Using GeoDA Software for Geographic Data Analysis and Exploration Developed by Luc Anselin Arizona State University School of Geography and Planning geodatacenter.asu.edu.

Using GeoDASoftware for Geographic Data Analysis

and ExplorationDeveloped by Luc Anselin

Arizona State University

School of Geography and Planning

geodatacenter.asu.edu

Briggs Henan University 2010

1


2

Software for Spatial Analysis and Statistics• ArcGIS 9 The most common GIS Software, but $$$$!

– Spatial Statistics Tools for point and polygon analysis– Spatial Analyst tools for density kernel– GeoStatistical Analyst Tools for interpolation of continuous surface data

• OpenGeoDA, Geographic Data Analysis by Luc Anselin now at Arizona State– Download from: http://geodacenter.asu.edu/– Runs on Vista and Windows 7 (also MAC and UNIX)– Earlier version called GeoDA runs only on XP (0.9.5i_6)– Easy to use and has good graphic capabilities

• CrimeStat III download from http://www.icpsr.umich.edu/NACJD/crimestat.html– Standalone package, free for government and education use– Calculates values for spatial statistics but no GIS graphics– Good documentation and explanation of measures and concepts

• R Open Source statistical package, – originally on UNIX but now has MS Windows version– Has the most extensive set of spatial statistical analyses– Difficult to use– Need to learn it if you are going to do major work in this area

• S-Plus the only commercial statistical package with good support for spatial statistics– www.insightful.com

GeoDA Overview• GeoDA is a package for exploratory analysis of geographic data. • Primarily analyzes polygon data, but can also do some things with point

data• Has major capabilities not easily available elsewhere including: • --creates spatial weights matrices with multiple options • --linking and brushing between maps, histograms, scatter plots• --calculates and maps Local Indices of Spatial Association (LISA or local

Moran’s I).• standard multiple regression full diagnostics for spatial effects• spatial autoregressive model for both spatial lag and spatial error models• Free. ArcGIS not required, but it does require a shapefile for data input.


3

Obtaining GeoDA Software

• The GeoDA program is on my Web site at:

www.utdallas.edu/~briggs or go to

http://geodacenter.asu.edu/

You will have to create a new user account• download, unzip, and click the file OpenGeoDA.exe to

start the software– This version (OpenGeoDA) runs on Vista and Windows 7– Earlier version (GeoDA095i) only runs on XP

• it does have some “bugs” so some things may not work or it may crash!


4

Help and Documentation for GeoDA

• For help using OpenGeoDA, go to

http://geodacenter.asu.edu/

Click on Support tab

• For printable manuals, go to

www.utdallas.edu/~briggs and download geoDAdoc.zip– Geoda_quickstart : 25 page quick start guide to using geoda (read first)

– Geoda_spauto a quick guide to spatial autocorreletion measures (read next)

– Geoda93_manual is a 125 page manual which fully documents the software

– Geoda 95i_updates is a 64 page manual which covers bug fixes and enhancements in the latest release

– Note, all the above are written for the earlier version GeoDa9.3, not OpenGeoDa but differences are small

Briggs Henan University 2010 5

OpenGeoDa Interface: 1 of 2

Display and Create

• File—open a shape file: it should also contain the data to analyze

• Edit—copy maps, and open new maps to compare

• Tools—create spatial weights matrices (very good)

create shapefiles: Thiessen polygons, centroids, etc

create shapefiles from .dbf containing X,Y coordinates

• Table—Open a table (>Promotion), joins, variable manipulation, joins, etc.

To access more options, right click on any open window


6

OpenGeoDa Interface: 2 of 2

Analyze

• Map—create many types of choropleth maps

• Explore—creates various non-spatial graphs of data

• Space—calculating Spatial Autocorrelation measures

• Methods—standard and spatial simple and multiple regression

• Options—lists options for the currently active window.

To access options, right click on an open window


7

Data for Demowww.utdallas.edu/~briggs


8

geoDAdata.zip

china.zip

1. Use GeoDA to find the Centroids of the Provinces of China


(Need ArcInfo to do this in ArcGIS, which is expensive. GeoDA is free. )--Input the provinces shapefile: File>Open Shape File China.shp--Open the data table: Table>Promotion to see what is there--Create centroids for each province: Options> Add Centroids to Table Place check mark in X coordinates and Y coordinates box, click OK --X and Y centroid coordinates are added to the table--to keep them permanently you need to save as new shapefile Table> Save to Shapefile as China_Centroids.shp--to close these files and start something new: File>Close All

