Lab 12 - Geographically Weighted Regression (GWR)
Geographically Weighted Regression (GWR) incorporates a "spatial" component to regression models by explicitly recognizing geography and location. This type of model is beneficial and can improve upon non-spatial regression models, like the ones we learned about in the previous two labs (Zandbergen, Morgan).
For our lab assignment, we carried-out an OLS (Ordinary Least Squares) and GWR analysis; and then compared the results of the two models using location data of all crimes reported in one year in Mecklenburg County, North Carolina, as well as demographic data from the U.S. Census. The dependent variable I decided to analyze was the 'Hit and Run' offense (as a crime rate), against the independent variables of black race as a % of a total population, and residents of Hispanic ethnicity as a % of total population.
On my OLS results, neither of my results had a positive or negative coefficient, however the black race was a significant variable for p-value < 0.01. In my GWR results, the adjusted R-squared value increased, but the AIC value decreased (usually these two diagnostics are in sync). Finally, the spatial pattern for the Hit and Run crime rate regression coefficient is dispersed with a z-value of -1.7. This pattern makes sense based on the map of the OLS residuals because it emphasizes that I did not provide sufficient explanatory variables in my regression model. Screenshots of my work are provided below.
For our lab assignment, we carried-out an OLS (Ordinary Least Squares) and GWR analysis; and then compared the results of the two models using location data of all crimes reported in one year in Mecklenburg County, North Carolina, as well as demographic data from the U.S. Census. The dependent variable I decided to analyze was the 'Hit and Run' offense (as a crime rate), against the independent variables of black race as a % of a total population, and residents of Hispanic ethnicity as a % of total population.
On my OLS results, neither of my results had a positive or negative coefficient, however the black race was a significant variable for p-value < 0.01. In my GWR results, the adjusted R-squared value increased, but the AIC value decreased (usually these two diagnostics are in sync). Finally, the spatial pattern for the Hit and Run crime rate regression coefficient is dispersed with a z-value of -1.7. This pattern makes sense based on the map of the OLS residuals because it emphasizes that I did not provide sufficient explanatory variables in my regression model. Screenshots of my work are provided below.
Comments
Post a Comment