This implementation of SimSeg Lite is geared to exploring "Schelling segregation effects" similar to those reviewed in two recent articles in the journal Urban Studies (Laurie and Jaggi 2003; Fossett and Waren 2005).
In all, there are three implementations of SimSeg Lite.
SimSeg Lite A. Version "A" models three important segregation dymanics in American urban areas: discrimination, economic inequality, and ethnic preferences.
It provides a simple interface for running eight, pre-designed simulation experiments.
SimSeg Lite B. Version "B" models three important segregation dymanics in American urban areas: discrimination, economic inequality, and ethnic preferences.
It provides a more complex interface which allows users to run a wide range of simulation experiments.
SimSeg Lite C. Version "C" models "Schelling effects". These are segregation patterns produced by the interaction of ethnic preferences and ethnic demography.
It provides a complex interface which allows users to run a wide range of simulation experiments.
Programming contributions and/or research assistance was provided by Srikrishna Gurugubelli, Ping Wang, David Dietrich, and Andrew Yinger.
Additional support has been provided by the Department of Sociology and the Racial and the Ethnic Studies Institute at Texas A&M University, College Station, Texas.
SimSeg Lite is an agent-based, computer simulation of residential segregation dynamics in urban areas. It is geared for educational use and is intended to be a tool for helping students gain a better understanding of certain aspects of social processes and conditions that contribute to the creation and maintenance of residential segregation in urban areas.
SimSeg Lite allows users to explore prevailing theories of residential segregation by designing and running simulation experiments that implement processes and conditions emphasized by different theories. The results of the simulation experiments provide insights into questions about how different social dynamics and urban-demographic conditions may (or may not) affect residential segregation (at least in the context of the SimSeg Lite model). An overview of how to run experiments can be found at the following link:
About SimSeg Lite
SimSeg Lite is a limited-capability adaptation of SimSeg -- a more complex, stand-alone program that runs under Windows.
SimSeg Lite is limited by design. It is a subset of the full SimSeg model. Its purpose is to give users a "taste" of the potential of using simulation methods to explore segregation dynamics. It assumes the user has a limited background in segregation measurement and segregation theory.
SimSeg Lite's has two chief advantages. It is readily accessible via the world wide web for anyone with a java-enabled browser. And it is easy to learn and use because it implements a relatively simple simulation model.
The chief disadvantage of using SimSeg Lite is that its capabilities are limited in comparison with the capabilities of stand-alone versions of SimSeg. This is dictated by two considerations. One is that browser-based programs (i.e., java applets) are inherently slower than stand-alone programs. Thus, SimSeg Lite implements a "lean" model with only the computational complexity needed to illustrate the power of simulation methods.
The other is that SimSeg Lite lacks many useful features found in the stand-alone versions of SimSeg. For example, the full program provides the user the capability to: observe a more detailed visual depiction of the evolution of the city landscape over the course of the simulaton experiment; create and run new simulation scenarios; save and "replay" simulation results; examine a much wider range of simulation results; create sophisticated graphical and text-based reports; create much larger and more complex simulated cities; and run multiple simulations (including repeated trials of the same simulation) in "batch" mode.
SimSeg Learning Edition
SimSeg Learning Edition is a commerical-grade version of the SimSeg program. Developed and distributed by Amber Waves Software of Lancaster, Pennsylvania, it is geared to undergraduate and graduate education. It implements a more complex simulation model and provides exceptional capabilities for generating graphical and text-based reports on simulation outcomes. Information about SimSeg Learning Edition can be found at the following web site:
[SimSeg Learning Edition Home Page]
SimSeg Research Edition
SimSeg Research Edition is a version of SimSeg developed for use in academic research. Developed by Mark Fossett of the Department of Sociology at Texas A&M University, it is geared to academic research investigating theories of segregation dynamics and methodological questions regarding the measurement of segregation. It has been used as the basis for numerous academic papers exploring computational models of residential segregation. SimSeg Research Edition is not geared for general distribution. Information about the program can found at the following web site:
[SimSeg Research Edition Home Page]
Take the following steps to run a simulation experiment.
1. Use the selection boxes to specify the settings for the simulation scenario.
2. Click on Reset to initialize the scenario.
3. Click on Begin to begin the simulation.
4. If needed, click on Continue as many times as needed to continue running the simulation until the segregation pattern becomes evident. (This will be a matter of judgement, but in many cases the pattern is evident within 30 cycles.)
Program Guide Document providing an overview of the SimSeg Lite program. (Note, the document is in Adobe Acrobat format.)
Results Form Form for systematically recording the results of simulation experiments. (Note, the document is in Adobe Acrobat format.)
Analysis Guide Document providing guidelinesfor performing simple analyses to explore hypotheses about the effects factors such as preferences, vision, and group size on ethnic segregation. (Note, the document is in Adobe Acrobat format.)
SimSeg is a stand-alone computer simulation program for Windows that models selected factors thought to influence patterns of residential segregation in US urban areas.
SimSeg is SimSeg Lite's "big brother". It is a much more capable and ambitious program that implements a more nuanced and sophisticated simulation model. It models the role of factors such as:
1. Demographic structures including the city's ethnic mix, the shape of the status distribution, and inter-group inequality in socioeconomic status;
2. Urban structure in the form of the spatial distribution of high-quality housing;
3. The role of household-level location decisions as guided by preferences for housing quality, neighborhood status, and ethnic mix and constrained by means; and
4. The role of various forms of racial discrimination in housing markets.
Further description of the SimSeg program can found at the following link:
SimSeg Home Page
Version 2.0 of the stand-alone version of the SimSeg program may be downloaded for personal and educational use. The distribution files can be obtained by going to the following link:
SimSeg Home Page
Version 2.0 is the last version of SimSeg Research Edition to be placed in the public domain. Fosset and Waren used Research Edition Version 4.0 to perform the analyses reported in their 2005 article in Urban Studies. SimSeg Learning Edition, a commercial grade version of SimSeg geared to undergraduate and graduate instruction, is currently under development with funding support from NIH. Dr. Richard Senft of Amber Waves Software of Lancaster, PA and Dr. Mark Fossett of Texas A&M University-College Station, TX are leading this effort.
The final version of the new implementation (final version) is scheduled to be completed by 2006. It will be licensed and distributed commercially for educational and professional use. In addition to implementing a more sophisticated simulation model, it also will be more "polished" and easier to use. It also will be supported by Amber Waves Software, a professional software development company with expertise in developing and supporting scientific simulation software.
20% Black. Under this setting, 20% of the households in city are designated as Black.
20% Hispanic. Under this setting, 20% of the households in the city are designated as "Hispanic". Thus, the city is 20% Black, 20% Hispanic, and 60% White. (Note: The percentage for Whites is determined based on the percentage of households that are not Black and not Hispanic.
Ethnic Preferences. Ethnic preferences are set to correspond to levels reported in surveys. Whites seek 90% in-group contact and do not specifically seek out-group contact. Blacks seek 50% in-group contact and 30% out-group contact. Hispanics seek 50% in-group contact and 30% out-group contact.
Agent Vision (N1). Vision is set to "area + surrounding areas". Under this setting households examine the ethnic mix of the immediate bounded area (on the city landscape's neighborhood grid), plus the ethnic mix of the sourrounding areas.
Overview. The default scenario provides no systematic forces for producing ethnic segregation. Thus, the only segregation that will emerge under this simulation scenario is that produced by random forces.
Use the selection boxes to change the simulation scenario and explore what kinds of segregation patterns are produced under different scenario settings.
These pre-designed scenarios can be interesting in their own right (they are chosen to combinations of settings of particular interes). Alternatively, they can serve as a convenient first step toward specifying a user-defined scenario.
Three-Group Segregation. The default scenario (A) is an extension of a simulation reported in Fossett and Waren (2005). The city has a 60/20/20 ethnic mix, preferences for in-group and out-group contact are comparable to findings from surveys, and vision is "high". The simulation produces high levels of segregation among three groups.
L&J Replication 1. Scenario B replicates a simulation reported in Laurie and Jaggi (2003). The city has a 50/50 ethnic mix and in-group preferences are set to 0. The simulation produces no segregation. This is consistent with Schelling (1971).
L&J Replication 2. Scenario C replicates a simulation reported in Laurie and Jaggi (2003). The city has a 50/50 ethnic mix, in-group preferences are set to 50, and vision is "low". The simulations produces clear small-scale segregation. This is consistent with Schelling (1971).
L&J Replication 3. Scenario D replicates a simulation reported in Laurie and Jaggi (2003). The city has a 50/50 ethnic mix, in-group preferences are set to 50, and vision is "high". The simulations produces clear medium scale segregation. This is consistent with Schelling (1971).
L&J Replication 4. Scenario E replicates a simulation reported in Laurie and Jaggi (2003). The city has a 50/50 ethnic mix, in-group preferences are set to 30 (20 points below population representation), and vision is "low". The simulations produces clear small scale segregation. This is consistent with Schelling (1971).
L&J Replication 5. Scenario F replicates a simulation reported in Laurie and Jaggi (2003). The city has a 50/50 ethnic mix, in-group preferences are set to 30 (20 points below population representation), and vision is "high". The simulations produces little or no segregation. Laurie and Jaggi see this as inconsistent with Schelling (1971). They conclude segregation is low under weak preferences for in-group contact if vision is high.
F&W Replication 1. Scenario G replicates a simulation reported in Fossett and Waren (2005). The simulation is the same as scenario F in having low in-group preferences (IGP=30) and high vision. The key difference is it has an imbalanced ethnic mix (80/20) as is common in real cities. The simulation produces more segregation than Scenario F. Fossett and Waren see this as consistent with Schelling (1971).
The result highlights the "paradox of weak minority preferences".
Weak preferences can promote segregation when groups are small.
Generally, the result illustrates that the effect of in-group
preferences depends on the ethnic demography of the city.
F&W Replication 2. Scenario H replicates a simulation reported in Fossett and Waren (2005). The simulation is the same as scenario G but with in-group preferences for Whites set to 0. The simulations produces segregation similar to Scenario G. This highlights the importance "paradox of weak minority" preferences.
F&W Replication 3. Scenario I replicates a simulation reported in Fossett and Waren (2005). The simulation has in-group preferences for each group set at 10 points below their population representation. These preferences are logically compatible with integration, but the simulation produces segregation. The level of segregation increases the longer the simulation runs.
The result highlights that "tolerance" of integration permits
integration, but does not specifically promote it. Some other
mechanism is needed to promote integration.
F&W Replication 4. Scenario J replicates a simulation reported in Fossett and Waren (2005). The simulation has in-group preferences for each group set at 10 points below their population representation as in Scenario I. In this case, however, groups are both given preferences for out-group contact. Whites seek 20% out-group contact and Blacks seek 50% out-group contact. The simulation produces very low segregation.
The result highlights that preferences for diversity can directly
F&W Extension 1. Scenario K extends Scenario I to three groups. The ethnic mix of the city is set to 60% White, 20% Black, and 20% Hispanic. The simulation has in-group preferences for each group set at 10 points below their population representation as in Scenario I. Also as in Scenario I, households do not seek out-group contact. The simulation produces high levels of segregation.
The result highlights that "tolerance" of integration permits
integration, but does not specifically promote it. Some other
mechanism is needed to promote integration.
F&W Extension 2. Scenario L extends Scenario J to three groups. The ethnic mix of the city is set to 60% White, 20% Black, and 20% Hispanic. The simulation has in-group preferences for each group set at 10 points below their population representation. In addition, all groups seek out-group contact. The simulation produces low levels of segregation.
