1. Urban Segregation as a Complex System An Agent-Based Simulation Approach
Flávia da Fonseca Feitosa
1ª Oficina de Intercâmbio INPE, CEDEPLAR/UFMG e FEA/USP
2 a 13 de Agosto/2010 1

2. An Urban Age
Since 2008, the majority of the world’s population lives in urban areas
Source: UN-Habitat, 2007

3. “Cities are not the problem; They are the solution!”
An Urban Age
Is this a problem?
“Cities are not the problem; They are the solution!”
(Jaime Lerner)
Potential as engines of development

4. An Urban Age Inclusive Cities Promote growth with equity
A place where everyone can benefit from the opportunities cities offer

5. A barrier to the formation of inclusive cities
Urban Segregation
A barrier to the formation of inclusive cities

6. Obstacles that contribute to perpetuate poverty
Impacts of Segregation
Obstacles that contribute to perpetuate poverty
Policies to minimize segregation demand:
A better understanding of the dynamics of segregation and its causal mechanisms

7. Segregation displays many of the hallmarks of complexity
Complex Nature of Segregation
Segregation displays many of the hallmarks of complexity

8. Complex Nature of Segregation
The Process Matters!
Require bottom-up simulations  Agent-Based Model
Agent-Based Models (ABM)
Focus on individual decision-making units (agents), which interact with each other and their environment
Natural way to explore the emergence of global structures

9. Multi-Agent Simulator for Urban Segregation
MASUS
Multi-Agent Simulator for Urban Segregation
Scientific tool to explore alternative scenarios of segregation
Support planning actions by offering insights about the impact of policy strategies
Purpose
Improve the understanding about segregation and its relation with different contextual mechanisms

10. MASUS Conceptual Model

11. City of São José dos Campos
São José dos Campos, Brazil
City of São José dos Campos
Study Area
São Paulo State

12. MASUS: Process Schedule

13. Decision-making sub-model
ALTERNATIVES
Not Move
Move within the same neighborhood
Move to the same type of neighborhood (n alternatives)
Move to a different type of neighborhood (m alternatives)
Higher probability to choose alternative with higher utility

14. Nesting Structure of the Model
Decision-making sub-model
Nesting Structure of the Model

15. MASUS: Process Schedule

16. Operational Model

17. Simulation Experiments
Comparing simulation outputs with empirical data
Testing theoretical issues
Testing anti-segregation policy strategies

18. Comparison with Empirical Data
Initial condition: São José dos Campos in 1991
Import GIS layers (households, environment)
Set parameters
Run 9 annual cycles
Compare simulated results with real data (year 2000)

19. Comparison with Empirical Data
Dissimilarity Index (local scale)
Initial State
(1991)
Simulated Data
( )
Real Data
(2000)
0.51
0.30
0.19
0.54
0.51
0.31
0.30
0.15
0.19

20. Comparison with Empirical Data
Isolation Poor Households (local scale)
Initial State
(1991)
Simulated Data
( )
Real Data
(2000)
0.51
0.51
0.54

21. Comparison with Empirical Data
Isolation Affluent Households (local scale)
Initial State
(1991)
Simulated Data
( )
Real Data
(2000)
0.19
0.19
0.15

22. Testing a theory How does inequality affect segregation? Experiment
Relation between both phenomena has caused controversy in scientific debates
Experiment
Compare 3 scenarios
Scenario 1: Previous run (baseline)
Scenario 2: Decreasing inequality
Scenario 3: Increasing inequality

23. Testing a theory Inequality (Gini) Proportion Poor HH
Proportion Affluent HH
Dissimilarity
Isolation Poor HH
Isolation Affluent HH
And here are the graphs: In blue is the baseline scenario (inequality increases a bit, but it is almost constant), the green represents the scenario where inequality decreases; and in the red, inequality increases.
These graphs show the proportion of poor households and affluent households, because they influence the results of the isolation indices.
The dissimilarity graph shows the the index really follow the inequality trends
The same happens with the isolation of poor households, but this graph is not so revealing, because, as I told before, the isolation index tends to follow to proportion of the groups… so, if there are more poor families, the isolation of poor families is higher
But what is really revealing is the graph with the isolation of affluent families, because it challenges the natural trend of the index.
We see that in the low-inequality scenario, we have a higher proportion of affluent families, but we have a LOWER isolation of this group. Which is quite impressive, something I have never seen, and reinforces the idea that YES, inequality does promote segregation. Cities with lower inequality are more likely to be less segregated.
Scenario 1 (Original)
Scenario 2 (Low-Ineq.)
Scenario 3 (High-Ineq.)

