correct: general waste), and noodle boxes as paper waste (correct: general waste; Árnadóttir et al., 2019). Thus,
participants’ knowledge about the appropriate way to separate the waste was insufficient, even when they had a moderate to high intention to recycle their waste. While intention usually is a good indicator of behavior, when people do not have enough knowledge or skills to perform that particular behavior in the correct way, they will not exhibit that particular correct behavior (Árnadóttir et al., 2019) as they are either executing the behavior in a(n) (unintentionally) wrong manner or because they lack self-efficacy beliefs. Clearly, the rules concerning waste separation for private households are complex, and hard to grasp, and this
calls for better knowledge provision in an easy and accessible way. The goal of 80% waste separation in the Netherlands can only be reached if people understand and know how to accurately separate their waste - and perform this behavior.
One way to increase knowledge could be using persuasive digital technologies to change people’s behavior. In particular, the use of mobile applications (in short: apps) can be promising: Globally, 2.7 billion people own a smartphone, and 1.35 billion people own a tablet (Buildfire, 2019); the Google Play Store offers 2.8 million apps to download, and the Apple App Store 2.2 million (Buildfire, 2019); and over twenty percent of the Millennial age group opens an app 50 times a day or more (Buildfire, 2019). One of the advantages of apps is that they can foster behavior change.
Whereas green apps have not been often studied, studies into health and
education-related apps have proven to be effective in fostering behavioral change (Brauer et al., 2016), for example weight loss apps (e.g. Alnasser et al., 2019; Ogden, Maxwell & Wong, 2019). A national study in the U.S. showed that 58% of mobile phone users had downloaded a health-related mobile app, and most participants used them at least daily (Krebs & Duncan, 2015). While the persuasive effects of health-related apps have been studied, the field of apps that encourage green behavior is relatively new. This raises the question if green behavior can be stimulated by apps, in particular for recycling and waste separation behavior. For this, two elements are essential. The first is whether using a green app can increase knowledge and subsequently green behavior. This ultimately depends on the second element: whether people actually use the app.
Two main persuasive communication theories could potentially explain if, and why, people would successfully use a green mobile app, and in particular an app that encourages waste separation. In persuasive communication, the Theory of Planned Behavior (TPB;
Ajzen, 1985, 1989, 1991) model has been one of the main models to explain behavior (e.g. for waste separation behavior; Árnadóttir et al., 2019; Ayob, Sheau-Ting, Jalil, & Chin, 2017; Xu, Ling, Lu & Shen, 2017; Zhang, Huang, Yin & Gong, 2015), while the Technology Acceptance Model (TAM; Davis, 1985) focuses specifically on technology use (e.g. mobile tourism applications, Chen & Tsai, 2017; mobile game-based learning, Ghani et al., 2019; and e-learning, Abdullah, Ward & Ahmed, 2016). Few studies have investigated and compared the predictive value of the TAM with the TPB in information technology (e.g. computer programs, Mathieson, 1991; telemedicine, Chau & Hu, 2002), but no previous studies have compared the TAM with the TPB to explain green mobile app use.
In sum, the current study has two aims: 1) investigating if using a mobile application that provides knowledge on recycling improves recycling behavior, and 2) investigating what factors influence green app use, specifically testing whether the TPB model or the TAM is more accurate in predicting green app use.
Theoretical background Knowledge and Self-efficacy
In order to stimulate correct recycling behavior to combat environmental issues, it is important that people believe themselves to be capable to successfully execute the behavior that is required to produce the desired outcome, which refers to the concept of self-efficacy (Bandura, 1977). An extensive body of research has shown the importance of self-efficacy as a determinant of behavior change (e.g. Ajzen, 1991). Countless studies have shown that there is a positive effect of self-efficacy on behavior within the Theory of Planned Behavior (TPB) context, for example on online purchasing behavior (George, 2004), physical activity
(Armitage, 2005), choice of travel mode (Bamberg, Ajzen & Schmidt, 2003), condom use (Asare, 2015), and healthy eating (Connor, Norman & Bell, 2002). Moreover, self-efficacy
has been shown to be significantly and positively related to general green behaviors (Meinhold & Malkus, 2005; Rainear & Christensen, 2017), green private-sphere behaviors (e.g. purchasing or using green products; Lauren, Smith, Louis & Dean, 2017), green
public-sphere behaviors (e.g. voting, donating, volunteering, contacting government officials, and protesting about climate change; Doherty & Webler, 2016), conservation behaviors (Kaiser, Hübner & Bogner, 2005), and the willingness to engage in climate change
discussions (Geiger, Swim & Fraser, 2017). Relevantly, the TPB has been used extensively to explain waste separation behavior (e.g. Árnadóttir et al., 2019; Ayob, Sheau-Ting, Jalil, & Chin, 2017; Xu et al., 2017; Zhang, Huang, Yin & Gong, 2015). Indeed, research has found that participants’ self-efficacy, subjective norms, and attitudes were related to the intention to recycle their waste, and self-efficacy was observed to have the largest effect (Árnadóttir et al., 2019). Furthermore, in trying to identify predictive individual, collective, and
organizational factors of recycling behavior to help organizations to increase the recycling rates in their communities, the results of a sample among 1501 residents showed that self-efficacy had a positive and direct effect on recycling behavior (Tabernero et al., 2015).
If increasing self-efficacy can lead to better recycling behavior, this begs the question how to increase self-efficacy. Importantly, research has suggested that increasing knowledge by knowledge-based interventions is effective in increasing self-efficacy (Geiger, Swim & Fraser, 2017). Where self-efficacy concerns the belief that it is possible to execute a behavior to produce a desired outcome, knowledge provides the tools for this by giving knowledge about the where, when, and how of recycling (Schultz, 2002). In research, this is often measured by asking participants which materials are recycled, and to code their correct answers. Studies have found overwhelming evidence that knowledge is a strong and
to a change in recycling behavior, supporting the knowledge-deficit theory that assumes that increasing knowledge will lead to behavior change (Schultz, 2002). The more knowledge one has about which materials can be recycled, and when and where waste collection takes place, the more likely one is to recycle. For example, in a study with adolescents, adolescents who demonstrated more green attitudes and greater green knowledge reported greater amounts of green behavior (Meinhold & Malkus, 2005).
