Abstract
This paper provides an empirical analysis of the multidimensional, spatio-temporal quality of life (QoL) trends followed by neighbourhoods in Charlotte, NC, between 2000 and 2010. Employing a combined geocomputational and visual technique based on the self-organising map, the study addresses which types of neighbourhood experienced the most change or stability, where (in attribute and geographical spaces) did neighbourhoods that began the decade with a particular set of characteristics evolve to, and where did neighbourhoods that concluded the decade transition from? Results indicate that the highest QoL neighbourhoods were most stable, while those with lower homeownership, closer to the city centre, exhibited the sharpest longitudinal trajectories. Lower-income neighbourhoods are found to be heterogeneous in terms of their social problems, dividing between high crime concentrations and youth-related social problems. An exchange of these social issues over time is observed as well as a geographical spread of crime to middle-ring suburbs.
1. Introduction
Understanding the ways in which urban neighbourhoods evolve over time is of great importance to city planners, policy-makers and community leaders who share a common interest in devising policies and action plans aimed at community development. Healthy neighbourhood initiatives are often multifaceted and pursue a two-prong strategy of maintaining or improving the vibrancy of living experiences of neighbourhood residents and establishing placed-based communities as economically competitive environments for businesses. Quality of life (QoL) in either of these contexts refers to the multidimensional notion of livability, desirability or competitiveness of a locale. From a community planning or policy-making perspective, quality of life indicators present alluring metrics for monitoring neighbourhood conditions over time in general and for developing targeted action plans in particular (Galster et al., 2005; Myers, 1998; Sawicki and Flynn, 1996). The concept of QoL is not new to the debate on urban geographies as it offers a framework for understanding processes of placed-based social change across urban landscapes (Helburn, 1982; Pacione, 2003).
A significant body of literature covers various aspects of the integrative and multifaceted social construct of neighbourhood quality of life. Much of this research pertains to measurement issues, while few studies have utilised the multidimensional, longitudinal datasets resulting from these QoL efforts to investigate neighbourhood trends. Exploring these data may provide confirmation of traditionally held conceptions about how neighbourhoods transition over time, while more detailed inspections of the trajectories followed by neighbourhoods may illuminate unanticipated paths of change across the array of attributes. Common methods of exploratory spatial data analysis (ESDA) are ill-equipped to extract emergent trends or patterns from complex, spatial-temporal and multidimensional data such as the aforementioned QoL indicators. This paper draws from recent advances in geovisual analytics which seek to blend computational or data mining techniques with highly visual outputs to explore such datasets. The outputs produced from this analytical technique portray non-linear relationships between QoL indicators, providing general statements on the overall spatial-temporal dynamics of neighbourhoods, while the visualisation of longitudinal change trajectories enables one to identify meaningful trends in the results. This methodology holds particular promise to urban leaders wishing to gain new insights into the dynamics of neighbourhood QoL and to communicate trajectories of change to the public.
Specifically, this paper utilises neighbourhoods within the city of Charlotte, North Carolina, as a case study to explore trends across a series of 17 biennially collected QoL indicators from 2000 to 2010. A combined computational and visual methodological approach based on the self-organising map is adopted to address four major questions: what is the typical QoL profile of neighbourhoods that experienced the most change or stability; how did neighbourhoods that began the decade with a given QoL profile change 10 years later; conversely, where in attribute space did neighbourhoods that concluded the decade transition from; and, did geographically adjacent neighbourhoods undergo similar transformations in terms of QoL? The visual analytical approach further allows for the identification of neighbourhood trajectories that have clearly differed from other neighbourhoods with similar characteristics, or outlier trajectories.
The remainder of the paper is structured as follows. In section 2, the theoretical and empirical background on neighbourhood change is presented, followed by an overview of the study area, data and methodology in section 3. Results are presented in section 4 and the discussion and conclusions insection 5.
