• Lab: spatial statistics

  In Geoff's presentation we see a selection of techniques applied to some example data. We'll be trying to replicate his success today, as we use the same software he did - Stat! and GeoMed - to answer question concerning the example data sets that come with the software.

  The objectives of this lab are as follows: you should

  1. become familiar with Stat! and GeoMed;
  2. learn how to use Stat's and GeoMed's Statistical Advisor;
  3. analyze a spatial cluster using Stat!;
  4. analyze a spatial/temporal cluster using GeoMed;
  5. consider how to use these programs in the context of your projects; and
  6. give feedback about the software.

  Lab

  We'd like you to work with a partner, so that you have someone to compare notes with. You'll be asked to work your way through the software, answering some questions by following the suggestions and clues we give below. If you have trouble following the directions, please make a note of the problem in your evaluation which you give at the end of this exercise. We appreciate and value your feedback.

  We'll be glad to answer questions, but ask that you attempt to answer them first using whatever help facilities the software offer: this will help us gauge how well the software works. Good luck!

  Stat!

  You may need to download trial versions of Stat! (ask!) and GeoMed for this lab.

  UMich students: unzip the trial version of GeoMed to c:\temp, and Stat! to c:\temp\stat. The executables are called GeoMed.exe, and stat.exe

  Exercise 1: Using Stat!'s advisor (Fkey F6) select appropriate methods for:

  1. Detecting global clustering in spatial data using case and control locations.
  2. Detecting global spatial clustering in disease rates when denominators are unknown.
  3. Detecting global spatial clustering in disease rates when denominators are known.
  4. Testing Tobler's `First law of geography' in times of diagnosis when place of residence (x,y) and time of diagnosis (t) are known, and you wish to identify the scale at which space-time interaction occurs (ignoring any latency period).
  5. Answering question (d) using rates through time.
  Notes:
  • Double left clicking on a highlighted phrase switches to that screen.
  • Clicking on the upper left hand corner of the Advisor box restores the prior window: you don't have to keep closing and invoking the Advisor.
  • When you are done with the advisor, use the Escape key to exit.

  Exercise 2: In the 1980's Lyme disease was an emerging public health problem. The disease is caused by a Spirochete and is vector borne. Its residence in the wild and eventual transmission to humans involves deer mice, white tailed deer (animal reservoirs) and the deer tick (vector). The spatial distribution of the disease is inverse-density dependent: it occurs primarily in rural areas with small human population sizes. The following files are provided with the software (the links simply allow you to examine the files):

  • Lyme.dat - specifies the rates in each county
  • Lyme.con - specifies spatial weights such that adjacent counties are `connected'. County 1 (first row in Lyme.con, Lyme.dat, and LymeIpop.dat) has 4 adjacencies: it is adjacent to counties 7, 8, 10 and 13.
  • LymeIpop.dat - specifies population and counts per county
  Notes:
  • Pressing the F1 key while in the input data dialog describes the format of the files.
  • Help on a particular test is also available: highlight the test, then press the F1 key.
  • The Stat! manual will be available in lab, in case you need to consult it.
  1. Part 1: Using ONLY Lyme.dat and Lyme.con, determine whether the Lyme disease rates in Georgia follow Tobler's first law of geography.

  2. Part 2: Using Lyme.con and LymeIpop.dat, address the question in Part 1, but accounting for differences in population sizes across the counties. Compare the results from Parts 1 and 2.

  Exercise 3: Use GeoMed to perform analyses unavailable in Stat!.

  Exercise 4: Don't forget to evaluate this lab. Geoff will be especially interested in your comments concerning Stat!, as he produces and maintains this software. Your comments can help him improve Stat! We will also consider your comments when it comes time to select software for the course next time. Leah will want your feedback on GeoMed.
  

• Software: Stat!, GeoMed

• Featured techniques: Kulldorff's Spatial Scan Statistics, Cuzick and Edward's test, Moran's I, Moran's I adjusted for population size (Ipop), k-NN test, and Grimson's test.

• Readings:
  Principal Readings:
  1. Jacquez, GM, R Grimson, LA Waller, DE Wartenberg. 1996. The analysis of disease clusters Part II: Introduction to Techniques. Infection Control and Hospital Epidemiology 17:385-397.
  2. Kulldorff, M. 1998. Statistical methods for spatial epidemiology: Tests for randomness. In GIS and Health (Eds T. Gatrell and M. Loytonen). Taylor and Francis.

• Data: We'll use the data provided with the Stat! and SaTScan software, although we encourage you to use your own data to test these routines as well.

• Links:
  • SaTScan, Martin Kulldorf's "Software for Calculating the Spatial, Temporal and Space-Time Scan Statistics". If you're interested in trying it out, try this "mini-lab".

Page by Andy Long. Comments appreciated.