Ecological Economics 204 (2023) 107651 Contents lists available at ScienceDirect Ecological Economics journal homepage: www.elsevier.com/locate/ecolecon Methodological and Ideological Options A framework for agent-based models of human needs and ecological limits Joël Foramitti ∗ Institute of Environmental Science and Technology, Universitat Autònoma de Barcelona, Spain Institute for Environmental Studies, Vrije Universiteit Amsterdam, The Netherlands ARTICLE INFO ABSTRACT Keywords: The social and ecological challenges of our time require a better understanding of the complex interactions agent-based modeling between the multiple dimensions of human well-being and environmental impacts. This article introduces the social simulation Needs and Limits (N&L) framework, a theoretical and computational foundation for agent-based simulations quality of life of heterogeneous individuals who try to increase their quality of life through the satisfaction of human needs. human needs Based on psychological research, human needs are described as satiable, adaptive, and interdependent with the bio-physical constraints social and bio-physical environment. The N&L framework represents a generic foundation that can be applied to a broad range of socio-economic and ecological scenarios. A comparison is provided with classical utility approaches. The framework is illustrated for the topics of income inequality and climate policy. 1. Introduction are able to represent the world as a complex adaptive system (Arthur, 2021). They allow for the simulation of societal dynamics from the Confronted with an increasing number of social and environmental bottom-up, based on the interaction of autonomous agents with lim- emergencies, humanity faces the challenge of enabling a high quality of ited information and heterogeneous characteristics (Farmer and Foley, life for all people while staying within ecological limits (O’Neill et al., 2009). These features have lead to an increasing application of ABMs 2018). A system that would achieve such a balance between a social to economic and environmental policy analysis (Dawid and Delli Gatti, foundation on the one hand and an ecological ceiling on the other has 2018; Castro et al., 2020). However, challenges remain to enhance the been termed a ‘safe and just operating space for humanity’ (Raworth, realism of such models through the integration of psychological theory 2012; Steffen et al., 2015). Moving into this space requires a more — especially in regards to the representation of human needs (Jager, effective use of our energy, time, and resources towards the aim of 2017). increasing human well-being (Brand-Correa and Steinberger, 2017; Roberts et al., 2020; Rao and Wilson, 2021; Hickel et al., 2021). The aim of this paper is to introduce a theoretical and computational Research on this topic is difficult due to the many interdependent foundation for ABMs aimed at addressing multiple dimensions of hu- dimensions that have to be taken into account. First, human well- man well-being and environmental impacts: the Needs & Limits (N&L) being depends on the satisfaction of multiple human needs (Maslow, framework. It describes the activities and choices of human individuals 1943; Max-Neef, 1991; Jackson et al., 2004; Royo, 2007; Gough, 2015; who try to improve their quality of life through the satisfaction of Sirgy, 2021). Second, each of these needs can be satisfied in different their needs while being subject to multiple bio-physical constraints. ways, with some being more resource-intensive than others. And third, The presented approach is flexible enough so that different theories of there are numerous bio-physical limitations to consider, including both human needs can be applied and provides a consistent way to describe resource constraints and planetary boundaries like climate change and how these needs translate into human behavior within a particular biodiversity loss (Steffen et al., 2015). social, institutional, and bio-physical environment. Social simulations are a useful tool to increase understanding within The N&L framework is designed to address research questions that this complexity while taking into account that there is fundamental un- focus on the understanding of mechanisms. It can be used to explore certainty about future outcomes (Arthur, 2021). Computational models ways in which social and economic changes can achieve a reduc- can provide a structured way to think about real-world dynamics, show tion of ecological impacts that is consistent with high levels of well- us the logical implications of our assumptions, help us to understand being (Creutzig et al., 2021) — and how those changes could come patterns of past events, and explore possible outcomes of future sce- narios (Kucharski, 2021). Agent-based models (ABMs), in particular, about in a complex social setting. It further allows for policy analysis ∗ Correspondence to: Institute of Environmental Science and Technology, Universitat Autònoma de Barcelona, 08193 Bellaterra, Spain. E-mail address: joel.foramitti@uab.cat. https://doi.org/10.1016/j.ecolecon.2022.107651 Received 24 November 2021; Received in revised form 26 August 2022; Accepted 14 October 2022 Available online 4 November 2022 0921-8009/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by- nc-nd/4.0/). J. Foramitti Ecological Economics 204 (2023) 107651 Table 1 Different categorizations of separate life domains that can be applied within the N&L framework. Theory Source Categories Basic needs — Theory Maslow (1943) Physiological; safety; love; esteem; self actualization of human motivation Basic needs — Human Max-Neef (1991) Subsistence; protection; affection; understanding; participation, scale development idleness; creation; identity; freedom Domains of life Cummins (1996) Material well-being; health; productivity; intimacy; safety; satisfaction community; emotional well-being Basic human functional Nussbaum (2011) Life; bodily health; bodily integrity; senses, imagination, and capabilities (see also Sen, 1999) thought; emotion; practical reason; affiliation; other species; play; control over one’s environment Basic needs — Theory Gough (2015) Physical health; autonomy of agency (mental health, cognitive of human need understanding, opportunities to participate); critical autonomy Decent living standards Kikstra et al. (2021) Nutrition; Shelter; Health; Socialization; Mobility that takes a large number of dimensions into account, identifying poten- C1. Motive. The improvement of one’s own QOL is a universal tial synergies and drawbacks that are overlooked in simpler models. In motive that governs most of human behavior (Sirgy, 2021). In line with current research agendas, this supports a focus on the intersec- other words, most choices and activities of an individual can be tion between environmental outcomes, equity, and well-being (Roberts explained by the aim to enhance their own well-being. Adopting et al., 2020; Rao and Wilson, 2021). an eudaimonic understanding, this includes the fulfillment of The remainder of the paper is organized as follows. Section 2 lays values and goals and thus also relates to the well-being of out the psychological foundation of this study. Section 3 presents an others (Ryan et al., 2008). overview of needs-based simulation models in the literature. Section 4 C2. Behavioral biases. Human behavior is subject to numerous behavioral biases that can diverge from this motive. A central provides a detailed description of the framework. Section 5 illustrates reason for this is that people face fundamental uncertainty about the framework for the topics of income inequality and climate policy. both themselves and their environment. This means that it is Section 6 discusses limitations and future applications. not possible to know how to best enhance one’s QOL. Most of our activities thus follow heuristics: routines, habits, and simple 2. The quality of life rules of thumb. However, this should not be seen as naive or irrational as the use of heuristics can lead to better results The quality of life (QOL) is used to describe the overall assessment than more complicated ways of decision-making while using less cognitive resources (Gigerenzer and Brighton, 2009). Another of human experience (Costanza et al., 2007). There are numerous reason is that the improvement of one’s QOL governs most but philosophical and psychological