From: Subject: Desktop and Internet GIS Development for Spatial Segregation Analysis Date: Mon, 14 Feb 2011 11:02:59 -0500 MIME-Version: 1.0 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Content-Location: http://geog.gmu.edu/seg/esri_paper/p0275.htm X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.5994 Desktop and Internet GIS Development for Spatial = Segregation Analysis

Desktop and Internet GIS Development for Spatial Segregation=20 Analysis

David W. S. Wong=20

Several geographical measures of segregation have been introduced in = the past=20 decade. Because they require certain types of spatial information in = their=20 formulations, it is natural to incorporate these spatial measures in a = GIS=20 environment. This paper documents an effort in using Avenue scripts in = ArcView=20 to implement a set of spatial segregation measures. The results of this=20 development (project files) are available for researchers and = practitioners. In=20 addition, segregation is often a concern of local communities. Given the = wide-spread access of Internet, tools and data for analyzing spatial = segregation=20 are made accessible to the public through Internet GIS.

1. Introduction

Most studies on ethnic and residential = segregation so=20 far have relied heavily on the dissimilarity index D advocated by = sociologists=20 Duncan and Duncan (1955). Regardless of how useful it is in measuring = the=20 evenness dimension of segregation (Massey and Denton, 1988), it fails to = distinguish population patterns effectively. A set of spatial = segregation=20 measures has been proposed during the past decade, however, these = measures have=20 not been used extensively. Part of the reason that spatial measures were = not=20 widely adopted is because they were no existing tools to compute these = measures.=20 Generic tools such as database programs and spreadsheets cannot handle = the=20 computation because spatial information is needed in the calculation. = Therefore,=20 using GIS is necessary to develop such tools. However, accessibility to = desktop=20 GIS technology, though improved over the last decade, is still not high = outside=20 of the academic and research communities. Even if tools for computing = spatial=20 segregation are developed using GIS, local communities and public = organizations=20 will still be inaccessible to the tools and technology. There is a need = to make=20 these tools to compute spatial segregation more accessible to the = public,=20 probably through Internet.=20

The purpose of this paper is to report a recent effort to implement a = set of=20 spatial segregation measures within ArcView GIS. These measures are = developed as=20 additional spatial analytical tools in the ArcView user interface. = Interested=20 users can download the project file (.apr) with the tools embedded from = the=20 website managed by the author ((http://geog.gmu.edu). In addition, = this=20 paper explores the possibility of delivering the tools to compute = spatial=20 segregation indices on Internet through ArcIMS such that the public can = compute=20 segregation measures for the areas they are interested in. The next = section=20 provides an overview of the set of spatial segregation measures. The = third=20 section discusses the implementation of these measures in ArcView GIS. = Then, a=20 framework to deliver the segregation analysis environment through = Internet is=20 proposed. It is followed by a summary and discussion section.=20

2. Spatial Segregation Measures

2.1. Two-group Measures

The most commonly used segregation = measure is=20 the dissimilarity index D advocated by Duncan and Duncan (1955). This = index is=20 easy to compute, and ranges from 0 to 1, indicating no segregation to = perfect=20 segregation, respectively. The specific definition of the index cannot = be found=20 in Equation (1) in Appendix 1.=20

A major deficiency of D from a spatial perspective is that it cannot = solve=20 the checker-board problem. As long as each areal unit within the study = area is=20 dominated by one group or the other exclusively, the D index will return = a "1",=20 indicating a perfectly segregated situation, even if some adjacency = areal units=20 are occupied by different groups. Rearranging populations among areal = units will=20 not change the D value because D as well as other traditional measures = are=20 aspatial measures.=20

To overcome this limitation, spatial versions of D were proposed. The = D(adj)=20 index introduced by Morrill (1991) is the original dissimilarity index = less the=20 amount of interaction across areal unit boundaries. The level of = interaction=20 between any pair of neighboring units is then determined by the = differences in=20 the racial mixes of neighboring units. D(adj) is formally defined in = Equation=20 (2) in Appendix 1. This measure was further modified by Wong (1993) in = several=20 directions. Based upon the premise that the intensity of interactions = across a=20 boundary is not a simple function of adjacency, but likely the length of = the=20 shared boundary, the D(adj) index is slightly rewritten to incorporate a = boundary-length component to moderate the interactions across areal = units. This=20 index, labeled as D(w), is defined in Equations (3) and (4) in Appendix = 1.=20 Furthermore, Wong (1993) argued that the intensity of interactions = between areal=20 units is also a function of the size, shape, or the compactness of the = two=20 adjacency areal units. To incorporate the geometric characteristics of = areal=20 units into the measures of segregation, a compactness measure based upon = the=20 perimeter-area ratio was used. Equation (3) is further modified to = formulate the=20 index D(s), which is defined in Equation (5) in Appendix 1.=20

