An impact evaluation of area-based interventions in Cape Town using multivariate regression analysis

Area-based initiatives are popularly applied to alleviate the spillover effects of neighbourhood poverty in deprived neighbourhoods. This study analyses the effects of two area-based initiatives on neighbourhood poverty in Cape Town between 2001 and 2011 in a controlled baseline study. The purpose of this study is to determine whether the changes were the product of policies themselves or wider structural changes in the national economy, and what were the specific outcomes of the policies. The study revealed that, despite some minor gains, these policies were ineffective in reducing poverty levels in the policy areas, and that poverty levels are primarily determined by the broader changes in the economic environment and in-migration.


INTRODUCTION
Area-based policies are developed to spatially target resources and policy interventions in selected deprived locations, in order to reduce poverty in areas where it is concentrated (Alcock, 2004: 90).However, many analysts are critical of the manner in which area-based initiatives (ABIs) are implemented, arguing that concentrated urban poverty is a spatial side effect of aspatial policies (Logan & Molotch, 1987: 56;Wilson, 1987: 13).They argue that the root cause of spatially concentrated poverty is structural policies, and that ABIs will be ineffective in addressing concentrated poverty.
In light of this debate, the study evaluates the effectiveness of two ABIs targeting specific neighbourhoods in Cape Town in terms of changes in neighbourhood poverty between 2001 and 2011.The first is the urban renewal programme (URP), a nationally driven pilot ABI implemented in Khayelitsha and Mitchells Plain between 2001 and 2011.The second is the spatially targeted municipal area-based investment in neighbourhoods designated as marginalised areas by the 2004 Cape Town Socio-Economic Index (henceforth known as municipal area-based initiatives or MABIs).Additionally, the study uses a control group of high-poverty neighbourhoods, where the ABIs The author declares no conflict of interest for this title or article.
Mr Herman S. Geyer Jr. Researcher: CRUISE Centre, Department of Geography, GeoInformatics, Urban and Regional Planning, Stellenbosch University.Room 2046, Kamer van Mynwese Building, 20 Van Ryneveldt Street, Stellenbosch 7600, South Africa.Phone: (+27) 218089223, email: <hsgeyerjr@sun.ac.za> were not implemented, in order to comparatively evaluate variations in changes in neighbourhood poverty between 2001 and 2011 in a controlled baseline study.
The study employs multivariate regression analysis to analyse the socio-economic characteristics of neighbourhood change in neighbourhoods with high poverty.Cape Town was selected as the study area because of the comparative use of different ABIs within comparable sample populations and because of the diversity of neighbourhoods in the municipality.

AREA-BASED APPROACHES
Poor neighbourhoods experience proportionally higher rates of deprivation, including unemployment, welfare dependency, crime, morbidity and educational underachievement.
The spatial concentration of these phenomena intensifies economic and social deprivation in these neighbourhoods in a milieu of ever-increasing affluence (Wilson 1987: 26).This poses a political problem in a welfareoriented state, as the concentration of poverty indicates a failure of policy to integrate populations and redistribute wealth (Musterd & De Winter, 1998: 669).In an attempt to minimise the negative externalities of concentrated poverty, ABIs spatially target additional resources in the most deprived areas.They also address these problems by increasing economic participation and strengthening social agency in specific locations.Thus, ABIs attempt to directly target a greater number of deprived households than can be targeted by way of structural policies (Smith, 1999: 13).
More recently, third-way approaches also included smaller social projects, including childcare support, job-placement programmes, neighbourhood policing, educational, health and sporting programmes to promote social agency as well as a move towards collaborative selfgoverning partnerships and thirdsector participation (Lawless, Foden, Wilson & Beatty, 2010: 262).Third-way approaches are based on subcultural theories, in which it is believed that social agency can be channelled to regenerate an area without displacement by utilising existing social and cultural ties to a neighbourhood.Strong community dialogue and social interactions within a community can increase the attachment of residents to the neighbourhood, thus motivating residents to invest in their own neighbourhoods and stabilise the community (Temkin & Rohe, 1996: 165;Andersson & Musterd, 2005: 380).
While the concentration of urban poverty is a problem in both developed and developing countries, this problem is particularly acute in South Africa.Spatially, urban poverty is still concentrated in former apartheid-era Black townships, indicating that post-apartheid policy had hardly any effect on changing the spatial structure of apartheid (Noble & Wright, 2013: 188).In response, the South African government has piloted ABIs targeting the distribution of public resources in selected deprived urban areas in South Africa.
The intention of area-based planning in South Africa is to change the character of a location by means of a range of institutional mechanisms involving private, governmental and local stakeholders, in order to achieve the development objectives of local and regional planning policies (Turok, 2004: 409).

