From jacquez@biomedware.com Mon Oct 22 18:42:54 2001 Received: from mailscan6.cac.washington.edu (mailscan6.cac.washington.edu [140.142.33.14]) by lists.u.washington.edu (8.11.6+UW01.08/8.11.6+UW01.08) with SMTP id f9N1gpN135786 for ; Mon, 22 Oct 2001 18:42:51 -0700 Received: FROM mxu1.u.washington.edu BY mailscan6.cac.washington.edu ; Mon Oct 22 18:41:22 2001 -0700 Received: from websrv1.biomedware.com (websrv1.biomedware.com [198.108.20.30]) by mxu1.u.washington.edu (8.11.6+UW01.08/8.11.6+UW01.08) with ESMTP id f9N1fLw04705 for ; Mon, 22 Oct 2001 18:41:22 -0700 Received: from CALYPSO.biomedware.com (pm552-16.dialip.mich.NET [198.110.22.170]) by websrv1.biomedware.com with SMTP (Microsoft Exchange Internet Mail Service Version 5.5.2448.0) id 4B48CY69; Mon, 22 Oct 2001 21:32:17 +0100 Message-Id: <5.0.0.25.0.20011022104153.00a73e30@websrv1.biomedware.com> X-Sender: Jacquez@websrv1.biomedware.com (Unverified) X-Mailer: QUALCOMM Windows Eudora Version 5.0 Date: Mon, 22 Oct 2001 10:52:00 -0400 To: waphgis@u.washington.edu From: Geoffrey Jacquez Subject: Re: Spatial-temporal clustering by chance In-Reply-To: Mime-Version: 1.0 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable A few years ago I proposed a k-NN test for space-time clustering (Jacquez 1996) that has several advantages over both the Mantel and Knox test, and that I believe would serve your needs.  Mantel's (1967) test is a regression of the waiting time between case-pairs on the spatial distances separating those case-pairs. The Knox test is the count of the number of case-pairs that are "near" in both space and time, where "near" is defined by the investigator.  The k-NN test is the count of the number of pairs of cases that are nearest neighbors in both space and time.  The test can be evaluated for any value of k, such as first nearest neighbors, second nearest neighbors and so on, and the significance at each level of k is evaluated.  So if you believe clustering at a certain level of k is (such as 2 or 3 cases) is occurring by chance, you can test this hypothesis directly by inspecting the probability value associated with k=3D2 and k=3D3.  This test directly quantifies the scale of space-time clustering in terms of nearest neighbors.

The test is available in the Stat! package, and will soon be added to the ClusterSeer software.

1996. Jacquez, G. M. =93A k-nearest neighbor test for space-time interaction.=94 Statistics
in Medicine, 15:1934-1949.

-G.

At 12:30 AM 10/22/2001 +0800, you wrote:
Hi everyone,

In Singapore, we define a dengue cluster (referring to an outbreak when
dengue fever is spreading contagiously from one person to the next) as at
least 2 cases (points) within 200 m of each other and within 3 weeks of
incidence of each other. With about 2,000 cases so far this year, I have
about about 80 clusters according to this definition.

However, I suspect that the occurence of some clusters (especially the ones
with just 2 or 3 cases) may be by chance, instead of an actual disease
transmission happening. How can I test this? Possibly, a Monte=20 Carlo
simulation of some kind will help. Is there any software out there that can
help to do this? SATScan has got some promising functions, unfortunately it
deals with areal or polygon data and not point data (I think).

Will sum answers.

Thanks for your attention. Cheers.

Basil
.