Literature Review
2.7 Applying Agent-Based Models to Migration Behaviour
2.7 Applying Agent-Based Models to Migration Behaviour
Described in a report by Buchanan (2009) as the social science analogue of the computational simulations now routinely used elsewhere in science to explore complex nonlinear processes such as the global climate, agent-based models are seen to have numerous applications within the social sciences. As a result of the adaptive behaviour with which agents can be endowed, Buchanan proposed that market behaviour could be predicted by allowing it to emerge from the properties and interactions of agents without any presupposition of the outcome. In support of this notion, Farmer and Foley (2009) suggested that agent-based models presented a potential way to model the financial economy as a complex system. However, they went on to refer to the difficulty of accurately specifying how agents behave, a process whereby rules are often created on the basis of, “common sense and guesswork, which is only sometimes sufficient to mimic real behaviour”. In conclusion, Farmer and Foley (2009) stated that, in order to make an ABM useful, the developer must proceed with model development in a systematic manner that avoids arbitrary assumptions by grounding and testing each part of the model against reality and only introducing additional complexity where it is needed.
In partial agreement with Farmer and Foley’s (2009) assertion regarding complexity of model design, papers by Acosta-Michlik and Espaldon (2008) and An (2011) highlighted the challenge of developing behavioural models that allowed a simple representation of the complexity involved in coupled human-environment, or human and natural systems, models. An described such complexities as being manifest through various components such as heterogeneity, non-linearity, feedback loops and emergence, with humans playing a crucial part in affecting such complex components and giving rise to changes in the environment which may, in turn, affect future human decisions and behaviour. A decade prior to the widespread acknowledgement of
complexity issues within the field (Acosta-Michlik and Espaldon, 2008, Farmer and Foley, 2009, An, 2011) Janssen and de Vries (1998) constructed one of the first multi-agent models of adaptive responses to climate change. The aim of the model was to assess the impact of different perspectives (of how the world functions and of how it should be managed) upon how different forms of adaptive behaviour could be included into models of global change under circumstances where agents were capable of learning from their observations. The authors tentatively concluded that their model enabled a more explicit treatment of the notions of societal change and scientific surprise events (such as changing evidence of the nature of the radiative forcing associated with chlorofluorocarbons) than other modelling approaches in the field at that time.
In their paper that acknowledged challenges relating to model complexity, Acosta-Michlik and Espaldon (2008) presented an ABM of the behaviour of farmers in three villages in the Philippines in order to assess the factors that contributed to their vulnerability to global environmental change. In a previous article, Acosta-Michlik and Rounsevell (2005) had proposed a framework that explicitly modelled human adaptive behaviour. Within this framework, vulnerability to global environmental change was described as a function of exposure, sensitivity and adaptive capacity, but also cognition (Ziervogel et al., 2005, Grothmann and Patt, 2005, Acosta-Michlik, 2005). On this basis, Acosta-Michlik and Espaldon (2008) promoted the notion of cognition as enabling adaptation through such agent-level processes as perception and evaluation of risks, identification and assessment of options, the ability to make decisions and take actions and to modify and update behaviour on the basis of the outcome of these actions. On top of social and economic capacity to respond to environmental changes, Acosta-Michlik and Espaldon therefore proposed that adaptive decisions and actions could be either constrained or promoted by cognitive processes. As a tool that can be used to analyse and simplify a complex problem, such as the constantly changing human and natural systems that play a part in shaping adaptive behaviour in response to global environmental change, the authors promoted the value of agent-based modelling.
As the basis for their ABM Acosta-Michlik and Espaldon (2008) developed an intervulnerability framework, the cognitive component of which was developed from the
“consumat approach” created by Jager et al. (2000) and further developed by Acosta-Michlik (2005) to study the interactions between a multitude of consumer agents and their environments through consideration of driving forces at both the collective (macro) and individual (micro) levels. Involving four cognitive strategies: deliberation; social comparison; repetition; and
imitation, the “consumat approach” was adapted by Acosta-Michlik and Espaldon (2008) to suit Philippine farming communities through four cognitive processes: observation; interaction;
social comparison; and income maximisation. Through application of the ABM developed in their paper, Acosta-Michlik and Espaldon found that lack of money and information were the most important reasons for failure to apply available technical adaptation measures, particularly among traditional and subsistence farmers. Furthermore, the authors found that social networks were an important source of adaptive capacity. In conclusion, the authors found that agent-based models were a useful policy tool for simulating the effects of different adaptation options because they allowed representation not only of the dynamic changes in climate and markets but also the dynamic adaptive process of different communities to the impacts of change.
Referring to such human-environment interactions in the context of coupled human and natural systems (CHANS), An (2011) used complexity theory and its application to review various decision models used in relevant agent-based simulations. Focussing on understanding complex adaptive systems, complexity theory was described by An as usually encompassing heterogeneous subsystems or autonomous entities which often feature nonlinear relationships and multiple interactions. In the paper, An (2011) suggested that agent-based models modelling human decisions or behaviour within the realm of CHANS and its associated complexity, ranged from highly empirically-based examples (derived through trend extrapolation, regression analysis, expert knowledge) to those that were more mechanistic or process-based (econometric or psychosocial). An concluded that without greater understanding of human decision-making it was very difficult to appreciate complexity at multiple dimensions or scales in order to achieve in-depth and dynamic coupling of the natural and human systems. Furthermore, in agreement with Farmer and Foley (2009), An (2011) suggested that, until there is greater process-based understanding of human decision-making in the context of natural systems, agent-based modellers should avoid unnecessary complexity through the inclusion of large numbers of trivial details.