2. Neighbourhood Change
Traditional models of neighbourhood change largely focus on shifts in a neighbourhood’s socioeconomic or racial composition, through a process of ‘invasion and succession’ as proposed by sociologists in the Chicago School in the 1920s (Burgess, 1925), as a function of housing age and deterioration according to filtering theories of decline (Grigsby et al., 1987; Hoyt, 1933; Leven et al., 1976), or as a product of residential location choices driven by a trade-off between commuting costs and land values as suggested by urban economic bid rent models (Muth, 1969). In the United States, the urban landscapes generated according to each of these models feature the highest concentrations of the poorest residents and minorities living close to the city centre in deteriorating homes, while the wealthiest reside in newer, suburban structures on the outskirts of urban areas. Recent modifications to these theories have proposed that middle-aged homes are most susceptible to decline as older buildings are prime for revitalisation (Brueckner and Rosenthal, 2009; Rosenthal, 2008). Indeed, empirical studies on neighbourhood income and poverty dynamics have recorded a general concentration of poverty in neighbourhoods situated in the urban core of cities throughout the country, as well as a spatial spread to older, inner-ring suburbs, especially in rapidly suburbanising metropolitan areas where housing supply continues to expand (Cooke and Marchant, 2006; Lee, 2011; Madden, 2003).
Recent work documenting the decline of older, first-ring suburban neighbourhoods addresses a debate on the evolution of suburban neighbourhoods initiated in the 1960s. Proponents of the idea of suburban persistence have argued that, even as people in suburban neighbourhoods change, their socioeconomic characteristics persist as these neighbourhoods consistently attract the same types of people. Conversely, others have held a more ecological view according to which these neighbourhoods do change over time, and the trend followed is one of gradual decline (Vicino, 2008). A number of empirical studies have since illustrated a changing diversity and decline of suburban neighbourhoods (Hanlon et al., 2006; Lee and Leigh, 2007; Lucy and Phillips, 2000; Vicino, 2008).
While neighbourhood economic conditions are important predictors or descriptors of a neighbourhood’s overall quality of life, they are only one aspect of this multidimensional concept, and focusing solely on this dimension ignores the greater complexity of social processes in which neighbourhoods evolve (Chow and Coulton, 1998). This is especially true in light of a number of studies which have suggested that neighbourhoods with high levels of poverty are heterogeneous in nature—so, while some may experience severe social problems, others may not fare as poorly across all dimensions (Longley and Tobón, 2004; Morenoff and Tienda, 1997). The link between concentrated poverty and social problems—including teenage pregnancy, high school dropout or education levels, as well as crime rates, is primarily attributed to Wilson (1987) and Massey and Denton (1993) who both argue, through different demographic processes, that concentrated poverty exacerbates social problems as youths growing up in these neighbourhoods are not exposed to role models, are isolated from employment both physically and via social job networks, and are detached from social norms and behaviours. According to these views, social problems are hypothesised to increase in concentration in the poorest neighbourhoods through time. Chow and Coulton (1998) address this hypothesis in a study of three categories of neighbourhood distress between 1980 and 1990 in Cleveland. Using a factor analysis, the authors reveal a greater interdependence among these distress factors over time and conclude that social conditions did in fact worsen during the decade. No mention is made of the spatial dynamics of change, however.
Morenoff and Tienda (1997) explore the spatial and temporal evolution of social change in Chicago neighbourhoods between 1970, 1980 and 1990 by applying a cluster analysis on socioeconomic variables to develop four neighbourhood typologies (stable middle class, gentrifying yuppie, transitional working class and ghetto underclass). Transitional working-class neighbourhoods with lower socioeconomic characteristics, educational attainment and homeownership were identified as the most likely to change, while neighbourhoods that became transitional working-class neighbourhoods by 1990 were associated with a rapid Hispanic population increase. Underclass and gentrifying/yuppie neighbourhoods exhibited the most stability across the time-period. Spatially, an increasing concentration of affluence and a spread of ghetto underclass neighbourhoods were also recorded.
Other studies on neighbourhood quality of life change have aggregated multiple indicators into a composite score, reported on the number of neighbourhoods that transitioned between groups and have created maps of neighbourhoods that have improved, declined, or remained the same (Kitchen and Williams, 2009; Randall and Morton, 2003). These studies, however, provide little insight into the ways in which neighbourhoods change across the multidimensional attribute space. The purpose of this paper is to provide a more comprehensive view of multidimensional neighbourhood change by incorporating a wider range of QoL variables capturing the social, economic, physical and crime dimensions, while simultaneously enabling the spatial dynamics of these changes to be examined.
