Testing Scenarios to Achieve Workplace Sustainability Goals Using Backcasting and Agent-Based Modeling

Data Collection

We used a mixture of qualitative and quantitative data sources to build and configure an ABM. Specifically, a questionnaire was used as a basis for building decision-making algorithms for the agents for a number of different pro-environmental behaviors and for designing the social network. A backcasting scenario development workshop formed the basis for generating scenarios to explore within the model. The other data used pertain to the impacts of the behavioral choices agents make in terms of carbon emissions. The data used for this were obtained from the Environmental Office of the University.

Measuring Psychological Factors and Environmentally Relevant Behavior

The individual factors explored as determinants of pro-environmental behavior in organizations within this research belong to three categories, considered as the most relevant in previous research, and these are knowledge, motivations, and ability (Ajzen, 1987; Bamberg & Moser, 2007; Bonnes, Uzzell, Carrus, & Kelay 2007; de Groot & Steg, 2007; García-Mira, 2009; Stern, 2000; Stern & Dietz, 1994; Uzzell, Pol, & Badenas, 2002). A questionnaire was used to gather data on these psychological factors and to define the individual profiles of the actors in the ABM. Answers were provided on a scale ranging from 1 (totally disagree) to 7 (totally agree), with the exception of the scale for values, which ranged from −1 (opposed to my values) to 7 (supreme importance). The questionnaire was filled in by employees on a voluntary basis using Qualtrics as a web platform. The online questionnaire was distributed to the whole university staff through an email asking for their collaboration in the research project. Almost 400 individual entries were counted (almost 20% of the University staff). In the final analysis and model, however, a sample of 255 complete questionnaires was considered, and the others were excluded due to incomplete data (11% of the total University staff), using the method of listwise deletion of cases.

Within the category of knowledge, we included worldviews; for motivation, both distant and more proximal antecedents were included, and these were values, identity, and personal and social norms. For ability, measures of both self-efficacy and outcome efficacy were selected.

Worldviews have been defined as a set of assumptions about the physical and social reality that may have powerful effects on cognition and behavior (Koltko-Rivera, 2004). Worldviews were measured with six items from the New Human Interdependence Paradigm (NHIP) scale (Corral-Verdugo, Carrus, Bonnes, Moser, & Sinha, 2008; α = .88).

In terms of motivation, the first factor we considered were values, defined as a set of trans-situational goals that serves as a guiding principle in the life of a person or social entity (Schwartz, 1994). Four main types of values were measured through a 16-item scale: biospheric, altruistic, egoistic, and hedonic (α = .88, .69, .69, .76; for example, de Groot & Steg, 2007, 2008; Steg, Perlaviciute, Van der Werff, & Lurvink, 2014).

For identity, we considered three dimensions within the workplace: environmental self-identity, environmental organizational identity, and organizational identification. Environmental self-identity reflects the extent to which you see yourself as a type of person who acts pro-environmentally (Van der Werff, Steg, & Keizer, 2014; Whitmarsh & O’Neill, 2010). Environmental self-identity (α = .90) was measured with three items (adapted from Van der Werff et al., 2014). Environmental organizational identity refers to the degree to which the organization is perceived as defining itself as pro-environmental. Environmental organizational identity (α = .92) and organizational identification (α = .86), defined as the degree to which a worker identifies with the organization (Mael & Ashforth, 1992), were measured with six items each.

Norms have been found to be among the most relevant determinants of pro-environmental behavior (Carrus, Nenci, & Caddeo, 2009; Fornara, Carrus, Passafaro, & Bonnes, 2011; García-Mira, Real Deus, Durán, & Romay, 2003; Nolan, Schultz, Cialdini, Goldstein, & Griskevicius, 2008; Schultz, Khazian, & Zaleski, 2008). Besides social norms, personal norms have also been found to be a predictor of pro-environmental behavior (Schwartz, 1992; Stern, 2000), and they have been defined as self-expectations regarding own behavior. More recently, local norms, derived from social interactions localized in specific places where the behaviors are performed, were also proposed as a relevant local predictor of pro-environmental intentions and behaviors (Fornara et al., 2011). Both descriptive and injunctive norms were included, at general and local levels.

