The spatial pattern of premature mortality in Hong Kong: How does it relate to public housing?

Abstract

Research into understanding the relationship between access to housing, health and wellbeing in cities has yielded mixed evidence to date and has been limited to case studies from Western countries. Many studies appear to highlight the negative effects of public housing in influencing the health of its residents. Current trends in the urban housing markets in cities of advanced Asian economies and debates surrounding the role of government in providing housing underscore the need for more focused research into housing and health. In this paper, we investigate Hong Kong as an example of a thriving Asian city by exploring and comparing the intra-urban geographies of premature mortality and public housing provision in the city. Using a fully Bayesian spatial structural model, we estimate associations between public housing provision and different types of premature mortality. We find significant geographic variations in premature mortality within Hong Kong during the five-year period 2005–2009, with positive associations between the residents of public housing and premature mortality risk. But the associations attenuate or are even reversed for premature mortality of injuries and non-communicable diseases after controlling for local deprivation, housing instability, access to local amenities and other neighbourhood characteristics. The results indicate that public housing may have a protective effect on community health, which contradicts the findings of similar studies carried out in Western cities. We suggest reasons why the association between public housing and health differs in Hong Kong and discuss the implications for housing policy in Hong Kong and other Asian cities.

Keywords

Asian cities housing premature mortality public housing spatial analysis urban health disparities

Background

Recent reports by the World Health Organization (WHO and UN-HABITAT, 2010) and the Global Research Network of Urban Health Equity (GRNUHE, 2010) confirm that pronounced health inequalities persist in cities all over the world. The reports reiterate the need to address the social determinants of health in cities (Corburn, 2009; Marmot, 2005, 2010; Stephens, 2011), including socio-economic and physical neighbourhood conditions as well as access to housing, all of which are thought to modify the material, behavioural and psycho-social pathways of health (Marmot and Wilkinson, 2005; Kawachi and Berkman, 2003).

Secure access to housing is identified as a key policy arena (GRNUHE, 2010: 17), particularly critical in cities whose economies strongly interconnect with the global economy. Such cities display advanced forms of inequality and generate polarised geographies of disadvantage (Corburn, 2009; Gusmano et al., 2010; Sassen, 2001), a result of a combination of market forces, informal practices and policy initiatives including the provision of public housing. In this context, public housing appears to be a valuable policy tool to address the social and spatial determinants of the health of its residents (Doran et al., 2013).

Studies investigating links between housing and community health have been predominantly conducted in Western cities. While there is some evidence to suggest that housing provision and stability have a protective effect on disadvantaged residents (Baker et al., 2013; Burgard et al., 2012; Pollack et al., 2010), results have yielded a very mixed picture. In this paper we consider the case of Hong Kong by comparing the spatial pattern of health disadvantage – measured by premature mortality risk (PMR) – to the spatial distribution of public housing in the city. The aim of the study is to investigate whether in Hong Kong’s unique context of housing provision and land shortage, neighbourhoods with high levels of social housing coincide with reduced or increased levels of PMR, once area deprivation, access to services and aspects of the built environment and housing instability have been taken into account.

Public housing and health: Empirical evidence

Research on the relationship between housing and health has primarily focused on the environmental or physical aspects of housing – including building conditions, exposure to noise and overcrowding (Dunn and Hayes, 2000; Jacobs, 2011; Krieger and Higgins, 2002); the role of housing tenure (Baker et al., 2013; Dalstra et al., 2006; Hartig and Fransson, 2006; Leech, 2010); the relationship between physical and tenure patterns; as well as neighbourhood characteristics (Ellaway and Macintyre, 1998; Hiscock et al., 2003; Macintyre et al., 2003; Windle et al., 2006). With the exception perhaps of Baker et al. (2013), these studies highlight the negative health effects on residents of rented accommodation – including social housing – compared with residents of privately owned properties.

Identifying the health effects of housing over and above physical and neighbourhood conditions is complex, since social housing is often of inferior design and construction quality compared with private accommodation, and typically (but not always) located in deprived and socially fragmented neighbourhoods. The conflation of housing and neighbourhood stressors can easily confound correlations between social housing and the poor health of its residents (Anthony and Robbins, 2012; Gibson et al., 2011; Mueller and Tighe, 2007).

A recent study by Lawder et al. (2014), for example, confirms that residents in Scotland’s urban areas display poorer health in neighbourhoods with a larger percentage of social housing. This finding matches those of similar studies, but its results challenge the assumption that by guaranteeing access to secure accommodation, public housing exerts some positive health effects on those residents who are unable to afford private housing compared with residents with a similar socio-economic profile who live in private rented housing. This assumption of an association between public housing provision and health effects – a logical by-product of the social determinants of health approach – seems particularly pertinent to urban environments where housing is scarce; such as Asia’s rapidly growing cities and other global centres, where one might expect the provision of public housing to emerge as a critical determinant of the health and wellbeing of a city’s population.

In an ecological analysis of health and social mixing at urban ward-level in the UK, Graham et al. (2009) observe a positive association between social housing and health disadvantage. This association, however, becomes weaker if not insignificant when the authors control for housing stressors such as overcrowding or lack of physical facilities (e.g. heating). The authors note that ‘… there is no reason to assume that the owners [of private accommodation] will necessarily be “better off” than the social renters’ (Graham et al., 2009: 161), a conclusion that mirrors findings by Burgard et al. (2012), who identified heterogeneous associations between different kinds of housing instability, types of tenures and poor health. For example, individuals who suffer from financial stress (and are behind in their mortgage payments) rate their health more poorly than other types of occupiers and are more likely to report anxiety attacks (Burgard et al., 2012: 2221); this questions common assumptions on the health advantage of home ownership relative to social renting. The relationship between tenure, access to housing and health therefore seems more complex and contingent than is often hypothesised, and frequently suggests a positive effect of social housing in certain contexts (also see Burrows, 2003; Taylor et al., 2006).

Public housing in Hong Kong

Much of the research on the relationship between housing and health has been carried out in the UK, other European countries and North America. Hong Kong provides a useful Asian comparator, since it is highly connected to the global economy and shows similar trends to other global cities with regard to housing scarcity and urban inequality (Lee et al., 2007; Tai, 2006). For example, Hong Kong has a relatively high GDP per capita of US$38,800 and a high GINI coefficient of inequality of 0.537 (Govt HK, 2013), where 0 indicates full equality and 1 maximum inequality. In addition, the city has unique characteristics that create further pressures on housing, health and wellbeing. Located on more than 200 islands and surrounded by protected mountain ranges, the city is topographically constrained. The response to these physical limitations to expansion is a hyper-dense urban form marked by high-rise residential buildings, many over 20 storeys high (Lai, 1993; Lai and Baker, 2014: 223; Yip et al., 2009), reinforced by the Hong King Government’s strict regulation of land supply which adds to the housing pressure (Peng and Wheaton, 1993). Hong Kong’s population of just over 7 million people live on 7% of the total area of 1100 km² (Yip et al., 2009). As a consequence, the functioning of Hong Kong relies heavily on the public transportation system: the MTR (Mass Transit Railway), bus and minibus services.

Public housing played a central role in Hong Kong’s urban development and has often been a driver in government-led urban expansion and construction of New Towns. Since the 1950s, the government has committed itself to a large-scale public housing programme to accommodate the influx of immigrants from Mainland China. This programme has had a major impact on Hong Kong’s urban morphology and the ‘geography of housing opportunities’ (Forrest et al., 2004), by moderating the impact of income deprivation in Hong Kong’s highly unequal society. Scholars investigating housing in Hong Kong point out that the city’s success and position in the global economy is partly due to its extensive public housing programme, which contrasts sharply with the laissez-faire policy regime often associated with Hong Kong (Delang and Lung, 2010; Forrest and Yip, 2014).

Lee and Yip’s (2006) detailed, qualitative investigation of the social impacts of public housing in Hong Kong prior to 1990 offers some explanation of how social renting and the long-term socio-economic development of families and households are causally connected; public housing offers security and acts as a quasi-material resource to increase life chances and disposable income. Unlike in Western contexts, the authors conclude, public housing provides a material foundation for residents with low incomes and constitutes a stepping stone for social mobility.

Public housing in Hong Kong comprises different kinds of benefits that individuals can access based on a number of eligibility criteria, notably annually reviewed maximum incomes (Hong Kong Housing Authority (HKHA), 2014). Public rental housing is managed by the Hong Kong Housing Authority and the Hong Kong Housing Society and is rented at reduced rates, with a maximum admissible rent of 10% of household income (Census and Statistics Department (C&SD), 2013). In addition, there are two subsidised home ownership schemes: the ‘Home Ownership Scheme’, which allows low-income households to purchase housing units at discounted house prices, and the recently suspended ‘Tenants’ Purchase Scheme’, which provided for public rental housing tenants to purchase their apartments. Public rental housing constitutes the largest part of Hong Kong’s public housing programme; in 2011, just over 2 million Hong Kong residents (30% of the population) lived in public rental housing and nearly 1.4 million (17%) lived in subsidised home ownership flats (C&SD, 2011b).

With almost half the population in public housing schemes – if subsidised sale flats are included – Hong Kong currently operates the largest public housing scheme of any city in the capitalist world, excluding China (Forrest and Yip, 2014). Single public housing estates can be large, often with more than 5000 units accommodating up to 15,000 residents (HKHA, 2015). The geographic distribution of public housing in Hong Kong is uneven: whereas central and well-served areas of Hong Kong Island are virtually free of public housing, some areas of Kowloon, Tsuen Wan, Kwai Chung, Wong Tai Sin or Kwun Tong and other New Towns in the New Territories have more than 60% of the population living in public housing (Figure 1).

Figure 1.

Hong Kong: Urban context and public housing.

The Government estimates that in 2012 its large-scale public housing programme reduced poverty in Hong Kong by approximately one-fifth, using one particular estimation technique (C&SD, 2013). Yet the health effects of public housing schemes have not been fully investigated, although this may have important implications for the status of public housing in Hong Kong and other prospering Asian cities with increasing linkages to the global economy.

Linking public housing to health in Hong Kong

While there are studies on housing and neighbourhood conditions in Hong Kong, they tend to focus on the social and economic impact on households and residents (e.g. Delang and Lung, 2010; La Grange, 2010; Lui, 2007) and, since very recently, suicides (Hsu et al., 2015). Links between public housing and health and wellbeing have still not been investigated systematically. But given the special role of public housing in Hong Kong’s political economy and historic urban development, an inquiry into health effects could provide lessons for other Asian cities and shed more light on inconclusive evidence gleaned from studies conducted in Western countries. More specifically, it is likely that public housing contributes to the better health and wellbeing of individuals relative to residents in other types of housing, if this factor can be isolated from correlating aspects of different housing types, income, housing instability and neighbourhood resources.

Data and methods

To test the relationship between health disadvantage and public housing, we carried out a small-area study of premature mortality resulting from certain causes, and estimated their associations with public rental housing provision, adjusted for other neighbourhood characteristics that were likely to influence health.

In this first comprehensive small-area study of health in Hong Kong, we drew on a variety of data sources (see below) to measure premature mortality and neighbourhood context.

Mortality data

Premature mortality has been identified as an effective measure of health disadvantage, because it is thought to reflect the cumulative effect of exposure to stressors, including potential neighbourhood or housing influences (Kawachi and Berkman, 2003). Mortality data were obtained from the Hong Kong Census and Statistics Department. We pooled data for 2005–2009 to estimate the five-year small-area Standardised Mortality Ratio (SMR) for premature mortality. This time period included the 2006 by-census year, the most recent period that allowed access to the complete mortality data at the time of the study. In England, premature mortality is defined as mortality before the age of 75 years (Public Health England (PHE), 2014). Given an average life expectancy in England and Wales of 79 for males and 83 for females in 2010 (Office for National Statistics (ONS), 2012), this definition appeared appropriate for Hong Kong too, where in the same year life expectancy at birth was 80 and 86 years for men and women, respectively (Centre for Health Protections, Department of Health, Hong Kong (CHP), 2014). Since we were interested in the long-term effects of public housing and residential environments, we focused on Hong Kong residents and therefore excluded all death records of visitors (people without a Hong Kong residence).

In Hong Kong, the death register holds information on the date of death, sex, age, cause of death, residential area and length of stay in the city, as well as previous country of residence. Information on individual socio-economic background is not collected, hence the analysis of premature mortality could not be carried out in a multi-level model. Multi-level models have the advantage of separating individual effects from area effects (Meijer et al., 2012; Subramanian and Kawachi, 2004), but since we were primarily interested in health in the context of communities, we would argue that a full ecological analysis of premature mortality still reveals overall tendencies as evidence for the existence of health effects of public housing at the aggregate level.

The death register codes causes of death according to the International Classification of Diseases, 10th revision (ICD-10) (WHO, 2012). This information is useful for exploring different patterns for distinct groups of diseases. Previous research on health trends in Hong Kong suggests that different disease groups relate to the social environment in different ways (Lau et al., 2012; Schooling et al., 2010). We therefore found it necessary to disaggregate all-cause mortality by at least the first level logic adopted by the WHO Global Burden of Disease Study (WHO, 2008), which groups deaths into three cause categories: infectious, maternal, perinatal and nutritional conditions (type I), non-communicable conditions (type II) and injuries (type III).

Reference to planning units is included in each death record based on residential address, allowing the calculation of area-based mortality risk. Data at a higher spatial resolution are currently not released because of confidentiality concerns.

Spatial units and area characteristics

The Hong Kong Census Department provided population census counts at the level of Small Tertiary Planning Unit Groups. In 2006, there were 204 units with a median population size of 16,000 residents or 5500 households. The variables extracted from the 2006 Hong Kong By-Census included income, economic activity, education, demographic and household information, rooms per person or housing tenure (see Table 1). Socio-economic variables were selected to capture aspects that have been shown to correlate with health and wellbeing at the neighbourhood level (Congdon, 2010, 2013; Hiscock et al., 2003). We standardised selected variables to relative frequencies using the Location Quotient (LQ), which forms a ratio between counts of a particular category relative to the expected frequency in each planning unit, thus centring area values at the Hong Kong wide average:

{LQ}_{i} = \frac{(n_{i} + 1) / (p_{i} + 1)}{\sum_{i} n_{i} / \sum_{i} p_{i}}

(1)

where $n_{i}$ is the frequency of observed counts in area i and $p_{i}$ is the population of area with 1 added to the denominator in order to avoid LQs of 0. The area LQs were then transformed to z scores to further standardise variables to comparable units (where 0 represents the mean and 1 the value of 1 standard deviation). This procedure was applied to all census variables except income, where the z score of median income was used, and a few other variables including public housing, whose LQ values were log-transformed to meet distributional requirements. We only included public rental flats in our definition of public housing, since flats in both the Home Ownership Scheme and Tenant Purchase Scheme were grouped together under subsidised sale flats in small-area Census tables.¹

Table 1.

