The goal of this exercise is to learn about spatial statistics that are particularly relevant for public health. You will use three software packages including GeoDa, GWR3, and SaTScan. They are loaded on the computers in Saunders 322.
Part 1: Spatial Lag Regression in GeoDa
In this exercise you will use GeoDa to create a spatial lag regression model for spatially autocorrelated data. First read pages 45-54 in the software release notes. You will use the GeoDa workbook. Read Chapter 24. You might also need to use the manuals which consist of the software release notes, the Users Guide, and the section on spatial regression in a paper on GeoDa by Luc Anselin. Use South.zip data as shown in Chapter 24 of the workbook and go through the exercise. Then once you have practiced using the example in the workbook, build a different model. You can use another dataset of your own or use one from the online GeoDa collection of datasets. Or you can use the South.shp file and use HR60 as the dependent variable and DV60, UE60, RD60, MA60 as independent variables. That means you would drop the variable called PS60 from the book example. With your dataset first run the Classic OLS model first with the 2nd order rook's weights matrix. Print the output by cutting and pasting into a Word document. What is the R-squared value? Is the Robust LM (lag) significant? What is the prob-value? Is the Robust LM (error) significant? What is the prob-value? Then run it again using the lag model? Print the output and then run the diagnostics. Is the fit better for the lag or classic model? Explain your answer in a paragraph or two.
Part 2: Geographically Weighted Regression in GWR3
In this exercise you will use GWR3 to explore spatially varying relationships. Click here and here for a brief introduction to the method. Then read the document entitled "Geographically Weighted Regression Concept" which is available on blackboard. If you want to get into more mathematical detail then read some of the papers listed here. Then read the entire document entitled "Software for Geographically Weighted Regression" which is also available via blackboard. Once you've read the ENTIRE document then follow along with the example using the following sample file of Georgia. Try to produce exactly the same GWR3 model output that the example shows in Chapter 2 of this tutorial during this practice session. Use the X and Y variables (which are in UTM) instead of Longitude and Latitude for the location variables. Then try to produce exactly the same output that the example shows in Chapter 3. When you get to the section where you do a spatial join then use the following Georgia shapefile, which is in UTM. Try to understand as much of the output as you can during this practice phase. Up until now you've just been learning the method concept, how to use the GWR3 software, how to visualize the results in ArcGIS, and how to understand what it all means Next, you will implement another model and visualize and interpret the results.
Now, create a Gaussian model with PctBach (educational attainment) as the dependent variable and percent poverty, percent rural, and percent black as the independent variables. Use an adaptive filter and Cartesian coordinates, AIC as the bandwidth selection method, and output an E00 file. In the model output file title put your name. Run the model and print the model output file. Import the E00 file into ArcGIS and do a spatial join with the Georgia shapefile so you can create choropleth maps of some of the output variables.
Creating a coverage from an E00 file is explained in the GWR3 tutoral in Chapter3. However, the current version of ArcGIS is a bit different. There are different ways to do it. You can follow these steps. (1) Open ArcCatologue, (2) Click the "View" menu and point to "Toolbars", and click "ArcView 8x Tools" (3). Click "Conversion Tools", and go to "Import from interchange file" (4). Then enter the input file and name your output file and click "OK." Another way to do it is to look up "Import From Interchange File" in ArcToolbox and you can implement the procedure from there.
Make 5 choropleth maps (NOT graduated circle maps) using the following variables: (1) the intercept, (2) the parameter estimate for percent poverty, (3) the parameter estimate for percent rural, (4) the parameter estimate for percent black, and (5) the standardized residuals. Describe both the model output and the maps in a paragraph or two. What does it mean? In your answer be sure to explain what we have learned from the maps, including the maps of the parameters and the residuals. Is there any spatial pattern to how well the model that explains educational attainment? As you try to explain the model results, follow along with the document entitled "Software for Geographically Weighted Regression."
Part 3: Cluster Analysis in SaTScan
In this exercise you will use the SaTScan software to identify spatial clusters of health events. First read pages 26-42 and pages 57-69 of the SaTScan manual. You can also read the paper by Emch and Ali (2003) that is available on Blackboard for this class to get a conceptual understanding of the method. Do a cluster analysis for one of the following two datasets: (1) Brain Cancer Incidence in New Mexico or (2) Lung Cancer Incidence in New Mexico. For brain cancer you will analyze the spatial and temporal distribution of brain cancer in New Mexico from 1973-1991 by age, race and/or sex. For lung cancer you will analyze the spatial and temporal distribution of lung cancer incidence in New Mexico from 1973 to 1991 by age, race and sex. Explain the parameters you used to create the model. Print out the model results and describe them in a paragraph or two. Are there any clusters? Explain the attributes of the clusters.
Lab Deliverable Summary: Print out all of the outputs, put your name on them, and give them to the instructor. They include: Part 1: GeoDa output including OLS diagnostics and Lag Model results, a paragraph that explains the model results and answers the questions specified in Part 1; Part 2: GWR3 model result output that has your name on it, 5 choropleth maps, a paragraph describing the results and interpretation; Part 3: model results printout and a paragraph describing the parameters you used to create the model and interpretation of the results.
If you want to use the software in another lab on campus the here's how to access the software:
(1) Run Afs client program (the small lock icon in right bottom status bar) and click "drive letters" tab, then click "add" button, and configure a new drive letter for /afs/isis/depts/geog/pkg/spatialStats/ (2) From desktop, click "My Computer", then "Folders", and the drive letter you just create. (3) To access programs, please click each EXE file to run: Geoda\Geoda095i.exe GWR3\GWR30.exe Satscan7\SaTScan.exe Note - GWR is the installation directory for GWR, and students should not click any files under this directory.