App Usage Leading to Behavior Change
While increasing knowledge can be a powerful tool to increase self-efficacy, this raises the question: how to increase knowledge? Previous research has come up with a possible solution, and suggested that mobile apps can increase knowledge (Rachels &
Rockinson-Szapkiw, 2018). Specifically, one study looked at the effectiveness of Duolingo, a computer and mobile app that teaches foreign languages, on self-efficacy and knowledge. The treatment group used the Duolingo app for Spanish language instruction, while students in the control group received their regularly scheduled English L1/Spanish L2 class learning activities. Pretest-posttest results showed no differences between students who used the Duolingo app and students who were taught with traditional face-to-face instruction on their Spanish achievement or in academic self-efficacy, meaning that the knowledge
transformation through the app was similarly effective on achievement and self-efficacy compared to traditional teaching (Rachels & Rockinson-Szapkiw, 2018).
Indeed, mobile apps have also been shown to encourage behavior change, for example in the field of health-related and education apps (Brauer et al., 2016). However, there is hardly any research investigating how apps can increase green behavior (Sunio &
Schmöcker, 2017). This raises the question if green behavior can be stimulated by apps, in particular for recycling and waste separation behavior. In conclusion, through raising
people’s recycling knowledge, literature suggests that mobile apps increase self-efficacy, which will subsequently influence their recycling intention and actual separation behavior. Thus, it is hypothesized that:
H1: Using an app has a positive effect on performing recycling behavior, mediated by knowledge, self-efficacy, and intention to perform the behavior.
Predicting App Use
If recycling behavior can indeed be stimulated by using an app, it is important to know what predicts app use, such that these factors can be targeted in order to increase app use and ultimately recycling behavior. However, in order for green apps to be persuasive, people should actually use them, and there is still a gap in knowledge on which theory best predicts green app use. Within the context of technology acceptance, the Theory of Planned Behavior (TPB) and the Technology Acceptance Model (TAM) are commonly used theories to predict technology usage (Davis, Bagozzi & Warshaw, 1989; Davis, 1993; Venkatesh & Davis, 1996; Venkatesh & Davis, 2000; Venkatesh & Bala, 2008; Cheon, Lee, Crooks & Song, 2012; Yang, 2013). However, it remains to be seen which of these models best predicts green app use and resulting from this, the question of which concepts should be focused on when stimulating app use still remains unanswered.
The Theory of Planned Behavior (TPB). The Theory of Planned Behavior (TPB) was first introduced by Ajzen (1985, 1989, 1991), and aims to predict behavior in a variety of settings, including technology use (Cheon, Lee, Crooks & Song, 2012; Yang, 2013). The
theory states that human behavior is determined by the intention to perform a behavior. That intention is guided by attitudes towards the behavior, subjective norms, and self-efficacy (sometimes also referred to as perceived behavioral control; Ajzen, 2006). Attitudes towards the behavior refer to the extent to which one has a(n) (un)favorable evaluation or appraisal of a particular behavior, in this case whether one has a favorable view of using a green app. Subjective norms are defined as the perceived social pressure to (not) perform a particular behavior, in this case whether important others are using a green app (descriptive norms) and/or think they should use green apps (injunctive norms). As mentioned earlier,
self-efficacy refers to the extent that people believe themselves to be capable of successfully executing the behavior that is required to produce the desired outcome (Bandura, 1977), in this case whether people are able to successfully use the app. The TPB assumes that, the more favorable the attitude and subjective norms concerning a particular behavior, and the greater the self-efficacy, the stronger a person’s intention to perform the behavior (Ajzen, 1991).
The TPB has been extensively used to predict waste separation behavior (Árnadóttir et al., 2019; Ayob, Sheau-Ting, Jalil, & Chin, 2017; Xu, Ling, Lu & Shen, 2017; Zhang, Huang, Yin & Gong, 2015), and app use (e.g. Cheon, Lee, Crooks & Song, 2012; Yang, 2013), but not for green app use. This study aims to test the accuracy of the TPB model in explaining the actual use of green mobile applications. With regard to using green apps, it is expected that having a favorable appraisal of using the app (i.e. attitude), feeling social pressure to use the app (i.e. subjective norms), and evaluating oneself capable of using the app (i.e. self-efficacy) would lead to a high intention to use the app which in turn would lead to actual application use. Thus, in line with the TPB model, it is hypothesized that:
H2: The more favorable the attitudes and subjective norms towards using the app, and the greater the self-efficacy, the stronger a person’s behavioral intention to use the mobile application, which will lead to a higher application use.
The Technology Acceptance Model (TAM). Alternatively, the Technology
Acceptance Model (TAM) could also potentially explain application use. The TAM was first introduced by Davis (1985), and aims to specifically explain the users’ acceptance of the technology. Davis (1985) based his model on the TPB model, but specifically adjusted it to model user acceptance of information systems, such as apps. The model rests on two main attitudes: perceived usefulness and perceived ease of use. Davis (1985) defined perceived usefulness as the user’s perceived probability that using the technology will increase one’s correct execution of an action, and perceived ease of use as the extent to which the user expects that using the technology will be effortless. Since then, it has become one of the most important models for explaining information systems adoption mechanisms (Li, Qi & Shu, 2008).
Comparing the TAM and TPB model. As Davis (1985) built the TAM based on the TPB model, naturally the models show some overlap. This overlap is shown in that both the models include the mediation structure of attitudes towards, in this case, the app usage behavior influencing the actual app usage behavior through behavioral intention to use the app. While the TAM primarily and specifically includes beliefs about usefulness and ease of use (Davis, 1985), the TPB model includes general attitudes towards a behavior (Ajzen, 1991). Similarly, while the TPB model includes the person-related measure of self-efficacy that focuses on performing a behavior successfully, the TAM includes the more specific perceived ease of use that focuses more on the easiness of a behavior.
However, there exist also clear distinctions between the two models. For example, while the TPB explicitly includes social variables (i.e. subjective norms), the TAM does not. However, in case the situation and theory calls for it, it can be decided to include subjective norms as an external variable into the model, which is believed to have an effect on both perceived ease of use as well as perceived usefulness (Li, Qi & Shu, 2008). That is, over the course of the years, there have been many variations on the original TAM (Davis, Bagozzi & Warshaw, 1989; Davis, 1993; Venkatesh & Davis, 1996; Venkatesh & Davis, 2000;
Venkatesh & Bala, 2008). In those extended TAM models, the basic TAM is extended by incorporating external variables (e.g. self-efficacy, subjective norms) that affect the perceived ease of use and perceived usefulness of the application, which in turn influence attitude towards using the application, and the behavioral intention to use the application, which eventually affect the actual application use.