approaches to this concept, and its not all of human behavior — which means that there are also relation to human needs remains contested (Jackson et al., 2004). A factors of behavior that are unrelated to this central motive. comprehensive overview can be found in Sirgy (2021). The two most Factors of behavior such as desires and wants are thus connected influential schools of thoughts are described in the following. to but not the same as the factors of our well-being. The Benthamite tradition is based on the concept of hedonic utility. C3. Human needs. The QOL depends on the satisfaction of multi- From this perspective, a high QOL describes a life that is pleasant — a ple human needs (Maslow, 1943; Max-Neef, 1991; Royo, 2007; state of contentment that contains a high amount of pleasure and a low Gough, 2015). These needs can be understood as ‘universal amount of pain. This school of thought has originated the principle of motivations [that] underlie human behavior’ (Jackson et al., ‘choosing the action that leads to the greatest happiness of the greatest 2004). A related view is that well-being requires fulfillment number’ (Sirgy, 2021, p. 7) and is closely aligned to the concept of among different life domains (Rojas, 2006). To combine these welfare that is prevalent in mainstream economic literature (Fellner two perspectives, we here adopt the simplified framing that the and Goehmann, 2017). fulfillment of a life domain depends on the satisfaction of human The Aristotlean tradition of ‘eudaimonia’, in contrast, focuses less on needs within that domain. In other words, life domains are seen outcomes and more on the process of living well (Ryan et al., 2008). as categories of human needs. From this perspective, a high QOL relates to a life that is engaging, The categorization of these domains can range from a small meaningful, and fulfilling. It describes a state where an individual has set of generic spheres of life to the almost infinite number of the capabilities to reach ‘their highest potential within the context of different aspects that affect the human experience (Rojas, 2006). their society’ (Brand-Correa and Steinberger, 2017, p. 44). Related con- The framework presented here is meant to be generic enough cepts are human flourishing, psychological well-being, and perfectionist that different theories can be applied. An overview of different categorizations in the literature that are compatible with this happiness (Sirgy, 2021). approach is given in Table 1. Similar to Seligman (2004) and Sirgy (2021), we here refer to the C4. Need satisfiers. There are different ways in which human needs QOL as subjective well-being in its broadest sense, relating not only can be satisfied. While needs themselves are universal, the set to that which makes life pleasant (hedonic utility), but also to what of possible need satisfiers and their effectiveness depends on makes it engaging and meaningful (eudaimonia). The aim of this work an individual’s personal characteristics as well as their cultural is to understand long-term patterns of well-being in regard to different and institutional environment (Jackson et al., 2004). This means social, economic, and ecological dynamics. The focus here is therefore that people are heterogeneous regarding their needs in the sense not on momentary sensations, but on people’s overall QOL over longer that what they require to satisfy them is different for each periods of time. individual. Most forms of satisfaction involve some sort of ac- In the following, we describe eight key characteristics of human tivity that requires resources and has an environmental impact. behavior and well-being (numbered C1–C8) that will serve as a psy- In other words, ’resources per se do not contribute directly to chological foundation for the N&L framework. wellbeing. It is how people use resources that may enhance 2 J. Foramitti Ecological Economics 204 (2023) 107651 wellbeing’ (Sirgy, 2021, p. 124). Note that these resources must us to live a good life. Finally, it is also often the case that the not necessarily be physical materials, but could also regard the social and ecological environment in which we live has changed use of time, energy, or mental capacity. Finally, the fulfillment of in a way that changes our capabilities for need satisfaction. personal values and goals can also contribute to the satisfaction The study of human behavior and well-being thus requires not of needs (Sirgy, 2021). only understanding about the satisfaction of human needs, but C5. Need satiability. Human needs are satiable from a temporary also about the psychological and social dynamics that determine perspective (Jackson et al., 2004; Royo, 2007; Galak et al., what is both needed and wanted in the first place. 2014). This means that within a specific period of time, there is an amount of need satisfaction that is sufficient to fulfill a 3. Needs-based models domain. Above this point, additional satisfaction will bring little additional fulfillment since that domain is already saturated. Human needs have so far received little treatment in simulations of For example, additional income will lead to little additional socio-economic dynamics. One reason for this is that the equilibrium well-being after a certain satiation point (Jebb et al., 2018). models of mainstream economic literature require rigid assumptions C6. Need deprivation. If needs are not sufficiently satisfied within of representative behavior, exogenous preferences, perfect rationality, a specific period of time, they will become deprived. Such de- and insatiable wants (Royo, 2007; Arthur, 2021). This makes it difficult prived needs have a stronger effect on QOL than satiated needs. to take a large number of dimensions into account. Nonetheless, there This means that the deprivation of a need increases the relative have been attempts to integrate a perspective of needs into classical importance of its life domain on the overall QOL. Basic needs economic approaches (Seeley, 1992; Woersdorfer, 2010; Lades, 2013; can then be defined as needs whose deprivation results in a Baucells and Zhao, 2021). low QOL independent of other factors. In other words, one A stronger emphasis on human needs and their satiability can be cannot have a high QOL if only a single basic need is not met. found in the literature of ecological and post-keynesian economics Once such basic needs are met, their relative importance will (Jackson et al., 2004; Costanza et al., 2007; Lavoie, 2014). However, be reduced and ‘higher-order needs (e.g., self-development and not much work has been done to translate the theoretical debates social relationships) gain prominence’ (Sirgy, 2021, p. 267). and empirical insights around human needs into formalized simulation This relates to the idea of a hierarchy of needs (Maslow, 1943). models. Agent-based modeling, which has long put an emphasis on However, the distinction between basic and higher-order needs a richer representation of psychological theory in social simulations, is not clear (Jackson et al., 2004). The deprivation of higher- represents a promising method to fill this gap (Jager, 2017). order needs can also lead to a low QOL, and higher-order needs The best-known example of a needs-based ABM can be found in the can still play an important role even when basic needs are popular computer game series ‘The Sims’, which represent the behavior deprived. Therefore, we assume no fixed order between different and well-being of humans by describing them as needs-based artificial needs as their relative priority will depend not just on their intelligences (Zubek, 2010). Inspired by Maslow (1943), each need is general importance but also on their current level of deprivation. modeled as a reservoir that can be depleted and refilled through various C7. Interdependence. Our needs are interdependent with our so- actions. Bogdanovych and Trescak (2016) describe a similar model for cial and ecological environment. Some needs can be satisfied virtual reality reconstructions of historical sites. through pro-social behavior (Helliwell, 2014), some through A research application of this approach can be found in the model of being connected to nature (Pritchard et al., 2020), and some human activity patterns from Brandon et al. (2020). The agents’ needs through social comparison (Sirgy, 2021). Identity-related char- in this model