A multi-scaled agent-based model of residential segregation applied to a real metropolitan area
Introduction
As segregation has become an important feature of a modern city (Batty, 2010), our understanding of its causes, role and impact on the social and urban fabric of our societies remain relatively limited (Bruch & Mare, 2006). Yet, our ability to accurately model, measure, understand and anticipate segregation would be essential in having a more equitable distribution of public services and better social cohesion in the society.
The pioneering work of Thomas Schelling, 1969, Schelling, 1971 was an important milestone in investigating this multifaceted phenomenon (Clark, 1991). It built the foundation for an individually-based modelling (Crooks & Heppenstall, 2012) investigation focusing on the actions of agents (persons/ household embodiments) who made choices regarding where to relocate and live in the simulated world. Although many of these abstract models help us think about the “real world” (Fossett, 2011), the unrealistic and simplistic nature of artificial worlds in Schelling-style models has prompted “questions about how well they portray the neighbourhood dynamics of real cities” (O'Sullivan, 2009, p. 507). As a result, there are indications in the recent years that residential mobility and segregation modelling development point towards more realistic trends and their applicability to real urban areas.
This more realistic modelling approach is often comprised of four key dimensions: 1) more representative spatial characteristics; 2) use of real/empirical data; 3) more consistent and reliable evaluation (e.g. calibration and validation) against empirical benchmarks; 4) broader explanatory factors (determinants).
A model with more realistic characteristics has several advantages. Since “the outcomes of residential segregation models may strongly depend on the way that neighbourhoods are conceptualized and represented” (O'Sullivan, 2009, p. 508), it would be more consistent to use/ integrate real-world data which correspond to the same administrative spatial boundaries based on which data are collected (Rolfe, 2014).
Subsequently, the combination of “real data along their spatial characteristics is the ultimate form of model validation” (Stanilov, 2012, p. 258), as the evaluation of the model would naturally become more intuitive and reliable. Furthermore, inclusion of more explanatory factors in the model would enhance the overall realistic trait of the model, not the least because of the possibility of comparing the effects of the implemented mechanisms on the empirical and historical benchmarks. Overall, a more realistic model has higher potential to communicate its insights more effectively and engage easier with policy-makers (Stanilov, 2012).
Among agent-based models (ABM) in the residential segregation sphere, the pioneering work of Benenson, Omer, and Hatna (2002) has set a high standard for subsequent work (Bruch, 2014, Crooks, 2010, Feitosa et al., 2011, Yin, 2009).
However, a comprehensive assessment of changing patterns of residential segregation should preferably allow a thorough examination at different inter and intra levels of spatially nested entities (Parisi, Lichter, & Taquino, 2011). The geographical scale (e.g. divisions of a subdivided metropolitan area) can portray distinct dimensions of residential segregation (Reardon et al., 2009). A model with multi-scaled capability (illustrated in Fig. 1) will allow the investigation of segregation patterns on both macro-segregation (e.g. Metropolitan Area) and meso-segregation (e.g. Territorial Authority), based on their encompassed micro-spatial units' subdivisions (e.g. Area Units).
Similarly, the interpretation of shifting patterns of residential segregation and its social implications based on a single measure can be considered incomplete. Since there exist various paradigms and interpretations of segregation (Simpson, 2006), it is desirable to measure several dimensions of segregation (Massey and Denton, 1988, Reardon and O'Sullivan, 2004) in order to acquire more comprehensive portraits of the ethnic mosaic state in the meso and macro geographical entities of the urban area.
This research article presents an agent-based model of residential segregation which contributes to the same realistic modelling direction for analysing the effect of residential location decision of individual residents (agents) on the spatial ethnic mosaic pattern of the central Auckland region (New Zealand metropolis).
The following lists original features of the model. Firstly, the model deals with the entire population sizes based on census values, although only the relocating agents (informed by census mobility values for each ethnic group) are stochastically instantiated and make decisions about their residential location. Secondly, the model dynamic of residential location choice comprise of the main contextual mechanisms, including group and personal preferences (e.g. behaviours conditioned by bounded rationality), empirical vacancy rates (as proxy for combination of real estate market condition and (local) government policies related to new housing development), as well as economic conditions (by empirically informed proxy of residents' economic circumstances to relocate locally or globally). Thirdly, while intra-urban migration (movements by existing population within the boundaries of the urban/ metropolitan area and evidently its smaller spatial nested entities) takes place indigenously, inter-urban migration (movements between population of an external urban area and the simulated metropolitan area) has exogenous effect on the simulation dynamic, exhibiting an open urban system. Lastly, the effects of simulating residential decision-making of four major ethnic groups on various dimensions of segregation are measured and calibrated against their equivalent census-based benchmarks, before the simulations are projected into the future using Statistics New Zealand population growth projection estimates at meso-geographical scale as the base of segregation forecasting scenarios.
Thereby, the model is able to simulate future scenarios depending on changes in overall and ethnic-based population growth conditions and their distributions, including factors which are more susceptible to be influenced by macro (state, institutional) actors (such as control of international immigration, population birth rate, housing development/ vacancy rates), as well as micro (individual) actors (such as changing preferences of relocating residents).
In this article, we focus on presenting the following experiments and results. First, we show that the model is capable of generating patterns that are fairly comparable to the empirical benchmarks built from the application of multiple measures of residential segregation on several quinquennial periods of census data, notwithstanding detailed mechanisms regarding residential decision making are not fully present (implemented). Then, we use various experiments with the model to show that 1) higher population growth (and immigration) does not necessarily (automatically) exacerbate the intensity of residential segregation 2) segregation is not necessarily greater at the mesoscale than at the macroscale, 3) change of housing vacancy rate (in this case, tightening) in one of the meso-geographical units has an impact on the level of segregation in other meso-geographical units, as well as on the at the macro spatial unit (in this case, increase of segregation measured by the entropy-based information theroy index).
Section snippets
Study area and its characteristics
The Auckland region is the largest and most populous urban area in New Zealand and is located in the North Island (see Fig. 2). However, this study focuses on five central territorial authorities (TA) of Auckland City (AKL), Manukau (MKU), North Shore (NSH), Waitakere (WTK) and Papakura (PAK). In this article, we identify this macro geographical entity as a Metropolitan Area (MA), which encompasses a total of 316 selected area units (AU). The census data has been adjusted for multiple-ethnicity
Model implementation
Fig. 4 summarises the operation of the model along with scheduled processes. The following sections outline the details of how these are implemented and operationalized. Most of the scheduled processes are applied at micro-geographical level (often with consideration of empirical values available at mesoscale). However, the relocations are individually based and the moving agents will finally resettle in nearby or more distant micro-geographical locations with consideration of their contextual
Model calibration
The parameters involved in the calibration of a given model are normally those for which the exact empirical values are not available/ known or that (empirically informed) presupposed values are not considered fixed in specific scenarios and therefore open to further tuning (adjustment).
However, for the scenarios presented in this article, we (empirically or hypothetically) assigned specific values to the majority of the model parameters with the exception of tolerance (Θ) and ppersist
Model experimentation
Considering the lack of detailed information (data) on the process of residential location decision making and the subsequent limited realistic explanatory factors implemented in the model and the lack of feasibility in assigning empirical values with certainty to few parameters, in this article, we concentrate on presenting the outcomes of selected experimentations which are not vulnerable to potential uncertainties in the model parameters.
Moreover, like many social phenomena, the residential
Conclusion
This research article introduces a new multi-scaled spatiotemporal agent-based model of residential segregation. The new methodological contribution of the model can be described as a combined set of characteristics which are put into operation to investigate residential segregation dynamics with specific reference to empirical (census) data.
The model can deal with entire population sizes corresponding to census values. Similar to microsimulation models, the probability based mobility turnover
Acknowledgements
The authors would like to thank the anonymous reviewers for their valuable comments and suggestions that greatly contributed to improving this article. We would like to extend our gratitude to the Associate Editor for generous comments and support during this process. We would also like to express our appreciation to Statistics New Zealand for their helpful responses to our general and specific inquiries. Funding from the Royal Society of New Zealand's Marsden Fund (UOA0416) is gratefully
References (45)
Multi-agent simulations of residential dynamics in the City
Computers, Environment and Urban Systems
(1998)- et al.
Multi-agent simulator for urban segregation (MASUS): A tool to explore alternatives for promoting inclusive cities
Computers, Environment and Urban Systems
(2011) - et al.
Impact of urban planning on household's residential decisions: An agent-based simulation model for Vienna
Environmental Modelling & Software
(2013) - et al.
Race and space in the 1990s: Changes in the geographic scale of racial residential segregation, 1990-2000
Social Science Research
(2009) - et al.
