Varying the Agents Preferences

• Source Code for Basic Segregation model (729 KB)
• Shapefile (5 KB)

For the models presented thus far, all the agents wanted to live in an area where the agent itself is not in the minority. As with Schelling’s (1971) original model, agents only have preferences for their own group which were set to 50% (preferences for other groups will be explored in Preferences for Both). It is this preference that causes agents to seek different areas in the city. Clarke (1991) provides some empirical evidence to support Schelling’s abstract formulations, based on the fact that neither Black nor White households in several American cities would relocate into areas where they are the minority. However, in his work, preferences for specific composition of a neighbourhood varied among cities. .

This section will explore how the degree of segregation changes due to different preferences. The basic segregation model was used to explore how the preferences of individuals for their own group influence the degree of segregation seen within an area. For all the model runs, the same 1.5km2 area was used. The only model parameters that changed within the model were the agents’ preferences for the percentage of the same type to be located within its neighbourhood. Agents are satisfied within area if their preferences are achieved as reflected in the pseudo-code in Figure 1 (and see evaluateAndSetHappiness method in Resident Class). All the other parameters within the model were kept the same: 4000 agents were randomly placed, 2000 of each colour and neighbourhood size was set to 100m.

Figure 1: Pseudo-code for basic rule for an agent to be satisfied.

Figure 2 highlights the typical patterns of segregation that emerges from different preferences for neighbourhood composition. Animations of these simulation runs can be seen below. As the percentage of neighbours of the same type increases, the pattern of segregation becomes more noticeable. It is only when preferences become too high (90%) that agents are forced to leave the system as a result of their preferences are unable to be matched (Table 1). Not all the agents are removed as the system becomes less populated the number of agents in different neighbourhoods change. Where agents have been removed from the system, this removal only happens in the first iteration and for the first agents that move, as these agents are unable to find a suitable neighbourhood due to the initial random placement and mixed neighbourhoods at the start of the simulation. As these agents are removed, the area becomes less populated and the resulting agents can find neighbourhoods where their preferences can be satisfied. While it is possible to add these removed agents back into the system at the end of the simulation, it was felt more appropriate to leave them out. As it reflects the idea that as an area changes, residential groups are excluded from those areas.

Figure 2: Typical patterns of segregation with different preferences for neighbourhood composition.

Table 1 highlights the typical patterns of segregation that emerge from different preferences for neighbourhood composition. As the percentage of neighbours of the same type increases, the pattern of segregation becomes more noticeable. It is only when preferences become too high (>= 90%) that agents are forced to leave the system as a result of their preferences being unable to be matched (Figure 2.

Figure 2 highlights the typical patterns of segregation that emerge from different preferences for neighbourhood composition. As the percentage of neighbours of the same type increases, the pattern of segregation becomes more noticeable. It is only when preferences become too high (>= 90%) that agents are forced to leave the system as a result of their preferences being unable to be matched (Table 1). Not all the agents are removed for as the system becomes less populated, the number of agents in different neighbourhoods change. Where agents have been removed from the system, this removal only happens in the first iteration and for the first agents that move, as these agents are unable to find a suitable neighbourhood due to the initial random placement and mixed neighbourhoods at the start of the simulation. As these agents are removed, the area becomes less populated and the resulting agents can find neighbourhoods where their preferences can be satisfied. While it is possible to add these removed agents back into the system at the end of the simulation, it was felt to be simpler to leave them out, as this reflects the idea that as an area changes, residential groups are actually excluded from those areas.

Table 1 also highlights that by increasing the percentage of neighbours of the same type within the agents’ neighbourhood, more agents are forced to move at least once. For example when preferences are low (e.g. >= 10%), no movement occurs. However, as the preference for a minimum neighbourhood increases, more agents move (e.g. >= 40%) and the resulting pattern of segregation increases as highlighted in Figure 2.

Table 1: Comparison of neighbourhood preferences and model runs.

Although patterns can be deceiving and it is useful to have some measure of segregation, one possible measure is the average proportion of neighbours of like or opposite colour. By counting the total number of neighbours of different types for each of the agents remaining when all are satisfied with their neighbourhood, a greater understanding of the degree of segregation can be gained. At the same time, this allows for testing if neighbourhood and preference functions in the model are working correctly.

Table 2 presents the average neighbourhood composition at the end of each model run when all the agents have their preferences satisfied. This was calculated after the program had ended within a GIS package. As one would expect, the agents preference for a certain composition of a neighbourhood increases and the degree to which neighbourhoods are segregated also increases (e.g. from 40% onwards). The most noticeable variation is at 50% where the degree to which neighbourhoods are segregated rises the most. This is also visible in Figure 2. Variation of the values around the mean also decreases with higher preferences suggesting a more homogenised neighbourhood composition. A noticeable error seen within Table 2 is that at 100% preferences for a neighbourhood of the same type, a small percentage of agents have a number of agents of the different type within their neighbourhood. This is a result of the accuracy of converting from decimal degrees to metres in the GIS (e.g. 1 Decimal Degree is equivalent to 111319m, while 100m is 0.0008983153 Decimal Degrees. This conversion results in rounding errors within the GIS). Those agents whose neighbourhoods contain agents of a different type do so at their margins (Figure 3).

Table 2: Comparison of mean percentages of neighbourhood compositions for different preferences when all agents are satisfied.

Figure 3: Agents which have one or more agents of a different type in their neighbourhood when preferences are for 100% and all agents are satisfied.

[TOP]

Animations of simulation runs

Download zip file of images from model runs.

[TOP]

References

Clark, W.A.V. (1991), 'Residential Preferences and Neighbourhood Racial Segregation: A Test of the Schelling Segregation Model', Demography, 28(1): 1-19.

Schelling, T.C. (1971), 'Dynamic Models of Segregation', Journal of Mathematical Sociology 1: 143-186.

[TOP]