Urban renewal programme
The URP was an intergovernmental co-ordinated 10-year ABI implemented in eight former townships throughout South Africa.Selection criteria included locations with high poverty levels, high crime rates, low economic opportunities, low social capital, high unemployment rates, low education rates and low skill levels (DPLG, 2004).The purpose of the URP was to crowd in investment in previously disenfranchised neighbourhoods, mobilise partnerships with local investors, and strengthen public participation (Forster, Leon & Menguele, 2006: 16)

Municipal area-based initiatives
In addition, the Cape Town Metropolitan Municipality also implemented targeted area-based investment in MABIs.Although not designated as ABIs, the marginalised areas were identified using the 2004 Cape Town Socio-Economic Index to identify subplaces within the municipality that had the greatest levels of deprivation (CoCT, 2014).
Based on this Index, national census subplaces, representing neighbourhoods by proxy, were ranked in terms of household service levels, education rates, housing type and household income.The ranking was weighted towards neighbourhoods that had lower service levels and less adequate housing, not absolute income poverty levels per se, and the purpose of the policy was to lower all forms of poverty in the selected areas (Laldahprasad, Geyer Jr & Du Plessis, 2013: 42).Accordingly, different departments within the municipality coordinated to target services in areas delineated as deprived.
Cape Town municipal expenditures were significantly higher in areas that required the greatest need for public intervention, indicating an efficient implementation of the MABIs (Laldahprasad et al., 2013: 44).The MABI projects replicated many of the URP outcomes throughout the municipality, and collaborated with the URP within its demarcated area; marginalised areas were effectively prioritised (CoCT, 2011b: 47).These replications included public housing, public transportation, hospitals and clinics, public retail facilities and basic services infrastructure.However, tasked with a broader range of objectives, and being held accountable by city managers, the MABIs also included smaller initiatives such as the development of tourism information offices, fresh-produce markets, informal trading markets, sports facilities, arts facilities, libraries, community centres, and recreational parks (CoCT, 2011c: 33).

CHALLENGES FACING AREA-BASED INITIATIVES
From the outset, the performance of these ABIs highlight two basic challenges in area-based approaches, the first being the challenge of defining poverty and the spatial delineation of poverty.Poverty is generally defined in terms of the mean standard of living of persons in a pluralistic society (Goedemé & Rottiers 2011: 78).On that basis, poverty is related to the availability of choices and opportunities to individuals; poverty is thus multidimensional.However, because basic needs such as employment, health, education, housing, public services and a safe environment are ultimately attributed to income variations, household income is nearly universally employed to measure poverty.Defining spatial distribution of poverty is an equally difficult task.While neighbourhood poverty is generally analysed in terms of administrative boundaries, the neighbourhood is not a closed system; it is nested within larger social and economic networks with a fluid population.However, as the aggregate measures of socioeconomic indicators are similar at different levels of spatial aggregation, the official administrative spatial delineation of neighbourhoods is a useful proxy (Galster, 2001(Galster, : 2113)).
A second serious challenge is how to determine the effectiveness of ABIs in reducing poverty.Commercial property development in impoverished areas rarely attracts outside industries, due to negative externalities in terms of location to markets, suitable skilled labour, and suppliers.The majority of local relocations consist of the pitchshifting of existing local enterprises vacating derelict property in favour of upgraded property.The number and quality of jobs produced are marginal, and primarily consist of part-time, low-wage, unskilled jobs at the loss of locally owned and managed enterprises and the declining circulation of local disposable income (Turok, 1992: 369).Furthermore, the physical restructuring of the neighbourhood through retail-led regeneration does not address the key problems of unemployment, criminality and welfare dependency (Andersson & Musterd 2005: 384).
The displacement of poor residents to the worst neighbourhoods often occurs through the overt redevelopment and relocation of existing low-income communities, ostensibly through low-income housing redevelopment, crimemitigation programmes or environmental revitalisation programmes (Smith, 2002: 438).
Prestige projects tend to be concentrated in the most profitable locations and, although located in close proximity to the disadvantaged, there is very little synergy or interdependency other than the most superficial interaction.

