URISA (1994), p900-911, copyright Urban and Regional Information Association
Darrell Packer Christopher Shea Department of City Planning City of Pittsburgh 200 Ross Street, Fourth Floor Pittsburgh, Pa. 15219
Abstract:The City of Pittsburgh City Planning Department has developed an Analytic Atlas for display and dissemination of Census Data. Using the PAGIS system staff can select any set of demographic, economic, or administrative data for census tracts and create thematic maps for public display, meetings, grant applications or staff studies. Analytic techniques highlight comparative measures such as concentration of populations and specialization of neighborhoods. This extends the common thematic map presentation of simple numbers or percentages. The Atlas was developed in cooperation with the Department of Urban and Environmental Planning at the University of Virginia and thereby represents cooperative ventures possible in an era of electronic communications and public sector downsizing.
The City of Pittsburgh is located within Allegheny County in southwestern Pennsylvania. The city and county have both experienced population decline over the past decade and while economic adjustments have been particularly difficult for the City, areas of poverty and challenges of meeting service needs in the region extend beyond the city boundary into the county. Developing a strategy for public and private actions to address the social and economic health of the region requires insights that transcend the competition and simple cooperation among political jurisdictions in the region. The City Planning Department recognizes the importance of visually representing the demographic and economic diversity of the region as an element in policy exploration and strategic planning.
[End Page 900]
The Analytic Atlas developed out of the desire to use the cooperatively developed PAGIS information system to support regular staff activities. An interactive desktop analysis and mapping environment was envisioned where the full range of census information for the region could be explored. This base system would be expanded to include other spatial data developed by PAGIS.
PAGIS is the Pittsburgh Allegheny Geographic Information System developed by Pittsburgh. The distributed workstation configuration relies heavily of a few highly skilled GIS professionals for almost all of the production work. Undertaking the development of a general analytic tool to be used on the desktops of staff in city management, planning, housing and other departments exceeded resources available during a period of governmental downsizing. The cooperative arrangement between the Department of Urban and Environmental Planning at the University of Virginia and Pittsburgh City Planning Department permitted the detailed development and documentation of tools necessary to implement the Analytic Atlas.
The background for a computer based census atlas, the conceptual basis for what constitutes an Analytic Atlas, implementation details, and an evaluation of cooperative arrangement constitute the subject of this paper.
In this age of desktop mapping and geographic information systems the production of maps showing census information is commonplace. Quick thematic mapping is even being added to common spreadsheet programs. The apparent ease of mapping belies the long development paths for quality mapping of census information. Space does not permit here a full examination of those developments or appropriate mention of quality census maps generated by many cities across the country. Rather two notable efforts deserve mention because of their use of computer technology in their respective eras and as benchmarks to compare and contrast the elements of the Pittsburgh Analytic Atlas.
In 1974 the U. S. Bureau of Census and Manpower Administration published a series of map volumes (U.S. Census, 1974). Each Standard Metropolitan Statistical Area volume had twelve pairs of maps. Each pair showed the same 1970 census variable for the central city and the metropolitan area as a whole. Thus, the pattern for the city jurisdiction could be compared with the entire region. The variables included:
[End Page 901]
Each theme had standard breakpoints for the attribute classes and a consistent "color ramp" for the subject. Some subjects used the same color ramps. For example, Median Housing Value and Median Family Income both used a dark green to yellow ramp for dollar values. The innovative and now well known dual color ramp for Interrelationship of Family Income and Educational Attainment combined a beige to red ramp for Educational Attainment with a beige to dark blue ramp for Median Family Income to obtain a sixteen color matrix for the two variables.
The Atlas was produced using prototype hardware and software at the Lawrence Berkeley Laboratory. The census tracts were digitized using a laser beam, line-following digitizer. The maps were plotted using the CARTE mapping package developed at the Laboratory and output on Computer Output on Microfilm which was then enlarged to create the printing composites. The high quality printed maps and, what was at the time, innovative technology set the standard for modern census atlases.
The Department of City Planning in New York City produced an atlas of 26 variables from the 1980 Census of Population and Housing using their in house computerized mapping system (Department of City Planning, NYC 1985). The 2,216 census tracts were mapped for the City and each of its five boroughs. The resulting 156 maps were reproduced in black and white using standard hatching patterns. The City maps provided some comparison of patterns across the region while the Borough maps gave locational detail. (This herculean effort was matched by a similar mapping of land use by parcel for the city.) This project reflects the attention to detail but also the power of repetitive production possible with computer generated maps.
