S Q Muth1, J J Potterat, R B Rothenberg. 1. El Paso County Department of Health and Environment, Colorado Springs, CO 80910-3123, USA. smuth@uswest.net
Abstract
BACKGROUND: Comparability of study participants with non-participants is customarily assessed by contrasting the distributions of sociodemographic characteristics. Such comparisons do not necessarily provide insight into whether or not participants of a given subgroup are similar to non-participants of the same subgroup. A geographical information system (GIS) may provide such insight by visually displaying the spatial distributions of participants and non-participants. In a previously reported study of heterosexuals at elevated risk for human immunodeficiency virus (HIV), traditional methods suggested distributional differences in the demographic characteristics of participants and non-participants. METHODS: Based on residential address co-ordinates for each subgroup member, we used the subgroup's centroid as the origin and constructed a 360 degrees series of overlapping box plots of the distance of subgroups members to the origin, thereby producing closed polygons for each of the box plot demarcators. RESULTS: These rotational box plots revealed similar geographical distributions for most participant and non-participant subgroups, with the exception of African-American men and women. CONCLUSIONS: Observed differences resulted in part from the study design, and provided some insight into sampling problems encountered in social network studies. Based on Tobler's supposition that 'nearby things tend to be alike', the rotational box plot is a useful additional tool for investigating sample bias.
BACKGROUND: Comparability of study participants with non-participants is customarily assessed by contrasting the distributions of sociodemographic characteristics. Such comparisons do not necessarily provide insight into whether or not participants of a given subgroup are similar to non-participants of the same subgroup. A geographical information system (GIS) may provide such insight by visually displaying the spatial distributions of participants and non-participants. In a previously reported study of heterosexuals at elevated risk for human immunodeficiency virus (HIV), traditional methods suggested distributional differences in the demographic characteristics of participants and non-participants. METHODS: Based on residential address co-ordinates for each subgroup member, we used the subgroup's centroid as the origin and constructed a 360 degrees series of overlapping box plots of the distance of subgroups members to the origin, thereby producing closed polygons for each of the box plot demarcators. RESULTS: These rotational box plots revealed similar geographical distributions for most participant and non-participant subgroups, with the exception of African-American men and women. CONCLUSIONS: Observed differences resulted in part from the study design, and provided some insight into sampling problems encountered in social network studies. Based on Tobler's supposition that 'nearby things tend to be alike', the rotational box plot is a useful additional tool for investigating sample bias.