Voss, Paul, Katherine Curtis, and Roger B. Hammer
Working paper no. 2004-04
Abstract
This paper consists of two parts. The first reviews the historical role that space and place has played in the discipline of demography in the United States. We argue that until approximately the middle of the 20th century nearly all of demographic analysis could be labeled “spatial demography.” Beginning in the 1940s this pattern changed, as an increasing number of microdata files from large sample surveys began to provide attitudinal and behavioral data for individuals and families. The trend was further accelerated by release in the early 1970s of census data in the form of public use microdata sample files (PUMS). We argue that in addition to data availability, the drift away from analysis of aggregated census data was prompted by a conscious desire on the part of researchers to avoid the troublesome issues of aggregation bias and what came to be known in sociology as the ecological fallacy. Sometime around the 1950s to 1960s most population analysis in the U.S. shifted away from macro- (spatial) to micro-level research, although we acknowledge and document that in two small subfields of demography (rural demography and applied demography) fascination with aggregate (spatial) data analysis persisted.
In the second part of the paper, we argue that there has emerged in very recent years a renewal of interest in aggregate demographic data. Part of this re-emergence of interest in spatial demography is driven by awareness of developments in the fields of spatial econometrics and regional science that bring fresh approaches to the specification of traditional regression models and new software tools for estimating parameters in the presence of spatial externalities. By way of illustration, we provide a brief data analysis as various aspects of these new developments are discussed. Finally, we close the paper with an overview of how the earlier split between macro and micro approaches to data analysis are now being bridged with multilevel models that simultaneously consider both individual-level variables and aggregated contextual level variables for those areas where the individual lives or works.