The twenty six variables were:
[End Page 902]
*Indicates mapped as percentage of the appropriate reference population for each census tract.
Several lessons gleaned from these efforts are worth highlighting. Computer mapping has moved from the specialized laboratory to the individual city department and now to the individual desktop. Standardized format and attention to details on color or shading help produce understandable products. Content is better understood and interpreted when comparisons to a larger context is available. Computers can generate a multitude of maps with seemingly little effort. As a correlary, but one not indicated by these examples, errors can be propagated with similar ease.
Also apparent is the focus on the "characteristics of the place". Each map of the two atlases displayed information about the census tracts. The number of persons, the density, the median dollar value, the participation rates, or the percentage of the appropriate population with a characteristic was mapped for each census tract. Other than perhaps density, there is little information conveyed about the spatial distribution or concentration of special populations or housing variables. Indeed, eight of the twelve variables in the Urban Atlas and eighteen of the twenty six variables in the New York City Atlas were simple percentages for the census tract as the unit of analysis. This conveys a very useful picture of the character of each area, but does not indicate concentration or specialization of specific populations.
[End Page 903]
The Pittsburgh City Planning Department wanted to move beyond the standard portrait of the region focused on simply the characteristics of each sub-area. The spatial dispersal or concentration of specific populations or housing types concern policy analysts designing housing programs, planning the delivery of human services or targeting economic development. Further, the Department wanted to sharpen their understanding of the region by focusing on similarities and differences among the census tracts. The Department believed transformation beyond simple percentages would assist the community of census users in advancing their own analytical skills and advance a common understanding of change within the region. The City Planning Department was also interested in producing quality products with high readability and clear information content.
These motivations dictated several essential elements for the analytic atlas. A standard set of formats for map products would ensure quality products even for quick mapping exercises. A series of templates and well thought out color and gray-tone palates were essential. Maps would have to go beyond the customary but useful "Percentage" portraits to highlight concentration of populations and to highlight specialization of areas. Portraying a wide range of census variables would require analytic tools in the design of appropriate class intervals for thematic maps since predetermined classes would not be adequate. Census data would have to be preprocessed to ensure accurate linkage to each geographic area and to avoid the harsh clashes in map composition resulting from sliver tracts with small or no population or housing skewing classifications. The vision of interactive mapping available to a wide range of users heightened the importance of each of these elements and exacerbated the pressures on the already taxed professional GIS staff.
Data for the Pittsburgh Census Atlas Project was taken from the 1990 Census of Population and Housing and was derived from the STF3A CD-ROM. The project therefore includes data from the 100 % enumeration of households and the sample data collected by the Census Bureau. One hundred and forty two (142) of the 198 data tables from the Summary Tape File have been pre-processed to merge sliver tracts with their corresponding "mother" tracts and eliminate various geographic irregularities. In addition, a dozen variables from the 1980 census were converted to 1990 census tracts for comparative maps.
For many of these 142 Data Tables several if not all six of the following calculation would be appropriate for spatial analysis and display. Map templates have been created for both the City and the entire County for each of these measures. Each measure, rather than each topic, has a standard color ramp for display. This permits comparison across the entire study area placing the City in its regional context.
[End Page 904]
Data items number of persons or number of housing units can be mapped directly or as density maps. Other items such as median age or median housing value, can be mapped directly. Similarly the change in the number of persons or housing units between the decennial censuses can be mapped directly or as a percentage change. Each map represents the magnitude of the variable for each census tract. A standard yellow to dark green color ramp and gray scale is used. When both ends of the distribution are of interest, as in income or housing value, the color ramp is extended at the yellow end by adding orange and red.
The "percentage" calculation shows what percent of a tract's population or housing has the particular characteristic being studied. For example, the percentage calculation determines what percentage of a tract's housing stock is owner occupied or what percent of the families are headed by single parents. The percentage color ramp goes from light tan to dark brown.
The "share" calculation shows what percentage of the City's or County's total number of a sub-population is contained in a given census tract. This measure focuses on the distribution of a population throughout the region. Where would you expect to find a sizable number of this group? For example, the share computation would be used to determine what percent of Pittsburgh's total owner occupied housing is contained in a given census tract. The share color ramp goes from light to dark orange.