Both the original TAM, and extended TAMs have been used extensively to explain users’ adoption of mobile applications, for example for mobile tourism applications (e.g. Chen & Tsai, 2017), mobile game-based learning (e.g. Ghani et al., 2019), and e-learning
(Abdullah, Ward & Ahmed, 2016). In the current study, the focus lies on predicting the actual use of green apps, thereby conceptually replicating and extending previous research by investigating the TAM in a different field. To accurately compare the TAM with the TPB model, subjective norms and self-efficacy were included in the TAM as external variables. Thus, in line with the TAM it is hypothesized that:
H3: The stronger subjective norms or self-efficacy, the higher the perceived ease of use of the mobile application (PEOU). This leads to a higher intention to use the application, which is (partially) mediated by perceived usefulness (PU), and this leads to a higher application use.
So whereas the original TAM in its core is far more easy and straightforward than the TPB model, it is possible to extend the core variables perceived ease of use, perceived usefulness, behavioral intention and actual system use with self-efficacy, and subjective norms. To sum up, previous researchers have pointed out that the original TAM is more simple and easy to use, but the TPB is more specific and provides more details to explain the behavioral intention because the TPB is a more complex model with several independent variables that can capture different aspects of a person’s belief (Chuttur, 2009; Mathieson,
1991). Furthermore, the extended TAMs offer the possibility to combine the best of both models by including external variables as self-efficacy and subjective norms. In this study, the two models will be compared in how suitable they are to specifically predict green app use, where the TPB, TAM, and extended TAMs (one where PEOU and PU are preceded by self-efficacy, and one where they are preceded by subjective norms) will be compared. Previous research has shown that the TAM had a slight empirical advantage over the TPB by explaining more variance in a user’s intention to use an information system (Mathieson, 1991). Therefore, it is expected that the first will be more accurate to explain behavior (especially when extended with external variables). This leads to the following hypothesis:
H4: The TAM (especially the extended TAMs) is more suitable to accurately predict app usage compared to the TPB.
Method Design
This experiment used a within-subjects design to study (1) the effects of using a mobile recycling application on recycling knowledge, self-efficacy, intentions, and behavior and (2) which model (TPB, original TAM, or extended TAMs) predicts the usage of a recycling app best. The data was collected by using the experience sampling methodology (ESM): a data collection method in which participants at certain moments in time respond to repeated questionnaires while they just continue living their natural lives (Shiffman, Stone & Hufford, 2008). The specific type of ESM incorporated in this study was interval-contingent sampling, which means that participants had to complete self-reports after a designated interval, in this case daily (Scollon, Prieto & Diener, 2009). Participants had to complete assessments on a daily basis over the course of 14 days, see Figure 4. The data collection period was split up in two weeks: Week 1, baseline, where daily recycling behavior was
assessed before participants downloaded the recycling app, and Week 2, intervention, in which participants could use the app and where recycling behavior was once more assessed. ESM is exceptionally suitable for this particular study, because it allows the comparison of recycling behavior prior to and after downloading the app. Such comparisons facilitate the understanding of within-person processes (Scollon, Prieto & Diener, 2009). In addition, it increases the ecological validity of this study by taking the study out of the laboratory and into real-life (Scollon, Prieto & Diener, 2009). ESM also solves methodological problems, such as memory biases (Scollon, Prieto & Diener, 2009).
While the ESM aimed to measure participants’ actual recycling behavior and
application use, this study also focused on the role of knowledge and self-efficacy concerning recycling behavior, and the measurements of these variables were included prior to the study at T0, at Day 7 (T1), and Day 14 (T2). In addition, the variables necessary for model
comparison of the TPB, the TAM, and extended TAMs concerning application use were measured at T1.
Summarizing, the baseline data was collected through an extensive questionnaire prior to the study (T0) that measured the knowledge, self-efficacy, and intention variables concerning recycling behavior, and control variables. Next, data collection took place in 14 consecutive daily waves, with the extensive T1 questionnaire and task to download the mobile application at Day 7, and the final extensive follow-up T2 questionnaire at Day 14. For an overview of the timeline, see Figure 4.
Participants
Participants were recruited through three types of channels. First, several Facebook posts with calls to participate were posted in Facebook groups (i.e. voluntary response sampling). Particularly worth mentioning is the Facebook recruitment post that Milieu Centraal posted in its Facebook recycling group. Second, the researcher relied on her own network to further recruit participants on social media (e.g. Facebook, and WhatsApp; convenience sampling). These participants were subsequently asked to recruit members from their own networks as well (i.e. snowball sampling). Lastly, undergraduates from the
University of Amsterdam voluntarily participated in the study in exchange for credits toward a course requirement.
Within-subjects designs require less participants compared to between-subjects design to detect effects of comparable sizes and have therefore more power than
between-subjects design of a similar sample size (Thompson & Campbell, 2004). The questionnaires were administered in two separate groups. Group 1, starting on 11 May 2020, contained 243 participants, and group 2, starting on 18 May 2020, contained 28 participants.
To be included in the final sample, participants had to complete the three extensive questionnaires at T0, T1 (Day 7), and T2 (Day 14), and at least 4 out of the 7 short, daily ESM questionnaires for week 1, as well as for week 2. Of the 271 Dutch-speaking
participants, 159 of these completed all three extensive T0, T1, and T2 questionnaires. After excluding participants who had not finished at least 4 out of 6 daily questionnaires in both week 1 and week 2, 139 participants were included in further analysis. The manipulation consisted of downloading the Waste Separation Index app on one’s smartphone. This was checked by asking participants to check the app for the correct waste category for three types of waste. The manipulation was determined to be successful, if 1) a participant correctly answered at least 2 out of 3 of the control questions, and 2) if the time to answer these questions did not exceed 180 seconds. Of the N = 139 participants that had completed their participation in this study, N = 21 failed the manipulation and were excluded from further analysis. Therefore, N = 118 successfully completed the manipulation and were included in the analysis.
Of this sample, 74.6% was female (n = 88) and 25.4% male (n = 30), age ranged between 15 and 69 (M = 31.92, SD = 14.24). Most of the participants (50%) had at least completed a higher secondary education (in Dutch: Havo- en VWO-bovenbouw, MBO-2-4) (further sample characteristics are shown in Table 1, and in Appendix A).
Procedure
The questionnaires were administered through Qualtrics between 9 May 2020 and 31 May 2020. When signing up to participate in the study, participants opened a link to the baseline (T0) questionnaire. After providing informed consent, participants filled out their email address for receiving the follow-up questionnaires. Subsequently, the knowledge, self-efficacy, and intention variables concerning recycling were measured, as well as
demographic variables, including a control question on changes in recycling behavior due to COVID-19 to check whether people’s recycling behavior was not affected too much by the current COVID-19 pandemic, which was not the case (M = 2.26, SD = 1.53, 1-7 scale). Finally, participants were asked to check their email inboxes daily for the subsequent questionnaires.
