To cite this abstract / Pentru a cita rezumatul:
Petrisor AI, Drane JW, Jackson K, Dragomirescu L (2001), Detecting Spatial Clusters Using the DAC Statistics. Abstract, 31st Annual Meeting of the South Carolina Chapter of the American Statistical Association, Swearingen Engineering Center, Columbia, SC, 2001

Detecting Spatial Clusters Using the DAC Statistic

Alexandru I. Petrisor, MSPH1; J. Wanzer Drane, PE, PhD1; Kirby L. Jackson, AB1; and Liviu Dragomirescu, PhD2
1 - School of Public Health, University of South Carolina, Columbia, SC, USA
2 - Department of Systems Ecology, University of Bucharest, Bucharest, Romania


Empirical distributions are not strangers to the biostatistician, but binary spatial distributions constructed from random locations indexed by longitude and latitude might be. This study uses the longitude and latitude as the (X1, X2) coordinates of the homes of mothers in Spartanburg County, SC who gave birth to their babies in either 1989 or 1990. Mathematically, the above coordinates have an arbitrary origin as well as arbitrary orientations of their axes. What difference does it make to change either the location of the origin or the orientation of axes? This question is herein addressed. Since empirical and theoretical cumulative distributions are unaffected in shape by translations, this study addresses the effects of those distributions by rotating the axes.
The DAC statistic is the difference between the empirical cumulative distribution of cases and that of population at a particular point (x1, x2). For any size of a random sample of locations taken from those of the 6434 live births there is a noticeable variation of the location of the DAC statistic with random rotations within a given sample, when transformed back to original longitude and latitude. A simulation exercise indicated that the location of the maximum DAC statistic is not unique, moreover there is a geometrical locus of it, and this varies as the orientation of the axes changes. Therefore, the DAC statistic is not the measure of choice when using empirical spatial distributions, but may be helpful in determining spatial clusters. Obviously, more and deeper investigations are the order of the day.