2.2. Multi-group Measures

The study of segregation in North = America has=20 been dominated by the two-group model as blacks and whites have been the = dominant groups historically. However, recent immigration history has = made the=20 North America societies more multiethnic. Therefore, there is a need to = measure=20 segregation in multi-ethnic settings. Based upon the concept of the=20 dissimilarity index D, Morgan (1975) and later Sakoda (1981) introduced = a=20 multi-group version of D. It's definition, D(m) is in Equations (1) and = (2) of=20 Appendix 2.=20

This multi-group measure of segregation can accommodate more than two = groups,=20 but shares the same limitations with the dissimilarity index D. That is, = the=20 measure is aspatial and rearranging populations among areal units will = not=20 change the overall level of segregation. To inject the spatial dimension = into=20 this multi-group measure, a spatial version was proposed (Wong, 1998). = The=20 formulation of the spatial version of D(m) is based upon the concept of=20 composite population counts. The composite population count of areal = unit i for=20 group j is defined in Equation (4) of Appendix 2, which is based upon = the=20 premise that within the neighborhood of i, people belong to different = ethnic=20 groups can interact as if they are in unit i. After the composite = population=20 counts for all areal units are computed, they are used to calculate D(m) = as if=20 they are the original population counts. Therefore, the spatial version = of D(m)=20 is SD(m) as defined in Equations (3) and (5) of Appendix 2. The major = difference=20 between the D(m) and SD(m) is in the population count. Using the = composite=20 population counts in SD(m) accommodates the interactions of population = groups=20 within the neighborhood, which are not accounted for in the = conceptualization of=20 D(m). Therefore, D(m) and SD(m) have the same mathematical properties.=20

Based upon the concept of spatial congruence, an ellipse-based = measure was=20 suggested to evaluate spatial segregation (Wong, 1999). Given the = locations of=20 people of a given group, an ellipse is used to fit the locations to = capture the=20 overall spatial distribution characteristics of the group. After = multiple=20 ellipses are derived for different groups, they are then overlaid to = derive an=20 index of segregation based upon the ratio of the intersection and union = of all=20 ellipses. Mathematically, the S index is defined in Equation (6) in = Appendix 2.=20

3. Implementation of Spatial Segregation Measures

GIS are useful = in=20 segregation studies in general (Wong, 1996). However, most analysts of=20 population and segregation studies probably take advantage of only the = mapping=20 capability of GIS without fully exploiting the potential of GIS in = assisting=20 population and segregation studies. One definite potential of GIS is to = support=20 spatial analysis. However, most standard spatial analytical functions in = GIS=20 fail to support the types of index calculation described in the previous = section. Still GIS can play a critical role in the computation of any = spatial=20 measures, including the set discussed above by providing or deriving = spatial=20 information necessary for the calculation of spatial measures.=20

A previous effort in using GIS to implement spatial measures of = segregation=20 has been successful (Wong and Chong, 1998). However, that implementation = exercise was not very efficient and rather costly because it had to rely = on a=20 GIS package (pre-Arc/Info 8 versions) and S-Plus, a statistical modeling = package. In addition, the process was not a GIS process in the sense = that it=20 requires using only database functions to search through the Polygon = Attribute=20 Table (PAT) and the Arc Attribute Table (AAT) to derive most of the = spatial=20 information required.=20