The daily ESM questionnaires were sent out everyday at 5 PM. In case participants had not yet started, or had not completed the questionnaire, they would receive a reminder at 8 PM, and if necessary, a final reminder at 7 AM the next morning. All the questionnaires included in the final sample were filled out within one day, and on four rare occasions within
two days. The daily questionnaires of week 1, that is Day 1 to Day 7, were short
questionnaires that took approximately 3-5 minutes to fill out. Participants were asked to indicate for several waste categories how often they had separated those types of waste today. In addition, participants were asked to indicate which sources they had used that day to learn about the correct way to separate waste, including the use of mobile applications.
On Day 7, in addition to the daily ESM questionnaire, participants would receive the T1 questionnaire. Participants were instructed to download the Milieu Centraal Waste
Separation Index app (in Dutch: Afvalscheidingswijzer), and were provided with directions to download the app. The Waste Separation Index app falls in the category of waste
management apps providing people with information on in which category to separate their waste. Milieu Centraal is a Dutch information organization founded in 1998 by the former Dutch ministry of Environmental Management (Ministerie van Milieubeheer, VROM). Their main goal is to inform the public about making green choices by providing them with
practical information based on scientific knowledge. The app has a straightforward interface with one main screen that provides users with the option to fill in a product and receive information on how to recycle this product, for example, ‘box of a delivery pizza’ belongs to the general waste once stained with grease. The app shows waste categories like general, waste, paper, plastic, and glass waste, or shows that the waste should be taken to a disposal location (e.g. domestic chemical waste). Furthermore, the app provides information on why the (packing of the) product should be recycled along with that particular type of waste, and general information on the specific waste category (for screenshots of the app, see Appendix B).
After downloading the app, participants were asked to check the app for the correct waste category for three types of waste: 1) a hairspray can; 2) a tea light cup; and 3) a
window envelope. Serving as a manipulation check, to control whether they downloaded and understood how to use the app, they were asked to remember the correct answers, and fill them out in the next screen as quickly as possible. Next, the questionnaire contained the TAM and TPB measures on application use. In addition, the measures on recycling behavior, recycling intention, recycling self-efficacy, and recycling knowledge were measured again. Then, participants were instructed to keep the application installed on their smartphones, and were told that they were allowed to use the application as they preferred.
Subsequently, week 2 (i.e., Day 8 to Day 14) of the daily ESM questionnaires
followed. These daily questionnaires were similar to the ones that were administered in week 1. On Day 14, next to the daily ESM questionnaire, the final questionnaire (T2) was
administered, that is, the measures on recycling behavior, recycling intention, recycling self-efficacy, and recycling knowledge were measured for a final time. Lastly, participants were thanked for their participation, and then the study ended. All materials and data can be found on the Open Science Framework (OSF),
https://osf.io/fguc5/?view_only=7fa535f3d305483b99ef786755121088.
Measurements
Recycling behavior per week (ESM). Every day participants were asked to indicate for eight waste categories (e.g. ‘organic waste’, ‘tinplate waste’, ‘plastic waste’) how often they had separated waste that day on a 5-point Likert scale (1 = never; 5 = always). For each waste category, a scale was created for each week (i.e. day 1-7, and day 8-14) to measure recycling behavior prior to and after downloading the application. A total scale was created per week by calculating a grand mean for all waste categories. For the Cronbach’s alphas, Means and SD’s of the scales, please see Table 2. For an overview of all items, see OSF.
Application use (ESM). In case participants had indicated that for a specific day they had separated their waste for at least one waste category (see the measurement ‘Recycling behavior per week’), they were asked whether they used a source for knowledge on how to correctly separate waste of which one option was whether they had used an app. This was used as an unobtrusive measure for recycling application use. For each week (i.e. Day 1 to 7, and Day 8 to 14) a percentage was calculated to represent how many times participants had used an app during that week given the amount of times they had filled out the ESM questionnaires (i.e. 4, 5, 6, or 7 times). These percentage scores were used to compare application use before and after downloading the Waste Separation Index app as a dependent variable in the TAM and TPB models.
Measurements concerning recycling behavior. For the Cronbach’s alphas, Means and SD’s of the scales, please see Table 3. All items were measured on a 7-point scale from 1 (Strongly agree) to 7 (Strongly disagree), unless mentioned otherwise. For an overview of all items, see OSF 1.
Recycling behavior was measured with four items like ‘I actively recycle and 1
separate my waste’ (cf. Zhang, Huang, Yin & Gong, 2015). Recycling intention was
measured with four items like ‘I plan to regularly separate my waste’ (cf. Ayob, Sheau-Ting,
1The variables below have been measured and analyzed (see Appendix C) for exploratory purposes, but were later decided to not include in this research: 1) Educational value of the app was measured with two items (i.e. ‘To what extent did you learn from the Waste Separation Index app?’ and ‘Did you learn through the Waste Separation Index app that you previously have been separating your waste incorrectly?’) on a 7-point scale from 1 (Strongly agree) to 7 (Strongly disagree). These items were found to be significantly correlated (r = .55, p < .001). 2) Recycling attitude was measured with four items on a 7-point Semantic Differential scale (e.g. ‘Separating my waste is…’; 1 = Extremely bad, 7 = Extremely good; cf. Ajzen, 2006). 3) Recycling subjective norms were measured with five items. Two items were based on the scale by Ajzen (2006; e.g. ‘Most people who are important to me think that I should separate my waste), and three items were based on the scale used by Zhang et al. (2015; e.g. ‘My friends expect me to separate my waste).
Jalil & Chin, 2017). Recycling self-efficacy was measured with four items like ‘If I wanted to I could correctly separate my waste’ (cf. Ajzen, 2006). A reversed coded item led to a low reliability at T1 (⍺ = .64), and was therefore deleted. The final scales therefore consisted of 3 items.
Recycling knowledge was measured by asking participants to select the correct waste separation category (i.e.. ‘general waste’, ‘paper waste’, ‘plastic waste’, or ‘glass waste’) for 8 different waste products (e.g. ‘empty bag of chips’, ‘empty pizza box with fat stains’) based on the method used by Árnadóttir et al. (2019). However, while conducting the study the advice for one of the waste products (i.e. ‘garbage bag made from black plastic’) was changed in the Waste Separation Index app. Therefore, it was decided to exclude this
measure from further analysis. The amount of correct answers for the remaining 7 items were summed up to constitute a 0-7 scale.