The result highlights that preferences for diversity can directly
F&W Extension 3. Scenario M modifies Scenario K by giving minority groups preferences for in-group contact that are mild (only 30%), but above their population representation. The Scenario produces very high levels of segregation.
F&W Extension 4. Scenario N modifies Scenario M by giving minority groups preferences for in-group contact that are mild (only 30%), but above population representation. However, it also gives all groups preferences for out-group contact.
The scenario produces much lower levels of segregation than Scenario M.
Note that most integrated neighborhoods are "binary" (i.e, they have
two groups, not three).
F&W Extension 5. Scenario O implements preferences similar to those documented in surveys. The model has three groups. All three groups seek at least 50% in-group contact with Whites seeking 90%. Minorities also seek substantial out-group contact (50%)
The scenario produces very high levels of segregation.
User-Defined Scenarios. Any scenario that does not exactly match one of the eight pre-designed scenarios listed above.
The literature on measuring residential segregation
recognizes five separate dimensions of segregation:
uneven distribution, isolation, clustering, centralization,
Uneven Distribution. This reflects the extent to which two groups are distributed unevenly across areas of the city. The key consideration is each group's "exposure" to the pair-wise percentage for one of the groups in the comparison. The pair-wise percentage is the group's percentage out of the combined count of both groups. (It doesn't matter which group is considered, the outcome is the same either way.)
If both groups live in neighborhoods with pair-wise percentages that match the pair-wise percentage for the city, the groups are evenly distributed.
On the other hand, if the groups live in neighborhoods with pair-wise percentages that differ from the pair-wise percentage for the city, the groups are unevenly distributed.
Simseg assesses uneven distribution using the index of relative pair-wise isolation (R). This index registers systematic departure from random distribution. It takes a value of 0 when the pair-wise percentages for neighborhoods are the same, on average, for both groups. It takes a value above zero when the averages for neighborhood pair-wise percentages are different for the two groups. The maximum of 100 is reached when members of two groups never reside in the same neighborhood.
The index of relative pair-wise isolation (R) has a simple interpretation. It is the difference between the two group's average exposure to the pair-wise percentage for one of the groups. For example, segregation between blacks and whites can be measured by the difference between their average exposure to pair-wise percentage black for neighborhoods. If the two groups are evenly distributed, the averages will be the same for both groups and the difference will be zero. If the two groups are unevenly distributed, the average exposure to pair-wise percent black for neighborhoods will be higher for blacks and lower for whites and the difference will be a non-zero value between 0 and 100. For example, if the average for blacks is 60 and the average for whites is 10, R will be 50.
Scores on R can be interpreted as follows.
0 minimum uneven distribution
20 low uneven distribution
40 medium uneven distribution
60 high uneven distribution
80 very high uneven distribution
100 maximum uneven distribution
The index of relative pair-wise isolation (R) is similar to
another well-known measure, the index of dissimilarity (D).
However, R is more attractive than D for present purposes.
This issue is discussed in detail in technical papers
on measuring segregation in simulation studies available
at the SimSeg web site. The key advantage of R is that
it is "well-behaved" statistically and easy to interpret
when segregation is measured using small neighborhoods
such as the ones used in simulation studies.
Isolation. This reflects the degree to which members of a group are residentially isolated from other groups. This is measured by the P* (i.e., "P-star") index of same group contact. This index registers the average residential contact members of a group have with other members of their group. Its maximum is 100 (which occurs when all contact is within group). Its minimum is the group's percentage in the population (which occurs when the group is evenly distributed).
Isolation is affected by two factors - uneven distribution
and relative group size. If a group is unevenly distributed,
it will tend to be isolated from other groups. However, if
a group is a large percentage of the population, it may be
somewhat isolated even when the group is evenly distributed.
For example, in the SimSeg Lite simulation, whites are 60%
of the population, so it is difficult for whites to get
a low isolation score. Under even distribution whites
would have at least 60% contact with whites and only 40%
contact with others.
Clustering. This reflects the degree to which the areas where a group is predominant occur near each other in urban space and form "clusters" or "ghettos" - regions of contiguous neighborhoods where the group is predominant. The opposite of clustering is "checkerboarding" - a pattern wherein ethnic neighborhoods are randomly distributed throughout the city.
SimSeg Lite measures clustering in terms of the group's percentage of households that reside in "ethnic clusters". Households are classified as residing in an ethnic cluster based on two criteria. One is that the co-ethnic presence (i.e., same group representation) in the area must be at least 60%.
The second criteria concerns the ethnic mix in surrounding areas. Here one of two conditions must be met. At least three contiguous areas must each separately meet the 60% co-ethnic presence standard. Or, there must be at least 50% co-ethnic representation in the combined population of all contiguous areas.
These measures may range from 0 (no clustering) to 100 (maximum
possible clustering. Like isolation scores, clustering scores
are shaped both by patterns of uneven distribution and by
relative group size. Thus, since whites are 60% of the population,
it is difficult for the group to get a low clustering score.
Centralization. This reflects the degree to which members of a group are concentrated in neighborhoods near the center of the city.
SimSeg measures centralization on a scale ranging from -100 to 100 with the following interpretation
100 maximum centralization
50 very high centralization
20 high centralization
0 neutral distribution
-20 low centralization
-50 very low centralization
-100 minimum centralization
Centralization scores are calculated in the following way.
1. Each housing unit is assigned a percentile score for distance from the city center (e.g., 60 = farther from the city center than 60% of all housing units).
2. The mean (Y) for these percentile distance scores is calculated separately for each ethnic group.
3. These mean scores (Y) are converted to centralization scores (CS) using the following formula:
CS = (2*100)*(50-Y)/(100-P))
where P is the group's percentage in the city population.
The formula can be explained as follows. The term (50-Y) captures deviation from neutral distribution (since 50 is the expected percentile distance score under random distribution).
The term (100-P) is included in the formula to adjust for the fact that the minimum and maximum scores for Y are P/2 and 100-(P/2), repectively. Thus, the maximum range of Y for any group is 100-P.
The term (2*100) is a scaling adjustment to convert the result into
a number between -100 and 100. (Without the adjustment, the range
would be -0.5 to 0.5.)
Concentration. This reflects the degree to which a members of a group are concentrated in a small geographic area due to disproportionate representation in high-density neighborhoods.
Since all neighborhoods in the SimSeg Lite
simulation have the same density (i.e., all neighborhoods are
identical in size and have an identical number of housing
units), this dimension of segregation is not measured here.
Segregation can be computed for different spatial domains. One option is to use "bounded areas" (BA). This is relevant when location dynamics and patterns of social interaction are correlated with boundaries in space (e.g., boundaries for school districts, subdivisions, municipalities, etc.).
Segregation can also be computed for site centered (SC) regions which contain the nearest neighbors surrounding a household. This is relevant when location dynamics and patterns of social interaction revolve around nearby neighbors.
SimSeg Lite reports both kinds of segregation scores.
Bounded Areas (BA). Bounded area (BA) segregation scores are computed using data for the areas delimited by the city's neighborhoood grid. This reflects the assumption that group contact should be evaluated within bounded areas.
When agent vision follows grid areas, these segregation
scores will tend to be higher than site centered segregation
Site-Centered Regions. Site centered (SC) segregation scores are computed using data for neighborhoods that are defined separately for each household in the city based on their immediate neighbors (i.e., the eight immediately adjacent neighbors on the lattice). This approach reflects the assumption that group contact should be evaluated relative to nearby neighbors.
When agent vision follows site centered regions, these segregation
scores will be higher than bounded area segregation scores.
