Abstract
Estimated measures of positional accuracy must be used in the interpretation of results of analyses that investigate relationships between health outcomes and exposures measured at residential locations. Although specific results from this study cannot be generalized to all new situations, the methods used to determine the sensitivity of results can be applied broadly. Analyses similar to those employed in this paper can be used to validate the interpretation of results from empirical analyses that use geocoded locations with estimated measures of positional accuracy.
Introduction
In environmental epidemiology, researchers frequently rely on geocoding to translate street addresses into geographic coordinates. These coordinates are then used to estimate an individual's exposure to environmental pollutants, such as emissions from industrial facilities or agricultural operations. However, geocoding is rarely perfect. Depending on the method used, a person’s recorded "location" may be hundreds of meters away from their actual residence. This study investigates how this spatial "noise" impacts the ability of researchers to detect and accurately measure links between environmental exposures and health outcomes.
Methods and Data
The researchers conducted an experimental simulation using real-world geography from Carroll County, Iowa. The study focused on 1,581 unique rural residential structures and their proximity to 55 Concentrated Animal Feeding Operations (CAFOs).
Three levels of geocoding accuracy were compared:
- Orthophoto (G°): The "Gold Standard," where locations were manually identified using high-resolution aerial photography.
- E-911 (GE): Points used by emergency responders, typically located at driveway entrances.
- TIGER (GT): Automated matching to street centerlines using U.S. Census files, tested with various "offsets" (0 to 800 feet) to simulate moving the point toward the likely house location.
Exposure levels were calculated using AERMOD, a standard plume dispersal model, to estimate relative air contaminant levels at every geocoded point. The researchers then assigned a known health risk (e.g., an Odds Ratio of 1.2) to the population based on the Gold Standard locations and ran 10,000 simulations to determine if this relationship could be recovered using the less accurate E-911 and TIGER coordinates.
Key Results
The Attenuation Effect
The study found that as geocoding accuracy decreases, the estimated Odds Ratio (the strength of the link between exposure and disease) consistently shrinks toward 1.0. This "attenuation" means that poor spatial data can make a significant health risk appear smaller or statistically insignificant.
Statistical Power and Positional Error
Positional errors varied significantly by method. TIGER geocoding had a median error of 693 feet, while E-911 performed better with a median error of 151 feet. These errors directly impacted statistical power:
- Orthophoto (Gold Standard) achieved 93.2% power to detect a modest health risk (OR = 1.2).
- E-911 power dropped to 87.0%.
- TIGER (with 0 offset) plummeted to 57.5%, indicating a high likelihood of failing to detect a real health link.
The TIGER Offset Paradox
In rural areas, standard TIGER geocoding places points directly on the street centerline, which is often far from the actual house. The researchers discovered that adding a 400–800 foot offset—moving the point away from the road—actually improved the recovery of health relationships by bringing the coordinates closer to the actual residential structures.
Discussion and Conclusion
Rural geocoding errors are typically much larger than urban ones, making rural health research uniquely vulnerable to spatial inaccuracies. This study demonstrates that lower positional accuracy leads to attenuated results, potentially masking real public health threats.
Researchers are encouraged to perform sensitivity analyses similar to those used here before beginning large-scale studies. By simulating their specific geographic environment and exposure models, they can determine if their chosen geocoding method provides sufficient accuracy to reliably detect the health risks under investigation.
Summary updated