ecological fallacy:  (a) mistakenly attributing an average or modal characteristic across individuals in a group to an individual in a group;  (b) mistakenly attributing the combination of characteristics that are most common within a group to any individual who has one of those characteristics.

• A neighborhood with very high per capita household income may contain households with low incomes.
• If a neighborhood has above-average proportions of high-income households and households with older heads, it may nonetheless contain relatively few high-income households with older heads.  (For example, if house prices have been appreciating faster than average incomes, newer, younger home-buyers may have substantially higher incomes than the older folks who bought their houses 15-20 years earlier).

spatial analysis:  examination of the spatial relationships among cases and their attributes.

• A non-geographic approach to the relationship between education and income might observe these two variables for 100 individuals, and look for a statistically valid relationship between them (perhaps, controlling for other variables such as sex and years of working experience).
• A geographic approach might observe these two variables for 20 individuals in each of 5 regions, and look for a statistically valid relationship between the variables, controlling for region (because the prevailing income is likely different in different regions).
• A spatial-analytic approach might assess the extent to which each variable or the relationship between the two variables depends on the way in which the regions are defined, or might look for spatial autocorrelation in the average values of each variable across regions (or in the strength of the relationship between regions).  A spatial-analytic approach might assess the impact of the location of each sampled individual within the regions:  does it matter whether the samples are drawn randomly within each region?  Regions are not only used as cases, but the spatial characteristics and relationships of the regions and of the observations form a basis for analysis.

boundary problems:  shortcomings in data analysis when the regionalization scheme doesn’t match the geography of the processes.

• A particularly glaring example would be trying to relate economic trends (e.g., the relationship between changes in employment and population) by using municipal data (relate employment change and population change as two variables of each city and town in Washington State) rather than metropolitan or location data.  (What’s a typical way around this?  Metropolitan-area data, in which the regionalization scheme has something to do with the processes at hand — commuting patterns)
• GIS analysis might allow the researcher to relate the population change of each city to the employment change in each other city, weighted by distance.

• When is this more vs. less of a problem?  (Less of a problem when the regional unit is given and of some theoretical or administrative importance).
• How to work with it?
a) Ask at what geographic scale the phenomenon occurs;  e.g., some poverty is a result of region-wide lack of economic activity;  some is a result of neighborhood-specific issues (poor housing, poor access, poor schools);  some is a result of individual circumstances.
b) Be able to defend your regional boundaries as being relevant to the process you're trying to investigate.  (For example, if you're trying to study neighborhood social issues, then you'd want to have neighborhood boundaries that reflect actual social interaction).
b) Use GIS to estimate the effects of aggregating small areas in various ways to form your regions.

GIS  IN  MARKETING

What makes a marketing task geographic?  (Not all are).

• Site analysis:  the need to locate a distribution or a retail outlet that you expect to experience distance decay in its demand.
• Direct mail:  the desire to target potential customers by postal code (because you don’t have household-specific information or because you want to target potential customers of a particular outlet)
• Market analysis:  the desire to target potential customers by geographic media market
• Sales planning:  the need to develop non-overlapping sales territories for staff or contractors.

Site analysis

We’ve already reviewed a methodology for selecting new outlet locations using the criteria of:

a) non-overlapping market areas
b) high and proximate concentrations of small areas with desirable market characteristics

1. Use GIS to determine the 0-15-minute, 15-30-minute, 30-45-minute, and 45-60-minute drive-time radii around an existing outlet and the location of a proposed outlet.
2. Use G-D data and GIS to determine the number of households in each small-area geodemographic “type” within each drive-time radius of each site.
3. Use survey data to determine the number of customers (or volume of sales) at the existing site that come from each geodemographic zone and drive-time.
4. Assuming that the proposed site will obtain the same market penetration ratio by zone and radius, calculate the sales at the new location.

The brief case in Martin and Longley (in L&C, 1995) at the bottom of page 22 (and Fig. 2.5) describes a use of geodemographic data for targeted, direct marketing.  Be able to encapsulate the four steps entailed:
 
 

GIS  IN  PUBLIC-SERVICE  PLANNING

What makes a public-service task geographic?  (Not all are).

• The need to establish a physical presence “close” to the people who need the service (emergency services, daily-use services).
• The desire to affect specific regions more than others, presumably because the target population is concentrated there.

Note the Cardiff case [Martin & Longley 1995 in L&C book, pages 24-27]:  a scheme for identifying targets of home-repair grants.  What makes this task explicitly geographic:  why should localized neighborhoods be targeted?

• Additional public funds (for public-infrastructure repair) may be available if a large proportion of houses in a neighborhood are being improved.
• There are localized externalities of home-repair, benefits that accrue to other homes and property owners near the repaired property:
• Increased property values
• Private mortgage availability
• Psychological benefit that may lead to others to engage in privately financed home repair

• The program targets older houses:  the age of housing units is spatially autocorrelated — if you find several older houses, there are probably others nearby.
• The program targets single-family houses:  because of zoning, this is also spatially autocorrelated.
• The program targets lower-income households:  ditto.
• The program targets households that are likely to stay in place:  there is some geographic pattern to residential mobility.

Why might one want to use a GIS, since these data are already available by small area?
 

REFERENCES

Birkin, M.  1995.  Customer targeting, geodemographics and lifestyle approaches.  Ch. 6 in Longley, P. and Clarke, G., eds.  GIS for Business and Service Planning.  New York:  John Wiley & Sons.

Longley, P. and Clarke, G., eds.  1995.  GIS for Business and Service Planning.  New York:  John Wiley & Sons.

Martin, D. and Longley, P.  1995.  Data sources and their geographical integration.  Ch. 2 in Longley, P. and Clarke, G., eds.  GIS for Business and Service Planning.  New York:  John Wiley & Sons.