Households are the "active" agents in the simulation.
Each household is assigned an ethnic status and preferences
governing their goals for achieving specific levels of
in-group and out-group residential contact.
Housing units are located throughout the city. In this implementation, they are all similar in quality and desirability. They vary in only one respect, the ethnic mix of the neighborhood they are in.
Housing units are arranged in a spatial grid or city "landscape". The city landscape is divided into equal-sized, square subareas with distinct boundaries. These are termed "bounded neighborhoods" and are roughly comparable to census block groups.
Bounded nedighborhoods are the areal units that households consider when they evaluate "neighborhood". These are also the areal units that are used to compute city-wide segregation segregation measures (e.g., the index of dissimilarity).
Legend for City Landscape
The legend below illustrates how SimSeg Lite uses color to indicate the ethnicity of the households residing in occupied housing units.
(Note: The legend also shows socioeconomic status, but that is not relevant in version "C" of SimSeg Lite.)
During each cycle a certain fraction of households are randomly chosen and given the opportunity to "search" for new housing.
The searching household is "shown" a random selection of 12 vacant housing units. The household then identifies the "best" available unit and makes a decision to move to that unit or remain in their present unit.
Use this selection box to choose what characteristic of the housing unit or the resident household will be displayed in the city landscape image. There are two options:
Using Colors. Depicts resident households using color to indicate the household's ethnic status. This is the default option.
Using Grays. Depicts resident households using shades of gray to indicate the household's ethnic status.
Households are assigned membership in one of three ethnic groups - white, black, or Hispanic. In the context of the simulations, these ethnic group labels are merely that - arbitrary labels devoid of content other than that explicitly outlined by the simulation scenario.
The only characteristics that households have (beyond ethnic status) are ethnic preferences. This is the only dimension on which ethnic groups may differ.
The ethnic demography of the city can vary in two respects. One is that the city may have two or three ethnic groups. The other is that the size of the groups may vary. These settings are controlled by two selection boxes for ethnic group size.
One selection box sets the size of the Black population. Another sets the size of the Hispanic population. The size of thhe White population is determined by these two settings.
If the Hispanic population is set to "None", the city will have two ethnic groups - Whites and Blacks.
If a non-zero value is selected for the Hispanic population, the city will have three ethnic groups - Whites, Blacks, and Hispanics.
A household's decision to move is based on an evaluation of the available housing identified through search.
Each unit is evaluated on the basis of the ethnic mix of the neighborhood (defined by agent vision).
An overall "satisfaction" score is computed for each housing unit the household examines.
If the best available house identified through search is more attractive than the household's current house, the household attempts to move.
In a certain fraction of searches, the household moves to
the best available house identified through search even if
it is less attractive than the household's current house.
This simulates the fact that households frequently "must"
move. For example, they may lose their lease; they may be
forced to relocate due to a job transfer; they experience
a major change in household structure (e.g., divorce, death),
Households in the simulation are given "preferences" that guide their evaluation of their current residence and also available alternatives (when the household is searching for housing).
Preferences relating to ethnic mix are determined by the setting of the selection boxes for in-group and out-group preferences. Areas are evaluated based on the agents vision.
"Cycles" and "stages" represent the time dimension in the simulation model.
Cycles are measured and reported. A cycle is a period of time during which a certain fraction of the households in the city are selected at random and given the opportunity to engage in search and possibly move. The fraction is chosen to produce movement that approximates the amount of movement seen in real cities over a period of 6 months to one year. (Early in the simulation the amount of movement is actually higher because all households are required to move at least once.)
Choosing a higher number of cycles helps insure that the simulation will run long enough for segregation patterns to emerge and become evident. Choosing a lower number makes it easier to "pause" and observe intermediate results as the pattern begins to emerge.
A "stage" is a span of cycles that are run using the same scenario (i.e., the same settings on the model variables). If the scenario is changed and new cycles are run, a new stage has begun.