are rest, hunger, income, and travel. The satiation of acteristics like values and goals are formed through social in- each need drops over time. During the simulation, agents continuously teraction as well (Schachter, 2005). Note that this can also be choose the activity that best satisfies the most urgent need. The aim undesirable in the sense that social interaction can lead to the of this model is to understand short-term patterns, i.e. the length and formation of values and goals that are in conflict with other timing of different activities during a day. needs or harmful towards the well-being of others.1 Kangur et al. (2017), in contrast, regard need satisfaction from C8. Adaptation. Both factors of behavior and factors of well-being a long-term perspective. Their model simulates decisions of car use can change over time. Based on the view that needs are universal and purchase. Here, agents make use of different cognitive processes (C3), the factors that can adapt are the set of available need to seek information, which include optimization, repetition, imitation, satisfiers and their effectiveness as well as our behavioral biases. and inquiry. They further regard four different types of needs: fi- This can cause needs to be insatiable from a long-term perspec- nancial, functional, social, and environmental. To reach consumption tive (Royo, 2007). Increases in income, for example, do not only decisions, each of these needs are evaluated through custom functions fulfill basic needs but also lead to an increase in material de- that describe car-related satisfaction. sires (Jebb et al., 2018). This upwards adaptation is often called A more comprehensive setting is considered in the urban develop- the ‘hedonic treadmill’, describing how the satisfaction gained ment planning model of González-Méndez et al. (2021). Here, agents from an activity can drop once people get used to it (Redden, aim to satisfy the basic human needs described by Max-Neef (1991, 2015).2 see Table 1). Based on perceived information from their environment, It is an important difference whether this adaptation regards agents choose a decision rule that they think will best satisfy their desire (factors of behavior, C2) or actual needs (factors of well- needs, and then make their decisions according to this rule. In contrast being, C3). In the first case, our behavior changes in a way to the other models, this framework also includes a spatial dimension. that is less beneficial to our well-being. In the second case, we The work presented here extends this literature with a more general ourselves have changed in a way that makes it more difficult for framework that is not limited to a specific application with a pre- defined set of needs or decisions. Drawing from the psychological 1 insights of the previous section, it aims to provide a more generic In self-determination theory, such undesirable goals are often linked to foundation of how human needs translate into both human behavior extrinsic motivations like wealth or fame that rely on external indicators of worth (Fellner and Goehmann, 2017). and well-being. In addition, it connects the fields of social simulation 2 The hedonic treadmill is sometimes also referred to as satiation. In and ecological economics by combining this multi-dimensional perspec- contrast to the satiaton of needs discussed in C5, this describes the reduction tive on the quality of life with a detailed treatment of bio-physical of the effectiveness of a need satisfier. constraints. 3 J. Foramitti Ecological Economics 204 (2023) 107651 4. Model description The N&L framework describes a generic agent-based model, which can be fitted to different applied scenarios. Section 4.1 introduces the model’s possible dimensions. Section 4.2 defines the model’s variables. Section 4.3 presents the chain of events during each time-step of a simulation. Section 4.4 describes the computational implementation of this model. Fig. 1. Conceptualization of human activities and their impacts. 4.1. Dimensions The model is built upon the following dimensions: 4.2.2. Resources The resource inventory 𝜌𝑖,𝑟,𝑡 ∈ R≥0 describes the stock of a given • Human individuals 𝑖 ∈ 𝐼 = {1, … , 𝑛𝐼 } are the agents of the model. resource 𝑟 in possession of an individual 𝑖. It depends on a default flow Their goal is to adapt their behavior in order to satisfy their needs 𝜌̂𝑖,𝑟,𝑡 as well as the impact of each activity and choice. and thus improve their overall quality of life. ∑ ∑ • Activities 𝑎 ∈ 𝐴 = {1, … , 𝑛𝐴 } represent the various actions that 𝜌𝑖,𝑟,𝑡 = 𝜌̂𝑖,𝑟,𝑡 + 𝑅 𝛼𝑖,𝑎,𝑡 𝛿𝑎,𝑟 + 𝑅 𝜁𝑖,𝑐,𝑟,𝑡 (𝜙𝑖,𝑐,𝑡 ) (1) agents can take. Examples are the use or acquisition of resources, 𝑎∈𝐴 𝑐∈𝐶 trade, recreation, or social interaction. If a particular resource 𝑟′ can be kept from one time-step to the next • Choice sets 𝑐 ∈ 𝐶 = {1, … , 𝑛𝐶 } describe discrete decisions (i.e. it is non-perishable), the default flow can be defined as 𝜌̂𝑖,𝑟′ ,𝑡 = that agents can take from a set of available options 𝑜 ∈ 𝑂𝑐 = 𝜌𝑖,𝑟′ ,𝑡−1 . The first type of bio-physical limitations within this framework {1, … , 𝑛𝑂 𝑐 }. Examples are choices of employment, commitments, is represented by the fact that resource inventories must always be or investments. positive. • Life domains 𝑑 ∈ 𝐷 = {1, … , 𝑛𝐷 } describe distinct aspects of life that correspond to a specific category of human needs. Examples 4.2.3. Environmental impacts are given in Table 1 of Section 2, C3. The environmental impact 𝜀𝑒,𝑡 ∈ R describes the human influence • Resources 𝑟 ∈ 𝑅 = {1, … , 𝑛𝑅 } represent any input or output on a given environmental dimension 𝑒. It depends on a default flow 𝜀̂ 𝑒,𝑡 of an activity that can be available to a single agent. They do as well as the impact of all activities and choices. not necessarily have to be physical. Examples are money, time, [ ] energy, goods, or materials. ∑ ∑ ∑ 𝐸 𝐸 𝜀𝑒,𝑡 = 𝜀̂ 𝑒,𝑡 + 𝛼𝑖,𝑎,𝑡 𝛿𝑎,𝑒 + 𝜁𝑖,𝑐,𝑒,𝑡 (𝜙𝑖,𝑐,𝑡 ) (2) • Environmental dimensions 𝑒 ∈ 𝐸 = {1, … , 𝑛𝐸 } describe factors 𝑖∈𝐼 𝑎∈𝐴 𝑐∈𝐶 of the social and ecological environment that can be impacted The difference between resources and environmental factors is that by human activities. Examples are emissions, biodiversity, or the former are a stock that is accounted for on the individual level inequality. while the latter can be any kind of systemic variable. Different kinds • Time-steps 𝑡 ∈ 𝑇 = {1, … , 𝑛𝑇 } describes the discrete steps of of environmental factors can be defined as functions of 𝜀𝑒,𝑡 to capture the simulation. Based on the model’s application, it can represent interdependencies between the agents’ activities and the environment. different units of time. Note that within each time-step, time can Such feedback loops are the second type of bio-physical limitations that also be a resource. can be represented in this framework. 4.2. Variables 4.2.4. Satisfaction of needs The satisfaction of needs 𝑠𝑖,𝑑,𝑡 ∈ R≥0 within a given life domain The state of the model 𝑋 𝑡 denotes the vector of values for all 𝑑 refers to an intermediate state that describes the total impact of variables in the model at a given time-step 𝑡. An agent’s perceived state satisfiers. It depends on its default satisfaction 𝑠̂𝑖,𝑑,𝑡 and the impacts of ̃ The perceived state 𝑋̃ 𝑖,𝑡 thus of a variable is denoted by a tilde (𝑥). different activities and choices. describes what an agent believes the current variables of the model to ∑ ∑ 𝑆 𝑆 be. Variables that are not defined in the following subsections are free 𝑠𝑖,𝑑,𝑡 = 𝑠̂𝑖,𝑑,𝑡 + 𝛼𝑖,𝑎,𝑡 𝛿𝑖,𝑎,𝑑,𝑡 + 𝜁𝑖,𝑐,𝑑,𝑡 (𝜙𝑖,𝑐,𝑡 ) (3) 𝑎∈𝐴 𝑐∈𝐶 to be specified in custom ways for different applied scenarios. The fulfillment of a life domain 𝑞𝑖,𝑑,𝑡 ∈ [0, 1] describes the resulting 4.2.1. Activities and choices psychological state that arises from need satisfaction in a domain 𝑑. It is defined as a function of the current satisfaction 𝑠𝑖,𝑑,𝑡 and the satiation The activity intensity 𝛼𝑖,𝑎,𝑡 ∈ R≥0 describes the amount of which rate 𝑘𝑖,𝑑,𝑡 ∈ R>0 . each activity 𝑎 is performed by an individual 𝑖 at time 𝑡. Choices differ from activities in the sense that they can only be chosen from a discrete 𝑞𝑖,𝑑,𝑡 = 1 − 𝑒 − 𝑘𝑖,𝑑,𝑡 𝑠𝑖,𝑑,𝑡 (4) choice set 𝑐. The option 𝜙𝑖,𝑐,𝑡 ∈ 𝑂𝑐 denotes an agent’s currently active choice. This equation represents the satiability of needs (Section 2, C5). The As illustrated in Fig. 1, both activities and choices can have three resulting fulfillment curves for different satiation rates 𝑘𝑖,𝑑,𝑡 are shown different kinds of impacts (Section 2, C4). Activity impacts are denoted in Fig. 2. Note that this function transforms the unbounded space of by 𝛿 ∈ R, and choice impacts by 𝜁 ∈ R. The possible impacts are as satisfaction ([0, ∞]) into the bounded space of fulfillment ([0, 1]).3 follows: 𝑅 1. An impact 𝛿𝑖,𝑎,𝑟,𝑡 𝑅 (𝑜) on resource inventories. or 𝜁𝑖,𝑐,𝑟,𝑡 3 If applied to a scenario with potentially negative impacts, the satisfaction 𝐸 2. An impact 𝛿𝑖,𝑎,𝑒,𝑡 𝐸 or 𝜁𝑖,𝑐,𝑒,𝑡 (𝑜) on environmental factors. 