2. Create Thiessen Polygons forProvinces of China


--use point file of province centroids created --Start the tool: Tools>Shape>Points to Polygons Input File: China_Centroids.shp Output file: China_Thiessen.shp Bounding Box: leave blank (establishes outer edges)--click Create, then Close

--Display the Thiessen polygonsFile>Open Shapefile> China_Thiessen.shp

If a map window is already open, use: Edit>New map layer> China_Thiessen.shp

Result not good because of outer boundary problem--to close these files and start something new: File>Close All

3. Explore data with different mapsIlliteracy for Provinces of China

11

-- Input the provinces shapefile, with data: File>Open Shape File ChinaData.shp Map window opens showing China provinces --To see the data: Table>Promotion (variables are defined in the file: chinaProvinceData.xls)--To map the data, right click on the map window and select Map > Quantile Select variable to map: 1st variable: Illiteracy (% illiterate)(note: default variable via Edit>Select variable does not work) --Multiple different choropleth maps available: Quantile, percentile, box map, std dev, equal interval, natural break choropleth map: color polygons based on variable value--Draw a second map:

Edit>Duplicate map (to use the same data set)Edit>New map layer (to use a different data set)

• quantile (note the frequency counts in the map legend!)– classes have equal numbers (quantities) of observations

(equal areas under the frequency distribution) – If use 4 categories called quartile (quarter) map– Each has 25% of data

• equal interval (note the frequency counts!)– classes are equal width on variable– will have different numbers of observations

• standard deviation– categories based on – 1,2, etc, SDs above/below mean– Classes have different numbers of observations

Different Choropleth Map Types: Always examine different map types and number of classes!

0-1 1-2 2

34%14% 34% 14%

.68-.68

23%23%25%

0

(Assumes a Normal distribution)

25%

Standard Deviation

Equal area %s

Equal interval %s

Equal interval score

Equal area score• natural breaks – finds “natural groupings” by minimizing the variance within each class using Jenks optimization.

• Percentile– Similar to quantile with 4 cabut

• Box map

We are often interested in outliers: observations with very large or very small (extreme) values

Box map examines extreme data values Possibly no observations in the extreme categories

Map<Box with “hinge” = 1.5 :

• Similar to quantile map with 4 categories

• adds “extreme” categories for data with values which are 1.5 (or 3) times the interquartile range

(difference between 25% and 75% percentiles)

• Extremes here are based on the data value itself.

– Maybe no observations in the extreme categories

– always look at the frequency counts in the legend

Different Choropleth Map Types:Identifying the extremes: Box Map

We are often interested in outliers: observations with very large or very small (extreme) values

Percentile map examines extreme percentages of data Always have observations in the extreme categories

Map>Percentile with “hinge” = 1.5 (or 3):

• Similar to quantile map with 4 categories except

• Uses percentiles to identify extremes: top & bottom 1% & 10%.

– Extremes are the tails of the distribution.

• Extremes here are based on the data value itself.

– Always* have observations in these categories, but they may not be extreme (*in theory, but sometimes not!)

Different Choropleth Map Types:Identifying the extremes: Percentile Map

4. Box Plots and Frequency Distributions• Close all windows

• Explore>Box Plot

repeat for illiteracy, urban pop %, NatGrow05

• Explore>Histogram


The Box Plot:• all observations are positioned based on their value on the variable

• the green asterisk is the median observation

• The blue line is the mean

• the colored center section shows the 25-75% percentile

• the red T line in the upper part shows the location of upper “hinge”

• (value which is 1.5 times the interquartile range above the mean)

• the red in the lower part shows the location of lower “hinge”

(value which is 1.5 times the interquartile range below the mean)

• --sometimes both Ts are at the top & bottom of box (as in crime data), so no observations are beyond the hinge

• --sometimes no Ts show at all—if they are within the interquartile range


15

5. Linking between maps and plots• Edit>Duplicate Map to create map layer

• Right click, and select Map>Percentile


(ignore warnings)

widen the legend box so that you can see frequency count

arrange boxes as illustrated

note that <1% has 0 observations for Urban pop, NatGrow

--the reason for warnings

Linking

• click a province on the map :– it’s highlighted on other maps and plots! – click a data point in a plot, it shows on the map

• If not, maybe it’s too small to see (e.g. Hong Kong): use zoom


16

Warning about Missing Data• Often, value for some observations on some variables are missing

– e.g. for Macau, or the Taiwan islands near the Fujian coast

• Can cause big problems with results of analyses and with plots (such as the box plot)– Software often assumes value is zero

– Big mistake

• Observation should be:– Omitted

– Insert average for the variable

– Use an estimate (provided you have evidence)