METHODOLOGY
The study evaluates the effectiveness of ABIs in Cape Town in terms of changes in neighbourhood poverty between 2001 and 2011.Cape Town was chosen as the study area, due to the large number of neighbourhoods and the diversity of neighbourhood types in the municipality.The critical question in the study is whether the observations are attributable to ABIs or not.Thus the study takes the approach that appropriate ABI evaluations should evaluate policy outcomes independent of broader regional changes in neighbourhoods.
To accomplish this, the study uses best subset linear multivariate regression analysis in a differencein-differences (DD) methodology to determine the variables best correlated with policy effects.DD is a statistical technique that creates an experimental research design, using observational study data to mitigate the effects of extraneous factors and selection bias by comparing the change over time in the treatment group to the change over time in the control group (Abadie, 2005: 1).This maximises the internal validity of the findings and thus the reliability of the study results.

Data collection
The

Data analysis and interpretation
The first analysis estimates the correlation between the policy and changes in neighbourhood poverty between 2001 and 2011 in the entire municipality, in order to determine whether policy was a significant factor in neighbourhood poverty change in the study area during the study period.In the second analysis, the study determines what factors are significantly correlated to changes in neighbourhood poverty within the treatment and control groups between 2001 and 2011 using a DD methodology.Controlling for broader changes in regional attributes in the control group, the neighbourhood poverty change as a product of policy itself can be significantly correlated in the treatment groups.This is done to determine the effectiveness of ABI policies, independent of broader changes in regional attributes.
The study selected the subset of factors best predicting changes in neighbourhood poverty by reducing the number of variables while maximising the coefficient of determination.