𝑠𝑖,𝑑,𝑡 must be capped at a minimum value of zero. Since this can create a lack of gradient for the algorithm in Section 4.3.2, such scenarios may require 𝑞𝑖,𝑑,𝑡 𝑆 3. An impact 𝛿𝑖,𝑎,𝑑,𝑡 𝑆 or 𝜁𝑖,𝑐,𝑑,𝑡 (𝑜) on the satisfaction of needs. to be defined differently. One possible alternative would be to use a logistic function. 4 J. Foramitti Ecological Economics 204 (2023) 107651 4.3. Simulation procedure The order of events per time-step 𝑡 is given as follows: 1. Adaptation phase (perceived states) (a) Individuals update their (biased) perception. (b) Custom adaptive variables are updated. 2. Decision phase (desired states) (a) Individuals make discrete choices (if available). (b) Individuals decide on their desired activities. 3. Action phase (actual states) (a) Individuals perform their activities. (b) Activities cause impacts. Fig. 2. Domain fulfillment function for different satiation rates. 4.3.1. Adaptation phase (perceived states) In this first phase of the simulation, agents update their perception of the world. This is described by the perceived state of the world 𝑋̃ 𝑖,𝑡 , which describes what state the agent expects different variables of the model to be in. The way in which these perceptions are formed is not defined here so that it can be specified in custom ways for different scenarios.4 Usually, they are a function of past states of the world: 𝑋̃ 𝑖,𝑡 = 𝑓 (𝑋 𝑋 𝑡−1 , 𝑋 𝑡−2 , …) (7) Any custom adaptive variables of a particular scenario are also meant to be evaluated in this phase. This can be used to represent the Fig. 3. Difference between CES utility and the N&L framework. interdependence and adaptivity of psychological factors (Section 2, C7 & C8). 4.2.5. Quality of life 4.3.2. Decision phase (desired states) The quality of life 𝑄𝑖,𝑡 ∈ [0, 1] is described as a bounded variable, The agents’ factors of behavior are represented through the objec- representing a range from the worst possible life (𝑄𝑖,𝑡 = 0) to the tive function 𝑄̃ 𝑖,𝑡 . This adjusted form of the QOL is calculated like 𝑄𝑖,𝑡 best possible life (𝑄𝑖,𝑡 = 1). Following Rojas (2006), the relationship (see Eq. (5)), but with the agent’s perceived values (denoted by a tilde) between the separate life domains and QOL is described by a constant being used for each variable. This means that agents aim to increase elasticity of substitution (CES) function. their QOL (Section 2, C1), but can diverge from this objective due to behavioral biases (Section 2, C2). ∑( ) 1 𝑄𝑖,𝑡 = 𝜔̄ 𝑖,𝑑,𝑡 𝑞𝑖,𝑑,𝑡 𝜎𝑖,𝑡 𝜎𝑖,𝑡 (5) Agents are able to think about this objective in relation to any 𝑑∈𝐷 hypothetical context 𝑋 ′ . Their cognitive challenge represents a problem The factor 𝜎𝑖,𝑡 ∈ R>0 denotes the degree of substitution between of non-linear multivariate constrained optimization.5 The independent domains. Note that an unfulfilled domain will gain greater relative variable of this problem is the potential activity pattern 𝛼 ′ , which importance if 𝜎𝑖,𝑡 is low, representing the effects of need deprivation describes a vector over all potential activity intensities. Constraints are given by the facts that activity intensities and expected resource described in Section 2, C6. The factors 𝜔𝑖,𝑑,𝑡 ∈ R>0 denote the relative inventories 𝜌̃𝑟,𝑡 must be positive. importance of each life domain. To stay within the bounds of 𝑄𝑖,𝑡 , they are transformed into weights: maximize 𝑄̃ 𝑖,𝑡 (𝑋 𝑋 ′, 𝛼′) 𝜔𝑖,𝑑,𝑡 subject to 𝛼𝑎′ ≥ 0 𝑓 𝑜𝑟 𝑎 ∈ 𝐴 (8) 𝜔̄ 𝑖,𝑑,𝑡 = ∑ (6) 𝛼′) ≥ 0 𝜌̃𝑟,𝑡 (𝛼 𝑓 𝑜𝑟 𝑟 ∈ 𝑅 𝑑 ′ ∈𝐷 𝜔𝑖,𝑑 ′ ,𝑡 As illustrated in Fig. 3, this application of the CES function is In the computational model, this cognitive process is implemented different from its use for classical CES utility functions (Appendix A.4). through sequential least square programming Traditional models assume that utility is directly gained or lost from (SciPy Community, 2021).6 This algorithm works as follows. An agent starts from the last rounds’ activity pattern 𝛼𝑖,𝑎,𝑡−1 . They then take different kinds of consumption, resulting in potentially infinite values. discrete steps through the multi-dimensional space of activity combi- Here, the CES function is applied to the fulfillment of life domains. This nations that are feasible within the given constraints, trying to follow results in a bounded scale and allows for qualitative interpretation. 4.2.6. Networks 4 See Dosi et al. (2020) for examples of different heuristic expectation rules. 5 Finally, agents can be connected to each other through one or Optimization here refers to an individual’s process of trying to improve multiple networks. A network describes a graph in which the nodes are an expected outcome, which is different from the traditional economic notion of optimality that would imply the result to be a perfectly known optimum usually the agents themselves and the edges are connections between for both the agent and the system. agents (Hagberg et al., 2021). However, there could also be networks 6 This algorithm has been chosen for its robustness and the possibility to between resources or environmental dimensions. The definition of such represent generic behavioral biases which makes it suitable for many different networks is not specified further here as it can vary greatly depending applications. Performance can be enhanced by creating a custom algorithm for on the application. a particular scenario. 5 J. Foramitti Ecological Economics 204 (2023) 107651 the gradient of the objective function. The algorithm stops when a maximum number of steps or a satisfactory threshold is reached.7 Different settings can be applied to this algorithm to represent additional behavioral biases. First, a threshold value can be set where the algorithm will stop looking for improvements. This reflects that there is a level of well-being above which the agent will be satisfied and not look for further improvements. This is also called satisficing. Second, the maximum number of steps within each use this algorithm can be used to represent the limited cognitive resources of agents. Agents can apply this cognitive process to choose an option 𝑜 from a discrete choice set 𝑐. For each option, they calculate the desired activities 𝛼 ′ (𝑜) for each hypothetical expected context 𝑋 ′ (𝑜) where the Fig. 4. Scenario configuration. option 𝑜 is taken. Agents choose the option 𝜙𝑖,𝑐,𝑡 that leads to the highest value of 𝑄̃ 𝑖,𝑡 (𝑋 𝑋 ′ , 𝛼 ′ ). The related 𝑋 ′ (𝜙𝑖,𝑐,𝑡 ) then becomes the new perceived context 𝑋̃ 𝑖,𝑡 .8 5.1. Scenario description After these discrete choices, agents apply the same process to decide on their desired activity pattern 𝛼𝑖,𝑎,𝑡 ∗ for the current round. To do this, An overview of the demonstration scenario is given in Fig. 4. This they evaluate 𝛼 ′ for the current perceived context 𝑋̃ 𝑖,𝑡 . scenario is meant to explore general mechanisms in an abstract setting, and does not represent a particular place or time. Following Klein and 4.3.3. Action phase (actual states) van den Bergh (2021), we focus on the interplay between the two Agents want to follow their desired activity pattern 𝛼𝑖,𝑎,𝑡 ∗ . However, resource dimensions of money and time. Each agent receives one unit of since agents do not necessarily know the exact impacts of their activi- time per time-step (i.e. a default flow), while money has to be earned ties (Section 2, C2), this desired pattern can turn out to be infeasible. through paid work. The amount of work is a discrete choice and has They can therefore be forced to make changes. However, agents still two options: part-time and full-time. Agents can further perform three different kinds of activities: aim to stay as close as possible to their desired pattern. In the computational model, this process is implemented through 1. Recreation & rest uses time and satisfies immaterial needs. linear constrained optimization with the simplex algorithm (SciPy Com- 2. Green consumption uses money and time, and satisfies material munity, 2021). The independent variables are the actually enacted needs. activity intensities 𝛼𝑖,𝑎,𝑡 . The algorithm is configured to solve the follow- 3. Brown consumption is similar to green consumption, but requires ing problem to keep the difference between desired and actual activity no time and instead causes emissions. The amount of money intensities minimal while staying within feasibility constraints. required for this activity can be increased through the carbon ∑ ∗ −𝛼 |𝛼𝑖,𝑎,𝑡 𝑖,𝑎,𝑡 | tax level 𝜏. minimize 𝑎∈𝐴 ∗ 𝛼𝑖,𝑎,𝑡 subject to 𝛼𝑖,𝑎,𝑡 ≥ 0 𝑓 𝑜𝑟 𝑎 ∈ 𝐴 (9) The impacts of both activities and choices are described in Table 2 𝜌𝑟,𝑡 (𝛼𝑖,𝑎,𝑡 ) ≥ 0 𝑓 𝑜𝑟 𝑟 ∈ 𝑅 and Appendix A.1. This configuration is meant to roughly reflect two real-world aspects. First, that the use of time for leisure has a strong 4.4. Software positive effect on QOL (Sirgy, 2021). And second, that sustainable activities often require more time than their polluting substitutes, as The computational implementation of the N&L framework is written for example in regard to different modes of transportation. in Python 3. The software is build with the AgentPy package for agent- The scenario includes an additional life domain that describes the based modeling in Python (Foramitti, 2021), and makes use of multiple social need to follow perceived norms (Helliwell, 2014; Konc et al., algorithms from the SciPy optimization module (SciPy Community, 2021). This assumes that people gain satisfaction from activities that 2021) and NetworkX (Hagberg et al., 2021). The source code and appear to be popular. The more an activity is performed by their documentation of this software is publicly available under an open- friends, the more it contributes to satisfaction in the life domain of source license. The repository can be found under https://github.com/ social norms. The impact functions of this dynamic are also given in Appendix A.1. JoelForamitti/NeedsAndLimitsFramework. The scenario is populated with 100 individuals.9 As described in Appendix A.2, the agents are connected to each other through a small- 5. Application examples world network of friendships. Their income rates follow the world’s income distribution in the year 2016 (Helliwell et al., 2017).10 Note This section demonstrates the range of applications that are possible that this only describes the default setting of the scenario. In some of within the N&L framework. Section 5.1 introduces a simple demon- the following experiments, income rates can be subject to growth and stration scenario, which refers to a particular configuration of the redistribution. framework with a specific set of dimensions and variables. Sections 5.2– Finally, the agents’ needs are assumed to be heterogeneous. The 5.5 present various numerical experiments based on this scenario. An satiation rates 𝑘𝑖,𝑑 of each agent and life domain are drawn from interactive notebook able to replicate all of the presented results can a normal distribution. As described in Appendix A.3, the mean and be found in the software repository. standard deviation of these distributions are calibrated so that the QOL 7 9 In complex situations, this algorithm might get stuck in local or temporary While this is a small number of agents, it is able to capture the mech- maxima and thus lead to imperfect decisions. Note that this can also happen anisms of interest and display the heterogeneity between the agents without under heuristic decision-making (Section 2, C2). requiring too many computational resources. 8 10 The structure of this decision-making process is similar to the Actor-Critic This does not mean that these agents represent the global population. The method of reinforcement learning (Lowe et al., 2020). The calculation of a intention is only to display the same degree of inequality as in the real world desired activity pattern could be seen as the actor, and the estimation of and to demonstrate how data can be used to parameterize a model within this well-being under this pattern as the critic. framework. 6 J. Foramitti Ecological Economics 204 (2023) 107651 Table 2 Dimensions and impacts of the demonstrated scenario. Non-constant values are defined in Appendix A.1. Impact on Type Index Activities Choices Brown Green Recreation Paid consumption consumption & rest work 𝑎=1 𝑎=2 𝑎=3 𝑐 =1 Money Resource 𝑟=1 −1 − 𝜏 −1 0 𝑅 𝜁𝑖,𝑐,𝑜,𝑟,𝑡 Time Resource 𝑟=2 0 −1 −1 𝑅 𝜁𝑖,𝑐,𝑜,𝑟,𝑡 Material needs Life domain 𝑑 =1 1 1 0 0 Immaterial needs Life domain 𝑑 =2 0 0 1 0 𝑆 𝑆 𝑆 Social needs Life domain 𝑑 =3 𝛿𝑖,𝑑,𝑎,𝑡 𝛿𝑖,𝑑,𝑎,𝑡 𝛿𝑖,𝑑,𝑎,𝑡 0 Emissions Environmental 𝑒=1 1 0 0 0 Fig. 5. Distributions of income and well-being within a single run. outcome of the model follows a distribution that is similar to real-world needs and social interdependence in this case, there is a stronger cor- values of life-satisfaction. relation to income as under the QOL. In addition, utility can endlessly Note that some of the indices from Section 4 are omitted if a variable be increased through additional income. is constant over a specific dimension. For example, the index 𝑡 is not Note that the agents’ characteristics in this single simulation run are used for the satiation rates 𝑘𝑖,𝑑 here since they are assumed to be randomly drawn from a given probability distribution (Appendix A.3). constant over time in this particular scenario. All of the following experiments will be based on multiple runs, with results being given as averages and standard deviations over 20 such 5.2. Single run random configurations. In this first numerical experiment, we perform a single simulation 5.3. Growth and redistribution run under the default setting. The left half of Fig. 5 presents the main results. There is a general trend that higher income leads to higher QOL We now explore the effects of changes in the economic system with a diminishing rate of return. However, significant variations exist within this demonstration scenario. We consider two kinds of changes: due to the fact that people’s needs are heterogeneous. In line with the data that the model was calibrated to, the majority of people experience 1. A proportional growth of income for all agents. a QOL in the middle of the spectrum (Helliwell et al., 2017). 2. A redistribution of income between agents. These results match two empirical observations. First, that a certain amount of income is necessary for a high QOL, but not sufficient (Ke- Fig. 6 presents the effects of these two interventions on well-being. sebir and Diener, 2008). And second, that ‘once people have high In addition to QOL, comparative results are also presented for the incomes [...], additional increases in wealth have a very small influence measures of income and CES utility (Appendix A.4) — which are often on [QOL] suggesting that added income beyond modest affluence used as a proxy for well-being. The first row presents the average well- no longer helps answer important desires and needs’ (Diener and being among agents. The second row presents the Gini coefficient in Biswas-Diener, 2002, p. 145). regard to well-being, with a low coefficient describing high equality of For comparison, the right half of Fig. 5 shows how the same well-being between agents. distribution looks like under a classical CES utility function which Income growth leads to an increase of average QOL. At the same assumes that utility requires both money and time. The definition of time, the inequality of QOL is reduced as the well-being of low-income this function is given in Appendix A.4. Since there are no heterogeneous agents is increased to a larger extent than for high-income agents. The 7 J. Foramitti Ecological Economics 204 (2023) 107651 Fig. 6. Effects of income growth and redistribution on different measures of well-being. increase of average QOL declines as material needs become saturated. redistribution that is caused by the recycling of tax revenues is then The same is true for the increase of QOL equality. In other words, the defined as 𝜏 𝜓.11 effects of additional income display a decreasing rate of return to both The left half of Fig. 7 presents the average and the inequality of QOL average and QOL equality. both emissions and QOL. It shows that the carbon tax level reduces Income redistribution similarly increases average QOL as the posi- emissions. However, its effectiveness per unit of additional tax level tive difference for low-income agents is larger than the negative effect declines for higher tax values. This