With the goal to maximize the number of potential users and minimize = the cost=20 of using these spatial segregation measures, ArcView was chosen as the = GIS=20 platform to implement the spatial measures. Besides its relative=20 user-friendliness and popularity among GIS users and social scientists=20 (sociologists, economists, political scientists and planners), several=20 characteristics of ArcView, including the powerful OO programming = environment=20 with Avenue, customizable GUI, the use of more compacted non-topological = shapefiles, make it a desirable candidate for the implementation. Even = though=20 shapefiles data structure does not stored topological information in the = database (Burrough and McDonnell, 1998), but given the object-oriented=20 environment in ArcView, topological information, geometric = characteristics and=20 spatial relations among geographic features can be computed quite = efficiently.=20

3.1. Two-group Measures

All two-group measures consist of two = major=20 components: the traditional D dissimilarity index, and the spatial = component=20 which moderates D. The computation of D is rather straightforward = because it=20 requires only the population data which are usually stored in the = feature=20 attribute table. An algorithm written in Avenue scripts was developed to = calculate the traditional D index without using any spatial operation in = ArcView.=20

To compute the spatial component, several items are needed to derive = from GIS=20 in addition to the population data. Adjacency is used to compare the = differences=20 in ratio mixes. To identify neighbors of a given areal unit in ArcView,=20 conceptually it is the same as to identify areal units with distance =3D = 0 to the=20 referenced areal unit. Specifically, in the ArcView environment, first = the=20 referenced areal i is selected. Then, issuing the SELECTBYTHEME request = to the=20 theme and set the selection type to #FTAB_RELTYPE_ISWITHINDISTANCEOF = with a=20 distance of "0", areal units in the theme adjacent to i, or with a = distance of=20 zero from i, will be selected. This selection process essentially = executes the=20 cij element in Equation (2) in Appendix 1 and permits the = comparison=20 of attributes of neighboring units in Equations (3), (4) and (5) in = Appendix 1.=20 By changing the distance parameter of the SELECTBYTHEME request, the=20 neighborhood definition becomes flexibly enough to accommodate other = criteria=20 besides the adjacency criterion.=20

When neighbors of i are selected, geometric parameters or attributes = of i and=20 each of its neighbors are derived. Perimeters (Pi and = Pj)=20 of each polygon pair can be obtained by sending the request RETURNLENGTH = to the=20 polygon objects. Similarly, the areas (Ai and Aj) = of the=20 polygon pair can be obtained by sending the request RETURNAREA to the = polygon=20 objects. After obtaining the perimeter and area information, the = compactness=20 measure (P/A) can be derived. In order to identify the line segment or = segments=20 representing the share boundary of the two polygons, the request=20 LINEINTERSECTION can be issued to the two polygons i and j. This request = will=20 return the intersecting line segment or segments of the two adjacent = polygons.=20 By issuing the RETURNLENGTH request again to the common boundary line = object,=20 dij is then extracted.=20

Regardless whether adjacency information (cij in Equation = (2) of=20 Appendix 1) or the length of a shared boundary (wij and=20 dij in Equations (3) and (4) of Appendix 1) is needed, the=20 information can be stored for later access. Perimeter and area measures = of each=20 polygon can be stored as additional items in the feature attribute table = to be=20 used in later computation. However, storing and retrieving these = different types=20 of spatial information in tables are not very efficient. First, storing = these=20 tables or matrices, especially for large spatial systems, will occupy a = lot of=20 memory space and in turn will slow down the computation. Second, = accessing=20 tables in ArcView is not a very efficient process. Instead of using the=20 traditional approach to store all types of spatial information in = tables, the=20 current approach is to take advantage of the efficient spatial indexing = system=20 which provides very efficient spatial selection operations.=20

Basically, the operation proceeds from one polygon to the next until = all=20 polygons are enumerated. In detail, the operation includes the following = major=20 steps:=20

1) select a polygon i in the study area.=20

2) select neighbors of i (i.e., j).=20

3) between i and each j, derive the racial mixes (zi and=20 zj), and their absolute differences. This=20

step will produce the numerator of Equation (2) for D(adj).=20

4) derive the perimeter and area for each neighboring pair. This step = permits=20 the computation of the numerator of the last part of Equation (5) for = D(s).=20

5) given perimeter information, intersection operation on neighboring = polygons is used to derive the shared boundary to compute dij = and=20 wij in Equation (4) and for D(w). 6) select another polygon = in the=20 study area to repeat Steps #2 to #5 until all polygons are enumerated.=20

7) select the MAX(P/A) from P/A ratios for all polygons.=20

8) compute the spatial components for D(adj), D(w) and D(s), and then = subtract them from D.=20

In addition, the algorithm also allows users to use another MAX(P/A), = which=20 is likely the global MAX(P/A) among all study regions, to compute D(s). = The=20 D(s)'s based upon the global MAX(P/A) provide meaningful comparison of=20 segregation levels among regions with different spatial partitioning=20 characteristics (Wong, 1993). After the spatial components of spatial = measures=20 are computed, they are subtracted from the aspatial measure D.=20

3.2. Multi-group Measures

The implementation of the spatial = version of=20 the multi-group dissimilarity index is very similar to the two-group = spatial=20 measures. The key process is to derive the composite population count = for a=20 population group in each areal unit (Equation (4) in Appendix 2). A = composite=20 population count is defined by a neighborhood function. For each areal = unit i,=20 neighboring units are selected using the theme selection request as in = the=20 two-group implementation process. Then the composite population counts = for=20 different groups are derived for the referenced areal unit i by = aggregating=20 population counts in the referenced unit and surrounding areal units for = each=20 group. Composite population counts are then derived for all population = groups in=20 all areal units. Based upon the computed composite population counts, = rows and=20 columns non-spatial operations for tables are used to compute SD(m) in = Equation=20 (3) in Appendix 2.=20

For the multi-group ellipse-based segregation index, the process is = quite=20 different from previous indices. To fit an ellipse to the spatial = distribution=20 of a population group, the first step is to extract the x-y coordinates = of each=20 areal unit or population point. Because the calculation of an ellipse is = entirely based upon the x-y coordinates of the point locations or = centroids of=20 polygons, therefore, for each polygon unit, the centroid as a point is = first=20 derived by issuing the RETURNCETER request to the polygon and its x-y=20 coordinates are extracted using the GETX and GETY requests issued to the = centroid point. Because each polygon or point location may have = different=20 population counts or weights, these weights or counts are then = multiplied to the=20 coordinate readings. For each population group and based upon the = locations (x-y=20 coordinates) of the population group, four parameters of the ellipse are = derived. These four parameters are the center (in terms of x-y = coordinates), the=20 two lengths of the perpendicular axes, and the angle of rotation of the = ellipse.=20 These parameters serve as input to create the ellipse object in ArcView=20 (Ei). After an ellipse is created for each population group, = it is=20 converted into a polygon object for subsequent operations which support = polygon=20 shape objects but not ellipses. Intersection of ellipses is obtained by = the=20 request RETURNINTERSECTION issued to the two corresponding polygon = objects.=20 Similarly, union of ellipses is obtained by the request RETURNUNION = issued to=20 the two corresponding polygon shapes. These steps are repeated for all = ellipses=20 to obtain the intersection and union of all ellipses. Then requests are = sent to=20 the intersecting polygon and the union polygon to obtain the areas in = order to=20 derive the numerator and denominator of Equation (6) in Appendix 2 and = thus=20 compute the S index. The algorithm provides the option for users to = store the=20 ellipses as polygons for all population groups into shapefiles, and = store the=20 intersection and union as polygons for display to provide a visual = impression of=20 the segregation level.=20

Algorithms developed include the traditional measures of segregation = for=20 two-group and multi-group situations (D and D(m)), spatial measures for = the=20 two-group cases (D(adj), D(w), and D(s)), spatial measure for = multi-group cases=20 (SD(m)), and the ellipse-based measure for multi-group cases. Scripts = written to=20 implement these algorithms are compiled and are linked to new menu items = under a=20 new menu labeled as "Segregation" in the ArcView GUI (Figure 1). = These new=20 functions and their corresponding items in the ArcView GUI are saved in = a=20 project file. Under the new menu, four items corresponding to the four = scripts=20 for the algorithms are added. Interested users can download the project = file=20 from the website managed by the author (http://geog.gmu.edu/seg). An = example using=20 the ellipse-based S index on population counts of white and black for = counties=20 in Georgia is also shown in Figure 1.=20