3. Study Area, Data and Methodology
3.1 Study Area
Charlotte, North Carolina (Figure 1) is the largest city in the state and one that has experienced rapid population growth and urban expansion in recent years, particularly over the course of the previous decade. According to US census estimates, between 2000 and 2010, the city of Charlotte’s population rose from 540 828 to 731 424 residents, an increase of 35.2 per cent. Mirroring other ‘New South’ cities, Charlotte’s neighbourhood geography has been dynamic and transformational during this period. In particular, four significant land use, demographic and public policy shifts have shaped neighbourhood development. First, expansive suburbanisation has pushed new greenfield urbanisation to the edges of the county. Broadly speaking, these new neighbourhoods are upper-middle to upper income and largely White. A second characteristic is strong gentrification trends in the city centre and selected pre-World-War-II suburbs surrounding the urban core. Affluent, White residents dominate the gentrifiers’ profile. In the city centre, older public housing projects have been replaced by high-density condominium projects, while poor and minority renters have relocated to lower-priced neighbourhoods west and north of downtown. Thirdly, Charlotte has become a Hispanic hypergrowth metropolitan area (Suro and Singer, 2002), absorbing a 128.2 per cent increase in Latino residents since 2000. Middle-ring suburban neighbourhoods (post-World-War-II) are the overwhelming destination for new Latino settlers, where they have succeeded the original White and subsequent African American middle-class dominance (Smith and Furuseth, 2008). Finally, overlaying Charlotte’s robust physical growth and demographic restructuring was a unique shift from the traditional busing of public school students since the 1970s to a neighbourhood school assignment plan implemented in 2002 that affected the desirability of housing markets midway through the research period (Smith, 2010). Recognising the role that perceived school quality performance plays in neighbourhood desirability, the neighbourhood-based school assignment process has undoubtedly affected neighbourhood attractiveness. Yet how much effect this policy shift had in the second half of the study period in realigning the trajectory of gentrifying inner-city neighbourhoods is unclear and certainly warrants further investigation.
The study area, Charlotte, North Carolina.
3.2 Data
Data for this study come from the Charlotte Neighborhood Quality of Life Study Group (Metropolitan Studies, 2010), a biennial compilation of primary and secondary indicators representing the economic, social, physical and crime dimensions for 173 neighbourhood statistical areas (NSA), units of analysis similar to US census block groups, but customised for the Charlotte region based on community feedback. Beginning in 2000, and continuing every other year, 17 attributes were collected for each neighbourhood describing its social, physical, crime and economic conditions. Most of the variables were compiled from exhaustive data collected over time, in contrast to the biennial American Community Survey from the US census which relies upon sampling. Table 1 describes each of the QoL attributes collected between 2000 and 2010, resulting in a 6-year panel of data.
QoL indicators
3.2 Analytical Methodology
In order to investigate trends in this multidimensional, multitemporal and spatial dataset, this research draws on recent work in the field of geovisual analytics, which seeks to combine computational and visual methodologies to facilitate exploratory space–time analysis (for example, Andrienko et al., 2010a; Andrienko et al., 2010b; Guo et al., 2006; Yan and Thill, 2009). Geocomputational methods typically impose fewer distributional assumptions regarding data structures and relationships as compared with traditional statistics, while visualisation techniques exploit the ability of human vision and intelligence in recognising patterns, relationships, trends and anomalies (Yan and Thill, 2009). As an intermediary between purely computational and visual analytical methods, the self-organising map (SOM) offers advantages of both. It is a neural network-based computational method for reducing the dimensionality of large datasets and revealing embedded structures; it generates an inherently visual output for exploring results (Andrienko et al., 2010a; Guo et al., 2005). The SOM further enables the identification of non-linear relationships among QoL variables, thus giving it an advantage over traditionally utilised dimensionality reduction methods such as principal components or factor analysis.
Recently, in the context of the analysis of intraurban neighbourhoods, SOM has been used to examine geographical positions of neighbourhoods with similar demographic attributes (Spielman and Thill, 2008), to define or examine changes within housing sub-markets (Kauko, 2004, 2009a, 2009b) or of indicators related to neighbourhood deprivation (Pisati et al., 2010). A host of applications of self-organising maps within the geographical discipline can be found in the edited volume by Agarwal and Skupin (2008). Skupin and Hagelman (2005) propose a methodology based on the self-organising map to visualise attribute change of census tracts by creating trajectories across the SOM attribute space, thus eliminating the need for multiple maps depicting the pattern of each attribute at each period of time, which becomes increasingly ineffective as the number of time stamps increases.