General descriptive norms (α = .83) and local descriptive norms (α = .81) were measured with four items each reflecting to what extent respondents believed that a certain reference group acts pro-environmentally at work (cf. Ajzen, 2006). The same was done for general injunctive norms (α = .85) and local injunctive norms (α = .79). Personal norms (α = .83) were measured with four items based on Steg and de Groot (2010).

Norms are transmitted through social networks. To serve the purposes of implementing a process of norm transmission within the ABM, we included a measurement of norm transmission, defined as the self-reported frequency with which workers encouraged others to act pro-environmentally in the workplace (the categories of “others” we included were co-workers, subordinates, supervisors, and the management team). The procedure for calculating rates of norm transmission and norm reception for workers and the description of the implementation of this process in the model have already been described elsewhere (Sánchez-Maroño et al., 2014).

Norm transmission (α = .91) was measured by asking the following question: How often do you encourage the following people to act pro-environmentally at work? The respondents rated the frequency of this behavior on a scale from 1 (never) to 7 (always), for four categories of people: subordinates, co-workers, supervisors, and the management team. Also, a question was introduced on the perception of having an exemplary role within the organization, with two answer options (yes, no).

Finally, the ability to act pro-environmentally was conceptualized as efficacy in this study. We distinguished between self-efficacy, defined as the confidence and perceived control that people experience to execute sustainable behavior (Ajzen, 2006), and outcome efficacy, defined as the extent to which people think they can do something about environmental problems by acting pro-environmentally (Schwartz, 1977). Both self-efficacy (α = .85) and outcome efficacy (α = .85) were measured with three items each (cf. Steg & de Groot, 2012).

Finally, the questionnaire measured the frequency of a list of environmentally relevant behaviors at work. In total, 27 items were used to measure behaviors chosen from three categories of environmentally relevant behavior: 11 items for consumption of materials and energy, eight items for waste generation and management, and eight for work-related mobility. Sociodemographic data were also collected.

Obtaining Different Policy Pathways Through the Use of Backcasting Scenario Development

To reach the objective of defining and testing several policy pathways to promote the adoption of pro-environmental behavior in the workplace, we used a methodology of backcasting scenario development, which allows options to be produced by real organizational actors.

A process-oriented, participative, and iterative backcasting scenario approach was used. Process-oriented scenarios (Quist & Vergragt, 2006; Robinson, 2003) are centered on ensuring effective participation of stakeholders and to produce, besides images of desired end-states, possible pathways to reach them and specific agendas for their implementation. It also used a stepwise approach to help participants generate the scenarios, guided by a series of structured questions (Dumitru et al., 2013). Formulations of the scenarios were narrative, as these are easier to handle by stakeholders than abstract representations.

Two workshops were carried out and a combined methodology was used, with mixed focus groups to develop the scenarios, inspired in part by the one used by Svenfelt, Engström, and Svane (2011) but significantly adapted to fit the objectives of this research, and the stepwise approach of Kok, Van Vliet, Bärlund, Dubel, and Sendzimir (2011), to orient the process and help stakeholders in getting disengaged with the present, which is considered to be one of the hardest aspects of backcasting scenarios both with stakeholders and experts (Svenfelt et al., 2011). As behavior change in organizations was a key aspect of this research, scenarios included a focus on who should change, which is often ignored in backcasting studies (Wangel, 2011).

The first workshop focused on the creation of the visions for the future of the organization and followed three steps: problem definition and establishment of targets, a brainstorming phase, and the final articulation of the main elements of the future images. Participants were provided with input on the specific country calendar for reducing emissions according to the Kyoto agreements, and these targets were established as system boundaries for the scenario development. The year 2050 was chosen as a target year, as it is the projection year most commonly used in the European Commission’s policy documents due to the fact that it is relatively easy to envision and also far enough away as to allow for radical change to happen (Vergragt & Quist, 2011). Information was provided on the contribution of each area of practices to overall emissions for the University, and this was used as a basis for deciding reduction targets in each of the areas of practice. This information is publicly available and provided by the organizational Environmental Office. Results of the questionnaire were not fed into the workshops, and these two methodologies were kept separate. Three future visions were produced, ranging from a more conservative to a more radical one.

