Use this URL to cite or link to this record in EThOS:
Title: Putting 'Geo' into Geodemographics : evaluating the performance of national classification systems within regional contexts
Author: Alexiou, Alexandros
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Geodemographics is an academic field that engages in identifying socio-spatial patterns through the process of organizing areas, typically referred to as neighbourhoods, into categories or clusters that share similarities across multiple socio-economic attributes (Singleton and Longley, 2009). Geodemographics can thus provide a simplified measure of socio-spatial structure through discrete segmentation of geographic space. In nomothetic terms, the basis of the spatial aggregations is based on societal homophily, the tendency of people to associate themselves with similar people. In this sense, people who live close by are bound to have more in common than a random group of people. While geodemographic analysis can be viewed as an established methodology, the simplistic nature of the theoretical framework along with the lack of a single global optimization function produces a lot of uncertainty regarding the success of national geodemographic classifications, i.e. whether they can actually provide good representations of socio-spatial patterns. A review of the relevant literature has shown that little has been done within geodemographic research in the last 30 years as a response to issues of classification uncertainty and system-wide accuracy (Openshaw et al., 1980; Twigg et al., 2000; Voas and Williamson, 2001; Petersen et al., 2011; Reibel and Regelson, 2011). Evaluation constrains are further enhanced by the lack of classification transparency, that would otherwise enable replication and modification which are necessary in order to advance the field (Longley, 2007; Fisher and Tate, 2015). This Thesis focuses on the issue of system-wide accuracy, specifically whether national classification systems can capture spatial variation of socio-spatial patterns at a regional level. Arguably, classification methods are a function of scale; therefore, patterns that are important locally are not necessarily captured in a data-driven national taxonomy. In particular, methodological issues are raised when aggregations into categorical measures sweep away contextual differences between regions, so that final classifications assume that areas within the same cluster have the same underlying characteristics. With this ecological fallacy standard geodemographic classifications fail to incorporate near-geography effectively, and despite the term, geodemographics could be 'aspatial'. As a response to this problem, regional classifications are developed in order to adequately accommodate local or regional structures that diverge from national patterns. Such an example is the London Output Area Classification (LOAC) (Singleton and Longley, 2015). This Thesis tries to elucidate some of the inner workings of Geodemographics by systematically exploring the accuracy of national classification systems for various geographic scales. The main aim of this research is firstly to compare the classification similarity between national and regional patterns, and secondly introduce a methodological extension to conventional geodemographic analysis that accounts for spatial contexts, thus assuming better correlations between places and social identity. In order to provide a comprehensive approach to the issue, the Thesis initially provides a more in-depth review of the theoretical and methodological framework of geodemographic analysis. It demonstrates the evolution of geodemographics, from precursor studies aiming to measure socio-spatial segregation, to a contemporary exploratory and analytical tool with many applications. It also demonstrates the evolution of tools and techniques used in Geodemographics, since G.I.Science and the computational power currently being offered has made a variety of methods available. The Thesis provides a review of such methods, with an emphasis on clustering techniques that are typically used with such socio-economic, quantitative data. It also practically demonstrates the methodological framework of geodemographics through a bespoke classification regarding the build environment and morphology of British neighbourhoods. Based on these methodological frameworks, the analysis explores the issue of accuracy vis-à-vis scale. A number of administrative and functional zones are used in order to delineate various geographic contexts. Comparisons are then carried out between a national classification, which acts as a baseline model, and a series of regional and local classifications at the UK level. The analysis uses arc cosine similarity to evaluate similarity levels between cluster centres and the Rand Index to evaluate a measure of cluster assignment, similar to spatial correspondence (Openshaw et al., 1980). In order to evaluate hundreds of regional contexts simultaneously, an automated process within the R programming language has been developed. Results indicate considerable divergence from national socio-spatial patterns across the UK on a case by case basis. Exploration results showed that, excluding several large conurbations, middle-sized urban areas perform better, while smaller Local Authorities and rural towns score lower. Outcomes suggest several policy implications regarding the applications of geodemographics; areas that national classification seems to perform worse are the same areas that would benefit the most out of a national geodemographic system, considering they are more likely to lack resources and expertise to carry out classifications at their local level. Furthermore, economically lacking and remote areas are prospective targets of national socio-economic policies, and as such, discrepancies are seriously undermining the usefulness of national classification systems, as spatial identification might actually be misleading in regions where it is needed the most. A second step of the analysis is the methodological extension to the traditional geodemographic methodology that accounts for spatial context within the clustering process. The methodological framework is based on Webber's (1980) response to national classification critics, suggesting that national classifications do not work locally because they operate on different attribute means and standard deviations. Based on this observation, geographic dependencies are built within attribute values by means of regional standardisation, enabling classifications to be more sensitive to local variation of attributes. In particular, the model introduces a geographic factor 'g' that adjusts the level of impact of contextual geography to attribute values, for various levels of regional geography - Regions, Travel-to-Work Areas and Local Authority Districts. Model results for various level of 'g' show that the intensity and nature of cluster transitions between neighbourhoods is highly cluster-dependant, while also suggesting that Regional classifications seem to outperform other contexts in terms of neighbourhood representation and cluster cohesion. This research is not developed as a critique to Geodemographics, but rather tries to systematically evaluate certain aspects of classification methodology. Although results are of tentative nature, a model where attribute values are conjoined spatially can help mitigate scale effects. The limitations of the approach are mainly the selection of the extents of near-geography, i.e. the contextual geography used to standardise values, and the value of the 'g' factor, which are both biased parameters and as such should reflect the theoretical rationale and purpose of the classification creator.
Supervisor: Singleton, Alexander ; Williamson, Paul Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral