Spatial Regression Models for the Social Sciences. Jun Zhu

Science Research Institute and Population Research Institute. Dr. Chi is an environmental demographer. His research examines the interactions between population change and the built and natural environments. He pursues his research program within interwoven research projects on climate change, land use, and community resilience, with an emphasis on environmental migration and critical infrastructure/transportation and population change within the smart cities framework. Most recently, Dr. Chi has applied his expertise in big data to study issues of generalizability and reproducibility of Twitter data for population and social science research. He also studies environmental migration, including projects on coupled migrant-pasture systems in Central Asia, permafrost erosion and coastal communities in the Arctic, and ecological migration in China. Dr. Chi’s research has been supported through grants from national and state agencies, including the National Science Foundation, National Institutes of Health, National Aeronautics and Space Administration, and U.S. Department of Transportation. He has published more than 50 articles in peer-reviewed journals. His research on gasoline prices and traffic safety has been highlighted more than 2,000 times by various news media outlets, such as National Public Radio and Huffington Post.Jun Zhu is Professor of Statistics at the University of Wisconsin–Madison. She is a faculty member in the Department of Statistics and the Department of Entomology, as well as a faculty affiliate with the Center for Demography and Ecology and the Department of Biostatistics and Medical Informatics. The main components of her research activities are statistical methodological research and scientific collaborative research. Her statistical methodological research concerns developing statistical methodology for analyzing spatially referenced data (spatial statistics) and spatial data repeatedly sampled over time (spatio-temporal statistics) that arise often in the biological, physical, and social sciences. Her collaborative research concerns applying modern statistical methods, especially spatial and spatio-temporal statistics, to studies of agricultural, biological, ecological, environmental, and social systems conducted by research scientists. Dr. Zhu’s methodological and collaborative research projects have been supported by the Environmental Protection Agency, National Institutes of Health, National Science Foundation, U.S. Department of Agriculture, U.S. Department of Defense, and U.S. Geographical Society. She is a Fellow of the American Statistical Association and a recipient of the Distinguished Achievement Medal in its Section of Statistics and the Environment.

      1 Introduction

      Learning Objectives

       Understand the current status of spatial social science research.

       Understand basic concepts and terminologies related to spatial effects.

       Familiarize yourself with the primary data example of population change that is used throughout this book.

      This is a book for social scientists who want to learn spatial regression models with relative ease. But first of all, why should social scientists care about studying spatial regression models? Many statistical methods could be useful to social science research; learning any of them could take one considerable effort and also sometimes involve a steep learning curve.

      We believe that quantitative social scientists, especially those who deal with aggregated quantitative data, can benefit from learning and using spatial regression models from at least three perspectives—theoretical, methodological, and practical. First, theoretically, many phenomena of the social sciences exhibit spatial effects. This has been explained both explicitly and implicitly by many theories and examined in many empirical studies. Second, methodologically, the standard linear regression analysis may not be reliable when the independence assumption of the model is not adequately met; if spatial effects are not properly accounted for, estimation and statistical inference may be unreliable (e.g., the effects of explanatory variables may be overstated or understated). Third, practically, the past twenty years have experienced a rapid increase in the use of spatial regression models for social science research—first human geography and regional science and then other social science disciplines, including anthropology, criminology, demography, economics, political science, urban studies and urban planning, and sociology. The bar for quantitative social science research has been raised, and conducting spatial regression analysis can be methodologically more rigorous in many research areas.

      Why has there been a rapid increase in the use of spatial regression models for social science research? Because in addition to the recognition of their usefulness in social science research, spatial regression models have also become more accessible for social scientists to explore (Chi & Zhu, 2008). Multiple factors are associated with this increased accessibility, such as the following:

       the upsurge in the availability of geographically referenced data,

       the development of user-friendly spatial data analysis software packages, and

       increased computing power combined with affordable computers.

      Geographically referenced data from geographic information system (GIS) sources and remote sensing images are often useful for social scientists and can be easily added to geo-referenced social science databases. Three resources are especially useful for researchers conducting spatial demographic analysis: the topologically integrated geographic encoding and reference (TIGER) system products; census summary files of 1980, 1990, 2000, and 2010, as well as the American Community Survey data; and sociological and demographic survey databases companioned with the geocoding technique. Moreover, development of spatial statistical software packages has been rapid in the past two decades, via products such as ArcGIS, GeoDa, GWR, MATLAB, Python, SAS, SpaceStat, S-plus, SPSS, Stata, and WinBugs. Improved computing power together with inexpensive personal computers has made conducting spatial regression analysis more affordable for social scientists. Furthermore, greater opportunities for studying spatial regression have arisen due to the larger number of textbooks, journal articles, and conference presentations advancing or using spatial data analytical tools (Entwisle, 2007; Florax & Van der Vlist, 2003; Logan, 2012). And this number is only growing.

      Considering the opportunities for and development in spatial regression methods, we assert that it is an opportune time for social scientists to make better use of spatial regression methods and apply them to social science research. We believe that spatial regression models and methods can be learned with relative ease, and this book is intended to help readers do just that.

      1.1 Spatial Thinking in the Social Sciences

      Many phenomena of the social sciences exhibit spatial effects; this has been addressed both explicitly and implicitly in many theories and empirical studies. It would be useful to review the spatial thinking and theories as well as the empirical studies in the social sciences to build a foundation and make a case for spatial regression models in social science research. However, doing so is challenging considering that there is a large amount of spatial theories and empirical studies in the social sciences, and certain theories and studies are limited to specific social science disciplines. With both the benefit and the challenge in mind, and considering that this book is about spatial regression models rather than spatial theories, in this section, we provide a brief overview of the current status of spatial theories and empirical studies across social science disciplines.

Most social science disciplines and their subdisciplines have addressed the spatial dimension of their phenomena of study, although the extent to which they have done so varies greatly (Goodchild & Janelle, 2004). The Center for Spatially Integrated Social Science (CSISS) Classics provides a list of classic readings that contribute to spatial thinking in the social sciences.¹ Overall, spatial thinking in social science disciplines and subdisciplines can be addressed from multiple perspectives.

¹ www.csiss.org/classics

      First,