Although geocomputational methods of data reduction such as SOM offer a number of advantages over traditional statistical analyses, they do suffer from some critiques. One major objection to the SOM algorithm is in being a-theoretical, letting the data drive results, rather than testing a stated theory as is the case in confirmatory statistical analyses. However, as with other exploratory data and spatial data analyses, the purpose is to generate, rather than test, hypotheses.
Self-organising map
A self-organising map is an artificial neural network developed by Kohonen (1990) that projects multi-dimensional input data onto an output attribute space of lower dimensionality (normally 2 dimensions) so that similar observations across the multiple input attributes are placed in proximity to one another on the output space. It is a non-linear generalisation of PCA. A SOM consists of a set of input and output nodes, also referred to as neurons. Each neuron, k, is represented by a series of unstandardised weights, or an n-dimensional vector such that
where, n is the dimensionality of the input space.
Output nodes are connected to surrounding nodes via a neighbourhood relation which establishes a topological structure to the output grid. Training the SOM is an iterative process: at each step, a random input vector, x, is selected and presented to the output neuron grid, where the nodes ‘compete’ for x based on the similarity of the input’s vector of attributes and each neuron’s weight values. Similarity is computed as the Euclidean distance between x and all of the weight vectors. When the best-matching unit or neuron is identified, its weights and the weights of its neighbours are updated resulting in an ordered output grid so that neighbouring neurons have similar weight vectors (Skupin and Agarwal, 2008; Vesanto et al., 1999). This ordering is a distinguishing feature between SOM and the k-means algorithm, which also utilises a Euclidean similarity measure, but no relation is formed between clusters, or observations within clusters. The size of the output grid is determined a priori, with a small number of output nodes forcing the SOM to behave solely as a clustering technique, and a very large number of nodes (exceeding the number of input observations) enabling the emergence of structures. For the purpose of this paper, an intermediate number is selected (20 x 8 grid), smaller than the number of input nodes to allow clustering where observations are very similar, but creating enough output space to visualise longitudinal change. The SOM toolbox for Matlab (Vesanto et al., 2000) is used for the SOM training portion of the study; details of the algorithm can be found in the software documentation. All variables are standardised before entering the SOM procedure to ensure an equal weighting of the dimensional attributes.
In order to utilise the SOM procedure to visualise change, each neighbourhood is included in the initial input dataset six times; once for each time-period. The input data are then trained according to the aforementioned procedure so that each neighbourhood is assigned to a node on the output space six times. Finally, trajectories of change are created for each neighbourhood by tracing its position across the output nodes; its location at each time stamp serves as the vertices of the directed line. The procedure is illustrated in Figure 2 and the result is a decennial trajectory for all 173 neighbourhoods.
SOM training and development of trajectories.
4. Results of Analysis
4.1 Cross-sectional View
One primary output of the SOM procedure are so-called component planes, which are visual depictions of the relative contribution of each QoL attribute to the overall sorting of neighbourhoods in the final layout of the SOM output space. These planes reveal non-linear and partial correlations between variables and thus provide an interesting cross-sectional view of the 17 input variables. Figure 3 illustrates the resulting component planes and shows a distinct ordering of observations across the SOM output space. For example, the first four components in Figure 3—income, homeownership, kindergarten and competency exam scores—all exhibit a similar pattern of high scores towards the top of the planes, descending towards low scores in the bottom portion of the planes. Conversely, attributes commonly associated with lower QoL have a largely opposite pattern of low levels of food stamps, appearance violations, high school dropout rates, teen births and crime rates at the top of the space, and increasing in value towards the bottom.
Component planes.
Partial correlations can also be identified from the plots, including a high concentration of youth social problems (HsDo, TeenBirths) and physical deterioration indicators (Appear, Infrastr) in the lower right-hand corner, while high crime is largely concentrated on the opposite, lower left-hand corner. The plots also reveal that, while neighbourhoods located along the top of the output space generally fair well across all QoL dimensions, those towards the bottom are not the mirror image, scoring poorly across all QoL dimensions; a heterogeneity in social, physical and crime conditions exists.