Factors describing small-area characteristics, used as control variables in models.

Variables from HK Census Department^a	N’hood affluence	Housing instability	Elderly residents	Social fragmentation (reversed)
Age < 15	−0.055	−0.106	−0.772	0.312
Age ≥ 65	−0.336	0.038	0.809	0.001
Median income^b	0.756	−0.292	−0.118	0.276
High status occupations	0.888	−0.164	−0.150	−0.070
Unemployment	−0.627	0.423	−0.097	−0.078
High education	0.969	−0.208	−0.073	0.064
Single household	−0.025	−0.288	0.240	−0.570
Single elderly household^c	−0.594	0.101	0.335	−0.233
Household size ≥ 5	0.219	−0.254	−0.080	0.910
Rented accommodation	−0.197	0.706	0.042	−0.044
Crowding^b	−0.375	0.906	0.172	0.052

Eigenvalue	3.39	1.82	1.51	1.40
Proportion variance	0.308	0.166	0.137	0.127
Cumulative variance	0.308	0.474	0.611	0.738

Variables from HK Lands Department^e	Centrality (reversed)	Variables from HK Lands Department^f		Land-use intensity

Distance to clinics	0.911	Floor area ratio		0.998
Distance to hospitals	0.752	Surface coverage ratio		0.941
Distance to markets	0.912	Open space ratio		−0.999
Distance to supermarkets	0.934	Net population density		0.593
Distance to recreation centres	0.905
Distance to parks	0.853
Distance to restgardens	0.766
Distance to sports grounds	0.840

Eigenvalue	5.94	Eivenvalue		3.23
Proportion variance	0.742	Proportion variance		0.808

Other variables from HK Census Department
N’hood daily fluctuation^d	Control variable (not used in factor analysis)
Public housing^c	Exposure of interest (not used in factor analysis)

Notes: ^aAll variables transformed to z score of location quotient, except where noted. ^b z score of absolute values. ^c z score of log-transformed location quotient. ^dDefined as people not working in the same neighbourhood. ^eAll variables z score of log-transformed distance value. ^f z-scores of absolute values.

In addition, we considered access to local amenities and services. Only a few studies take into account supply-side determinants of health, although they constitute an important aspect of community deprivation (McLennan et al., 2011). We used geographic point coordinates available for clinics, hospitals, parks, supermarkets and sports grounds, which were obtained from the Hong Kong Lands Department. For each planning unit, we calculated population-weighted road network distances to each amenity.

In addition, we measured an urban design aspect which we found specifically relevant in Hong Kong’s dense urban context: land-use intensity. Based on building footprint data obtained from the Hong Kong Lands Department, we calculated surface area ratio, open space ratio, surface coverage and net population density (census population on built-up land) as recommended by Berghauser-Pont and Haupt (2005).

Spatial structural model

For each small area we determined the observed number of deaths, based on the death register data, and calculated the expected number of deaths, which was derived by multiplying the whole Hong Kong sex- and age-specific mortality rate by the corresponding small-area population data (in 5-year age bands) extracted from the 2006 Hong Kong By-Census. We then derived small-area age-standardised mortality rates (SMRs) by dividing the observed number of deaths by the expected number of deaths.

Small-area estimates are prone to statistical uncertainty owing to small counts and are also subject to spatial autocorrelation: rates in geographically close areas are likely to be related (Best et al., 2005; Congdon, 2012; Elliott and Wakefield, 2001). A widely used model in spatial epidemiology that takes account of these properties is the so-called BYM model, a Bayesian hierarchical model proposed by Besag et al. (1991; applications in e.g. Chang et al., 2011; Cheung et al, 2012; Congdon, 2013). The model estimates area relative risk based on a Poisson regression as follows:

Y_{i} ~ Pois (e_{i} θ_{i})

(2)

\log θ_{i} = μ + x_{i} β + v_{i} + u_{i}

(3)

where $θ_{i}$ is the relative risk of area $i$ , $μ_{i}$ the intercept, $x_{i}$ a vector of covariates, $β$ a vector of regression coefficients, $v_{i}$ an unstructured, normally distributed error component, which accounts for global variability in risk across the whole territory, and $u_{i}$ , a spatially structured component, which accounts for local variability and correlation between neighbouring areas.

Smoothed estimates of small-area SMRs could be calculated using this model, when no covariates are included. Smoothed SMRs can be understood as a weighted average of the observed area SMR, the global mean, and the SMR in neighbouring areas, with weights based on estimated levels of global and local variability. Hong Kong’s fragmented urban form, located on hundreds of islands with dispersed, topographically separated settlements, challenges basic spatial conceptualisations of neighbour-relations. In order to overcome false adjacency based on area boundaries, manual adjustments with ancillary data (road network and ferry connections) were necessary to connect all areas to a single component and permit Bayesian estimation with correct neighbour-relations.

We estimated premature mortality risk ratios for all male and female residents through a series of models: first, a null model of smoothed SMR with no covariates; second, uni-variable models with each of the covariates studied; and third, a full multivariable model including all variables. Since area covariates typically correlate, we ran factor analyses to derive latent variables representing different aspects of area characteristics. We used the Deviance Information Criterion (DIC) and the effective number of parameters (pD) to assess model fit and calculated the additional variance the full models accounted for, compared with the null models with no covariates. In order to test evidence for global spatial patterning of mortality risk, we used the Moran’s I test of spatial autocorrelation.²

The Bayesian estimation was implemented using Integrated Nested Laplace Approximation (INLA), a deterministic alternative to the computationally intensive Multiple Chain Monte Carlo (MCMC) sampling methods (see R Development Core Team 2011; Rue et al., 2009). We used the Open Source R-INLA package (www.r-inla.org) to estimate the models but back checked these INLA-based models by estimating the same model using the software package WinBUGS (Spiegelhalter et al., 1999), which uses the conventional MCMC sampling methods. The results were nearly identical.

Results

Between 2005 and 2009, there were 198,734 mortality cases in Hong Kong. Among them 80,770 (41%) were premature, excluding 1768 cases where information on mortality cause was missing. 83% of premature deaths occurred owing to non-communicable conditions (type II), 8% owing to communicable, maternal, perinatal and nutritional conditions (type I) and 9% owing to injuries (type III) (Table 2). In all types of premature deaths, the male figures were more than double the female, indicating a considerably higher PMR among men.

Table 2.

Premature deaths in Hong Kong 2005–2009.

Premature mortality	All	%	Male	%	Female	%
Type I (infectious, maternal, perinatal and nutritional conditions)	6711	8.3	4838	9.0	1873	6.9
Type II (non-communicable conditions)	66,844	82.8	44,067	82.0	22,777	84.3
Type III (injuries)	7215	8.9	4835	9.0	2380	8.8

All causes	80,770	100.0	53,740	100.0	27,030	100.0

Notes: For classification of mortality into categories of death causes, see WHO 2008.

Source: Authors’ calculations on data from Hong Kong Census and Statistics Department.