As a result of the comprehensive review of CHANS agent-based modelling approaches, An (2011) proposed that a protocol similar to those promoted by Grimm et al. (2005, 2006) for ecology and Hare and Deadman (2004) for environmental management was required to formalise model development. Such a demand was described by An (2012) as resulting from significant growth in the use of ABM within different branches of the social sciences, including modelling migration. The reported rise in the application of agent-based models to the field of migration studies has come about as a result of assertions by authors such as Edwards (2008)
and Kniveton et al (2008). In a paper assessing the computational tools that can be used in predicting and assessing forced migration, Edwards (2008) suggested that computational models, and more specifically agent-based models, possessed considerable potential for predicting when and where groups of individuals might flee given some displacement generating event, particularly in the context of otherwise counter-intuitive flight.
A useful example of an ABM developed to model migration comes from Naivinit et al. (2010) who presented a participatory approach to constructing an ABM of rice production and labour migrations in one village of northeast Thailand. Consisting of three interacting entities (water, rice, and household), daily decisions were undertaken by the household element that could result in labour migration of individuals. Used to deepen the understanding of local farmers on the nature of interrelations between labour migration and rice production, the model focussed upon the possible economic gain to be made from migration rather than the cognitive process that leads individuals to migrate in response to stimuli. As an economically driven model, climate stress could be considered to have an implicit impact upon migration as manifest through, for example, a reduction in yield. However, by focussing upon economic stimuli alone and avoiding a full representation of migration decision-making, Naivinit et al.’s model has limited potential to extrapolate the impact of climate stress upon migration. Furthermore, the model is not suited to the realm of quantifying the influence of environmental change upon migration as a result of its development for just one Thai village.
Perhaps more relevant to the research presented by this thesis, Silveira et al. (2006) described an ABM constructed to analyse the phenomenon of rural-urban migration within the context of an economy in the early stages of industrialisation. Simulations using the model developed by the authors showed some emergent properties common to those often seen in developing countries: transitional dynamics characterised by continuous growth of the urban population, followed by the equalisation of expected wages between rural and urban sectors, as per the equilibrium condition proposed by Harris and Todaro (1970). However, like the model of Naivinit et al. (2010), Silveira et al.’s (2006) model did not adopt an approach whereby migration was considered from the context of the full psychosocial decision itself but used what the authors described to be a “statistical mechanics” approach. As such, agents were modelled to take their decision to migrate or not based upon a consideration of only the differential of expected wages between their present sector and the sector they intended to go to.
Although described by Piguet (2010) as having been used in migration studies as early as the 1970s through Schelling’s (1978) examination of the segregation process leading to intra-urban migration, applications of agent-based modelling within the field are seen to have been few, and rather disparate. Furthermore, most have considered agents to act largely on the basis of wage differentials with little or no consideration of cognitive aspects such as the bounded rationality of actors (Kant and Thiriot, 2006). Piguet (2010) went on to suggest that only very tentative studies had, at that time, used ABM in the field of environment-migration relations with no convincing results published thus far. Although citing two potential issues with the method:
limited availability of data on reactions of different groups to environmental stress and; the sudden and previously un-experienced nature of such stimuli, Piguet proposes that ABM retains clear potential within the field.
2.7 Summary
This chapter has outlined the key literature produced as a result of the academic evolution of interconnected disciplines that has led to this research. Starting with a description of the originally distinct fields of migration studies and climate change, the convergence of the two disciplines has been described and the nexus between changes in the environment resulting from climate change, and migration investigated. It is clear from analysis of the literature that has led to our current understanding of environmental migration that there are considerable degrees of complexity and uncertainty inherently involved in any attempt to predict the numbers of people likely to migrate in the future as a result of changes in climate. This complexity is seen to result from numerous components, not least of which is our current understanding of when relocation by an individual can be deemed to be environmentally induced, and whether or not such a move should be considered a proactive adaptation strategy or last resort.
Having highlighted the need for further research into the quantification of environmental migration the review turned to literature relating to climate change adaptation and migration in the case study location of Burkina Faso. With a historically mobile population that can be considered to be highly vulnerable to changes in rainfall, Burkina Faso is highlighted as a valuable opportunity for research into the influence of environmental change upon migration.
This assessment is supported by the wealth of literature relating to both migration and climate change within the country, in particular a paper produced by Henry et al. (2004a) which provides clear evidence of a statistically derived link between rainfall and migration in Burkina Faso.
After defining the area of research targeted by this thesis and reviewing literature relating to the case study location this chapter then highlights the wealth of literature that has worked to conceptualise decision-making and, more specifically, migration decision-making from both theoretical and modelling perspectives. By drawing upon previous contributions to the field of migration decision-making, the incorporation of computational advances in social simulation is documented and the advantageous properties of agent-based models as tools for simulating complex systems portrayed. On the basis of the potential advances that may be garnered from the application of an agent-based approach to research in the social sciences, the agent-based modelling methodology adopted by this research is justified. The review then closes by outlining recent literature relating to the development of agent-based models within the fields of climate adaptation and migration.
By bringing together contributions from numerous disciplines this research therefore aims to develop an ABM that appropriately represents the complex and multifaceted process of migration decision-making in the context of Burkina Faso. As a result, a rigorous means of investigating the role of changes in rainfall variability upon migration flows is sought. By simulating both past and future migration flows resulting from different future rainfall scenarios the ability of the model to replicate and predict migration can be assessed and the future role of changes in rainfall variability upon migration within and from Burkina Faso quantified. The following chapter draws upon further existing literature to develop a conceptual model of migration adaptation to rainfall change in Burkina Faso.