Aperçu

Résumé (français)

Due to privacy concerns, individual-level census data is unavailable for public use. Aggregated census data – compiled at the census tract and dissemination area levels – is used as a proxy to portray socioeconomic conditions within these administrative units. Literature shows that despite widespread usage of aggregated census data, researchers and policymakers fail to critically assess the limitations and embedded assumptions of this method. How does the aggregation of data to arbitrarily defined geographic units affect the socioeconomic portrait they produce? My research elaborates on findings from Prouse et al’s (Forthcoming) report for the Neighbourhood Change Research Partnership (NCRP) exploring Halifax’s geography of income inequality and polarization. The report revealed mixed trends in the overall CMA, with no strong evidence of increasing income polarization at the census tract level. In many cases, trends were ambiguous and diverged from hypotheses derived from local understandings of the lived reality of these spaces. Though Halifax is consistently portrayed in literature as a relatively egalitarian city compared to larger Canadian CMAs, observed circumstances – including concentrated poverty in Halifax’s public housing projects and gentrification in the North End – suggest otherwise. Prouse et al hypothesized that the study parameters – using census tracts as the units of analyses – could explain discrepancies between census data indicators and qualitative observations of socioeconomic conditions. Hence, in this study, I explore the dynamics and nature of sociospatial polarization at the dissemination area level. In particular, I sought to determine whether greater evidence of sociospatial polarization is evident at the DA level than at the CT level: thus determining which scale is more appropriate to observe Halifax’s socioeconomic conditions. Using the Modifiable Areal Unit Problem (MAUP) as a theoretical lens, I analyzed differences between conditions at the CT and DA levels. The MAUP is a phenomenon occurring when census data is collected for individuals but is reported for administrative units possessing modifiable boundaries. Data aggregation mutes extreme values and obscures diverse socioeconomic conditions occurring within these units. The MAUP is a particular issue for smaller municipalities and rural areas, since administrative units are formed on a larger scale than in big cities with higher population densities. Hence, homogeneous clusters of individuals often form at a scale smaller than the administrative unit boundaries, causing diverse clusters of socioeconomic conditions to form within them. I conducted statistical and spatial analyses on ten socioeconomic indicators, comparing their characteristics, relationships, and spatial patterning at the CT and DA levels. Descriptive statistics showed that for all indicators, DA level data is more dispersed from the CMA average than at the CT level; many DAs have extreme values that are muted when these values are aggregated with adjacent DAs to form CTs. Percentages of visible minorities across DAs had the greatest difference in dispersion of all indicators with DA level proportions being 81% more dispersed than at the CT level. I also compared differences in relationships between indicators using Pearson’s Bivariate Correlations and Ordinary Least Squares Regression tests. Results from these tests were consistent with those in literature, thus affirming the influence of the MAUP on Halifax’s census data. At the CT level, we obtain stronger correlations between variables and a more robust regression model than at the DA level. The muted CT values follow more consistent trends than engendered by the extreme outliers at the DA level where it becomes more difficult to generalize relationships with definitive conclusions. However, the tests show a more complex portrait of socioeconomic conditions and relationships at the DA level; more relationships are deemed ‘statistically significant’ – we can confidently ascertain a linear relationship exists between them – than at the CT level. I observed similar trends in the spatial analysis. Stronger dichotomies between contrasting conditions emerged in the CT level maps, with categories split at a large scale. The DA level maps show a diverse mosaic with DAs displaying adjacent contrasting socioeconomic conditions. At the DA level, we observe polarized adjacencies: spatially proximal concentrated clusters of contrasting conditions. Polarized adjacencies are obscured when the extreme DA values are combined to create values for the overall CT. Sociospatial polarization emerges much more frequently through polarized adjacencies at the DA level. Though the CT and DA level values have relatively similar frequency distributions across indicators, the extreme cases at the DA level are crucial determinants of the nature and severity of Halifax’s sociospatial polarization. They contribute to polarized conditions in many of the city’s CTs, causing areas exhibiting extreme deprivation and poor socioeconomic conditions to appear less severe. Thus, the CT model is suitable for economists and statisticians who seek a stronger general model with a more parsimonious causal structure, or for the NCRP researchers wishing to derive general comparative paradigms for neighbourhood change. However, for the purposes of policy makers, scholars, and practitioners concerned with socioeconomic inequality and polarization trends, polarization is portrayed much more intricately through the DA level. All indicators, their relationships with each other, and their spatial manifestations are recognized, even if their impact is relatively small when tests are conducted for the CMA as a whole. When we restrict analysis to conditions within the CT, these weaker relationships encourage extreme polarized adjacencies between DAs and have significant implications on the lived experiences of residents. Therefore, researchers and policymakers must be wary of the embedded limitations of using administrative unit data to represent individual-level conditions. In urban policy, census data is used for informing policy changes, forecasting growth projections, allocating community infrastructure, amenities, and services, and creating sustainable municipal visions for the future. Misrepresentation of these data yields deleterious consequences, including the misallocation of services. Findings emphasize the importance of robust data collection measures for small geographic units.