The second workshop focused on defining the strategic pathways to reach sustainable future images and the social actors that should be involved. Participants were first reminded of the targets set in the first workshop and presented with the three narrative descriptions of the sustainable organization in 2050. Participants were then guided to reach a consensus on the most desirable future vision to work toward and set specific targets for emissions reduction in the three areas of interest. As mobility accounts for approximately 50% of all University emissions, a lot of the discussion focused on establishing targets for work-related mobility. A reduction of car use to 20% of university staff and students was considered worthwhile and feasible for 2050. This meant that 80% of the university population would use more sustainable means of transportation: 20% would come on foot, 30% would use bicycles, and another 30% would use public transportation. A stepwise approach was again used to define pathways, intermediary milestones, and actors involved at each stage.

A group of 12 people participated in the two workshops. About half of the composition of the first workshop was maintained for the second, and additional members were then invited. An initial list of about 30 University staff members was drafted, belonging to different academic departments and whose work had an environmental aspect to it. They thus played the double role of university staff and experts. Both workshops lasted between 3 and 4 hr. Participants at the event were invited to attend via mail, and the invitation included a short description of the working technique.

Dynamic Simulations of Policy Implementation: Designing the Agent-Based Model

The profile of the agents was defined by using psychological variables that are known to influence pro-environmental behavior in the three areas of interest. Instead of using theoretically derived assumptions, we grounded the profiles in empirical data obtained through a questionnaire. Workers’ decision-making processes were simulated using empirically grounded decision algorithms (for further information on the decision algorithms used, please refer to Online Appendix B and Figure S2), which gave us an opportunity to explore the relationship between individual trait factors such as values, or identity, and the importance of social influence processes (e.g., social norms) on the direction of behavior. The theoretical model, used as indication in the preprocessing stages of setting up the ABM, was the Value-Belief-Norm (VBN) model of Stern (2000), adapted by de Groot and Steg (2009) to include other relevant variables such as identity. However, once the theoretical decision-making model has been designed, the results obtained were almost purely data-driven.

The interaction between workers was modeled by taking into account both the formal structures of the organization (a hierarchical structure) and the more informal connections among workers, which have proven to be significant as sources of influence in organizational psychology research (Mehra, Kilduff, & Brass, 2001; Oh, Chung, & Labianca, 2004; Sparrowe, Liden, Wayne, & Kraimer, 2001). To our knowledge, the combination of questionnaire-based methods for data collection, backcasting scenario development, and ABM is novel in the study of pro-environmental behavior.

Several policy tracks and interventions were developed from the scenario exercise data, to be implemented in the organizational simulations, using ABM (Matthews, Gilbert, Roach, Polhill, & Gotts, 2007; Sánchez-Maroño et al., 2012). These policies were tested in different combinations to see their effects on the performing of certain behaviors and related emissions levels. The ABM is a prototypical simulation which included different types of individuals with different organizational roles, their interactions—represented as a social network following rules of interaction that are based on both hierarchical and horizontal relationships—and the environment of the organization.

For the organization under study, the ABM had the following goals:

Formal representations of the organization provided a test bed with which to examine the various ways scenario-based interventions affect overall system behavior as an emergent property of interactions. These provide a “tool to think with” (see, for example, McIntosh, Jeffrey, Lemon, & Winder, 2005; McIntosh, Seaton, & Jeffrey, 2007; Torrens, 2003), which may be used as part of wider discussions on the effectiveness of proposed measures. One advantage of ABM approaches in this regard is that it can be used to try out policy options that would be costly, or entail political risks if tried out in the real world. The results of such tests, if evaluated with respect to the assumptions embedded in the construction of the model, could be useful in determining whether real-world trials should be attempted, or whether the expected effect of the proposed measure is likely to be insufficient to merit exploration outside the domain of the simulation. Therefore, ABMs have been broadly used for sustainability research, with examples ranging from the exploration of the role of awareness in the consumption of a scarce resource (Sissa & Damiani, 2015) to the assessment of the performance of buildings given realistic occupants (see Andrews et al., 2011; Azar & Menassa, 2012). For more details on the implementation of the decision-making processes, how the ABM was validated, and how uncertainty was addressed, please refer to Online Appendix B.