17


18


19

6. Moran’s I and Lisa


20

6.1 Create Spatial Weights Matrix• Create File: Go to Tools>weights>create

Input file: chinadata.shp

Queen contiguity

Click Add ID Variable (using existing variable does not work)

Enter new variable name: Poly_ID

Click Save to DBF Click—Yes, its safe

Click Create and name the file: ChinaData.gal A new file ChinaData.gal is saved in the folder with

• Check File: Go to Tools>Weights>Properties– Enter name of weights file– Histogram (frequency distribution) showing number of neighbors– Polygons with zero neighbors are potential problems (4 in this case)– Click on zero column and they are highlighted on map (Linking)– Open table (Table>Promote) and they are highlighted in table


21

Format of .gal File• .gal file is a .txt file:

– open with Notepad


22

First line: 4 items: 0, Number of observations, filename, IDvariableAll subsequent lines are in sets of two:

ID, number of neighbors List of neighbor IDsID, number of neighbors List of neighbor IDs

0 35 ChinaData POLY_ID1 0

2 1303 625 14 11 6 5 44 523 9 6 5 35 630 14 13 9 4 3

Hainan

Macau

6.2 Calculate Moran’s I Calculate Moran’s I: Space>Univariate Moran• Variable: Illiteracy Click OK• Select Weight: ChinaData.gal Click OK• Moran Scatterplot opens

– W_Illiteracy on vertical (Y) axis (neigbors)

– Illiteracy on X axis

• Moran’s I is .2047


23

6.3 Statistical Significance via SimulationCheck Statistical Significance via Simulation:

• Right Click on scatterplot and select Options>Randomization

Select 999 permutations Click Run for additional simulations and to check sensitivity of results

• If p-value < .05 then statistically significant


24

Note numbers at bottom:

– I: 0.2047: Morans I

– E(I) -0.294: Expected value

for Moran’s I if random (no SA)

• same for every simulation

– Mean: of the sampling distribution

– Sd: Standard Deviation of Sampling Distribution (Standard Error)

– Change each simulation

I=.2047

6.4 Calculate Anselin’s LISA (Local Moran’s I)


25

• Calculate LISA: Go to Space>Univariate LISA

• Variable: Illiteracy Click OK

• Weights: chinadata.gal Click OK

• Place checks in top 2 boxes

• We discussed these maps in

our last lecture

6.5 Saving Results of LISA AnalysisSave spatial lag and standardized (z scores) for variable analyzed

• Right-click Moran scatterplot and go to Save Results.

• Check the boxes you want

• Optionally, change the default

variable names


26

Save LISA scores, relationship type*, and probability level

• Right-click significance or cluster map and go to Save Results.

• Check the boxes you want

• Optionally, change the default

variable names*1: high-high, 2: low-low, 3: low-high, 4: high-low

To permanently add the new variables to the table, right-click on the table and go to Save Shape File As....

6.6 Recomputing Moran’s I for selected observations The Moran's I slope and value can be recomputed for all observations excluding the ones

selected:

• Right-click on Moran scatterplot and choose Exclude Selected

Exclude selected observations

• Click an individual observation or drag a box and Moran’s I is recomputed excluding selected observation(s)– New value shown in red on top right– Exclude observations also highlighted on maps

Exclude groups of observation by brushing

• Hold Ctrl key and draw a rectangle; release mouse, then release Ctrl key; rectangle flashes;

• Use mouse to move rectangle across screen

• Moran’s I recalculated excluding observations within rectangle

Note: not a true Moran's I since lag-X not adjusted for excluded observations.


Hints on getting your data into geoda • Data (variables) must be in a shapefile, or in a .dbf which you

join to the shapefile using Table>Join tables• a shapefile also stores data in a .dbf file which you can edit to

add variables

How do I edit a .dbf file to add data?• Use Excel 2003 or earlier

– You can save files from Excel in .dbf format

– Excel 2007 or later will read but not write .dbf files

• Use OpenOffice from Sun/Oracle

www.openoffice.orgAn almost exact replica of Excel which is free


28

Spatial data creation


29

• geoDA also contains some capabilities for creating shapefiles: see Tools>shape

What have we learned today?

How to use geoDA for

• general exploration of spatial data

• analysis of spatial autocorrelation

Next time

• spatial regression

• Then, using geoDA for spatial regression


30

Using GeoDA Software for Geographic Data Analysis and Exploration Developed by Luc Anselin Arizona State University School of Geography and Planning geodatacenter.asu.edu.

Documents

spatial analysis

geoda software

software geoda

geoda overview geoda

spatial lag

spatial statistics tools

geoda program

spatial weights matrices