Spatial pattern of premature mortality in Hong Kong

Overall, there were significant geographical disparities of PMR in Hong Kong. Two corridors of increased risk of all-cause PMR emerged in Hong Kong’s geography (Figure 2): one running across the Northern New Territories, from Sai Sang Tsuen eastward to Fanling, and another one along central Kowloon, from Sham Shui Po to Jordan. The former high-risk corridor indicates a general health disadvantage in Hong Kong’s northern New Towns. Some of them are directly linked through common development and transport infrastructure: the MTR West Rail Line as well as light rail links serve Tuen Mun, Tin Shui Wai (for Sai Sang Tsuen) and Yuen Long. Fanling and Hang Tau are physically separated from Yuen Long and form their own cluster of high-risk areas. The corridor in central Kowloon stretches along the city’s busy Nathan Road which is characterised by high commercial activity and mixed residential land use. Population and building density are high in these areas, with limited access to open space.

Figure 2.

The spatial pattern of all-cause PMR in Hong Kong.

Sham Shui Po and some of the northern New Towns had a PMR of more than 1.5 the Hong Kong average. West Kowloon and most planning units of Hong Kong and Lantau Island were below the Hong Kong average. Most of the high-risk areas showed a risk ratio significantly above the Hong Kong average: in the New Towns Tuen Mun, Yuen Long and Fanling, the Kowloon neighbourhoods around Sham Shui Po, as well as the areas on Hong Kong Island, the probability that type I PMR risk exceeded 1 was above 0.95. This indicator is a way of measuring statistical confidence within the spatial structural model.

In order to obtain a more robust estimate of the magnitude of difference, we calculated the risk difference between the 5th and 95th risk percentile of areas. For all-cause premature mortality, disparities in risk were wide: they ranged from 0.13 to 6.06 (crude) and from 0.31 to 5.44 (smoothed), with a 3.47-fold relative risk difference between the 95th and the 5th percentiles. This difference reflects an absolute risk difference of 259 cases (364 minus 105) per 100,000 person-years of residents younger than 75 years.

The geography and magnitude of health disparities differed by type of mortality cause (see Appendix for cause-wise maps). Crude risk ratios for type I causes (communicable, maternal, perinatal and nutritional diseases) ranged from 0.10 to 11.1 in their extremes (excluding observed counts of 0). This difference in magnitude reduced to 0.40 and 5.13 after smoothing. For type I causes, the absolute risk difference was 23 (34 minus 11) cases per 100,000 person-years or 3.04 in relative terms. Between the 5th and 95th risk percentile of areas, we found a more than three-fold difference in type I PMR.

There was statistical evidence for weak spatial autocorrelation in mortality risk for type I causes, with a Moran’s I of 0.95 (p = 0.011). Areas associated with high risk were concentrated in the north, east, and south coast of Hong Kong Island, Kowloon, and central areas in the North Territories. In contrast, areas of low risk were located in Nam Cheong, West Kowloon, Sha Tin and on Lantau Island.

PMR resulting from non-communicable conditions (type II) ranged from 0.14 to 6.18 (crude relative risk) or 0.33 to 5.33 (smoothed relative risk) between planning units in Hong Kong. The 95th versus 5th percentile ratio was 3.2 for the smoothed values, reflecting an absolute risk difference of 203 (295 minus 92) premature deaths per 100,000 residents after smoothing. Type II PMR closely mirrored the geography of all-cause PMR since, as shown above, the type II premature mortality cases constitute 83% of all cases. With no statistical evidence of clustering, areas associated with high risk were scattered across the whole territory: in central parts of the city Kowloon, as well as Yuen Long, Tai Po and Sai Kung Town in the New Territories. Areas of low risk were located across northern Hong Kong Island, west Kowloon and some parts of the New Territories. Yuen Long and San Sang Tsuen in the New Territories, and Sham Shui Po in north west Kowloon emerged as areas of increased risk with risk ratios between 1.6 and 1.9. On Hong Kong Island too we found a few areas where residents are more likely to die prematurely because of non-communicable conditions. The high-risk areas covered larger housing estates in Aberdeen and Chai Wan, as well as dense and high-activity neighbourhoods adjacent to Hong Kong’s business district.

PMR resulting from injuries (type III) also ranged widely from 0.12 to 3.45 in crude terms; this difference narrowed to 0.32 and 2.04 after smoothing. The observed 95th versus 5th percentile ratio of 2.79 corresponds to an absolute risk difference of 21 (32 minus 11) premature deaths per 100,000 person-years.

While overall spatial clustering of type III PMR was insignificant, a low degree of spatial concentration could be observed around the new towns of Tuen Mun, Yuen Long and certain northern neighbourhoods in Kowloon. The high-risk corridor, starting in Kowloon’s Sham Shui Po and in the Western New Territories around Tuen Mun, appeared again for type III PMR. Here, all types of health disadvantage coincided. Among the northern New Towns, only Tuen Mun showed increased risk, while the remaining ones did not indicate high type III PMR risk.

Associations of PMR and public housing

Table 1 shows all the factors derived from the data described above. The four factors that could be identified from the Census were neighbourhood affluence, housing instability, elderly residents and social fragmentation, based on the loadings of individual variables. Road distance to a variety of services correlated very strongly and formed a single factor that can be interpreted – in reversed form – as centrality of areas. Similarly, the variables measuring land-use intensity correlated and formed another single factor: the higher the value, the greater the intensity of land use.

An additional variable describing neighbourhood context – neighbourhood daily fluctuation – did not load significantly in factor solutions with other census variables and the variable was therefore separated from factor analysis. All factors and derived variables, along with public housing as exposure of interest, were included in Bayesian models. We first ran uni-variable (unadjusted) models for each covariate and compared them with the fully adjusted multivariable models of all-cause PMR and cause-specific PMRs. We did this for the whole population as well as the female and male subpopulations. Table 3 shows the unadjusted and adjusted results for cause-specific PMRs among women and men.

Table 3.

Contextualisation of premature mortality by type of risk in Hong Kong (2005–2009).^a

		Female unadjusted			Adjusted			Male unadjusted			Adjusted
		Mean	Lower	Upper	Mean	Lower	Upper	Mean	Lower	Upper	Mean	Lower	Upper
	TYPE I PMR	(n = 1873)						(n = 4838)
1	N’hood affluence	−0.140	−0.241	−0.043	−0.234	−0.359	−0.106	−0.205	−0.315	−0.097	−0.283	−0.423	−0.140
2	Housing instability	0.032	−0.045	0.110	−0.015	−0.139	0.110	0.071	−0.016	0.160	0.054	−0.099	0.211
3	Elderly residents	0.115	0.018	0.211	0.099	0.008	0.189	0.046	−0.061	0.150	−0.003	−0.107	0.098
4	Social fragmentation	0.100	−0.005	0.207	0.055	−0.062	0.173	0.079	−0.026	0.184	0.002	−0.115	0.119
5	Land-use intensity	0.127	0.007	0.249	0.054	−0.110	0.216	0.087	−0.032	0.209	−0.062	−0.255	0.129
6	Centrality	0.207	0.083	0.332	0.146	−0.030	0.324	0.200	0.083	0.322	0.246	0.052	0.444
7	N’hood fluctuation	−0.036	−0.117	0.044	0.064	−0.048	0.178	−0.085	−0.174	0.003	0.059	−0.066	0.184
8	Public housing	0.015	−0.048	0.079	−0.026	−0.116	0.065	0.021	−0.055	0.099	−0.107	−0.225	0.008
9	Intercept				−0.100	−0.183	−0.021				−0.152	−0.232	−0.074