As the theoretical model postulates that different value profiles will be associated with different behaviors, the 16 value items (refer to section “Measuring Psychological Factors and Environmentally Relevant Behavior”) collected through the questionnaire were used to identify different profiles by using a clustering technique. Five different clusters were identified, three of them representing “pure” profiles—biospheric, altruistic, and hedonic—while the remaining two were hybrid clusters (biospheric–altruistic and hedonic–egoistic). For the sake of brevity, the details of how these clusters have been identified are discussed in Online Appendix A. All simulated university employees were distributed in one of these five groups, and each cluster would exhibit different behaviors in the model (please refer to Table S1 in Online Appendix A).

The other aspect to be designed in an ABM is the interaction among agents, as they interrelate and change their behavior over time based on the observation of the behavior of others and/or the environment. Therefore, a social network must be designed (Ronald, Dignum, Jonker, Arentze, & Timmermans, 2012). Social networks are an important quality of formal organizations, as previous research has shown that employees tend to be more cooperative and productive when their formal contacts are accompanied by informal ties (Mehra et al., 2001; Oh et al., 2004; Sparrowe & Liden, 1997; Sparrowe et al., 2001). Our empirical findings suggest that activity- and place-dependent social networks do have a role to play in influencing everyday pro-environmental practices in organizations (García-Mira & Dumitru, 2014), which is also coherent with the findings of Barr et al. (2011). Within this network, the dynamics can determine the way of thinking of the individuals and their behavior.

Due to the specific characteristics of the studied organization, in which individuals have more autonomy (understood as the degree of control over how to perform job-related tasks) than in other more “formal” or hierarchical organizations, two networks working in parallel were used. The vertical or hierarchical network is the one employed for modeling the transmission of rules and policies of the governing bodies of the organization at different levels (research groups, departments, etc.). The horizontal network is the one modeling the relationships of companionship and friendship, and thus the transmission of norms and behaviors through social relations beyond the hierarchical structure of the organization, that a priori seem of utmost importance, especially in the case of a university. The horizontal network is constructed from the vertical network, linking the individuals in the bottom part of the vertical network with their peers in the research group (if teaching/research personnel) or administration service (if administrative personnel), taking into account their reported number of interactions on average, which was reported in the questionnaire by respondents.

A more detailed description of the ABM can be consulted in Online Appendix B, and the modeling of social networks is described in more detail in Sánchez-Maroño et al. (2015).

Experiments

Based on the results of the backcasting scenario development workshops, different interventions were tested for each of the three categories of behavior included in the study. Interventions to promote sustainable work-related mobility behavior were given priority. We will provide examples for three different types of interventions tested, with details on the experimentation included in the following section. The following types of interventions were tested: (a) targeting the increase in the proportion of University staff having a biospheric value profile; (b) targeting structural limitations for car use, thus using top-down “choice editing” strategies; and (c) targeting the increase in the use of alternative transportation means such as bicycle use.

Hiring Staff With a Biospheric Profile, Increasing It by 50% as Compared With Current Levels, in 2050

The first policy option we explored had to do with modifying the staff structure of the organization over time, by including a preference for biospheric values in the hiring policy of the university, ending in an increase of 100% in the clusters which contain this profile (pure and hybrid). It is clear how such a policy might be difficult to experiment with in a real setting, yet experimenting with it in a simulated environment can give us insight into its influence over behavior in the workplace. Also, it is conceivable, given educational efforts to instil pro-environmental values in young people, that future generations might lean more toward self-transcendent value profiles. Figure 1 shows a graph representing the agents per profile in the five profiles that were obtained for the organization. The subfigure on the center reflects the situation at present, and the resulting situation in 2050 if the present ratios are maintained. In the subfigure in the left, the ratio of biospheric personnel has been increased up to 50% as compared with the present situation, simulating new staff recruitment. Similarly, the figure on the right represents a situation in which the university staff is composed exclusively of staff with a biospheric value profile. The changes in the proportion of employees for each value profile are easily perceptible in the figure. For effects of interventions on carbon emissions, refer to Figures S4 and S5 in Online Appendix B.

OPEN IN VIEWER

Figure 1.

Number of agents per value profile in the present situation with no exogenous change (top center) and in the simulation of the increase of numbers of agents with a biospheric profile (50% of staff in the left and 100% on the right).

