GWR4 was developed by the same scholars that created Geographically Weighted Regression (GWR) (Brunsdon, Fortheringham, and Charlton). In brief, GWR runs local regression models on each geographic feature vs. ordinary least squares (OLS) regression which globally runs one model on all features. In GWR, coefficient estimates are allowed to vary geographically and can be mapped.
- Questions that GWR can help answer:
- General: Do the effects of demographics race, income, education vary geographically on your outcome of interest, after statistical adjustment?
- Specific: Does the percent of smokers on prevalence of asthma vary geographically or stay the same?
- Since smoking is a major risk factor for asthma, we would expect all local coefficients to be positive--increasing the prevalence or risk of asthma.
If you are still confused, click the map of Tokyo below--based on GWR4 and the sample data. Hopefully, this will help clarify the significance of GWR models. (I ran out of time or I would have classified the layer better. Keep in mind, odds > 1 each unit increase in the unemployment rate is associated with higher mortality, odds < 1 each unit increase in unemployment rate is associated with lower mortality.)"GWR 4 can be used to explore geographically varying relationships between dependent/response variables and independent/explanatory variables."
|Odds of the effects of the unemployment rate on mortality among the working age population. Each of the 262 municipalities in Tokyo has an estimate of unemployment's contribution--adjusted for other variables in the model.|
Visit here to download GWR4: http://www.st-andrews.ac.uk/geoinformatics/gwr/gwr-software/
|GWR's Interface and Workflow: Load data->Identify Dependent and Independent Variables-->|
--> Choose kernel, bandwidth, and selection criteria-->Output and Execute!