	TYPE II PMR	(n = 22,777)						(n = 44,067)
1	N’hood affluence	−0.082	−0.139	−0.026	−0.129	−0.207	−0.051	−0.169	−0.246	−0.095	−0.183	−0.273	−0.092
2	Housing instability	0.089	0.039	0.140	0.150	0.056	0.246	0.147	0.083	0.211	0.167	0.056	0.279
3	Elderly residents	0.018	−0.042	0.077	0.003	−0.055	0.061	0.049	−0.027	0.125	0.018	−0.049	0.084
4	Social fragmentation	0.047	−0.010	0.105	0.031	−0.031	0.093	0.031	−0.040	0.102	−0.021	−0.089	0.046
5	Land-use intensity	0.043	−0.020	0.109	−0.114	−0.224	−0.004	0.053	−0.028	0.138	−0.189	−0.316	−0.063
6	Centrality	0.127	0.060	0.197	0.161	0.052	0.272	0.195	0.114	0.280	0.273	0.149	0.399
7	N’hood fluctuation	−0.038	−0.088	0.013	0.115	0.041	0.188	−0.118	−0.181	−0.056	0.064	−0.020	0.148
8	Public housing	0.043	−0.002	0.089	−0.064	−0.133	0.005	0.088	0.028	0.149	−0.098	−0.182	−0.014
9	Intercept				−0.115	−0.161	−0.070				−0.160	−0.214	−0.108

	TYPE III PMR	(n = 2380)						(n = 4835)
1	N’hood affluence	−0.188	−0.289	−0.094	−0.328	−0.439	−0.216	−0.311	−0.411	−0.219	−0.404	−0.513	−0.293
2	Housing instability	0.101	0.026	0.178	0.123	0.006	0.242	0.129	0.049	0.211	0.100	−0.020	0.222
3	Elderly residents	0.133	0.043	0.223	0.078	−0.006	0.161	0.217	0.119	0.313	0.149	0.071	0.226
4	Social fragmentation	0.128	0.030	0.227	0.066	−0.043	0.175	0.113	0.012	0.216	0.108	0.011	0.207
5	Land-use intensity	0.175	0.069	0.287	0.035	−0.112	0.182	0.070	−0.042	0.190	−0.025	−0.172	0.121
6	Centrality	0.273	0.159	0.391	0.164	0.005	0.325	0.199	0.089	0.314	0.134	−0.012	0.281
7	N’hood fluctuation	−0.098	−0.176	−0.022	0.076	−0.026	0.178	−0.168	−0.249	−0.088	0.071	−0.027	0.169
8	Public housing	0.006	−0.055	0.068	−0.140	−0.224	−0.056	0.092	0.021	0.163	−0.067	−0.156	0.021
9	Intercept				−0.145	−0.223	−0.071				−0.180	−0.246	−0.115

		Female adjusted (vs. unadjusted)						Male adjusted (vs. unadjusted)

	Model fits^b TYPE I	ΔDIC=	−1.3	ΔpD=	−18.8	Var=	0.685	ΔDIC=	3.1	ΔpD=	−6.27	Var=	0.256
	TYPE II	ΔDIC=	−2.8	ΔpD=	−5.41	Var=	0.218	ΔDIC=	4.9	ΔpD=	−6.0	Var=	0.350
	TYPE III	ΔDIC=	−21.8	ΔpD=	−28.4	Var=	0.756	ΔDIC=	−6.5	ΔpD=	−30.6	Var=	0.612

Notes: ^aAuthors’ calculations based on data from Hong Kong Census & Statistics Department – Vital Events Register. ^bModel fits were assessed by comparison with models without covariates (ΔDIC = DIC difference, ΔpD = difference of effective number of parameters) and estimation of variance accounted for (Var).

The correlation of area characteristics with PMR differed by type of PMR and between women and men. We identified significant associations as those which excluded 0 from the coefficients’ lower and upper limits – known as credible intervals – which is measured as the range between the 2.5th and 97.5th percentile of the distribution of coefficient values.

Neighbourhood affluence was inversely associated with PMR in all models, i.e. the higher the affluence of a neighbourhood, the lower the risk of dying prematurely regardless of mortality cause. The intensity of the association as measured by the regression coefficient increased after all other variables were accounted for. For instance, the mean correlation coefficient in type I PMR for women is 0.14, which indicates that if we were to increase neighbourhood affluence by one standard deviation, we would observe a reduction of PMR of 13%. After model adjustment, this crude effect rose to 21%.³ These associations in all PMR were stronger for men than for women, which may reflect the higher life expectancy of women.

As for cause-specific models for type I (communicable) PMR, the older age composition of neighbourhoods was associated with increased risk for women. Centrality – the proximity to services including hospitals – was positively associated with type I PMR for both women and men, although the credible intervals for women just included 0. Public housing was not significantly associated with type I PMR. Yet, the direction switched from a positive association to a negative one after model adjustment. The association was almost significant for men, i.e. the credible interval stretched very little beyond 0.

As for type II (non-communicable) PMR, housing instability (which included crowding) emerged as a significantly positive covariate. As with neighbourhood affluence, the association intensified after model adjustment. For public housing, we again observed a switch in directions: the presence of public housing was associated with increased type II PMR, but after controlling for all other area characteristics, notably neighbourhood affluence and housing instability, the association reversed. For men, both of these associations were significant; for women they seemed more moderate yet nearly significant. The negative association between public housing and type II PMR in the entire population – i.e. women and men combined – reached a value of 0.085 (not shown in table). The daily fluctuations in a neighbourhood were correlated with type II PMR among women: the higher the share of residents working outside the neighbourhood, the higher the incidence of type II PMR. Land-use intensity showed the opposite trend to centrality: the higher the land use, the lower the incidence of type II PMR.

Type III (injuries) PMR, too, was higher in less affluent and more central areas. Injury risk was increased in areas with more elderly residents. Among men, there was a higher type III PMR in more socially fragmented areas. Again, public housing was inversely correlated with risk only after adjustment. This time, the inverse relationship turned out to be significant for women but not for men, while the reverse applies for the unadjusted association. The small number of premature mortality cases may not allow for definitive conclusions, but the general trends observed with other types of PMR can still be identified. This becomes evident when combining men and women and looking at the entire population: here, both the unadjusted, positive association and the adjusted inverse association are significant (not shown in table).

When all mortality cases regardless of cause were considered, the observed associations re-appeared and closely followed the pattern for type II cases, since these constituted 83% of cases. In general, it could be observed that public housing was significantly, positively associated with most types of PMR for both women and men in the unadjusted models, and inversely associated with PMR after full adjustment. The coefficient for public housing for type II PMR in the entire population was −0.085.

Discussion

Public housing as a social determinant of health in Hong Kong and beyond

In our assessment of geographic health disparities in Hong Kong, we identified clusters of small areas that experience increased premature mortality risk across different categories of causes. Some of these areas have a high share of public housing – indeed, they were developed with this as a central element of planning – and thus suggest a positive link between public housing and mortality risk. But by disaggregating the individual parameters normally conflated as being representative of public housing clusters – deprivation, poor housing conditions and social renting – our analysis revealed different undercurrents in the relationship between public housing and health. The appearance of a positive association between public housing and risk in unadjusted models and their consistent reversal in adjusted models may offer some explanation for the inconsistency of findings in the literature.

While our analysis supports the hypothesis that public housing contributes to health, our study also provides empirical support – at the aggregate level – for Lee and Yip’s (2006) continued asset role of public housing. Drawing on our findings and research by others on housing in Hong Kong, we suggest four potential pathways linking public housing and health.

First, it is possible to view access to public housing as an in-kind benefit that represents a transfer from market-level rents to subsidised rents (Lui, 2007). This indirectly increases households’ disposable income and potentially expands material assets relevant to health and wellbeing. Indeed, in Hong Kong, private renters have been subjected to rent increases in the past decade while public housing tenants never pay more than 10% of their income. In private housing it can be up to 50% (C&SD, 2011a). The disposable incomes among low-income groups not living in public housing has therefore diminished relative to those living in public housing.

The income disadvantage of private tenants increases even more when one considers that the government waives the rent of public housing tenants for up to three months for families in receipt of additional social benefits (under a scheme known as Comprehensive Social Security Assistance). In addition, as Yip et al. (2009) note, the public-assisted home ownership scheme has enabled households to accumulate capital directly within the public housing sector. Although we did not explicitly include subsidised sale flats, Home Ownership Scheme flats are indirectly included in our analysis through their co-varying location quotient. Households who own their property under the Home Ownership Scheme can re-sell their properties to households that are eligible for public housing or they can pay the land premium cost in selling the flat in an open market.

Second, in Hong Kong, public housing estates are usually maintained and managed by the Housing Authority to high functional and sanitary standards, particularly in newly developed estates in suburban areas such as the northern districts in the New Territories. The inferior quality of the low-end private rental market may trigger material pathways between housing conditions and health disadvantage, resulting in health advantages among public housing tenants as revealed by the models. Minimal space standards in public housing compared with often overcrowded private flats, which include subdivided and cage dwellings, may further contribute to health disadvantage in the private rental sector.

The positive associations between public housing and health disadvantage found in Western studies may indeed result from poorer conditions of public housing estates, the reverse of the Hong Kong situation. The evidence from Hong Kong suggests that for public housing to have positive impacts on its residents, it is not sufficient to simply provide the buildings and infrastructure; the public housing stock must also be maintained and improved in terms of safety, hygiene, environmental and structural conditions and quality of maintenance – an important observation in the context of current debates on the funding and management of public housing in some Western countries (Doran et al., 2013; Fenton et al., 2013; Hamnett, 2010).

Third, the widespread presence of large units of public housing in central and high-quality locations, alongside the targeted upgrading of older building stock, reduces the stigma that social tenants typically experience in some Western contexts (Delang and Lung, 2010; Forrest and Yip, 2014; Li, 2005). Housing and location-related stigma may limit life chances and may activate psycho-social pathways in health (Anthony and Robbins, 2012). In Hong Kong, these pathways are likely to be attenuated by the relatively high estimation in which public housing is held – if they materialise at all. The additional psychological sense of security afforded by guaranteed access to public housing may further reinforce the ability of households to cope with social stress.

Fourth, there is evidence that social cohesion is relatively high in public housing estates in Hong Kong (La Grange, 2010), which again contrasts with the experience in many Western public housing projects. Hong Kong’s prioritisation of the family unit through its housing programmes – applicants who live with their families are given preferential treatment in the allocation and terms of their tenancies – may further contribute to greater social cohesion. Hence living in public housing offers possibilities for improved wellbeing through familial and social support and represents a significant advantage over the low-end private renting sector.

Taken together, the Hong Kong case offers particular insights for rapidly growing global cities into how strongly competing interests can be managed in a context where land is scarce and housing demand is high. The city’s longstanding commitment and continued investment in the provision of secure and stable access to public housing and public transport are likely to have greater impacts on shaping the social wellbeing and improving the life chances – and consequently physical and mental wellbeing – of a large proportion of its population. Testing whether this association can be further substantiated in Hong Kong will provide important lessons for other cities in Asia and the West.

The role of urban context

There are a few unexpected results that deserve further discussion and further study. Our results indicate that land-use intensity is associated with lower premature mortality risk because of non-communicable diseases. Land-use intensity is high in the central and commercial areas in both Hong Kong Island and Kowloon. While these areas have a better supply of services not directly included in the centrality PCA factor (such as primary healthcare), the result raises intriguing questions. Since other factors already reflect deprivation and accessibility and are thus controlled for, there appears to be an independent association of density with better health. Here, we may be observing the effects of regeneration carried out by Hong Kong’s Urban Renewal Authority in central areas, encouraging inner city gentrification that displaces elderly and more deprived residents and attracts wealthier and healthier ones.

The centrality factor (as derived from PCA), on the other hand, is consistently and significantly associated with higher premature mortality risk. Because centrality implies better access to a range of services, including shops, professional services, community facilities, hospitals and clinics, and leisure facilities, we are probably observing another selection effect. Wealthier people – and possibly less healthy people at later life stages – move to central areas in order to be closer to the services they need to support them. Depending on their individual income and material circumstances, these more fragile residents end up living in smaller, more cramped and overcrowded accommodation, making trade-offs between the amount and quality of space versus access to services. This pattern would explain the independent contribution of centrality in fully adjusted models.

We also observed that neighbourhoods experiencing a high level of commuting are associated with health disadvantage. These areas can be found across the urban region, both in peripheral locations and the New Territories, but also in central areas of Hong Kong Island. Given that commuting has significant effects on residents’‘time budget’, the length of commuting time may reduce opportunities for relaxation, sleep and recovery, which in turn compromises mental and physical wellbeing.

Since the data suggest that the association between neighbourhood fluctuations and health applies predominantly to women, the impact of commuting on time budgets cannot be the only explanation. While the work participation rate of women has risen in the last decade, women have a range of additional tasks in relation to household, family and child-raising. The difficult task of balancing private and work commitments may translate into lifestyle patterns that have a particular health impact on women. A highly fluctuating neighbourhood would then be a quasi-biographical indicator of lifestyle patterns, if premature cases have occurred during economic activity or if women have remained in their neighbourhood after retirement. While this may be a plausible explanation in the Hong Kong context, this is a rather speculative point with some risk of ecological fallacy. Yet future exploration of this point may be highly relevant for questions about Hong Kong’s working culture, gender roles and their impact on health and wellbeing.

Conclusions

Study limitations and avenues for future research

Although one may be tempted to attribute causality to observed associations, it should be remembered that this ecological analysis represents a contextualisation rather than an explanation of risk. The cross-sectional design of our study further challenges our ability to identify causal relationships with certainty.

As with any ecological, cross-sectional analysis, there may be a risk of residual confounding – that is, the observed associations may be caused by unobserved covariates that are directly or indirectly associated with public housing. For this reason we have drawn on diverse data sources to include a wide range of area characteristics. Using factor analysis, we sought to avoid colinearity among inter-correlated variables and reveal small-area characteristics that are specific to Hong Kong and yet comparable with neighbourhood effect studies carried out by others.

Since, conceptually speaking, health is an outcome at the individual level, approaches such as multi-level models would be useful in disentangling individual characteristics from contextual information (Kawachi and Berkman, 2003). But Hong Kong’s mortality register does not provide much detail about individual cases beyond age, sex and residential area; and therefore a purely ecological study seemed appropriate at this stage. In addition, many public housing estates are and often accommodate the majority, if not the entirety, of the population of a spatial unit, thus reducing the risk for ecological fallacy.

Despite these shortcomings, it is our view that the study succeeds in highlighting – for the first time in Hong Kong – markers of evidence as a starting point for more focused, aetiological research of housing, neighbourhoods and health under the four potential pathways set out above. Looking ahead to how further applied work might build on our study, we consider that a geographically focused evaluation of Hong Kong’s housing policies, similar to the work carried out by Lee and Yip (2006) or La Grange (2010), may shed some light on questions concerning whether state-subsidised housing contributes to the higher availability of assets relevant to health and wellbeing, fosters a higher level of social cohesion or moderates the health consequences of people at more disadvantaged levels of society.

Future research should comprise not only studies that draw on a wider range of data at a higher spatial resolution (as they emerge), but also targeted neighbourhood studies that apply a mix of quantitative and qualitative methods to further investigate the history and trends of residents’ health and wellbeing in high-risk areas; it should also instigate joint community and government interventions, such as those proposed by Corburn (2009). In this sense, in addition to informing urban policy substantively, our study highlights geographically varying challenges and localised information needs that can be addressed effectively at a strategic level.

Final remarks

Despite the known limitations of cross-sectional analysis, our study reveals the characteristics of the intra-urban geography of health disadvantage in Hong Kong and discusses the role of a central instrument of urban policy: public housing. We thus hope to contribute to the literature on housing and health, not least by expanding the geographical reach of this field of inquiry. The locally specific guise of pathways linking public housing and health in Hong Kong may be manifest in a larger group of Asian cities, where living in public housing becomes a crucial determinant of urban health and wellbeing in the context of increasing urban inequalities, pressure on public services, and globalised housing markets.

Footnotes

Appendix

Acknowledgements

The authors would like to thank Myfanwy Taylor and Antoine Paccoud for their comments and support during the course of this study as well as Alexandra Gomes for her feedback and Rachel Lewis for proof-reading. The paper also benefitted from Frances Law’s review and suggestions. We would like to express our gratitude to the three anonymous reviewers for their extremely useful comments.

Funding

The research was conducted at LSE Cities as part of the Urban Age conference programme, supported by Alfred Herrhausen Society, the International Forum of Deutsche Bank. S-S.C. was supported by the Hong Kong Research Grants Council General Research Fund (HKU784210M and HKU784012M) and the Chiang Ching-kuo Foundation, Taiwan (grant number RG014-P-12). The project was also in part supported by the Chief Executive Community project and the Hong Kong Jockey Club Charities Trust.

Notes

References

Anthony

Robbins

(2012) A latent class analysis of resilient development among early adolescents living in public housing. Children and Youth Services Review 35(1): 82–90.

Baker

Bentley

Mason

(2013) The mental health effects of housing tenure: Causal or compositional? Urban Studies 50(2): 426–442.

Berhauser-Pont

Haupt

(2005) The Spacemate: Density and the typomorphology of the urban fabric. Nordisk Arkitekturforskning 18(4): 55–68.

Besag

York

Mollié

(1991) Bayesian image restoration with two applications in spatial statistics. Annals of the Institute of Statistics and Mathematics 43: 1–20.

Best

Richardson

Thomson

(2005) A comparison of Bayesian spatial models for disease mapping. Statistical Methods in Medical Research 14(1): 35–59.

Burgard

Seefeldt

Zelner

(2012) Housing instability and health: Findings from the Michigan Recession and Recovery Study. Social Science & Medicine 75(12): 2215–2224.

Burrows

(2003) Poverty and Home Ownership in Contemporary Britain. Bristol: The Policy Press.

Census and Statistics Department (C&SD) (2011a) 2009/10 Household Expenditure Survey and the Rebasing of the Consumer Price Indices. Census and Statistics Department, Hong Kong Special Administrative Region, People’s Republic of China.

Census and Statistics Department (C&SD) (2011b) 2011 Population Census: Key Statistics. Census and Statistics Department, Hong Kong Special Administrative Region, People’s Republic of China.

10.

Census and Statistics Department (C&SD) (2013) Hong Kong Poverty Situation Report 2012, Hong Kong. Census and Statistics Department, Hong Kong Special Administrative Region, People’s Republic of China.

11.

Centre for Health Protection, Department of Health, Hong Kong (CHP) (2014) Life expectancy at birth (female and male), 1971–2012. Available at: http://www.chp.gov.hk/en/data/4/10/27/111.html (accessed January 2014).

12.

Chang

Sterne

Wheeler

. (2011) Geography of suicide in Taiwan: Spatial patterning and socioeconomic correlates. Health & Place 17: 641–650.

13.

Cheung

Spittal

Pirkis

. (2012) Spatial analysis of suicide mortality in Australia: Investigation of metropolitan-rural-remote differentials of suicide risk across states/territories. Social Science and Medicine 75(8): 1480–1488.

14.

Congdon

(2010) The spatial pattern of suicide in the US in relation to deprivation, fragmentation and rurality. Urban Studies 48(10): 2101–2122.

15.

Congdon

(2012) Modelling small-area inequality in premature mortality using years of life lost rates. Journal of Geographical Systems 15(2): 149–167.

16.

Congdon

(2013) Assessing the impact of socioeconomic variables on small area variations in suicide outcomes in England. International Journal of Environmental Research and Public Health 10: 158–177.

17.

Corburn

(2009) Toward the Healthy City. Cambridge, MA; London: The MIT Press.

18.

Dalstra

JAA

Kunst

Mackenbach

(2006) A comparative appraisal of the relationship of education, income and housing tenure with less than good health among the elderly in Europe. Social Science & Medicine 62(8): 2046–2060.

19.

Delang

Lung

(2010) Public housing and poverty concentration in urban neighbourhoods: The case of Hong Kong in the 1990s. Urban Studies 47(7): 1391–1413.

20.

Doran

Misa

Sha

(2013) Housing as health care – New York’s boundary-crossing experiment. The New England Journal of Medicine 369(25): 2374–2377.

21.

Dunn

Hayes

(2000) Social inequality, population health, and housing: A study of two Vancouver neighborhoods. Social Science & Medicine 51(4): 563–587.

22.

Ellaway

Macintyre

(1998) Does housing tenure predict health in the UK because it exposes people to different levels of housing related hazards in the home or its surroundings? Health & Place 4(2): 141–150.

23.

Elliott

Wakefield

(2001) Bias and confounding in spatial epidemiology. In: Elliott

Wakefield

Best

. (eds) Spatial Epidemi ology: Methods and Applications. Oxford: Oxford University Press. Available at: http://www.oxfordscholarship.com/view/10.1093/acprof:oso/9780198515326.001.0001/acprof-9780198515326-chapter-5 (accessed November 2015).

24.

Fenton

Lupton

Arrundale

. (2013) Public housing, commodification, and rights to the city: The US and England compared. Cities 35: 373–378.

25.

Forrest

Yip

N-M

(2014) The future for reluctant intervention: The prospects for Hong Kong’s public rental sector. Housing Studies 29: 551–565.

26.

Forrest

La Grange

Yip

(2004) Hong Kong as a global city? Social distance and spatial differentiation. Urban Studies 41(1): 207–228.

27.

Gibson

Petticrew

Bambra

. (2011) Housing and health inequalities: A synthesis of systematic reviews of interventions aimed at different pathways linking housing and health. Health & Place 17(1): 175–184.

28.

Global Research Network for Urban Health Equity (GRNUHE) (2010) Improving Urban Health Equity Through Action on the Social and Environmental Determinants of Health. Final report of the Global Research Network on Urban Health Equity, London.

29.

Govt HK (2013) Hong Kong Poverty Situation Report 2012. Hong Kong: Government of the Hong Kong Special Administrative Region.

30.

Graham

Manley

Hiscock

. (2009) Mixing housing tenures: Is it good for social well-being? Urban Studies 46(1): 139–165.

31.

Gusmano

Rodwin

Weisz

(2010) Health Care in World Cities. Baltimore, MD: The Johns Hopkins University Press.

32.

Hamnett

(2010) Moving the poor out of central London? The implications of the coalition government 2010 cuts to Housing Benefits. Environment and Planning A 42(12): 2809–2819.

33.

Hartig

Fransson

(2006) Housing tenure and early retirement for health reasons in Sweden. Scandinavian Journal of Public Health 34(5): 472–479.

34.

Hiscock

Macintyre

Kearns

. (2003) Residents and residence: Factors predicting the health disadvantage of social renters compared to owner-occupiers. Journal of Social Issues 59(3): 527–546.

35.

Hong Kong Housing Authority (HKHA) (2014) Annual Report 2013/14. Hong Kong: The Hong Kong Housing Authority.

36.

Hong Kong Housing Authority (HKHA) (2015) Complete List of PRH/TPS Estates. The Hong Kong Housing Authority. Available at: housingauthority.gov.hk/en/global-elements/estate-locator/completeList.html (accessed January 2015).

37.

Hsu

Chang

Lee

. (2015) Geography of suicide in Hong Kong: Spatial patterning, and socioeconomic correlates and inequalities. Social Science & Medicine 130C: 190–203.

38.

Jacobs

(2011) Environmental health disparities in housing. American Journal of Public Health 101(Suppl 1): S115–S122.

39.

Kawachi

Berkman

(2003) Neighborhoods and Health. Oxford, New York: Oxford University Press.

40.

Krieger

Higgins

(2002) Housing and health: Time again for public health action. American Journal of Public Health 92(5): 758–768.

41.

La Grange

(2010) Neighbourhood and class: A study of three neighbourhoods in Hong Kong. Urban Studies 48(6): 1181–1200.

42.

Lai

LWC

(1993) Hong Kong’s density policy towards public housing: A theoretical and empirical review. Third World Planning Review 15(1): 63–85.

43.

Lai

LWC

Baker

(2014) The final colonial regional plan that lingers on: Hong Kong’s Metroplan. Habitat International 41: 216–228.

44.

Lau

Schooling

Tin

. (2012) Income inequality and cause-specific mortality during economic development. Annals of Epidemiology 22(4): 285–294.

45.

Lawder

Walsh

Kearns

. (2014) Healthy mixing? Investigating the Associations between neighbourhood housing tenure mix and health outcomes for urban residents. Urban Studies 51(2): 264–283.

46.

Lee

Yip

N-M

(2006) Public housing and family life in East Asia: Housing history and social change in Hong Kong, 1953–1990. Journal of Family History 31(1): 66–82.

47.

Lee

Wong

Law

(2007) Social polarisation and poverty in the global city: The case of Hong Kong. China Report 43(1): 1–30.

48.

Leech

TGJ

(2010) Subsidized housing, public housing, and adolescent violence and substance use. Youth & Society 44(2): 217–235.

49.

(2005) The impact of social stigma: An examination of the public and private housing markets in Hong Kong. The Appraisal Journal 73(3): 305–317.

50.

Lui

(2007) The redistributive effect of public housing in Hong Kong. Urban Studies 44(10): 1937–1952.

51.

Macintyre

Ellaway

Hiscock

. (2003) What features of the home and the area might help to explain observed relationships between housing tenure and health? Evidence from the west of Scotland. Health & Place 9(3): 207–218.

52.

McLennan

Barnes

Noble

. (2011) The English Indices of Deprivation 2010. London: Department for Communities and Local Government.

53.

Marmot

(2010) Fair Society, Healthy Lives. London: The Marmot Review.

54.

Marmot

(2005) Social determinants of health inequalities. The Lancet 365: 1099–1104.

55.

Marmot

Wilkinson

(2005) Social Organization, Stress, and Health. In: Social Determinants of Health. Oxford: Oxford University Press, pp. 6–30.

56.

Meijer

Röhl

Bloomfield

. (2012) Do neighborhoods affect individual mortality? A systematic review and meta-analysis of multilevel studies. Social Science & Medicine 74(8): 1204–1212.

57.

Mueller

Tighe

(2007) Making the case for affordable housing: Connecting housing with health and education outcomes. Journal of Planning Literature 21(4): 371–385.

58.

Office for National Statistics, UK (ONS) (2012) Mortality in England and Wales: Average Life Span. London: The National Archives. Available at: http://www.ons.gov.uk/ons/dcp171776_292196.pdf.

59.

Peng

Wheaton

(1993) Effects of restrictive land supply on housing in Hong Kong: An econometric analysis. Journal of Housing Research 5: 263–291.

60.

Pollack

Griffin

Lynch

(2010) Housing affordability and health among homeowners and renters. American Journal of Preventive Medicine 39(6): 515–521.

61.

Public Health England (PHE) (2014) Longer Lives. Available at: http://longerlives.phe.org.uk/ (accessed January 2014).

62.

R Development Core Team (2011) R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing.

63.

Rue

Martino

Chopin

(2009) Approximate Bayesian inference for latent Gaussian models using integrated nested Laplace approximations. Journal of the Royal Statistical Society B71(2): 319–392.

64.

Sassen

(2001) The global city: New York, London, Tokyo. Heredity 93(1): 447.

65.

Schooling

Lau

EWL

Keith

YKT

. (2010) Social disparities and cause-specific mortality during economic development. Social Science & Medicine 70: 1550–1557.

66.

Spiegelhalter

Thomas

Best

(1999) WinBUGS Version 1.2 User Manual. MRC Biostatistics Unit.

67.

Stephens

(2011) Revisiting urban health and social inequalities: The devil is in the detail and the solution is in all of us. Environment and Urbanization 23(1): 29–40.

68.

Subramanian

Kawachi

(2004) Income inequality and health: What have we learned so far? Epidemiologic Reviews 26(1): 78–91.

69.

Tai

P-F

(2006) Social polarisation: Comparing Singapore, Hong Kong and Taipei. Urban Studies 43(10): 1737–1756.

70.

Taylor

Pevalin

Todd

(2006) The psychological costs of unsustainable housing commitments. ISER Working Paper, Insititute for Social & Economic Research.

71.

Windle

Burholt

Edwards

(2006) Housing related difficulties, housing tenure and variations in health status: Evidence from older people in Wales. Health & Place 12(3): 267–278.

72.

World Health Organization (WHO) (2008) The Global Burden of Disease. 2004 Update. Geneva: WHO Library Cataloguing-in-Publication Data.

73.

World Health Organization (WHO) and UN-HABITAT (2010) Unmasking And Overcoming Health Inequities In Urban Settings. Geneva: World Health Organization and UN-HABITAT.

74.

World Health Organization (WHO) (2012) ICD-10 Version: 2010. Available at: http://apps.who.int/classifications/icd10/browse/2010/en.

75.

Yip

N-M

La Grange

Forrest

(2009) Making space: Residential trajectories in Hong Kong. Urban Geography 30(7): 756–778.