Calibration of the SLEUTH urban growth model for Lisbon and Porto, Portugal
Computers, Environment and Urban Systems
(2002) - et al.
A spatial microsimulation model with student agents
Computers, Environment and Urban Systems
(2008) Local indicators of spatial association—LISA
Geographical Analysis
(1995)Network effects in Schelling's model of segregation: New evidence from agent-based simulation
Environment and Planning. B, Planning & Design
(2012)Towards a new science of cities
Building Research and Information
(2010)Agent-based modeling: From individual residential choice to urban residential dynamics
The third state of the Schelling model of residential dynamics
ArXiv
Entity-based modeling of urban residential dynamics: The case of Yaffo, Tel Aviv
Environment and Planning. B, Planning & Design
Hybrid geographical models of urban spatial structure and behaviour
Path dependence and the validation of agent-based spatial models of land use
International Journal of Geographical Information Science
How population structure shapes neighborhood segregation
American Journal of Sociology
Neighborhood choice and neighborhood change
American Journal of Sociology
Residential preferences and neighborhood racial segregation: A test of the Schelling segregation model
Demography
Introduction to agent-based modelling
Constructing and implementing an agent-based model of residential segregation through vector GIS
International Journal of Geographical Information Science
Ethnic preference, social distance dynamics, and residential segregation: Theoretical explorations using simulation analysis
Journal of Mathematical Sociology
Generative models of segregation: Investigating model-generated patterns of residential segregation by ethnicity and socioeconomic status
The Journal of Mathematical Sociology
Overlooked implications of ethnic preferences for residential segregation in agent-based models
Urban Studies
Cited by (67)
Ex ante inequality of opportunity in health, decomposition and distributional analysis of biomarkers
2020, Journal of Health EconomicsCitation Excerpt :The intuition behind the Pigou-Dalton transfer principle is not based on income as such but, more broadly, on individuals’ and society’s welfare, which is determined (at least partially) by income. In this utilitarian framework, it has been assumed that utility is an increasing and concave function of income, with diminishing marginal utility of income (Atkinson and Brandolini, 2015; Schwartz and Winship, 1980). For example, taking a dollar away from a richer person and giving it to a poorer person, we decrease the first person's welfare by less than we increase the poorer person's welfare and, thus, we achieve an increase in the total welfare (as the Pigou-Dalton principle assumes).
Economic growth, inequality, and well-being
2016, Ecological EconomicsCitation Excerpt :On the other hand, Neumayer (1999, 2000); Dietz and Neumayer (2006), and Jackson and Stymne (2000) see numerous problems with the standard ISEW/GPI method of accounting for inequality. First, Jackson and Stymne (2000) criticize the Gini adjustment because it does not satisfy the principle of diminishing transfers, which requires that the effect of a transfer lessens as the absolute level of income grows (Schwartz and Winship, 1980). Furthermore, Jackson and Stymne (2000) argue that the Gini coefficient is premised on hidden value judgments — implicitly valuing distributions closer to the center.
Measuring segregation in social networks
2014, Social NetworksCitation Excerpt :The axiomatic method has been fruitfully applied in the social sciences. Examples include such diverse domains as utility measurement (Suppes and Winet, 1955), measurement of inequality (Schwartz and Winship, 1980; Cowell and Kuga, 1981; Chakravarty, 1999), income mobility (Cowell, 1985), numerous problems in social choice theory such as the axiomatization of the simple majority rule (May, 1952) or various implications of the assumptions about measurability and comparability of individual utility functions (for example, d’Aspremont and Gevers, 1977, 1985). Regarding segregation, much of the progress in the social stratification research on segregation has been made through the employment of an axiomatic approach (or its associated elements) in the work of James and Tauber (1985), in the later work by Reardon and Firebaugh (2002a) and others (e.g., Egan et al., 1998; Massey and Denton, 1998; Grannis, 2002; Reardon and Firebaugh, 2002b), and recently in work by Alonso-Villar and del Río (2010).
Stability tests of urban physical form indicators: The case of European cities
2011, Procedia - Social and Behavioral SciencesGlobalization, labor market transformation, and metropolitan earnings inequality
2011, Social Science ResearchAN EMPIRICAL AGENT-BASED MODEL FOR RESIDENTIAL SEGREGATION, CASE STUDY: TEHRAN
2023, International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives