Methods and Challenges of Analyzing Spatial Data for Social Work Problems: The Case of Examining Child Maltreatment Geographically

Abstract

Increasingly, social work researchers are interested in examining how place and location contribute to social problems. Yet, often these researchers do not use the specialized spatial statistical techniques developed to handle the analytic issues faced when conducting ecological analyses. This article explains the importance of these techniques when analyzing spatial data, describes appropriate spatial statistical techniques, provides an illustration of how using inappropriate statistical techniques with spatial data can produce biased estimates of statistical tests, and discusses challenges to conducting spatial analysis. The study involved analyzing data for 941 census tracts for structural factors related to child maltreatment using traditional ordinary least squares (OLS) regression and generalized linear squares (GLS) spatial regression. When using OLS, results showed that immigrant populations/child care burden was negatively related to rates of child maltreatment but not related when using GLS, which controls for correlations between these spatial units. Relying on OLS regression techniques to create interventions to reduce child maltreatment in spatial areas could result in developing ineffective strategies that fail to reduce maltreatment. KEY WORDS: child maltreatment; GIS; neighborhoods; spatial analysis ********** Advances in geographic information systems (GIS) technology are leading social work researchers and practitioners to think about social welfare problems more spatially. For example, Albert and Catlin (2002) studied whether states adjusted their welfare benefits to be more like adjacent states to discourage welfare recipients from crossing state borders to receive more benefits. Essentially, their question was one of space: Are welfare benefits in states located next to each other more alike than states farther away? Social work researchers increasingly pose similar types of questions, yet the methods used in social work research to examine this greater awareness of space have not kept pace with the questions asked. One area where the biggest awareness of space in social work practice and research can be seen is through the increased use of GIS-generated maps, which show the distributions of problems across communities or neighborhoods (Noble & Smith, 1994; Queralt & Witte, 1998; Robertson & Wier, 1998). Although the use of these maps has proliferated, analyses of spatial data from these maps continue to be conducted using traditional analytic techniques, such as regression, analysis of variance, or tabulated frequencies of events across spatial areas including neighborhoods and communities. Such techniques do not account for several of the unique problems associated with analyses of spatial data. At the very least, the neglect of these techniques when studying neighborhoods or other areas may result in specification error and inconsistent findings across studies. At the very worst, interventions to reduce social problems may be recommended to policy-makers based on biased statistical tests. The purpose of this article is to explain the importance of using specialized statistical techniques for analyzing spatial data, describe these techniques, provide an example that illustrates how using inappropriate statistical techniques of spatial data can produce biased estimates of the statistical tests, and discuss challenges to conducting spatial analyses. WHY SPATIAL DATA ANALYSIS? Spatial data analysis refers to the examination of some process in space and its relationship to other spatial phenomena (Bailey & Gatrell, 1995) and can be used to describe relationships between points, lines, or areas. This article focuses on those techniques used to describe areal data--data collected from geographic areas such as cities, zip codes, and census tracts. In the past 10 years, there has been a proliferation of studies of neighborhood areas examining many different social problems (see Burton & Jarrett, 2000; Leventhal & Brooks-Gunn, 2000; Sampson, Morenoff, & Gannon-Rowley, 2002 for a review). …