Résumé (anglais)

Due to privacy concerns, individual-level census data is unavailable for public use. Aggregated census data – compiled at the census tract and dissemination area levels – is used as a proxy to portray socioeconomic conditions within these administrative units. Literature shows that despite widespread usage of aggregated census data, researchers and policymakers fail to critically assess the limitations and embedded assumptions of this method. How does the aggregation of data to arbitrarily defined geographic units affect the socioeconomic portrait they produce? My research elaborates on findings from Prouse et al’s (Forthcoming) report for the Neighbourhood Change Research Partnership (NCRP) exploring Halifax’s geography of income inequality and polarization. The report revealed mixed trends in the overall CMA, with no strong evidence of increasing income polarization at the census tract level. In many cases, trends were ambiguous and diverged from hypotheses derived from local understandings of the lived reality of these spaces. Though Halifax is consistently portrayed in literature as a relatively egalitarian city compared to larger Canadian CMAs, observed circumstances – including concentrated poverty in Halifax’s public housing projects and gentrification in the North End – suggest otherwise. Prouse et al hypothesized that the study parameters – using census tracts as the units of analyses – could explain discrepancies between census data indicators and qualitative observations of socioeconomic conditions. Hence, in this study, I explore the dynamics and nature of sociospatial polarization at the dissemination area level. In particular, I sought to determine whether greater evidence of sociospatial polarization is evident at the DA level than at the CT level: thus determining which scale is more appropriate to observe Halifax’s socioeconomic conditions. Using the Modifiable Areal Unit Problem (MAUP) as a theoretical lens, I analyzed differences between conditions at the CT and DA levels. The MAUP is a phenomenon occurring when census data is collected for individuals but is reported for administrative units possessing modifiable boundaries. Data aggregation mutes extreme values and obscures diverse socioeconomic conditions occurring within these units. The MAUP is a particular issue for smaller municipalities and rural areas, since administrative units are formed on a larger scale than in big cities with higher population densities. Hence, homogeneous clusters of individuals often form at a scale smaller than the administrative unit boundaries, causing diverse clusters of socioeconomic conditions to form within them. I conducted statistical and spatial analyses on ten socioeconomic indicators, comparing their characteristics, relationships, and spatial patterning at the CT and DA levels. Descriptive statistics showed that for all indicators, DA level data is more dispersed from the CMA average than at the CT level; many DAs have extreme values that are muted when these values are aggregated with adjacent DAs to form CTs. Percentages of visible minorities across DAs had the greatest difference in dispersion of all indicators with DA level proportions being 81% more dispersed than at the CT level. I also compared differences in relationships between indicators using Pearson’s Bivariate Correlations and Ordinary Least Squares Regression tests. Results from these tests were consistent with those in literature, thus affirming the influence of the MAUP on Halifax’s census data. At the CT level, we obtain stronger correlations between variables and a more robust regression model than at the DA level. The muted CT values follow more consistent trends than engendered by the extreme outliers at the DA level where it becomes more difficult to generalize relationships with definitive conclusions. However, the tests show a more complex portrait of socioeconomic conditions and relationships at the DA level; more relationships are deemed ‘statistically significant’ – we can confidently ascertain a linear relationship exists between them – than at the CT level. I observed similar trends in the spatial analysis. Stronger dichotomies between contrasting conditions emerged in the CT level maps, with categories split at a large scale. The DA level maps show a diverse mosaic with DAs displaying adjacent contrasting socioeconomic conditions. At the DA level, we observe polarized adjacencies: spatially proximal concentrated clusters of contrasting conditions. Polarized adjacencies are obscured when the extreme DA values are combined to create values for the overall CT. Sociospatial polarization emerges much more frequently through polarized adjacencies at the DA level. Though the CT and DA level values have relatively similar frequency distributions across indicators, the extreme cases at the DA level are crucial determinants of the nature and severity of Halifax’s sociospatial polarization. They contribute to polarized conditions in many of the city’s CTs, causing areas exhibiting extreme deprivation and poor socioeconomic conditions to appear less severe. Thus, the CT model is suitable for economists and statisticians who seek a stronger general model with a more parsimonious causal structure, or for the NCRP researchers wishing to derive general comparative paradigms for neighbourhood change. However, for the purposes of policy makers, scholars, and practitioners concerned with socioeconomic inequality and polarization trends, polarization is portrayed much more intricately through the DA level. All indicators, their relationships with each other, and their spatial manifestations are recognized, even if their impact is relatively small when tests are conducted for the CMA as a whole. When we restrict analysis to conditions within the CT, these weaker relationships encourage extreme polarized adjacencies between DAs and have significant implications on the lived experiences of residents. Therefore, researchers and policymakers must be wary of the embedded limitations of using administrative unit data to represent individual-level conditions. In urban policy, census data is used for informing policy changes, forecasting growth projections, allocating community infrastructure, amenities, and services, and creating sustainable municipal visions for the future. Misrepresentation of these data yields deleterious consequences, including the misallocation of services. Findings emphasize the importance of robust data collection measures for small geographic units.