Background
PartI
PartII
PartIII
Summary
Acknowledgement

Part III: Use spatial autocorrelation to test independence assumption

Introduction

In developing the Metro Vancouver LUR models, sampling sites were selected to optimize the variability in measured NO2 concentrations. To evaluate the assumption that these sampling sites are independent, spatial autocorrelation was tested on the residuals from the 2009/10 model using Moran’s I statistics. No evidence of spatial autocorrelation was found between model residuals, suggesting that simple linear regression was appropriate for the data. 

Methods

The predicted values were subtracted from actual measurements and saved as a field called "residuals" in the attribute table for the 112 sites (point data). Then "Spatial Autocorrelation (Moran's I)" under "spatial analyst" was run to generate statistics.

Results

Moran’s I Index:  0.158

Z-score:  5.814

P-value:  0.000

Conclusion

The spatial autocorrelation was found insignificant, i.e. the residuals are not spatially autocorrelated, suggesting that the model did not violate the independent assumption and that the linear regression model was appropriate for the data.

 



Content last updated: Dec 9, 2010