Qualitative and quantitative assessment of geographic clustering of population samples selected using different methods of random digit dialing

Abstract

Random digit dialing is a method commonly used to select random population samples in epidemiologic research. Although random digit dialing is generally presumed to provide representative samples, coverage and nonresponse errors may affect the degree to which the sample is representative. The present investigation was undertaken to determine whether the geographic distributions of samples selected using variations of the basic random digit dialing technique accurately reflect the underlying population distribution, and if not, whether such samples tend to be either more or less dispersed than the populations from which they were selected. Data regarding control groups from three case-control studies conducted from 1983-1986 were utilized. The residence addresses of 998 controls were located and assigned an X-Y coordinate and census tract designation within a three-county geographic area in northwest Washington State. Initially, the spatial distributions of controls were examined graphically in relation to the age-sex structure of the underlying population. No differences in geographic pattern were apparent. A more formal statistical evaluation was conducted based on centrographic techniques utilizing two measures to describe the spatial distribution of a set of points. Results indicate that the samples chosen were neither more nor less dispersed than the underlying populations. However, the geographic centers of samples selected using primary numbers tended to be shifted from the centers of their respective populations. Several possible explanations for such shifts are considered, and extensions of the analytical approach are suggested in relation to the further evaluation of population sampling and the investigation of space-time aggregations of disease.