is because green consumption gets on those with high incomes. As expected, the inequality of QOL is also in increasing competition with recreation & rest as they both require reduced. However, a full equality of income (i.e. 100% redistribution) time. does not result in full equality of QOL. This is because people’s needs The inequality of emissions is increased by a tax, meaning that are heterogeneous. Some agents are thus able to reach a higher QOL the tax has a higher emission-reducing impact on low-income agents than others with the same amount of income. than on high-income agents. This is because the former are forced to For comparison, Fig. 6 also shows how the same analysis would change consumption habits due to income restrictions. Some of the look like under the assumption that income can serve as a proxy for latter, in contrast, only have an incentive to change due to relative price well-being. Note that the results are not completely linear as income differences but can in principle afford to receive sufficient satisfaction is affected by work-time choices. Income growth has a potentially through brown consumption. infinite positive effect, and little effect on inequality. Redistribution, If no income redistribution is caused through revenue recycling, in contrast, has no effect on average income and leads to almost full average QOL is negatively affected by the tax. However, the effect equality. starts to become positive around a tax progressiveness of 𝜓 > 0.1. This The CES utility measure produces results that lie somewhere in- is because redistribution provides a higher income for those in most between the two other perspectives. Similarly to QOL, this function (material) need for consumption. Notably, this progressiveness does not is able to describe decreasing marginal returns of additional income increase the average emissions but instead makes the carbon tax even to well-being. However, utility has no upper bound and can reach more effective. potentially infinite values similar to the income perspective. The results The right half of Fig. 7 shows how the different activities and on inequality are also similar to the income perspective. Income growth choices are affected by the tax. As expected, brown consumption is has little effect on inequality, and redistribution leads to an inequality gradually replaced by green consumption. Progressiveness increases the close to zero. amount of green consumption even further. The intensity of recreation & rest is reduced, since it competes with green consumption for the use 5.4. Carbon pricing of time. Next, we look at the environmental dimension of emissions. We consider the effect of different carbon tax levels 𝜏 on both emissions 11 Note that this is not a full economic model that involves firms and trade, and QOL. Following Klein et al. (2021), we also look at the tax’s so the carbon tax is only assumed in terms of its effect. For example, if 𝜏 = 1 progressiveness 𝜓 that defines how much income redistribution is and 𝜓 = 0.1, consumption will get 1 monetary unit more expensive for each caused through the recycling of tax revenue. The effective income unit of embedded emissions, and 10% of total wealth will be redistributed. 8 J. Foramitti Ecological Economics 204 (2023) 107651 Fig. 7. Effect of a carbon tax on QOL and emissions (left half) and activity intensities and choices (right half). Finally, let us regard the choice to work either part-time or full- time. The last panel of Fig. 7 shows the fraction of agents who choose to work part-time. This fraction is slightly decreased by low levels of the carbon tax. However, after a threshold around 𝜏 = 0.5, the tax leads to a reverse effect and increases the percentage of people who choose to work part-time. There are two reasons for this. First, the carbon tax increases the need for money as brown consumption gets more expensive. And second, it also increases the need for time as people increasingly switch to green consumption. The non-linear results can be explained by the combination of these two dynamics, with the latter becoming more dominant for higher tax levels. 5.5. Social multiplier Following Konc et al. (2021), we now look at how the effectiveness of a carbon tax is affected by social interaction. As described in Sec- tion 5.1, the assumption in this scenario is that people want to behave similar to their friends. When the carbon tax level changes someone’s behavior, it thus indirectly affects the behavior of their friends. This dy- namic is also called the social multiplier of environmental policy (Konc et al., 2021). To explore this effect, we vary the perceived importance of so- cial norms (𝜔̃ 𝑑 =3 ). The higher this value, the more social norms are perceived as important compared to both material and immaterial needs. In addition, we also vary the weight 𝜗̃ 𝑎=3 (Eq. (A.3)), which Fig. 8. Effect of a carbon tax on QOL and emissions under different behavioral biases. defines how important recreation & rest is perceived to be for the fulfillment of social needs relative to the two other activities that regard consumption. A stronger perceived importance of recreation & rest (𝜗̃ 𝑎=3 > 0) – an Note that this experiment looks at how much people think they additional dimension that is not accounted for in Konc et al. (2021) – need to follow norms, not how important it actually is for their well- can reverse this effect. This is because when people are able to fulfill a being. This reflects a behavioral bias (Section 2, C2). The variables 𝜔̃ 𝑑 =3 large part of their need to follow social norms through recreation & rest, and 𝜗̃ 𝑎=3 are varied, while the actual states (without a tilde) are kept they are less inclined to spend their extra time on green consumption constant. In other words, we look at differences in people’s behavior as the two activities compete for time. (i.e. their wants and desires) without differences in their needs. The top panel of Fig. 8 presents the effects of these biases on The bottom panel of Fig. 8 presents the effects of these biases on emissions. When only consumption matters (𝜗̃ 𝑎=3 = 0), the effects of QOL, making it possible to account for the difference between observed social interaction lead to a positive social multiplier effect on emission behavior (wants) and actual well-being (needs). The results show that reduction. This means that the tax leads to a lower emission level than the behavioral biases of this experiment have a similar effect on QOL it would without social interaction. This is in line with results from the as they have on emissions. This suggests that the additional emission model of Konc et al. (2021), where agents similarly have the choice reduction that can be achieved through the social multiplier effect between two activities of green and brown consumption. could be connected to a reduction in well-being. 9 J. Foramitti Ecological Economics 204 (2023) 107651 6. Conclusions Data availability This paper has presented the Needs and Limits (N&L) framework, A link to a public repository is provided in the paper. a theoretical and computational foundation for agent-based models aimed at addressing multiple dimensions of human well-being and Acknowledgments environmental impacts. It describes the adaptive behavior of human individuals who are trying to improve their quality of life. This makes it This study has received funding through an ERC Advanced Grant possible to represent the connection between individual decision factors from the European Research Council (ERC) under the European Union’s and emergent patterns of the whole system. The framework is further Horizon 2020 research and innovation programme (grant agreement no able to account for the satiability of human needs, the effects of need 741087). deprivation, the influence of social interaction and behavioral biases, I am grateful to Franziska Klein, whose ideas have greatly benefited the adaptive nature of human well-being, and the existence of multiple the application examples. I further thank Jeroen van den Bergh, Ivan bio-physical constraints. Savin, Théo Konc, Manuel Scholz-Wäckerle, Tania Treibich, Jarmo This approach can be used to explore how complex social and eco- Kikstra, and Yannick Oswald for their helpful comments and support. nomic dynamics can affect people’s capabilities to enhance their quality of life within specific institutional and bio-physical