Comparing Methods for Predicting Air Pollution Levels
Author Information
Author(s): Igor Burstyn, Nicola M Cherry, Yutaka Yasui, Hyang-Mi Kim
Primary Institution: The University of Alberta
Hypothesis
How can we best use available information on sulfur dioxide concentrations to predict location-specific average exposure in rural western Canada?
Conclusion
Collecting some measurements or using a reasonable empirical mixed effects model is likely sufficient for most epidemiological applications.
Supporting Evidence
- The regression method with fixed and random effects had the best agreement with the alloyed gold standard.
- The Bayesian method using normal mixture prior was the second best in terms of correlation with the alloyed gold standard.
- The study utilized a large dataset from 2001 to inform predictions for 2002.
Takeaway
This study looked at different ways to guess how much pollution is in the air. It found that using some real measurements or a good model can help us make better guesses.
Methodology
The study compared various exposure modeling approaches using air quality measurements collected over two years at multiple locations.
Potential Biases
Potential biases due to measurement errors and the use of poor priors in Bayesian methods.
Limitations
The study lacks a true gold standard for evaluating the performance of different exposure assessment procedures.
Participant Demographics
Air quality measurements were taken across rural areas of western Canada associated with cattle ranching and oil/gas exploration.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website