"Concentration" provides a rank ordering and accumulation of the census tracts by their percentage "share" of an attribute. The resulting value for each tract is the percentage of the attribute contained in that tract or all larger tracts. When mapped, these values show the census tracts in each quartile of the city's or county's distribution of a variable. Thus for example, six of the city's 174 census tracts contain one-quarter of all of the black owner occupied housing. And nine tracts contain the next quarter. Thus, half of the black owner occupied housing is found in just fifteen tracts. (See Map 1) The concentration color ramp depicts the first quartile in a dark blue-purple, the second quartile in a dark blue, the third quartile in a cyan and the fourth quartile in a light blue.
The "specialization" index shows how a tract compares with the entire City or County. This index facilitates the comparison of the similarities and
[End Page 905]
[End Page 906]
differences of a tract with study area. The specialization index is calculated by dividing a tract's percentage of a characteristic by the city- or county-wide percentage of the same attribute. If the resulting index exceeds 1.0, the tract has more than its share of the attribute and the value of specialization index shows the order of magnitude of the difference. Similarly, if the value is less than 1.0, the tract has less than its share of the attribute. The same index can also be determined by dividing a tract's share calculation for an attribute by the city or county total share. This specialization index, also called the Location Quotient, has a long and extensive history of use in regional economic and demographic studies. (Florence, 1939 and 1943; Isard 1960; Willie 1960; Schnore 1964; Rees 1970; Chen, 1994) (See Phillips, 1976 for a history of this index).
For example, a census tract may have 10% of the city's owner occupied housing but may have only 5% of the total housing units. As a result, the specialization index would be 2.0. Specialization can be computed relative to either the City of Pittsburgh or Allegheny County.
Since the value 1.0 has direct interpretive value, the color ramp designed for specialization maps has light tan and green for values less than 1.0 and increasing shades of blue for values greater than 1.0.
"Rates" provide a measure of what proportion of a specific population participate in certain behaviors. The labor force participation rate indicates the proportion of those of working age who have or are seeking employment. This rate can be conditioned by focusing on male and female labor force participation rates. Further conditions on the specific population could lead to the labor force participation rate for females with children under 6 years of age. Maps of these rates are different from the "percentage" maps described above as they focus on behavior. (See Myer, 1992, pp. 256-260 for a complete description of this distinction.) The map colors for rates goes from light to dark blue and into a dark red/purple to highlight exceptionally high rates.
To illustrate the types of topics added to the traditional census atlas, the following is a partial list of the attributes mapped by the atlas.
[End Page 907]
The Analytic Atlas is composed of desktop automated mapping derived from the PAGIS system. Formal high-quality maps are constructed on the work station implementation of ARC/INFO using ARCTOOLS. Specific census data or resultant measures described above can be "related" directly to the census coverages and incorporated into the predefined templates or map layout files. Desktop interactive mapping is accomplished using the pc-based MS-Windows version of ARCVIEW. Census coverages first must be constructed on the work station and then exported and imported into ARCVIEW on the PC's. As data base relates become feasible, the need to route census data through the workstations will be eliminated.
Interactive exploration of the census data itself is accomplished using a spreadsheet program called COMPARE (Phillips in Klosterman & Brail, 1993). This program imports a census table, computes the "Percentages", "Shares", "Concentration", "Specialization Index" for the entire array of attributes for that table and all of the census tracts. The analyst can also examine histograms of the percent distributions (e.g. the population age pyramid or income distribution of a census tract) and the Lorenz Curves and Gini Concentration Ratios for the Shares (e.g. racial or ethnic segregation). Entire arrays or selected vectors of data for the census tracts can be extracted and ported to the mapping program.
A histogram spreadsheet is also available so any vector that is to be ported to the mapping program can be displayed and appropriate break points for the data classes defined. Notes and comments can be recorded in a wordprocessing window. Obviously, the interaction and integration of these tools on the PC is greatly facilitated by the window environment. As relational capability is developed in the ARCVIEW product, the importing of data can be accomplished bypassing the need to access the workstation environment.
[End Page 908]
The detailed preparatory work and documentation necessary to implement an analytic atlas for multiple users was beyond the time and personnel resources of the City Planning Department in Pittsburgh during a period of downsizing. Yet the need for a substantial understanding of the social and economic patterns and trends continues to be important in a period of economic and political transition, increased call for cooperation, and in the face of scarce and competing resources among local governments. The academic/research environment at the Department of Urban and Environmental Planning complemented the constraints in the City: State of the art technology in the GIS laboratory, on-going concerns with PC and workstation networking and communications by University computer personnel, faculty and students familiar with and conversant with census data and regional analysis, and a flexible, fragmented schedule that permitted detailed work to proceed despite the academic calendar. At the same time, coordination presented its own difficulties. The two sites had difficulties remaining on the same schedule with system and software updates. Thus problems that were solved in the software at one site might take several months to be resolved at the other. Coordination of the schedules of local government and academia also made decision making cumbersome at times. Fortunately, tolerance and graciousness carried the day.
Advances in computers, networking, and geographic information systems software have made simple tasks like thematic mapping of census information much more feasible and effective. The production of an atlas has moved from the specialized research laboratory in the early 1970's, to the individual city department in the 1980's, and to the individual desktop in the 1990's. Obviously, the speed with which analysis can be performed and maps produced allow for more thought to be expended on understanding the spatial patterns contained in the data and on what information would be appropriate to communicate in mapped form. The speed with which mistakes can be made and the opportunity for carelessness to create misrepresentations of information caution us to be deliberate in the planning and implementation of even these simple systems. The Pittsburgh Analytic Atlas Project has found one way to exercise that care through a cooperative arrangement between local government and academic research.
With progress nearing completion on a related PAGIS project which is creating an Electronic Atlas on CD-ROM of historic and archeological features, further developments in the Analytic Atlas hint at creating a similar Electronic Atlas on CD-ROM of census data for wider public dissemination. The cautions for maintaining the quality of the information and the quality of the potential maps are even of greater significance with that prospect.
[End Page 909]
Department of City Planning, City of New York, 1985. Atlas of the Census: A Portrait of New York City from the 1980 Census. New York.
Chen, Lijian, 1994. Modeling Housing and Demographic Diversity at Census Tract Versus Block Group Levels of Aggregation, URISA Journal, Forthcoming.
Florence, P. Sargent, 1939. "Political and Economic Planning," Report on the Location of Industry. London, March 1939.
Florence, P. Sargent, W. G. Fritz, and R. C. Gilles, 1943. "Measures of Industrial Distribution," Chapter 5, Industrial Location and National Resources. Washington, D.C.: U. S. National Resources Planning Board.
Isard, Walter, 1960. Methods of Regional Analysis, Cambridge, Mass. The M.I.T. Press.
Myers, Dowell, 1992. Analysis with Local Census Data, San Diego, Ca.:Academic Press.
Phillips, David L., 1976. Comparative Analysis Techniques: A Framework for Regional Analysis Based on Conditional Probability. Ithaca, N.Y.: Graduate Field of City and Regional Planning.
Phillips, David L., 1977. "A Framework for Analysis of Urban and Regional Data Sets", in Urban and Regional Information Systems Association Proceedings, pp. 279-290.
Phillips, David L., 1993. "COMPARE: Comparative Analysis Techniques" in Klosterman, Richard, Richard Brail and Earl Bossard, Spreadsheet Models for Urban and Regional Analysis, Center for Urban Policy Research, New Brunswick, New Jersey, Rutgers University.
Rees, Philip H., 1970. "Concepts of Social Space: Toward an Urban Social Geography," Chapter 10, Geographic Perspectives on Urban Systems, ed. Brian J. L. Berry and Frank E. Horton, Englewood Cliffs, New Jersey: Prentice-Hall Inc.
Schnore, Leo F., 1964. "Urban Structure and Suburban Selectivity," Demography, Vol. 1, No. 1, p. 172.
U.S. Bureau of the Census and Manpower Administration, 1974. URBAN ATLAS, Washington, D.C.:Government Printing Office.
[End Page 910]
Willie, C.V., 1960. "Age Status and Residential Stratification," American Sociological Review, Vol. 25, pp. 260-264.
In addition to the authors, other City staff involved in the Pittsburgh Analytic Atlas Project include: Paul Farmer, Deputy Planning Director, who initiated both the PAGIS and Analytic Atlas Projects, and Ed Wells, Principal Information Systems Planner, who directed the implementation of the PAGIS project, and Jane Downing, Planning Director, who provided support for the Atlas Project. Other University staff included Virginia Ashby Mapp and Julia Kerr of the Department of Urban and Environmental Planning.
ARC/INFO, ARCTOOLS, and ARCVIEW are copyrighted trademarks of Environmental Systems Research Institute, Inc.
MS-Windows is a copyrighted trademark by Microsoft Corporation.
[End Page 911]