environment. To Declaration of competing interest illustrate this, application examples have been presented for the topics of income inequality and climate policy. The results suggest that the The authors declare that they have no known competing finan- inclusion of additional dimensions can change or even reverse existing cial interests or personal relationships that could have appeared to results from simpler models. A key insight from these experiments in line with existing literature is that the availability and use of time influence the work reported in this paper. plays a central role in trade-offs between well-being and environmental objectives (Jalas, 2002). Appendix. Scenario configuration The main strength of the N&L framework is that it can be applied to a broad range of socio-economic and ecological scenarios. It can This appendix provides additional information about the demonstra- incorporate any number of agents, networks, life domains, activities, tion scenario described Section 5. choice sets, resources, and environmental factors. To provide some examples, the scenario presented in Section 5 could be extended by a A.1. Impact functions differentiation between distinct types of green and brown consumption, the inclusion of unpaid work in addition to paid work, the represen- This section describes the custom impact functions from Table 2. tation of environmental values as an additional life domain, or the The impacts of the choice of working-time on the resource of time presence of further environmental factors to account for environmental reflect a job of 30 and 40 h per week respectively. problem shifting (van den Bergh et al., 2015). { The framework’s capacity of such high levels of complexity is also 𝑅 −1 40 … 𝑖𝑓 𝑜=1 (𝑓 𝑢𝑙𝑙−𝑡𝑖𝑚𝑒) 𝜁𝑖,𝑐 =1,𝑟=2,𝑡 (𝑜) = ∗ (A.1) its main limitation. The large amount of variables and the generic 168 30 … 𝑖𝑓 𝑜=2 (𝑝𝑎𝑟𝑡−𝑡𝑖𝑚𝑒) structure of the model result in long computation times. These aspects The gained money per time-step is calculated as an agent’s income make it difficult to calibrate and validate any applied scenario, which rate 𝛾𝑖 times their chosen amount of work-time. can partly be overcome through the use of detailed empirical data like in Kangur et al. (2017). However, for the stated aim of understanding 𝑅 𝑅 𝜁𝑖,𝑐 =1,𝑟=1,𝑡 (𝑜) = −𝛾𝑖 𝜁𝑖,𝑐 =1,𝑟=2,𝑡 (𝑜) (A.2) mechanisms, relevant insights can often be found in stylized models with a small number of agents. Performance can further be increased The impact of activities on the domain of social needs depends on by replacing the generic optimization algorithm with custom code that the activity-specific weights 𝜗𝑎 and the average activity intensity among is optimized for a specific application. an agent’s set of friends 𝐽𝑖 . Another limitation is the fact that any quantitative approach to 𝑆 𝜗𝑎 ∑ human well-being represents an extreme simplification of the human 𝛿𝑖,𝑑 =3,𝑎,𝑡 = 𝛼 (A.3) |𝐽𝑖 | 𝑗∈𝐽 𝑗,𝑎,𝑡 𝑖 experience. From the empirical side, every existing measure of well- being provides an incomplete picture (Joshanloo et al., 2019). From A.2. Parameter values the theoretical side, behavior that improves well-being from a certain philosophical perspective might appear undesirable in a model that is The demonstration scenario describes an abstract setting. Parameter based on another. For example, the focus on the long-term perspective values are thus not calibrated to a particular time or place, but are that is taken here will categorize choices that prioritize the short-term chosen to be within realistic ranges. as a behavioral bias. This difficulty regarding time-related trade-offs Each agent is connected to a set 𝐽𝑖 ⊆ {𝑗 ∣ 𝑗 ∈ 𝐼} of other agents has already been pointed out by Jager (2017) and remains a challenge that they consider friends. This network is randomly generated using to be addressed. the Watts–Strogatz algorithm for small-world graphs (Hagberg et al., Future applications of this framework will be able to explore ways to achieve a higher quality of life for all people with a lower environ- 2021), with 2 default neighbors per agent and a rewiring probability mental impact. Putting human needs and well-being at the center of of 0.1. analysis makes it possible to shift away from increasingly controversial The simulation length 𝑛𝑇 is set to 3 time-steps, allowing agents objectives like economic growth (O’Neill et al., 2018; Hickel et al., to observe the behavior of their friends and adapt their perception 2021). At the same time, it provides a more realistic description of multiple times. The substitution degree 𝜎 is set to a constant value human behavior that can be used to improve the demand-side de- of 0.01. This low value means that the deprivation of needs in one of scription of economic models — leading to a fuller understanding the three life domains cannot be compensated by fulfillment in another regarding the social and ecological impacts of different policies. In (Section 2, C6). addition, the framework is not methodologically limited to consider The income rates 𝛾𝑖 follow the world’s income distribution in the only incremental changes to the current system. It can also be used year 2016, as reported in World Inequality Lab (2021). These income to explore fundamentally different cultural and institutional settings rates are normalized so that the average income per agent equals one that include alternative forms of ownership, employment, currency, unit of money if all agents work full-time. All relative importance production, and trade. factors 𝜔 and 𝜗 are set to a value of 1. 10 J. Foramitti Ecological Economics 204 (2023) 107651 Table A.1 Dawid, H., Delli Gatti, D., 2018. Agent-based macroeconomics. In: Hommes, C., Calibrated parameters for the satiation rates 𝑘𝑖,𝑑 . LeBaron, B. (Eds.), Handbook of Computational Economics. vol. 4, Elsevier, pp. Domain Index Mean Standard deviation 63–156. Diener, E., Biswas-Diener, R., 2002. Will money increase subjective well-being? Soc. Material needs 𝑑 =1 8.35 1.78 Ind. Res. 57, 119–169. Immaterial needs 𝑑 =2 1.33 2.92 Dosi, G., Napoletano, M., Roventini, A., Stiglitz, J.E., Treibich, T., 2020. Rational heuristics? Expectations and behaviors in evolving economies with heterogeneous Social norms 𝑑 =3 5.49 8.84 interacting agents. Econ. Inq. 58 (3), 1487–1516. Farmer, J.D., Foley, D., 2009. The economy needs agent-based modelling. Nature 460 (7256), 685–686. Fellner, W., Goehmann, B., 2017. Human Needs and the Measurement of Welfare. SRE Perceived states simply reflect values of the previous round: - Discussion Papers 3811976, WU Vienna University of Economics and Business. Foramitti, J., 2021. AgentPy: A package for agent-based modeling in Python. J. Open 𝑋̃ 𝑖,𝑡 = 𝑋 𝑡−1 (A.4) Source Softw.. Galak, J., Redden, J.P., Yang, Y., Kyung, E.J., 2014. How perceptions of temporal Note that these parameters regard default values, which can be distance influence satiation. J. Exp. Soc. Psychol. 52, 118–123. subject to change in some of the presented experiments. Gigerenzer, G., Brighton, H., 2009. Homo heuristicus: Why biased minds make better inferences. Top. Cogn. Sci. 1 (1), 107–143. A.3. Calibration González-Méndez, M., Olaya, C., Fasolino, I., Grimaldi, M., Obregón, N., 2021. Agent- based modeling for urban development planning based on human needs. Conceptual basis and model formulation. Land Use Policy 101 (105110). The satiation rates 𝑘𝑖,𝑑 for each of the three life domains are Gough, I., 2015. Climate change and sustainable welfare: the centrality of human needs. heterogeneous amongst agents, and drawn randomly from a normal Camb. J. Econ. 39 (5), 1191–1214. distribution that is truncated to include only positive values. The Hagberg, A., Schult, D., Swart, P., 2021. NetworkX Reference (Release 2.6.2). https: mean and standard deviation of these distributions are calibrated to //networkx.org/documentation/. Accessed: 2021-10-11. be in a similar range as global self-reported life-satisfaction between Helliwell, J.F., 2014. Social norms, happiness, and the environment: closing the circle. Sustain,: Sci. Pract. Policy 10 (1), 78–84. 