In order to examine longitudinal change across this composite output space, a trajectory for each neighbourhood is created as described earlier. Displaying the results of each of these lines simultaneously upon the output space creates an uninterpretable situation. Accordingly, to aid in the interpretation and to address the research questions, a clustering procedure is applied to the multidimensional weights assigned to each node in the output space to delineate homogeneous regions of nodes with similar characteristics. This procedure is performed in two steps: first, a hierarchical Ward’s clustering method is used to determine an appropriate number of clusters, followed by a k-means approach to assign nodes to clusters. In this case, a k = 6 solution exhibits the best discriminating power and is illustrated in Figure 4. The compactness and contiguity of the clusters are a direct result of the ordering of the nodes on the SOM output map; like observations are arranged near one another. Characteristics of the clusters are obtained by interpreting the component planes and examining their values within each cluster; they are briefly summarised next.
Cluster 1—highest qol neighbourhoods: high income, homeownership, education scores; low food stamp dependency, crime rates, high school dropout and teen birth rates.
Cluster 2—middle-QoL neighbourhoods, type I: middle-class suburban characteristics: median incomes, high homeownership rates, median education scores; low crime, low social problems, few appearance violations and low accessibility.
Cluster 3—middle-QoL neighbourhoods, type II: median incomes and homeownership rates. Slightly higher education scores than cluster 2 (especially for nodes at the top of the cluster; towards the bottom, the education scores become similar to cluster 4); demographically older, with greater access to transit and retail.
Cluster 4—lower-QoL neighbourhoods type I: lower income and education, median homeownership, appearance violations and high school dropout rates; above-average teen birth rates. Low juvenile and violent crime rates, but median property crime. Older population, high transit access, but low retail access.
Cluster 5—lower-QoL neighbourhoods type II: highest concentration of high school dropout rate and teen births, as well as physical deterioration. Median property and violent crime rates, low juvenile crime, but high crime ‘hot-spots’. Low income and homeownership, high food stamp dependency.
Cluster 6—lower-QoL neighbourhoods, type III: highest concentration of violent, juvenile, and property crime rates, median teen births and high school dropout. Low income and homeownership, high food stamp dependency.
Clusters of output nodes identified by the k-means technique.
4.2 Trajectories of Change
To analyse the longitudinal trajectories of neighbourhoods across attribute and geographical spaces, we now consider each of the clusters in turn. First, the decennial trajectories of all neighbourhoods whose starting position in 2000 is in a node belonging to the first cluster are displayed on the output space, while the corresponding, geographical location of these neighbourhoods is highlighted on a second plot. Figure 5(a) illustrates this for the first cluster; according to the figure, neighbourhoods that began the decade in this highest QoL cluster have a geographical concentration in the southern ‘wedge’ of the city, expanding from close to the city centre outward to the city limit boundary. The majority of these neighbourhoods remained within the same group, with the exception of five neighbourhoods whose trajectories indicate a downward trend towards the more moderate income characteristics of the second group, and one other that moved into the third cluster, marginally declining in the concentration of homeowners; these declines are all slight, as evidenced by their ending position (in 2010) in nodes in the first three rows below their starting position in 2000.
Visualisation of cluster 1 and longitudinal trajectories: (a) 2000; (b) 2010.
In Figure 5(b), the trajectories of neighbourhoods that concluded the decade in cluster 1 are displayed, illustrating that neighbourhoods that transitioned into this highest QoL group came from nearby in attribute space and exhibit an apparent geographical pattern: they are adjacent to existing neighbourhoods in the southern wedge and along the outermost periphery of the city. Overall, neighbourhoods in this group have a large degree of decennial stability in quality of life.
The geographies of neighbourhood in the second cluster (Figure 6) reveal a very suburban pattern along the outermost periphery of the city, corroborating the middle-class suburban attribute descriptions. The trajectories of neighbourhoods that began the decade in this group and evolved away from it follow two distinct paths. One is a path of improvement in QoL indicators, joining the first cluster, while the other is marked by decline, clearly depicted by the downward facing arrows. While neighbourhoods that declined began the decade in the same group as those that improved, their starting positions within the cluster were towards the bottom. Neighbourhoods that transitioned into this group were few; the declines from the first group and one neighbourhood with a geographical location far from the others, towards the city centre. Its trajectory is also distinct, moving from a starting position in cluster 3, but increasing in homeownership, while otherwise maintaining its moderate education scores, and generally low social problems.