TAM and TPB measurements concerning application use. For the Cronbach’s alphas, Means and SD’s of the scales, please see Table 3. All items were once more measured on a 7-point scale from 1 (Strongly agree) to 7 (Strongly disagree), unless mentioned
otherwise. For an overview of all items, see OSF.
Application perceived usefulness was measured with three items like ‘The Waste Separation Index app can help me to separate my waste more correctly’ (cf. Muñoz-Leiva, Climent-Climent & Liébana-Cabanillas, 2017). Application ease of use was measured with four items like ‘I find the Waste Separation Index app easy to use’ (cf. Muñoz-Leiva et al., 2017). Application use subjective norms were measured with two items, one measured injunctive norms (i.e. what we think other people approve of) and one measured descriptive norms (i.e. what we think other people do; cf. Ajzen, 2006). However, the two items, ‘Most people who are important to me would approve it if I would use the Waste Separation Index
app or similar app’ (M = 5.20, SD = 1.47) and ‘Most people I identify with use the Waste Separation Index app or similar app’ (M = 3.05, SD = 1.46), were not significantly correlated (r = .08, p = .381). Therefore, no scale could be made for subjective norms. According to Ajzen (2002), injunctive norms are more consistent with the concept of subjective norms. Therefore, the item ‘Most people who are important to me would approve it if I would use the Waste Separation Index app or similar app’ was used to represent application use subjective norms. Application use self-efficacy was measured with four items like ‘If I wanted to I could correctly use the Waste Separation Index app’ (cf. Ajzen, 2006). Application use attitude was measured with four items on a 7-point Semantic Differential scale (e.g. ‘Using the Waste Separation Index app is...’; 1 = Extremely bad, 7 = Extremely good; cf. Ajzen, 2006). Lastly, application use intention was measured with three items like ‘I plan to regularly use the Waste Separation Index app’ (cf. Ayob, Sheau-Ting, Jalil & Chin, 2017).
Analysis
The data files were downloaded from Qualtrics, and merged together using R (for the code see R-file in OSF, https://osf.io/fguc5/?view_only=7fa535f3d305483b99ef786755121088). Next, the data was processed, and analyzed using IBM SPSS Statistics Version 25. Hayes’ PROCESS macro for SPSS version 3.5 was used to test the mediation models. Correlations between the variables can be found in Appendix E.
Results and Discussion Stimulating Recycling Behavior with Application Use
Hypothesis 1 stated that using an app would have a positive effect on performing recycling behavior, mediated by knowledge, self-efficacy, and intention to perform the
behavior. As expected, a paired-samples t-test showed that downloading the app increased the use of an app (as a ratio given the amount of days a participant had filled out the ESM
questionnaire) as a source to retrieve recycling knowledge, T1 (M = 0.48, SD = 4.15) compared to T2 (M = 9.47, SD = 14.31), t(117) = -6,91, p < .001.
Next, to test for differences in time in knowledge, repeated-measures ANOVAs were performed to test for differences between knowledge at T0 (before the start of the study), T1 (pre-downloading the app), and T2 (after downloading the app). When testing for differences in knowledge scores (amount of correctly answered questions) over time, Mauchly's test, χ2(2) = 2.04, p = .361, did not indicate any violation of sphericity. The repeated measures ANOVA showed that there was a significant difference between the knowledge score means over time, F(2, 234) = 13.79, p < .001. Post-hoc analysis showed that knowledge at the start of the study (T0; M = 4.87, SD = .11) was significantly lower compared to T1 (M = 5.25, SD = .11), and a week after downloading the app (T2; M = 5.34, SD = .11). This knowledge gained between T0 and T1 did not further increase between T1 and T2.
For recycling self-efficacy, Mauchly's test, χ2(2) = 2.12, p = .346, did not indicate any violation of sphericity. Furthermore, the results of the repeated-measures ANOVA showed that there was a significant difference in self-efficacy over time, F(2, 234) = 9.30, p < .001. Post-hoc analysis showed there was no significant difference between reported self-efficacy at the start of the study (T0; M = 5.68, SD = .90) and at T1 (M = 5.71, SD = 1.02). However, unexpectedly, downloading the app at T1 led to a decrease in self-efficacy at T2 (M = 5.41, SD = 1.21), which will be discussed in the General Discussion.
Next, when performing a repeated-measures ANOVA to test for differences in recycling intentions over time, Mauchly's test of sphericity was violated, χ2(2) = 23.84, p < .001, therefore the degrees of freedom were corrected Huynh–Feldt (ε = .854). The results showed, unexpectedly, that the reported recycling intentions did not significantly differ between T0 (M = 6.28, SD = 0.87), T1 (M = 6.20, SD = 0.85), and T2 (M = 6.29, SD = 0.81), F(1.69, 197.34) = 1.32, p = .267.
Application Use and Actual Recycling Behavior. Next, it was tested whether downloading the application led to an increase in actual recycling behavior. Paired-samples T-tests were performed to test for differences in recycling behavior before (week 1) and after downloading the app (week 2). Recycling behavior consisted of a total scale that was created per week by calculating a grand mean for recycling behavior for 8 waste categories. The results showed that, as expected, downloading the app led indeed to a significant increase of recycling behavior in general between week 1 (M = 3.53, SD = 1.14) and week 2 (M = 3.77, SD = 1.16), t(117) = -3.44, p = .001.
Because the general recycling behavior scale consisted of 8 different recycling categories, the differences between those categories were assessed as well. The amount of products that could potentially be recycled per week differed greatly between waste
categories (N), as did the recycling mean itself (M, see Table 4). Those differences between the categories can be explained, for example, by differences in the way the waste is collected (i.e. front of house waste collection vs. waste collection points; see Appendix B). Therefore, it was of importance to take the differences between these categories into account, and subsequent paired-samples t-tests were performed per waste category. Recycling behavior significantly improved between week 1 and week 2 for paper waste, and tinplate cans, but not for the other categories (e.g. plastic waste, organic waste, etc.; see Table 4).
Mediation Application Use through Knowledge, Self-efficacy, and Intention on Recycling Behavior. As mentioned above, Hypothesis 1 proposed that an increase in app use would have a positive effect on performing sustainable behavior, mediated by knowledge, self-efficacy, and intention to perform the behavior. A bootstrapping analysis with 5000 samples (using Model 6 Hayes, 2013) disconfirmed sequential mediation of application use through recycling knowledge, recycling self-efficacy and recycling intention on recycling behavior (indirect effect = 0.0003, SE = 0.0003, 95% CI [-0.0001; 0.0011]). The analysis showed that whereas using the app more did not lead to more recycling knowledge (b = 0.0069, se = 0.0065, t = 1.05, p = .296), recycling knowledge, however, had a positive effect on feelings of self-efficacy regarding recycling (b = 0.2643, se = 0.0833, t = 3.17, p = .002),
which subsequently positively affected the intention to recycle (b = 0.2526, se = 0.0520, t = 4.86, p < .001), and ultimately intention to recycle increased the actual recycling behavior (b = 0.5854, se = 0.1016, t = 5.76, p < .001). Therefore, hypothesis 1 was partially accepted.
Subsequently, this mediation was tested for recycling behavior of each of the eight waste separation categories by using PROCESS Model 6 (Hayes, 2013) to perform the bootstrapping analyses with 5000 samples. Sequential mediation of application use through recycling knowledge, recycling self-efficacy and recycling intention on recycling behavior for a specific waste category was confirmed for three of the eight waste categories (i.e. tinplate waste, beverage cartons, and glass waste; see Table 5). These indirect effects are, however, very small and seem to be driven by the rear part of the model (the effect of knowledge on self-efficacy, intention and behavior) as there was no significant effect of app use on recycling knowledge. The sequential mediation was not confirmed for the remaining five waste categories (i.e. bioplastic, organic waste, domestic chemical waste, paper waste, and plastic waste; see Table 5).
Predicting Application Use: TPB
Hypothesis 2, based on the TPB, stated that the more favorable the attitudes and subjective norms towards using the app, and the greater the self-efficacy, the stronger a person’s behavioral intention to use the mobile application which would lead to a higher application use. Prior to the analyses, the dependent variables were assessed on normality. Intention to use the application was non-normally distributed, with skewness of -0.82 (SE = .22 and kurtosis of 0.26 (SE = 0.44). Application use was also non-normally distributed, with skewness of 1.83 (SE = .22 and kurtosis of 3.69 (SE = 0.44). However, according to Hayes (2018) the normality assumption 1) is often not met in reality, and 2) is one of the least important assumptions in linear regression analysis. Therefore, it was decided to still conduct multiple linear regression analyses to test Hypothesis 2.
First, a multiple linear regression was conducted to predict behavioral intention to use the app based on the attitude towards the app, subjective norms, and self-efficacy. Results indicated that together, attitude towards the app, subjective norms, and self-efficacy statistically significantly predict the behavioral intention to use the app. Furthermore, the model explained 40% of the variance in behavioral intention (see Table 6). However, only attitude towards the app, but not subjective norms and self-efficacy, was a significant predictors of intention to use the app.
Next, a second multiple linear regression was conducted to predict application use (i.e. as a ratio given the amount of days a participant had filled out the ESM questionnaire) based on attitude towards the app, subjective norms, self-efficacy, and intention to use the app. Together these variables did not significantly predict the app usage behavior (see Table 6). The model explained 6% of the variance in app use behavior. However, in this model behavioral intention significantly predicted app usage (see Table 6). To further investigate the
effect of intention to use the app on actual application use, and explore whether there was also a relationship when not taking attitude towards the app, subjective norms, and self-efficacy into account, a single linear regression was conducted. Results showed that behavioral intention as a single variable did not significantly predict application behavior, with the model explaining 3% of the variance in app use behavior (see Table 6).
In conclusion, the results indicate that the TPB has predictive value for app use intentions, but not for actual app use. Therefore, Hypothesis 2 was partially accepted.
Predicting Application Use: TAM
In Hypothesis 3, it was proposed that the TAM would explain application use in such a way that perceived usefulness (PU) and intention to use the application would mediate the positive effect of perceived ease of use of the mobile application (PEOU) on application use. Subjective norms and self-efficacy, as external variables, would positively affect perceived ease of use of the mobile application.
Original TAM. First, the original TAM without any external variables was tested. The TAM proposes three sequential mediation paths: 1) sequential mediation of PEOU through attitude towards app use and app use intention on actual application use; 2)
sequential mediation of PEOU through PU and app use intention on actual application use; and, the main mediation: 3) sequential mediation of PEOU through PU, attitude towards app use, and intention to use the app on actual app use. Since there were no pre-programmed PROCESS models available, a custom model was built to specifically test the conceptual model including the three sequential mediation paths (for the custom model code, see OSF).
A bootstrapping analysis with 5000 samples disconfirmed sequential mediations of PEOU on application use (see Table 7). The analysis further showed, as predicted, that a higher PEOU leads to a higher PU and more positive attitude towards using the application. A higher PU leads to a more positive attitude towards using the app. Both a higher PU and more positive attitude lead to a stronger intention to use the app. There was no significant direct effect of behavioral intention on actual application use (see Figure 5).
To further investigate the mediation processes within the model excluding the
insignificant effect of intention on application use, the model was ran again with intention to use the app as the dependent variable. A bootstrapping analysis with 5000 samples confirmed sequential mediation of PEOU through PU and attitude on intention. A higher PEOU lead to a higher PU, which subsequently had an effect on attitude towards application use, which ultimately enhanced intentions to use the app (see Figure 5). Sequential mediations of PEOU through PU on intention, and of PEOU through attitude on intention were disconfirmed (see Table 7).
In conclusion, the mediation analyses show some evidence for the main mediation as hypothesized in the TAM when behavioral intention is the dependent variable (i.e. sequential
mediation of PEOU through PU and attitude on intention). However, the other proposed mediations (e.g. 1) PEOU through attitude towards app use on app use intention; and 2) PEOU through PU on app use intention), and the meditation model that had application use as the outcome, showed no significant results, and therefore no support for the theoretical mechanisms that underlie TAM.
Extended TAM: subjective norms. Next, subjective norms were added to the TAM as an external variable that would influence both PEOU and PU. Again, this model was tested by building a custom PROCESS model (for the custom model code, see OSF). A
bootstrapping analysis with 5000 samples disconfirmed sequential mediations of subjective norms on application use (see Table 7). Excluding application use as the outcome variable and replacing it with application intention, confirmed sequential mediation of 1) subjective norms through PU and attitude on intention; and 2) subjective norms through PEOU, PU, and attitude on intention. A stronger perceived subjective norm led to a higher PEOU and PU, PU subsequently had an effect on attitude towards application use, which ultimately enhanced intentions to use the app (see Figure 6).
Extended TAM: self-efficacy. Finally, self-efficacy was added to the TAM as an external variable that would influence both PEOU and PU. Again, this model was tested by building a custom PROCESS model (for the custom model code, see OSF). A bootstrapping analysis with 5000 samples disconfirmed sequential mediations of self-efficacy on
application use. Excluding application use as the outcome variable and replacing it with application intention, also led to disconfirmed sequential mediations of self-efficacy on intention to use (see Table 7 and Figure 7).
In conclusion, of all the models, only the sequential mediations with intention to use as the outcome variable were confirmed. No sequential mediation paths were confirmed for models with application use as the outcome variable. In contrast to self-efficacy, two
sequential mediation paths were confirmed when subjective norms was added to the model as an external variable. However, self-efficacy itself showed significant direct effects on both PEOU and PU. Thus, the evidence did neither completely support, or completely disprove the TAM. Therefore, hypothesis 3 was partially accepted.
Predicating Application Use: TPB vs. TAM
Hypothesis 4 stated that the TAM (especially the extended TAMs) would be more suitable to accurately predict app usage compared to the TPB. To compare the explained variance in both models, multiple linear regressions were carried out to assess the TAM as PROCESS is not yet able to give the explained variance for custom build models.
A simple linear regression was conducted to predict the effect of PEOU on PU. Results indicated that PEOU statistically significantly predicted PU, and the model explained 12% of the variance in PU (see Table 8). Next, a multiple linear regression was conducted to predict the attitude towards app use based on PEOU and PU. Together these variables significantly predicted the attitude towards the app use (see Table 8), and the model
explained 30% of the variance in attitude towards the app use. Subsequently, a multiple linear regression was conducted to predict the intention to use the app based on PEOU, PU, and attitude. Together these variables significantly predicted the intention to use the app, and the
model explained 41% of the variance in intention, F(3, 114) = 26.83, p < .001. In this model, PU (t = 2.88, p = .005) and attitude (t = 5.70, p < .001) were significant predictors of
intention. However, PEOU (t = 0.08, p = .938) did not significantly predict intention (see Table 8). Finally, a multiple linear regression showed, in line with the custom build PROCESS model, that PEOU, PU, attitude, and intention did not significantly predict application use, with the model explaining only 6% of the variance, F(4, 113) = 1.94, p = .109.
Next, the same sequence of multiple linear regressions was carried out for the TAM with subjective norms as an external variable. As hypothesized, subjective norms
significantly predicted both PEOU, and PU. The model consisting of subjective norms, PEOU, PU, and attitude towards using the app significantly predicted intention to use the application, and explained 42% of the variance in intention, F(4, 113) = 20.53, p < .001. In this model, attitude towards using the app (t = 5.50, p < .001) and PU (t = 2.70, p = .008) were significant predictors of intention, while PEOU (t = -0.01, p = .994) and subjective norms (t = 1.17, p = .246) were not. Lastly, a multiple linear regression showed, again in line with the custom build PROCESS model, that subjective norms, PEOU, PU, attitude, and intention did not significantly predict application use, with the model explaining only 8% of the variance, F(5, 112) = 2.06, p = .076 (see Table 9).
A similar procedure was carried out for self-efficacy as an external variable in the TAM. Again, self-efficacy significantly predicted both PEOU, and PU. The model consisting of self-efficacy, PEOU, PU, and attitude towards using the app significantly predicted
intention to use the application, and explained 42% of the variance in intention, F(4, 113) = 20.44, p < .001. In this model, again, attitude towards using the app (t = 5.60, p < .001) and PU (t = 2.53, p = .013) were significant predictors of intention, while PEOU (t = -0.39, p =
.695), and self-efficacy (t = 1.08, p = .284) were not. Lastly, a multiple linear regression showed, again in line with the custom build PROCESS model, that self-efficacy, PEOU, PU, attitude, and intention did not significantly predict application use, with the model explaining only 8% of the variance, F(5, 112) = 1.93, p = .095 (see Table 10).
When comparing the TPB, original TAM, and extended TAMs on their explanatory power regarding intention to use the app, all models explained variance in intention to use the app similarly well (but not actual app use). The extended TAM models (TAM + subjective norm: 42%; TAM + self-efficacy: 42%) had slightly higher scores compared to the basic TAM (41%) and TPB (39.8%). Therefore, hypothesis 4, stating that the TAM (especially the extended TAMs) would be more suitable to accurately predict app usage compared to the TPB, was partially accepted.
Conclusion and Discussion
The current study had two aims: 1) investigating if using a mobile application that provides knowledge on recycling improves recycling behavior, and 2) investigating what factors influence green app use, specifically testing whether the TPB model or the TAM is more accurate in predicting green app use.
App Usage Leading to Behavior Change
Hypothesis 1 proposed that using a green app would have a positive effect on performing sustainable behavior, mediated by knowledge, self-efficacy, and intention to perform the behavior. The results showed that, as expected, when participants downloaded the app this led to an increase in using the app as a source for retrieving recycling knowledge and to a significant increase in recycling behavior. However, downloading the app led to no differences in recycling intention, and, interestingly, to a decrease in self-efficacy. Also
notably, the results showed that knowledge scores at the start of the study were lower compared to right after downloading the app (T1), and one week after downloading the app (T2). However, there was no difference in knowledge between T1 and T2. In addition, using the app did not lead to more recycling knowledge. This suggests that knowledge did not increase further when the app was used more often. As expected, recycling knowledge, however, had a positive effect on self-efficacy beliefs regarding recycling, which
subsequently positively affected the intention to recycle, and ultimately the actual recycling behavior. Therefore, hypothesis 1 was partially accepted.
These findings provide important insights. First, there was an increase between T0 and T1 in knowledge, suggesting that keeping a diary of one’s recycling behavior and/or downloading the app and using it once already led to an increase in recycling knowledge, while using the app more often over time did not. An explanation for this could be that downloading the app and/or keeping a diary could have made recycling top of mind,
subsequently leading to more recycling due to priming of the topic. Priming effects leading to behavior effects are well-documented in the literature (e.g. of television commercials on eating behavior; Harris, Bargh & Brownell, 2009). Future research could investigate if this is indeed the case by, for example, letting participants keep track of their recycling behavior in a diary vs. not and compare these groups with a third group that also uses the app once (between-subject design). Another explanation is that participants indicated that, on average, in only 9% of the days that they had filled in the ESM questionnaires they had used the app. Therefore, it could be possible that the participants did not use the app enough to actually increase their recycling knowledge. It is assumed that the discrepancy between intention to use the app and actual app use shows that participants were interested in using the app, but did not see or did not have not a lot of opportunities to actually use the app. Naturally, this is
a limitation of this study. Although the study made use of a repeated-measures design over the course of two weeks, and therefore tried to enable opportunities for app use in its design, future research could investigate the actual app use for green apps that aim to change
behavior that can occur more often, such as grocery shopping, or travelling.
Secondly, the results show that using an app might actually decrease self-efficacy beliefs. This might be explained by people finding out that their current recycling behavior was actually not correct (see also Árnadóttir et al., 2019). At first sight this might seem a negative side-effect as low self-efficacy beliefs might hamper recycling behavior, however, this also indicates that people might have become aware of their incorrect recycling
behaviour and see opportunities to correct this. Future research could look into this.
Lastly, the results show that using a sustainability app is a good way to increase actual recycling behavior. It might not necessarily be the case that the specific app itself or the knowledge gained from the app can account for this positive effect. However, being occupied with the topic of recycling – either because of being more top of mind or because of other underlying mechanisms (e.g., knowledge, interest), leads to an increase in recycling. Sustainability apps therefore seem a good tool to stimulate sustainable behaviours Predicting App Use
Secondly, it was investigated what theories would predict app use best – as the positive effect of app use on recycling behavior is contingent on people actually using the app. Hypothesis 2, based on the TPB, stated that the more favorable the attitudes and subjective norms towards using the app, and the greater the self-efficacy, the stronger a person’s behavioral intention to use the mobile application which would lead to a higher application use. The results indicated that the TPB has predictive value for app use intentions, but not for actual app use, leading to partial acceptance of Hypothesis 2. Furthermore, in
Hypothesis 3 it was proposed that the TAM would explain application use in such a way that perceived usefulness (PU) and intention to use the application would mediate the positive effect of perceived ease of use of the mobile application (PEOU) on application use. Subjective norms and self-efficacy, as external variables, would positively affect perceived ease of use of the mobile application. Results showed that only the sequential mediations with intention to use as the outcome variable were confirmed. No sequential mediation paths were confirmed for models with actual app use as the outcome variable. Furthermore, the results showed support for the extended TAM that included subjective norms as an external variable, but not for the extended TAM with self-efficacy. Therefore, hypothesis 3 was partially accepted.
The results of this study showed strong evidence for both the TPB and the TAM to explain intention to use the app, but no effects were found for a relationship between
intention to use the app and actual app use. This is surprising, since in the TPB the intention to perform a behavior is the immediate antecedent of overt behavior (Ajzen, 1991). The same is true for the studied relationship between intention and behavior within the TAM. A review of thirty-four articles that studied the relationships among variables in the TAM reported ten significant relationships between behavioral intention and actual system use, but no
insignificant relationships, concluding that the relationship is stable and significant (Li, Qi & Shu, 2008). In addition, Venkatesh and Davis (2000) have indicated that they think that behavioral intention is a good independent variable to predict actual system use.
In the literature the inclusion of external variables in the TAM varies, depending on the context of that particular research. Examples of external variables are subjective norms, self-efficacy, enjoyment, and experience (Abdullah, Ward & Ahmed, 2016). The results of the current study showed support for adding subjective norms as an external variable to the
TAM, but no support for the model that included self-efficacy as an external variable. An explanation for the added value of subjective norms, could be that using a green app could be labelled as socially accepted: behavior that we want to display. Indeed, subjective norms can be effectively used in promoting sustainable behavior, and the effect is highest in cases that imitate real-life situations (Poškus, 2016). Self-efficacy, on the other hand, shows overlap with PEOU, since they both refer to the user’s required skills to effectively use the
application (Mathieson, 1991). This could explain that adding self-efficacy to the TAM, has no added value, when comparing those models to the TPB. The extended TAM with
subjective norms shows the most overlap between the original TAM and the TPB, making the model the ‘best of both worlds’.
When comparing the TPB, and TAM models on their explanatory power regarding intention to use the app, all models explained intention to use the app similarly well. This finding is in line with previous research that both the TPB and the TAM explain intention quite well (Mathieson, 1991). That study had a comparable conclusion as the current study: although the TAM explains (somewhat) more variance than the TPB, the differences are too small to draw the conclusion that the TAM is better than the TPB on a purely empirical basis (Mathieson, 1991). However, one could argue that for the particular topic of green app use, the TAM would be more suitable, since the TAM specifically focuses on technology use. Limitations
A limitation was that subjective norms concerning app use was measured by one item that regarded injunctive norms. Although injunctive questions are more consistent with the concept of subjective norms (Ajzen, 2002), research shows support for single-item measures when the underlying constructs are homogeneous, but this support is not strong enough to challenge the perspective that multiple-item scales are needed to reliably measure relatively
complex constructs (Loo, 2002). Therefore, future research that aims to replicate this study should use a multiple-item scale to measure subjective norms regarding app use.
In addition, when participants indicated that they had not separated their waste that day, they were not asked whether that day they had used an app as a source for recycling knowledge. In rare occasions, this could have meant that people that did not recycle that day but did use an app, but could not indicate that they had done so. Furthermore, the question was formulated in such a way that it asked whether participants had used an app in general, and not the Waste Separation Index app in particular – so to be able to keep the measurement the same in Week 1 and Week 2 and unobtrusively measure app use. This, however, implies that other app uses could have been included in the measure as well. Future research could therefore measure app use by specifically asking about the use of the app in question, and include that measure in each ESM questionnaire.
Practical implications
This study carefully concludes that, taking both empirical evidence and theoretical mechanisms into account, the TAM might be more suitable to predict the use of green apps. This provides app designers with a starting point on what variables to focus on (i.e. PEOU, PU, and subjective norms) and to take into account when designing a green app. It is thus important that the app is seen as useful to users, and easy to use. Furthermore, it would help if developers of an app could show that using the app is in line with subjective norms – for example, by using testimonials of people showing that using the app is something people should be doing.
Lastly, the finding that downloading a green app led to an increase in recycling behavior shows that there are promising opportunities for green apps that aim to increase green behavior. This study shows that, if a green app would succeed in significantly
increasing green knowledge, such an app could potentially lead to more behavior change. Notably, the potential for scalability is the main benefit of apps that aim to change behavior (Brauer et al., 2016). Therefore the (commercial) sustainability sector, NGOs, and
governments should consider the development of apps to enhance knowledge in their customers, and the public.
This study had interesting findings in a relevant and promising field. This study’s most important contributions show that using a green app can actually help improve
sustainable behavior. Furthermore, based on empirical evidence alone, the TPB and TAM are both suitable to predict green app use, and adding subjective norms as an external variable has added value to an extension of the TAM.