2014–2016, as reported in Helliwell et al. (2017). Helliwell, J., Layard, R., Sachs, J., 2017. World Happiness Report 2017. Sustainable The calibration procedure is set to minimize the difference between Development Solutions Network, New York. this desired QOL distribution and the outcome of the model, using Hickel, J., Brockway, P., Kallis, G., Keyßer, L., Lenzen, M., Slameršak, A., Stein- the Powell minimization algorithm (SciPy Community, 2021). The life- berger, J., Ürge-Vorsatz, D., 2021. Urgent need for post-growth climate mitigation satisfaction data consists of discrete values from 0 to 10, which are scenarios. Nat. Energy 6 (8), 766–768. Jackson, T., Jager, W., Stagl, S., 2004. Beyond insatiability – needs theory, consumption mapped to QOL as eleven equally sized ranges between 0 and 1. The and sustainability. In: The Ecological Economics of Consumption. Edward Elgar difference between the two distributions is measured with a chi-squared Publishing, Cheltenham, UK. statistic. The resulting values are shown in Table A.1. Jager, W., 2017. Enhancing the realism of simulation (EROS): On implementing and developing psychological theory in social simulation. J. Artif. Soc. Soc. Simul. 20 A.4. CES utility (3). Jalas, M., 2002. A time use perspective on the materials intensity of consumption. Ecol. Econom. 41 (1), 109–123. The CES utility measure 𝑈𝑖,𝑡 presented in Sections 5.2 and 5.3 Jebb, A.T., Tay, L., Diener, E., Oishi, S., 2018. Happiness, income satiation and turning assumes that individuals gain utility from the consumption of each points around the world. Nat. Hum. Behav. 2 (1), 33–38. resource (money and time) with equal weights. The substitution degree Joshanloo, M., Jovanović, V., Taylor, T., 2019. A multidimensional understanding of 𝜎𝑖,𝑡 is set to the same value as for QOL. prosperity and well-being at country level: Data-driven explorations. PLoS One 14 (10), e0223221. ( [ ]𝜎 ) 1 Kangur, A., Jager, W., Verbrugge, R., Bockarjova, M., 2017. An agent-based model for ∑ 1 ∑ 𝜎 diffusion of electric vehicles. J. Environ. Psychol. 52, 166–182. 𝑅 𝑈𝑖,𝑡 = 𝛼𝑖,𝑎,𝑡 𝛿𝑎,𝑟 (A.5) 𝑟∈𝑅 𝑛𝑅 𝑎∈𝐴 Kesebir, P., Diener, E., 2008. In pursuit of happiness: Empirical answers to philosophical questions. Perspect. Psychol. Sci. 3 (2), 117–125. Kikstra, J.S., Mastrucci, A., Min, J., Riahi, K., Rao, N.D., 2021. Decent living gaps and References energy needs around the world. Environ. Res. Lett. 16 (095006). Klein, F., van den Bergh, J., 2021. The employment double dividend of environmental Arthur, W.B., 2021. Foundations of complexity economics. Nat. Rev. Phys. 3, 136–145. tax reforms: exploring the role of agent behaviour and social interaction. J. Environ. Baucells, M., Zhao, L., 2021. Goods and Needs: A Satiation Approach. Darden Business Econ. Pol. 10 (2), 189–213. School Working Paper 2017/07, SSRN. Klein, F., van den Bergh, J., Foramitti, J., Konc, T., 2021. Agentizing a general van den Bergh, J., Folke, C., Polasky, S., Scheffer, M., Steffen, W., 2015. What if solar equilibrium model of environmental tax reform. Forthcoming. energy becomes really cheap? A thought experiment on environmental problem Konc, T., Savin, I., van den Bergh, J.C., 2021. The social multiplier of environmental shifting. Curr. Opin. Environ. Sustain. 14, 170–179. policy: Application to carbon taxation. J. Environ. Econ. Manage. 105 (102396). Bogdanovych, A., Trescak, T., 2016. Generating needs, goals and plans for virtual agents Kucharski, A., 2021. How modelling covid has changed the way we think about in social simulations. In: Traum, D., Swartout, W., P., K., S., K., S., S., A., L. (Eds.), epidemics. Guardian. Intelligent Virtual Agents. In: Lecture Notes in Computer Science, Springer, Cham. Brand-Correa, L.I., Steinberger, J.K., 2017. A framework for decoupling human need Lades, L.K., 2013. Explaining shapes of engel curves: the impact of differential satiation satisfaction from energy use. Ecol. Econom. 141, 43–52. dynamics on consumer behavior. J. Evol. Econ. 23, 1023–1045. Brandon, N., Dionisio, K.L., Isaacs, K., Tornero-Velez, R., Kapraun, D., Setzer, R.W., Lavoie, M., 2014. Post-Keynesian Economics: New Foundations. Edward Elgar Price, P.S., 2020. Simulating exposure-related behaviors using agent-based models Publishing. embedded with needs-based artificial intelligence. J. Expo. Sci. Environ. Epidem. Lowe, R., Wu, Y., Tamar, A., Harb, J., Abbeel, P., Mordatch, I., 2020. Multi-agent 30 (1), 184–193. actor-critic for mixed cooperative-competitive environments. Castro, J., Drews, S., Exadaktylos, F., Foramitti, J., Klein, F., Konc, T., Savin, I., van den Maslow, A.H., 1943. A theory of human motivation.. Psychol. Rev. 50 (4), 370. Bergh, J., 2020. A review of agent-based modeling of climate-energy policy. Wiley Max-Neef, M.A., 1991. Human Scale Development: Conception, Application and Further Interdiscip. Rev. Clim. Change e647 (May 2019), 1–26. Reflections. The Apex Press. Costanza, R., Fisher, B., Ali, S., Beer, C., Bond, L., Boumans, R., Danigelis, N.L., Dickin- Nussbaum, M.C., 2011. Creating Capabilities: The Human Development Approach. son, J., Elliott, C., Farley, J., Gayer, D.E., Glenn, L.M.D., Hudspeth, T., Mahoney, D., Harvard University Press. McCahill, L., McIntosh, B., Reed, B., Rizvi, S.A.T., Rizzo, D.M., Simpatico, T., O’Neill, D.W., Fanning, A.L., Lamb, W.F., Steinberger, J.K., 2018. A good life for all Snapp, R., 2007. Quality of life: An approach integrating opportunities, human within planetary boundaries. Nat. Sustain. 1 (2), 88–95. needs, and subjective well-being. Ecol. Econom. 61 (2–3), 267–276. Pritchard, A., Richardson, M., Sheffield, D., McEwan, K., 2020. The relationship Creutzig, F., Niamir, L., Bai, X., Cullen, J., Díaz-José, J., Figueroa, M., Grübler, A., between nature connectedness and eudaimonic well-being: A meta-analysis. J. Lamb, W., Leip, A., Masanet, E., et al., 2021. Demand-side solutions to climate Happiness Stud. 21 (3), 1145–1167. change mitigation consistent with high levels of wellbeing. Nature Clim. Change. Rao, N.D., Wilson, C., 2021. Advancing energy and well-being research. Nat. Sustain.. Cummins, R., 1996. The domains of life satisfaction: An attempt to order chaos. Soc. Raworth, K., 2012. A safe and just space for humanity: can we live within the doughnut. Indic. Res. 38, 303–332. Oxfam discussion paper, Oxfam International. 11 J. Foramitti Ecological Economics 204 (2023) 107651 Redden, J.P., 2015. Desire over time: The multi-faceted nature of satiation. In: Seligman, M.E., 2004. Authentic Happiness: Using the New Positive Psychology To Hofmann, W., Nordgren, L.F. (Eds.), The Psychology of Desire. Guilford Press. Realize Your Potential for Lasting Fulfillment. Simon and Schuster. Roberts, J.T., Steinberger, J.K., Dietz, T., Lamb, W.F., York, R., Jorgenson, A.K., Sen, A., 1999. Development As Freedom. Oxford University Press, Oxford and New Givens, J.E., Baer, P., Schor, J.B., 2020. Four agendas for research and policy York. on emissions mitigation and well-being. Glob. Sustain. 3 (e3), 1–7. Sirgy, J., 2021. the Psychology of Quality of Life - Wellbeing and Positive Mental Rojas, M., 2006. Life satisfaction and satisfaction in domains of life: Is it a simple Health, 3 Springer Nature Switzerland. relationship? J. Happiness Stud. 7 (4), 467–497. Steffen, W., Richardson, K., Rockström, J., Cornell, S.E., Fetzer, I., Bennett, E.M., Royo, M.G., 2007. Well-being and consumption: towards a theoretical approach based Biggs, R., Carpenter, S.R., De Vries, W., De Wit, C.A., et al., 2015. Planetary on human needs satisfaction. In: Bruni, L., Porta, P.L. (Eds.), Handbook on the boundaries: Guiding human development on a changing planet. Science 347 (6223). Economics of Happiness. Edward Elgar Publishing, Cheltenham, UK, pp. 151–169. Woersdorfer, J.S., 2010. Consumer needs and their satiation properties as drivers of the Ryan, R.M., Huta, V., Deci, E.L., 2008. Living well: A self-determination theory rebound effect: The case of energy-efficient washing machines. Papers on Economics perspective on eudaimonia. J. Happiness Stud. 9 (1), 139–170. and Evolution 1016, Max Planck Institute of Economics, Jena. Schachter, E.P., 2005. Context and identity formation: A theoretical analysis and a case World Inequality Lab, 2021. World inequality database. https://wid.world/. Accessed: study. J. Adolesc. Res. 20 (3), 375–395. 2021-06-03. SciPy Community, 2021. SciPy Reference Guide (Version 1.7.1). https://docs.scipy.org/ Zubek, R., 2010. Needs-based AI. In: Lake, A. (Ed.), Game Programming Gems 8. Accessed: 2021-10-11. Cengage Learning, Florence, KY. Seeley, E., 1992. Human needs and consumer economics: the implications of maslow’s theory of motivation for consumer expenditure patterns. The J. Socio-Econ. 21 (4), 303–324. 12