Visualisation of cluster 2 and longitudinal trajectories: (a) 2000; (b) 2010.
Neighbourhoods in cluster 3, characterised by lower levels of homeownership and greater accessibility to transit and retail opportunities as compared with the second group, are located within the city beltway, with a larger presence in the southern ‘wedge’ (Figure (7a)). Close inspection of the component planes for nodes within the cluster reveals that neighbourhoods towards the top of the group have higher learning achievement than those towards the bottom (the horizontal line in Figure 3 separating the 2nd and 3rd cluster serves as a cut-off). This is an important distinction when examining the neighbourhoods that left the group by 2010; those that experienced lower QoL all began with lower learning achievement scores and all increased in the number of youth social problems, whereas those that improved began the decade towards the top of the group. The trajectories of neighbourhoods that ended the decade in the 3rd cluster (Figure 7(b)) come from much greater distances across the output space as compared with the previous two cases, suggesting that many of the neighbourhoods with these characteristics in 2010 are very much in transition (primarily on an improvement trajectory). Given that the right side of the output space contains similar social and crime scores, but is distinguished by higher homeownership rates, the resulting trajectories suggest that improvements to these social and crime dimensions preclude increases in homeownership. In addition, a higher concentration of renters may also facilitate larger QoL changes as populations are presumably more fluid.
Visualisation of clusters 3 and 4 and longitudinal trajectories: (a) cluster 3, 2000; (b) cluster 3, 2010; (c) cluster 4, 2000; (d) cluster 4, 2010.
The fourth group of neighbourhoods has similar, median levels of homeownership as compared with the previous cluster, but has lower income levels and a higher concentration of social problems. These communities absorbed large numbers of inner-city residents displaced by gentrification. They were not, however, destinations for Latino settlement. Geographically, these neighbourhoods are in the middle-ring suburbs around the city. Their trajectories reveal considerable variability, showing some movement towards a higher concentration of renters with arrows pointing towards nodes on the left-hand side and a large amount of longitudinal fluctuation in QoL conditions (Figure 7(c)). Neighbourhoods that moved into this group were largely in decline, with the exception of two located south of the city centre, which follow an ascending trajectory, transitioning to the lowest nodes in the group (Figure 7(d)).
Finally, neighbourhoods in the two lowest-scoring clusters present disparate settlement histories. Specifically, cluster 5 broadly represents communities which had large resettlement streams resulting from gentrification. They display significant youth-related social problems. Cluster 6 neighbourhoods are communities with the highest crime rates and are strong destinations for Latino immigrant settlement. As is seen in Figures 8(a), 8(b), 8(c), and 8(d), over the decade, group 5 neighbourhoods that transitioned away generally moved towards the high crime group, with two exceptions; the first is a small neighbourhood north of the city, whose trajectory follows a path more akin to group 3, a neighbourhood well known for its revitalisation and gentrification during the decade; a second neighbourhood already highlighted in the previous group, moving to a bottom node in cluster 4. On the other hand, nearly all the neighbourhoods that increased in youth-related social problems began the decade with high crime concentrations. This is apparent in both the plots of neighbourhoods that transitioned into group 5 (Figure 8(b)), as well as in Figure 8(c), showing the trajectories of neighbourhoods that began in group 6. All neighbourhoods that exited group 6 moved to group 5; none followed paths of revitalisation. Geographically, all but one of the neighbourhoods that transitioned into cluster 6 were adjacent to a neighbourhood already in the group or one that also transitioned in, possibly indicating a spatial spillover of high crime (Figure 8(d)). Neighbourhoods in cluster 5 have a much more obviously geographical concentration than the high-crime neighbourhoods and have a greater presence closer to the urban core. Conversely, the high-crime neighbourhoods, especially by 2010, are much more dispersed in older suburban neighbourhoods, expanding eastward, whereas cluster 5 neighbourhoods are largely confined to neighbourhoods just north and west of the city centre.