In Figure 2, the number of agents using each transport type for commuting (car, bus, bicycle, train, motorcycle, walk) is shown. Similarly, the subfigure on the center upper part represents the present situation in terms of number of agents using each transport type, whereas the left and right subfigures indicate how the number of agents using each transport type changes if the proportion of staff with a biospheric profile changes by 50% and 100%, respectively. In Online Appendix B, we have also included the effects of such a policy simulation on daily carbon emissions (please refer to Figures S4 and S5). Taken together, these results show a very interesting trend. In spite of the important changes in value profiles of the agents as well as those in the type of transport chosen to commute from home to work (notice that the number of agents commuting by car is reduced by more than a 100 individuals), changes in emissions are almost imperceptible. It thus seems that changes in the value profiles of employees by themselves would not achieve sustainability goals, which is consistent with previous research results in environmental psychology, indicating that values alone account for relatively low percentages of the variance of behavior (Bamberg & Moser, 2007). This is due to the fact that values are general antecedents of behavior, and this low influence is even more prominent in organizational settings, where the autonomy to act on one’s own values is relatively limited.

OPEN IN VIEWER

Figure 2.

The number of agents using each type of transportation for the simulation of increasing recruitment of the biospheric profile.

Different simulations were tested varying the percentage of staff with a biospheric value profile; 10%, 20%, 50%, and 100% were considered, although only the most significant results were included in the previous figures. Experiments targeting a proportion of 10% or 20% of staff with biospheric profiles show a similar evolution to the ones presented here. Also, although results are slightly better for the 50% and 100% hiring rate, the difference in carbon emissions with lower rates diminishes over time (see Figures S4 and S5 in Online Appendix B).

A few important conclusions can be drawn from the simulations above. First, we can see that hiring more staff with a specific value profile that is more likely to carry out pro-environmental practices in the workplace does not significantly modify the number of agents using the different transport means. This is most likely due to the fact that acting upon one’s values is significantly limited by other personal, social, and structural factors in the organization (Ruepert et al., 2013). Further simulations support this reasoning in a dynamically simulated evolution over time. Emissions would be slightly reduced, but further simulations holding technological improvements constant have shown that these reductions disappear almost entirely when considering only the hiring of biospheric staff in isolation, which means that they are rather due to expected technological improvements of the environmental performance of vehicles. Finally, we can see in the last figure that an increase in the hiring of staff strongly endorsing biospheric values when replacing retiring staff by 100% gives slightly better results, yet the achieved reduction of GHG emissions diminishes over time, and are overall low (refer to Figures S4 and S5 in Online Appendix B for daily emissions reductions in the different scenarios). Simulations thus help us conclude that a hiring policy based on selecting staff with a pro-environmental value profile (the biospheric) would not achieve significant results in isolation, in spite of this profile being more motivated to act pro-environmentally. However, decision algorithms differ according to the value profile of the individual, suggesting that segmenting sustainability interventions by relevant traits of population clusters is a worthwhile avenue for policy, as previous research has also suggested (Abrahamse, Steg, Vlek, & Rothengatter, 2007).

Testing Both One-Off and Sustained Over-Time Policies for Reductions of Car Use for Commuting

Backcasting scenarios have also included restrictive measures for the use of cars, such as controlled reductions in parking space, or fees for parking that make commuting between home and work by car more costly. The effects of policies for the restrictions of car use were tested. These experiments affect the conditions of the environment in which the agents live. Experiments were run both for one-off policy measures, which would restrict the use of the car, and for stepwise, incremental policies that would be maintained over time up to the target year. One-off policies are common in organizations and social systems that rely on a democratic governance system, which includes limited terms of office. However, incremental policies have fewer political and psychological costs (in terms of citizen support, and adaptation) and have the advantage of creating an environment in which a culture favoring sustainable practices is likely to be created, as people start assuming the new conditions as part of their everyday life.

Simulations for the effects of restrictive policies of car-use reduction have taken into consideration of several scenarios: policies implemented at one point in time; policies being maintained over time with progressive interventions targeting reductions of 10%, 30%, or 50% of car use (in number of users) as compared with baseline numbers; and resulting emissions estimations for the three targets, with comparison among them.

