Scale, factor analyses, and neighborhood effects

Abstract

Studies of potential neighborhood effects have been constrained in most situations by the absence of small-area data generated to characterize the local contexts within which individuals operate. Using small-area data from the U.K. Census, this paper illustrates the creation of bespoke neighborhoods--local areas defined separately for each individual in a sample survey--at a variety of scales, and their characterization using factor analysis techniques. Theories of neighborhood effects are uncertain as to the spatial scale at which the relevant processes operate, hence the value of exploring patterns consistent with those processes at a range of spatial scales. One problem with such comparative study is the incommensurability of regression coefficients derived from analyses using factor scores as the independent variables. The work reported here adapts a procedure introduced for reconstituting partial regression coefficients to circumvent that problem, and illustrates that patterns of voting at a recent British general election showed neighborhood-effect-like patterns at two separate scales simultaneously--with individual voter characteristics held constant.