Visualisation of clusters 5 and 6 and longitudinal trajectories: (a) cluster 5, 2000; (b) cluster 5, 2010; (c) Cluster 6, 2000; (d) cluster 6, 2010.
5. Discussion and Conclusions
During the first decade of the century, Charlotte has experienced dynamic growth marked by explosive housing expansion in inner-city and suburban areas, with strong gentrification trends and vibrant immigrant newcomer streams. This research uses a quality of life template to present a case study for assessing existing urban theories and conventions regarding longitudinal neighbourhood change, while also enabling the emergence of new relationships and trends to be observed. Results of the analysis provide some affirmation of traditional theories of change as neighbourhoods that transitioned to the highest quality of life had a large spatial presence on the outermost ring of the city, and the highest QoL neighbourhoods proved to be the most stable across the decade.
In line with recent demographic research and house filtering theories, older, suburban neighbourhoods located in the city’s middle ring underwent significant transformations during the decade. The resulting neighbourhood trajectories, however, indicate a large degree of variability in the changes across the multidimensional attribute space. A number of neighbourhoods within this middle ring followed a path towards the highest crime concentrations in the city in conjunction with low incomes and high food stamp dependency (cluster 6), while others saw declines in educational attainment and increases in youth-related problems coupled with housing deterioration (cluster 4). These two categories encompass neighbourhoods experiencing large streams of disparate in-migrants: low-income residents displaced by gentrification (cluster 4) and Hispanic newcomers to Charlotte (cluster 6).
Another major finding that emerged from this work is a divergence in social problems between economically disadvantaged neighbourhoods. Two distinct groups were formed by the clustering analysis: those with the highest crime concentrations and those with more youth-related social problems (high school dropout rates and teenage pregnancy rates). Attribute change in these neighbourhoods generally occurred between the two groups, exchanging high crime for more social problems, or vice versa, lending some evidence to the notion that low-income, high-food-stamp-dependent neighbourhoods experience an increase in social problems over time. However, a contrasting geographical pattern was observed between the two groups over time. While the neighbourhoods with the greatest concentration of youth-related problems generally remained confined to an inner-city neighbourhood cluster, the highest crime rates dispersed outward towards older, suburban neighbourhoods by the end of the decade. Evidence of a spatial spillover of crime was also uncovered as many of the neighbourhoods that transitioned into the 6th group by the close of the decade were adjacent to neighbourhoods that were already in the group at the start of the decade, or to other neighbourhoods that also transitioned into it. Notably, none of the high crime neighbourhoods followed paths towards revitalisation, while sharp change trajectories depicting revitalisation could be identified from the visual plots of several neighbourhoods that entered the 3rd cluster (characterised by median homeownership rates and access to transit and basic retail opportunities). These neighbourhoods tended to improve along many social and economic QoL dimensions while maintaining median to low levels of homeownership, suggesting that increases in aggregate homeownership levels may follow other QoL improvements to a neighbourhood. Educational attainment was identified as a distinguishing characteristic between middle-income neighbourhoods that followed improvement or decline trends.
The analytical approach implemented in this study, which blends a computational, data projection and reduction procedure, the self-organising map, with visualisation techniques, provides a framework for exploring the multidimensional, multitemporal, quality of life data collected by municipalities. Clustering the SOM output as was done for this analysis enables general patterns of change for neighbourhoods with similar QoL profiles to be identified, while visualising longitudinal trajectories allows for a disaggregated view of the trends followed by each neighbourhood across the array of QoL attributes. Both visualisation scales equip policy-makers with communication tools for illustrating longitudinal QoL trends with constituents that would otherwise be impossible to depict from static change maps of the six time stamps. Furthermore, geographical maps portraying neighbourhoods with coinciding social problems also provide a means for pinpointing local, targeted initiatives.
The methodology used in this research is intended to be exploratory in nature, visualising patterns and trends within this complex spatial, temporal and multidimensional dataset. Future research will utilise confirmatory statistical approaches to test the hypothesised relationships and to tease out causal relationship. In particular, the impact of recent population influxes to the city and its neighbourhoods in shaping QoL changes should be evaluated, as should the influence of local policies including fundamental structural changes to the public school system, targeted policing and neighbourhood initiatives aimed at enhancing QoL.
Footnotes
Funding
This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.