4. Internet Extension of Spatial Segregation Analysis GIS

Levels = of=20 segregation are of great concerns in public policy formulation and = analysis.=20 Besides academics and policy makers, local residents, community = activists, and=20 political leaders are likely concerned about the level of segregation at = the=20 community or neighborhood levels. Even though GIS technology has = proliferated=20 tremendously in different levels of the society, the majority people of = the=20 nation still do not have access to the technology. In order to make = available=20 the spatial segregation analytical tools to the public, delivering the = tools and=20 data through Internet probably is the most desirable. Therefore, the = enhanced=20 ArcView functions for spatial segregation measures have to work through = the=20 Internet environment. This model can be generalized to increase the = public=20 access to other spatial analytical tools that are supported by GIS in = general=20 and ArcView in specific.=20

There are many possible approaches to port these spatial functions to = the=20 Internet environment. On one extreme, all codes for these functions = together=20 with other GIS capabilities, such as basic spatial query and selection = and=20 cartographic displays, could be rewritten in codes supported by the = Internet=20 environment. This approach starts from scratch and will result in a very = efficient and highly specialized environment. A definite difficulty is = that it=20 is costly in development. The other extreme is to rely on existing = resources to=20 minimize development effort. That is to couple existing tools or = packages to=20 deliver the segregation computation capability to the clients on = Internet. The=20 rest of this paper discusses a preliminary framework to explore this = later=20 approach.=20

The computational tools developed on ArcView are based upon Avenue, = which is=20 not a standard tool in today's development environment. ESRI has been = advocating=20 Visual Basic for Applications (VBA) which will work with ArcInfo8 and = ArcGIS. To=20 utilize the new GIS technology with VBA, all Avenue scripts have to be=20 converted. As the strategy is to minimize development effort, it is = preferred to=20 have the computation tools remained in ArcView 3.X. In terms of the web=20 interface, several ESRI products offer different options. ArcView IMS = probably=20 will work best in this case, but unfortunately ESRI has indicated to = discontinue=20 the support of this product. MapObjects IMS can offer the most = flexibility, but=20 it requires more development effort. Also, similar to ArcView IMS, = MapObjects=20 will not be the focus of ESRI future development effort. On the other = hand, ESRI=20 has been promoting ArcIMS as the Internet product to deliver maps and = spatial=20 data on the web. Therefore, the development strategy is to couple ArcIMS = and=20 ArcView 3.X together.=20

Figure = 2 is the=20 user front end based upon the conceptualization of the development = model. This=20 front end inserts an ArcIMS page on the right with clients' inputs on = the left.=20 There are three main components. The ArcIMS page on the right allows = clients to=20 select areas from different data layers. It serves the spatial query = function.=20 The second component is for clients to select population groups from = which=20 spatial segregation measures are derived. The third component is to = indicate=20 what aspatial and spatial measures are requested by the clients. After = all these=20 selections were made, information is transmitted back to the server = through=20 different channels for processing.=20

After the client selects the interested area for the analysis by = interacting=20 with the map rendered by the ArcIMS server, the selection information is = passed=20 back to ArcIMS. On the ArcIMS server, the extract server is setup to = subset the=20 selected features from the corresponding shapefiles. The shapefiles of = the=20 selected features will be created on the web server (Figure 3). = After the=20 client chooses the population groups and the index or indices for the = analysis,=20 these selection parameters are written into a text file document on the = web=20 server.=20

Next, the ArcView project file with the enhanced spatial segregation=20 computation capability is modified such that when the file is triggered, = it will=20 automatically retrieve parameter information from the parameter document = and=20 also the shapefiles extracted for the subset of features. ArcView will = use the=20 information stored in the parameter file to determine which population = groups=20 are used and which index or indices are computed. Then the results are = posted=20 back to the client.=20

However, implementing this framework encountered several obstacles. = First,=20 when the extract server extracts the selected features from the = shapefiles, the=20 shapefiles subset will be in compressed format. The file has to be = uncompressed=20 before ArcView can access the data and process the requests. Second, = after the=20 selection parameters are passed back to the server to create a document, = ArcView=20 has to be triggered. It is not clear how ArcView can be triggered. = CGI/Perl is a=20 promising option which has yet to be tested thoroughly. Third, when = ArcView=20 should be triggered to access the parameter file and the shapefiles = subset is=20 not clear. Because the shapefile subset and the parameter document are = created=20 through different processes that are not synchronized, it is possible = that when=20 ArcView was triggered, only one document was created. Finally, without = using=20 ASP, it seems to be difficult to pass the results from ArcView back to = the=20 client when ArcView finishes its computation. More research and = development are=20 needed to overcome these obstacles or to formulate a new framework to = develop a=20 spatial segregation analysis module on the web.=20

5. Summary

This paper provides a concise overview of spatial = segregation=20 measures and documents how these measures were implemented in ArcView = using=20 Avenue. A project file with all tools for computing spatial segregation = measures=20 is available to the public (http://geog.gmu.edu/seg). The paper also = proposes a=20 framework to develop a web user front end for the spatial segregation = GIS.=20 Currently, the least-development effort approach was used by coupling = ArcIMS and=20 ArcView to minimize coding development. However, several major obstacles = exist.=20 Some of these obstacles may be removed with more research, but it is = possible=20 that another framework which requires more intense development effort = has to be=20 formulated.=20

There are several lessons that we can learn from the later part of = this=20 paper. The issue is how spatial analytical techniques in general can = become=20 accessible through the web environment. The approach adopted by most = system=20 developers these days is hardcore coding. It is possible to develop = web-enabled=20 programs to perform spatial analysis and these programs will perform = efficiently=20 in the web environment. However, taking this approach implicitly = separates GIS=20 applications and development on the web from the desktop. Numerous = powerful=20 tools, including the spatial segregation measures reviewed in this = paper, have=20 developed on desktop GIS. With efficient spatial indexing systems and = spatial=20 data models, desktop GIS can perform spatial queries and selections = quite=20 efficiently. These are essential building-block functions for many = spatial=20 analytical techniques developed on desktop GIS. With the intent to take=20 advantage of the efficient spatial selection capability in desktop GIS, = this=20 paper explores the approach to rely on enhanced desktop GIS tools to = perform=20 spatial analysis while input information is gathered from users through = the web=20 interface. Unfortunately, this experiment has not been successful. It is = true=20 that many spatial analytical functions can be implemented on the web = environment=20 with programming, however, these functions will not perform as = efficiently as=20 those found on the desktop environment unless GIS data formats and = models used=20 in the web environment are comparable to those used on desktop = environment.=20 Currently, it seems that there is a big gap in the development of = spatial=20 analytical capability between the desktop and web environments.=20

Acknowledgment:

This project is partially supported by the = National=20 Institute of Health/Child Health and Human Development (NICHD) under the = National Institute of Health (NIH) grant number 1 R03 HD38292-01.=20

Appendices

3D"Appendix3D"Appendix=20

References

Burrough, P. A. and McDonnell, R. A. (1998). = Principles of=20 Geographical Information Systems. Oxford: Oxford University Press.=20

Duncan, O.D. and Duncan, B. (1955). A methodological analysis of = segregation=20 indexes. American Sociological Review, 20, 210:217.=20

Massey, D.S. and Denton, N. A. (1988). The dimensions of residential=20 segregation. Social Forces, 67, 281-315.=20

Morgan, B.S. (1975). The segregation of socioeconomic groups in urban = areas.=20 Urban Studies 12: 47-60.=20

Morrill, R. L. (1991). On the measure of geographical segregation. = Geography Research Forum, 11, 25-36.=20

Sakoda, J. N. (1981). A generalized index of dissimilarity. = Demography=20 18: 245-250.=20

Wong. D.W.S. (1993). Spatial indices of segregation. Urban = Studies 30:=20 559-72=20

Wong, D.W.S. (1996). Enhancing segregation studies using GIS. = Computers,=20 Environment and Urban Systems, 20(2), 99-109.=20

Wong, D.W.S. (1998). Measuring = multiethnic=20 spatial segregation. Urban Geography 19: 77-87=20

Wong, D.W.S. (1999). Geostatistics as = measures=20 of spatial segregation. Urban Geography 20(7): 635-647=20

Wong, D.W.S. and Chong, W. K. (1998). Using Spatial Segregation = Measures in=20 GIS and Statistical Modeling Packages. Urban Geography 19(5): = 477-485.=20

David W. S. Wong
Associate Professor
Department of Geography =
George=20 Mason University
Fairfax, VA 22030
U.S.A.
Tel: 703-993-1212 =
Fax:=20 703-993-1216
Email: dwong2@gmu.edu