ANALYSIS
The purpose of the first analysis is to determine whether ABI policies were a significant factor in neighbourhood poverty change in the study area.
The first analysis, presented in As the independent variables primarily contained 2001-2011 variances, there was hardly any interdependence within the independent variables.An optimal best subset regression with (VIF<2.1)and independent variable (p-values<0.05)was calculated.
The study has a reasonable fit (R 2 of 48.3% and Adj.R 2 of 48.1%).
β-value results of control independent variables in the regression are as expected: an increase in neighbourhood poverty is mostly correlated with factors linked to national socio-economic policies outside the scope of the neighbourhood, including decreasing mean monthly household incomes, increasing neighbourhood GINI inequality, decreasing employment rates, decreasing neighbourhood GVA, declining education rates (as a product of a failing national scholastic system), decreasing racial heterogeneity (stemming from selective outmigration), increasing dependency rates (due to high birth rates), and increasing ratios of female-headed households.However, the important variables are the dummy policy variables.In Table 2, the regression indicates that both the national URP and the MABI programmes are significantly correlated with increases in neighbourhood poverty, with neighbourhood poverty increases higher in MABIs (4.8%) than in the URP areas (2.3%).This does not necessarily imply that ABIs are causal factors in increases in neighbourhood poverty; it only indicates that these initiatives were ineffective in reducing neighbourhood poverty rates.
In Table 3, the second analysis presents DD multiple regression analyses of factors related to changes in neighbourhood poverty between 2001 and 2011 in both the study and the control areas.This analysis includes all the independent variables of the first analysis, except the ABI dummy variables.The analysis is conducted at a level of significance (p-values<0.05)with (VIF<2.5).All the studies have good fits, particularly the treatment groups.In all three areas, the change in neighbourhood poverty, evident as increasing neighbourhood poverty levels (as indicated in Table 2), is correlated with declining mean household income (as a percentage of the mean household incomes in the municipality) and increasing neighbourhood GINI inequality, factors that cannot be indirectly addressed by area-based approaches, but are related to the economic conditions in the country as a whole.
An important first observation is that state investment and increasing basic services provision was not significant in any of the treatment groups, indicating that the extent of ABI interventions was insufficient in effecting change in poverty levels.
In the uppermost URP regression in Table 3, policy effects in the treatment group include declining residential densities, declining crime rates and increasing informal dwelling ratios.Controlling for extraneous factors in control groups, the DD regression indicates that increases in GINI inequalities in the URP study areas were both significant and double that of control groups, further indicating a possible failure of policy.
Because the household income variances are similar between the URP treatment group and the control group, it can be reasoned that declining neighbourhood income is not a product of ABI policies.However, the significance of increasing in-migration and rapid demographic change indicates that increasing poverty could be the result of the in-migration of the poor rather than neighbourhood or policy effects.
In the middle MABI regression summary in Table 3, treatment effects, controlling for extraneous effects in control groups indicates that the only direct possible positive policy effect is an increase in employment.However, poverty levels also generally increased relative to increases in the employment rate.Increasing neighbourhood poverty is also significantly correlated with rapid demographic change, declining dependency ratios, declining in-migration rates, significant increases in GINI income inequality, and significantly declining household incomes, controlling for difference-in-differences.These indicate the possible occurrence of hypersegregation effects as a potential negative outcome of MABI policies.

DISCUSSION AND CONCLUSION
The study analysed the effect of ABIs on neighbourhood poverty in South Africa.The implementation of ABIs is a controversial issue, with urban managers and business interests favouring this mechanism to target resources in deprived neighbourhoods, in order to reduce poverty levels.However, many researchers and community groups oppose the implementation of ABIs as either ineffective or instrumental  In the MABI group, which specifically targeted neighbourhood poverty, the only direct possible positive policy effect is an increase in employment.However, this occurred with significant increases in GINI income inequality and declining household incomes, double that of control groups.Moreover, neighbourhood poverty was also significantly correlated with factors related to hypersegregation, including rapid demographic change, declining dependency ratios, and declining in-migration rates.These outcomes were the opposite of the social cohesive agenda that ABIs intend to engender.
In summary, it appears that ABIs were ineffective in lowering neighbourhood poverty, with increasing neighbourhood poverty attributed to exogenous national policies.There was an insufficient level of public investment in ABIs to lower poverty levels, with state investment insignificantly correlated in both study areas.Secondary policy gains achieved were limited, with the URP areas only experiencing decreased densities as a result of public housing.However, adverse effects included increased informality and declining economic growth, due to the adverse effect of retail development.In the MABI areas, possible policy effects could include increasing employment rates, despite increases in income inequality and declining household income, with possible adverse outcomes, including hypersegregation effects.Thus ABI policy outcomes are relatively mixed and no clear net positive outcomes are evident.

Table 2 :
Multiple regression analysis of factors related to changes in neighbourhood poverty,2001-2011.

Table 3 :
Multiple regression analyses of factors related to changes in neighbourhood poverty, 2001-2011, in the URP, MABI and control areas, respectively.CONTROL Best subset regression; DV: poverty trajectory 2001-2011 (positive value = increased poverty), R=.8503, R 2 =.7230, Adj R 2 =.7032, F(4,61) = 36.549,p< 0.0001, Std.Err.:.0609In addition, the empirical evaluation of the URP initiative is limited, as increasing in-migration and racial demographic change dilute the effects of policy treatments in URP study areas.This can also explain the correlation with increases in the informal dwelling ratio in this treatment group.