Evaluating Neighbourhood Weight Matrices in CAR Models
Author Information
Author(s): Earnest Arul, Morgan Geoff, Mengersen Kerrie, Ryan Louise, Summerhayes Richard, Beard John
Primary Institution: Northern Rivers University Department of Rural Health, The University of Sydney
Hypothesis
Does the choice of neighbourhood weight matrix affect the smoothing properties of Conditional Autoregressive (CAR) models in ecological studies?
Conclusion
Different neighbourhood weight matrices significantly influence the smoothing properties of CAR models, affecting their ability to recover observed risk in an area.
Supporting Evidence
- The study created four adjacency-based weight matrices and seven distance-based weight matrices.
- Distance-based matrices generally performed better than adjacency-based matrices in predicting relative risks.
- The 'Gravity' model had the highest sensitivity and specificity in classifying high-risk areas.
Takeaway
This study looked at how different ways of grouping areas can change the results when studying health data. It found that some methods work better than others.
Methodology
The study created various neighbourhood weight matrices and applied them to a dataset of births and birth defects in New South Wales, Australia, analyzing data from 1995 to 2003.
Potential Biases
Potential biases may arise from the choice of weight matrix and the ecological nature of the data.
Limitations
The study primarily focused on specific types of weight matrices and may not generalize to all ecological studies.
Participant Demographics
The study analyzed data from births and birth defects in New South Wales, Australia, covering a diverse population.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website