GIS in plain language¶
A geographic information system (GIS) stores information together with its location, so that separate data sets can be compared because they describe the same place. A list of ALS clinics and a table of population by county have nothing in common until both are placed on a map. Once they are, a question such as "how many people live more than 90 minutes from a clinic" has an answer.
This page covers the ideas behind every GeoALS map. It is an orientation, and the learning paths link to full courses.
Layers¶
A map is a stack of layers. Each layer holds one kind of thing: clinics, congressional districts, drive-time areas, census tracts. Analysis happens when layers are combined, for example by counting the points of one layer that fall inside the polygons of another.
Vector and raster data¶
| Type | What it is | Examples |
|---|---|---|
| Vector: points | A single location | A clinic, a monitoring station, a geocoded address |
| Vector: lines | A path | A road, a river, a transmission line |
| Vector: polygons | An area with a boundary | A county, a census tract, a congressional district, a lake |
| Raster | A grid of cells, each with a value | Satellite imagery, land cover, a modeled air quality grid |
Attributes and joins¶
Every feature in a layer has a row of attributes, like a spreadsheet row. A table join attaches outside data to a layer through a shared identifier, such as a county code. A spatial join attaches data through location alone, for example by giving each address the identifier of the census tract that contains it. Most health mapping depends on these two operations.
Geocoding¶
Geocoding converts an address into coordinates. It is the step where health data become mappable, and it is also the step with the greatest privacy consequence, because coordinates of a home identify a person. See Location privacy and governance before geocoding anything.
Geocoding quality varies. Rural routes, post office boxes and new construction match less reliably than urban street addresses, so a study should report its match rate and how unmatched addresses were handled.
Coordinate systems and projections¶
Coordinates only mean something relative to a coordinate reference system. Latitude and longitude from a phone or a web map are usually in the system called WGS 84. Because the earth is curved and a screen is flat, every map uses a projection, and every projection distorts distance, area or shape somewhere. Distances and areas should be calculated in a projection suited to the study region, and layers in different systems must be transformed before they are combined.
Geographic units¶
| Unit | Typical use | Caution |
|---|---|---|
| State and county | Policy, reporting, many federal data sets | Counties vary enormously in size and population |
| Census tract and block group | Neighborhood-level context | Boundaries change with each decennial census |
| ZIP Code | Often the only location in health data | ZIP Codes are mail routes, not areas; ZIP Code Tabulation Areas approximate them |
| Congressional district | Advocacy | Redrawn after each census, and sometimes between |
| Grid cell | Modeled environmental data | Resolution differs by source |
Core analyses¶
- Buffer: the area within a set distance of a feature, such as everything within 3 km of a site.
- Overlay: combining layers to find where conditions coincide.
- Proximity and drive time: distance or travel time from each person to the nearest service, measured along roads.
- Service area: the region reachable from a facility within a travel time, as in the GeoALS 90-minute clinic maps.
- Aggregation: counting or averaging points within areas, such as registry participants per district.
- Hot spot analysis: statistical tests for whether high or low values cluster more than chance would produce.
Reading a map critically¶
A map is persuasive, so it deserves the same scrutiny as a chart.
- Rates and counts differ. A map of counts mostly shows where people live. Compare rates, with a stated denominator.
- Small numbers are unstable. A county with two cases and a small population can show an extreme rate by chance. Small counts should also be suppressed for privacy.
- Boundaries shape the result. The same points grouped into different units can show different patterns. This is called the modifiable areal unit problem.
- Areas are not people. A pattern across counties does not establish that the same relationship holds for individuals. This is the ecological fallacy.
- Where someone lives now is one moment. For a disease with a long preclinical period, the address at diagnosis may say little about earlier decades, which is why residential history is central to exposure research.
- A blank area is not an absence of people. It may reflect who was reached by a survey or registry.