24. Poverty Dispersion vs. Wealth Dispersion
Testing policy strategies
Poverty Dispersion vs. Wealth Dispersion
Experiment
Compare 3 scenarios
Scenario 1
no voucher (baseline)
Scenario 2
200 – 1700 vouchers
Scenario 3
400 – 4200 vouchers
Poverty Dispersion: housing vouchers to poor families

25. Testing policy strategies
Dissimilarity
Isolation Poor HH % % % %
Scenario 1
No voucher (baseline)
Scenario 2
vouchers (2.3%)
Scenario 3
vouchers (5.8%)
Isolation Affluent HH % %

26. Poverty Dispersion vs. Wealth Dispersion
Testing policy strategies
Poverty Dispersion vs. Wealth Dispersion
Poverty Dispersion
Demands high and continous investment to decrease poverty isolation
Slows down the increase in segregation, but does not change the trends
So, evaluating the strategy of poverty dispersion, it is possible to say that :
(1) In a city like SJC, It demands massive and continous investment to decrease poverty isolation, to promote substantial change in the isolation of poverty.
(2) I continued the experiments, for the years , keeping the investment constant.
And it was possible to see that this continued investment slows down the increase in segregation, but does not change the segregation trends. Global indices continue to increase and the maps of segregation show the same patterns in comparison with the baseline scenario.
(3) Although this approach has been aplied, often sucessfully, in cities of developed countries, it does not seem realistic for cities in the developing world, where poor households represent a large share of the population.

27. Poverty Dispersion vs. Wealth Dispersion
Testing policy strategies
Experiment
Compare 2 scenarios
Scenario 1
(baseline)
Scenario 2
new areas for upper classes
Urban areas in 1991
Undeveloped areas for upper classes
Poverty Dispersion vs. Wealth Dispersion
Wealth Dispersion: Incentives for constructing residential developments for upper classes in poor regions of the city

28. Testing policy strategies
Dissimilarity
Isolation Poor HH
Isolation Affluent HH
Scenario 1
baseline
Scenario 2
new areas for upper classes

29. Poverty Dispersion vs. Wealth Dispersion
Testing policy strategies
Poverty Dispersion vs. Wealth Dispersion
Wealth Dispersion
Produces long-term outcomes
More effective at decreasing large-scale segregation
E.g Dissimilarity 2010
local scale (700m): %
large scale (2000m): %

30. Poverty Dispersion vs. Wealth Dispersion
Testing policy strategies
Poverty Dispersion vs. Wealth Dispersion
Wealth Dispersion
Positive changes in the spatial patterns of segregation
Baseline
2010
Wealth Dispersion
2010

31. Concluding Remarks MASUS: Multi-Agent Simulator for Urban Segregation
Virtual laboratory for testing theories and policy approaches on segregation
Does not focus on making exact predictions
Exploratory tool, framework for assembling relevant information
Oriented towards understanding and structuring debates in participative processes of decision support

32. Concluding Remarks For the next version of MASUS Decision-making
Dynamic representation of household‘s reasoning
Computational cost
Improve MASUS usability and effectiveness
Feedbacks from potential users
Additional experiments

33. Dimensions of Segregation
Exposure/Isolation: Chance of having members from different groups living side-by-side
Evenness/Clustering: Balance of the population groups distribution

34. TRADITIONAL PATTERN ‘CENTER-PERIPHERY’
Segregation Patterns
TRADITIONAL PATTERN ‘CENTER-PERIPHERY’
RECENT TRENDS:
NEW PATTERN OF SEGREGATION
WEALTHY AXIS EXPANDING INTO ONE DIRECTION
SLUMS (FAVELAS)
GATED NEIGHBORHOODS
WEALTHY CENTER
POOR PERIPHERY
Up to the 1980’s
“Center-Periphery pattern”
Macrosegregation
Nowadays
Not so simple
Sectorial: wealthy axis expanding into a single direction
At smaller scales
Slums (favelas)
Gated communities

35. MASUS
Methodological
Steps

36. Decision-making sub-model

37. NMNL: Affluent Households
Level
Choice
Variable
Coef.
Std. err.
1st
Move
Age of the household head
-0.040***
0.011
Renter
2.542***
0.425
Renter * household income
-9.4(10-5)
-7.5(10-5)
2nd
Move within the same neigh.
Constant
***
0.693
Move to the same type
of neighborhood
***
0.855
Type A neighborhood
0.477
0.661
Type B neighborhood
0.062
0.495
Kids * Type A
-0.368
0.636
Move to another
type of neighborhood
***
1.053
-0.256
0.732
1.760 ***
0.709
1.49 **
0.784
3rd
Generic
variables
Land price/ income
-0.084
0.053
Real estate offers
1.4(10-3) ***
5.1(10-4)
Distance from orig. neighborhood
-4.9(10-5) **
2.5(10-5)
Distance to CBD
2.3(10-5)
2.9(10-5)
Prop. of high-income families
0.960 **
0.503