When confronted with one-off restrictive policies targeting reductions in car use for home-work commuting, specifically 50% reduction (see Figure 3), agents are forced to adapt by choosing alternative transportation means. Nevertheless, after this initial change in behavior, the influence of the social network results in a progressive increase in car use, which over time would tend to reach baseline levels again (note that there are free parking areas near the campus). In Figure 4, we can see the effects of this policy when the social network (i.e., interactions between agents) is eliminated. We can observe that when controlling for the social network, or, said differently, if workers were isolated and not exposed to the influence of others in the organization, their choices of alternative transportation means would be maintained constant. The influence of the social network is thus very important in the final outcome, rendering one-off car-reduction policies inefficient.

OPEN IN VIEWER

Figure 3.

Effects of one-off policy interventions either restricting or making car use less attractive on the number of agents using a certain transport mode.

OPEN IN VIEWER

Figure 4.

Influence of the social network.

When policies are maintained over time, and include progressive interventions at certain intervals, results are far more promising, as agents start taking up other modes of transportation and adapt to the new reality, with the social network working in favor of sustainable goals. Figure 5 shows the results obtained in the case of maintained reduction policies aiming at a final reduction of 10%, 30%, and 50% of car use, respectively, by 2050. The model considered policy interventions working toward the final reduction target at 4-year intervals. Two important conclusions can be drawn from these simulations: (a) carbon emissions are considerably lower than in the isolated application of policy and (b) comparisons between the different reduction targets show that applying a more aggressive reduction target of 50% does not obtain significantly better results than for the less aggressive one of 30%, which indicates that milder policies can be sufficient for the emissions reduction targets of European organizations. Figure 5 shows that carbon emissions in the long term are almost the same for both reduction targets of 30% and 50%, which both obtained better results than the lowest percentage tested (10%), which in turn is better than the isolated application of the policy even when the influence of the social network is eliminated (see Figure 4).

OPEN IN VIEWER

Figure 5.

Comparative effects on emissions of car-reduction policies isolated in time (50%) and maintained over time, with targets of 10%, 30%, and 50% reductions by 2050 (tons of CO₂).

Testing Combined Policies for Transitions to Sustainable Commuting

Policy makers often find it appealing to combine policies targeting behavior change. This is especially so for commuting behaviors where policies tend to be unpopular due to the entrenched habits of car use and status and identity functions car use plays (Griskevicius, Van den Bergh, & Tybur, 2010). Testing and determining the effects of combined policies was thus a relevant objective of the present study. We tested combinations of policies targeting both car-use reduction and promoting the use of the bicycle (see Figure 6). The targets set were of a 100% increase in the number of bicycle users among university staff (a doubling of present-day numbers) and of 10% reduction in the number of car users. In the figure, it is visible that both policies maintain their desired effect in the long term and that the total emissions reduction obtained is similar to the situation in which the car-use reduction target is set to 30%. Thus, similar emissions reduction effects can be obtained by targeting a milder reduction in car use and combining it with other mobility policies such as the increase of the use of bicycles among workers to commute between home and work. Such policies are also less politically costly and thus easier to be adopted and implemented. They are also more likely to receive support from workers in an organization, as the measures targeting car-use reductions would be complemented with behavioral alternatives. Furthermore, such combinations of policies are likely to be especially effective in behavioral areas that have been notoriously difficult to change, such as commuting.

OPEN IN VIEWER

Figure 6.

Testing combined policies for transitions to sustainable mobility.

Restrictive policies need to be combined with the provision of alternatives for behavior in order for them to be effective in bringing about behavior change over time. Applying restrictive policies only, either one time or incrementally, might generate negative reactions, and more so in democratically governed public organizations. Furthermore, restrictive policies that only change the environment in which behavior takes place are likely to have only short-term effects as they do not affect the psychological antecedents of behavior such as personal norms or our identities, which have proven significant as determinants of pro-environmental behavior change (Bamberg & Moser, 2007; Steg & Vlek, 2009; Turaga, Howarth, & Borsuk, 2010; Whitmarsh & O’Neill, 2010).

Behavior change strategies that target the use of alternative means of transportation have the advantage of encouraging alternative choices without imposing them, thus reducing negative reactions to these policies. Also, such policy options (e.g., awareness-raising campaigns, bicycle use-day, etc.) target changes in personal factors (such as attitudes) or encourage experimentation with a new behavior, thus supporting the processes of habit-breaking (Verplanken & Wood, 2006) and pro-environmental identity formation, as mentioned earlier.

The main results of the agent-based simulation experiments demonstrated are summarized as follows: