Ecological Economics 167 (2020) 106401

Contents lists available at ScienceDirect

Ecological Economics
journal homepage: www.elsevier.com/locate/ecolecon

Methodological and Ideological Options

Exposure to green areas: Modelling health benefits in a context of study T

heterogeneity
⁎
Aline Chiabaia, , Sonia Quirogab, Pablo Martinez-Juarezb, Cristina Suárezb,
Silvestre García de Jalóna, Tim Taylorc
a
Basque Centre for Climate Change (BC3), Leioa, Spain
b
Department of Economics, Universidad de Alcalá, Spain
c
European Centre for the Environment and Human Health, University of Exeter Medical School, United Kingdom

A R T I C LE I N FO A B S T R A C T

Keywords: Although the beneficial health effects of green areas are gaining recognition, epidemiological studies show
Green areas mixed results with significance varying considerably by study and context, indicating that there is no unique and
Health risk reduction clear evidence. This relationship is influenced by multiple factors and characterised by high complexity not
Heckman Selection Model previously been incorporated in one single analysis. This study proposes a new application of the Heckman
Contextual factors
selection model to find evidence of key patterns emerging throughout the literature and identify main de-
Health inequalities
Adaptation
terminants affecting the relationship. The model aggregates outcomes of different studies and allows an as-
sessment of both significant and non-significant results from the literature in order to correct for unobserved
selection bias. Close attention is paid to the relevance of the background, particularly socioeconomic context.
The results show significant health benefits associated with increased exposure to green areas, where higher risk
reductions are observed for old and adult age groups, as well as in poorer countries, taking into account the
correction for the publication bias. This last issue points towards a redistributive impact of green areas in terms
of health and the importance of co-benefits arising from Ecosystem-based Adaptation, especially in poorer
neighbourhoods, translating in health care savings and reduced productivity loss.

1. Introduction and background Environmental factors pose serious risks to human health, as it is esti-
mated that 24% of the global burden of disease is attributable to en-
One of the trends constantly present during most of recorded history vironmental hazards (WHO, 2006), including air and water quality,
has been the increase in the population living in cities. Urbanization is a land use and urban design. Contacts with healthy environments are
process that has also accompanied industrialization all over the world. therefore central for promoting a better health in the population. This is
In such context, lack of contact with the natural environment is a even more important in the current trend characterised by an increase
growing concern (Antrop, 2004; Chen et al., 2008; Wright Wendel, of non-communicable diseases (NCDs) (e.g. cardiovascular and re-
2011; Wright Wendel et al., 2012). Nature has been identified as an spiratory diseases, diabetes, cancer) which, according to the WHO, will
important factor influencing human health. Among the potential ben- generate a cost of > 11 billion US$ to the world economy during the
efits that nature offers to individuals, improved health may be put 2011–2025 period (Mendis, 2014). Projections indicate a rise from 36
among the most important and a growing body of literature reflects this million deaths due to NCDs in 2008 to 44 million by 2020 globally,
(Gascon et al., 2016; Lee and Maheswaran, 2010; Lovell et al., 2014). especially in urban areas and among poorer groups (WHO, 2011).
The relationship between natural and semi-natural environments on Prevention of these diseases can be achieved through improved acces-
the one hand, and human health and wellbeing on the other, has been sibility to healthy environments and promotion of healthier lifestyles
considered not only by the scientific community but also by entities in (e.g. physical outdoor activity, recreational activities in rural areas and
charge of promoting health and protecting the environment. The World green spaces, etc.). Green areas can play a key role in this context, while
Health Organization (WHO) accounts for various environmental aspects providing also other benefits such as reducing health inequalities, im-
among the main determinants of health (World Health Organization, proving urban biodiversity and contributing to adaptation to climate
1986). The interactions between environment and health are complex. change (Chiabai et al., 2018).

⁎
Corresponding author.
E-mail address: [email protected] (A. Chiabai).

https://doi.org/10.1016/j.ecolecon.2019.106401
Received 16 November 2018; Received in revised form 14 May 2019; Accepted 16 July 2019
0921-8009/ © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

Different approaches have been taken in order to explain the in- purpose, including studies with quantitative results on the health ben-
teractions environment-human health and the associated benefits efits (in terms of risk reductions) associated with increased availability
(Martinez-Juarez et al., 2015). For example, trees can help mitigate of green areas, and a database has been constructed with all relevant
risks from air pollution by retaining contaminants present in the urban variables believed to influence this relationship.
atmosphere (Nowak et al., 2006), both chemical and acoustic (Stansfeld The paper is organized as follows. Section 2 presents the metho-
and Matheson, 2003). Exposure to green environments also interacts dological approach, including the process of constructing the database
with the human microbiome, which can lead to effects on the incidence with data obtained from the literature and external sources (Section
of inflammatory diseases such as allergies (Rook, 2013; Rook et al., 2.1), the definition of the variables and standardization process (Section
2013). Among other things, green areas can reduce surface runoff, 2.2), the Heckman selection model and its application to the current
hence contributing to reduced risks derived from flooding in urban study (Section 2.3), and the marginal effects associated (Section 2.4).
areas. Water cycle regulation provided by wetlands and other ecosys- Results are shown in Section 3, starting with the descriptive statistics
tems have important impacts on water supply and water quality. So- (Section 3.1) and following with the results of the analysis (Section
cioeconomic determinants have also been considered by some authors 3.2), while Section 4 presents a discussion of these results and the key
among the factors mediating in the relation between environment and conclusions.
health. Parks and other open green spaces may promote social cohesion
by providing meeting and leisure areas, which may have positive im- 2. Methods
pacts over mental health. The implications for social and economic
welfare go beyond this mediating effect. The health burden is particu- 2.1. Selection of previous case studies and database
larly strong for income deprived populations. Vulnerable populations
are more prone to poor health, and this relationship extends to various A literature review was conducted including peer reviewed pub-
aspects of health (Aschan-Leygonie et al., 2013; Mendis, 2014; Mitchell lications on the health benefits provided by green spaces, using a
and Popham, 2008; Roe et al., 2013; Ward Thompson et al., 2014, worldwide geographical coverage. This searching process included a
2012). Health inequalities may lead to poverty traps (Whitehead et al., systematic search through a set of selected keywords related to natural
2001), with worse health conditions often being accompanied by low environment and health which was described by Chiabai et al. (2018).
incomes. Worse eating habits, lower accessibility to health care, stress A detailed analysis of the outcomes and approaches was also conducted
and many other causes may lay behind this situation, but improved in order to incorporate the information into a common dataset which
access to green areas is among the possible mechanisms that have been was used afterwards for the econometric analysis. Chiabai et al. (sub-
proposed to reduce this aggravated impacts (Ward Thompson et al., mitted DiB) describes the steps taken from the literature review to the
2014). Urban green areas provide public open spaces that help vul- construction of a quantitative database summarising the main results
nerable segments of society to access to active leisure or benefit of extracted and used for the present analysis. As described in this last
cleaner air, as well as to improve social links. This may lead to a po- paper, the dataset included studies offering quantitative results linking
tential higher health improvement in deprived populations, hence green areas and human health. Table 1 reports the studies included in
contributing to health inequalities and the associated poverty trap the database in terms study location, methods, type of health outcomes,
(Mitchell and Popham, 2008). Socio-economic factors can therefore be health and green exposure indicators, number of observations available
seen as contextual variables contributing through different pathways to in each study as well as those with significant results in the undertaken
the expected health benefits of improved environments. analysis. It must be highlighted that the reviewed studies use avail-
In this context, making a comprehensive review of the existing lit- ability of green areas as a measure of exposure as it is discussed later in
erature is a complicated task. This is due to various reasons. The first is this section.
the fact that number of studies relating exposure to natural and semi- The observations reported in Table 1 are those extracted from each
natural environments and heath is still growing. Due to critical im- study to build the Heckman model and carry out the statistical analysis
portance of the issue, this field has attracted many investigators from as specified in the next sections. Each observation is recorded in terms
different backgrounds. This leads to the second challenge, the hetero- of a specific health indicator which measures the change in the health
geneity of methodologies and underlying assumptions. Methodological outcome due to increased availability of green areas, as indicated in
heterogeneity occurs at different points of the research, such as the Table 1 and discussed more specifically in Section 2.2. For example, in
variable measurement (e.g. health effect, exposure), the population Maas et al. (2009), the observations are in terms of changes in annual
selection, the inclusion of contextual factors, and the analytical tools prevalence rate (health indicator) in different diseases (health out-
employed. Issues of comparability among studies and the use of dif- come), associated to an increase in green spaces availability near the
ferent measures of health and green space complicate the identification respondents' residence. The health outcomes in Table 1 refer to the
of the underlying dose-response relationships. This leads to uncertainty specific effects on health, such as mortality, disease occurrence, per-
as to the “true” relationship between green spaces and health. While ceived health, as discussed in detail in Section 2.2.
there is a seemingly positive relation explored along the literature, the The literature was found to be quite diverse with respect to many
presence of non-significant and negative correlations has led some au- aspects, notably the methodological approach, the definition and in-
thors into questioning the validity of any generalization (Lee and dicators used for the health outcome and for exposure, which leads to
Maheswaran, 2010). significant statistical heterogeneity. As shown in Table 1, results are
Against this background, we aim to explore new approaches to deal mixed with significance varying considerably by study and type of
with the existing study heterogeneity, in order to extract generalizable health outcome, suggesting that there is no unique and clear evidence
conclusions from the literature linking green spaces and human health of the impact produced by green environment on human health. We
(Martinez-Juarez et al., 2015). This is a crucial step to facilitate briefly discuss hereby the main issues related to the diversity of the
knowledge transfer from academics to civil society on the importance of studies reviewed, and in a second step how the data from Table 1 have
green space. The hope is that it will also inform better interdisciplinary been standardized to construct a database for the econometric analysis
research in a field where various disciplines may interact. (Sections 2.2 and 2.3).
More specifically, the objective of this paper is to explore the po- The first point to highlight is the variety of methods and statistical
tential use of the Heckman selection model, as a way of identifying the techniques used in the literature to analyse the relationship health-
factors influencing the significance of the relationship throughout ex- green environment, depending on the type of data available, the pur-
isting studies and calculating the marginal effects of selected factors pose of the analysis and the health outcome analysed (for a discussion
found to be crucial. A literature review has been conducted for this see Chiabai et al., 2018). The studies reviewed can be categorized in

2
 A. Chiabai, et al.

Table 1
Studies included in the database.
Source: Chiabai et al. (submitted DiB).
Study Location Method Health outcomes Observations Health indicator Green exposure indicator

Total Significant
effects

Maas et al., 2009 Netherlands Multilevel logistic regression 24 outcomesa 58 26 (44.8%) Annual prevalence rate Percentage of green space in a radius of 1 and 3 km around
analyses the postal code of respondent's home
Maas et al., 2006 Netherlands Multilevel logistic regression Perceived general health 6 6 (100%) Percent of responses (5-point Likert Percentage of green space in a radius of 1 and 3 km around
analyses scale – very poor to very good) the postal code of respondent's home
Takano et al., 2002 Tokyo, Japan Multiple logistic regression All-cause mortality 21 8 (38.1%) Five-year survival rate for the Presence of walkable green spaces near the residence
analysis elderly (parks and tree lined streets) measured with qualitative
indicators
Mitchell and Popham, England Binomial regression model All-cause mortality, circulatory 15 9 (60%) Mortality incidence rate Population classified into 5 exposure groups based on the
2008 diseases and cancer proportion of green space of residence

3
Pereira et al., 2012 Perth, Australia Logistic regression Coronary heart disease 4 1 (25%) Hospital admissions and self- Neighbourhood greenness for a 1600 m service area
reported medically diagnosed cases around residence using remote sensing data
White et al., 2013 England Fixed effect regression Perceived general and mental 12 10 (83.3%) Percent of responses (5-point Likert Distance to the coast (0–5 km; 5–50 km; > 50 km) and
health scale – poor to excellent) percentage of green space
Dunstan et al., 2013 South Wales Multilevel logistic regression Perceived general health 3 2 (66.7%) Percent of responses (3-point Likert Neighbour measure of natural environment through
model. scale – not good-fairly good-good) Residential Environment Assessment Tool (REAT)
Tamosiunas et al., 2014 Kaunas, Lithuania Multivariate Cox proportional Cardiovascular disease, fatal and 21 0 (0%) Age-adjusted prevalence (%) Distance to city parks larger than 1 ha, categories
hazards regression non-fatal classified based on spatial land cover data
Pretty et al., 2005 Colchester, UK One-way ANOVA test Perceived mental health 4 3 (75%) Percent of responses (5-point Likert Exposure to visual stimuli (rural and urban photographic
(depression and anxiety) scale –not at all to extremely) scenes)
Marselle et al., 2013 England, UK One-way ANOVA test Perceived stress and depression 14 2(15%) Use of Major Depressive Inventory Questionnaire on participants' walking environments
and a 10-item Perceived Stress Scale
Roe et al., 2013 Dundee, UK Multiple linear regression Perceived mental health (stress) 4 2 (50%) Perceived stress score (5-point Percentage of green space (parks, woodlands, scrub and
Likert scale –never to very often) other)
Kerr et al., 2006 Japan Doubly multivariate profile Anxiety 2 1 (50%) Tension and Effort Stress Inventory Outdoor running session vs- outdoor running in natural
analysis (MANOVA) environment

a
Cardiovascular, respiratory, musculoskeletal, digestive, mental, neurological, miscellaneous.
Ecological Economics 167 (2020) 106401
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

Table 2
Description of variables.
Source: Chiabai et al. (submitted DiB).
Variable Description Data source Units

Health risk reduction % change in the health indicator due to an increase in exposure Reviewed studies % change
respect to a baseline defined as low exposure.
Exposure to green Availability of green spaces in the surroundings of people's living Reviewed studies Categorical variable (1 for low exposure, 2
areas environment, measured in terms of vicinity and/or % or density for medium exposure and 3 for high
of green. exposure)
Mortality Mortality versus morbidity impact. It allows measuring the Reviewed studies Dummy variable (1 for mortality, 0
differential effect between mortality and morbidity. morbidity)
Disease type General (all-cause, general health), mental, cardiovascular, Reviewed studies Categorical variable
respiratory, others (diabetes, cancer, etc.).
Female Proportion of female population over the total. Reviewed studies Percentage (of female on total)
Age Age groups: young < 16, adults 16 to 65, elderly > 65. Reviewed studies Percentage (of population in each age
group)
Subjective If the study relies on self-reported health, the observation is Reviewed studies Dummy variable (1 for the subjective
regarded as subjective, otherwise not. studies, 0 otherwise)
Income per capita GDP/population by country. Secondary source: IMF (http://www. GDP per capita
imf.org/external/pubs/ft/weo)
Hospital beds Hospital bed density (by country). Secondary source: CIA library Number hospital beds per 1000 people
(https://www.cia.gov/library/
publications)
Literacy Literacy rate, youth total (% of people ages 15–24, by country). Secondary source: World Bank (http:// Percentage
data.worldbank.org/indicator)
Urbanization % people living in urban areas (as defined by countries' Secondary source: World Bank (http:// Percentage
statistical agency) data.worldbank.org/indicator)

two main groups, “objective” and “subjective” studies. The first use benefits (dependent variable) and the increased exposure to green areas
health indicators computed with objective measures drawn from health (as explanatory variable). Given the diversity of indicators used for
registries (mortality rate, prevalence/incidence of specific diseases, these two variables, some assumptions for standardization are needed
hospitalization rate, life expectancy). The second rely on subjective to carry out the analysis under a common measurement framework.
measures such as opinions and individual perceptions on health status, Our first order of business was therefore to create standardized in-
quantified in survey-based questionnaires with qualitative measures dicators for a common measure allowing for comparison among the
using the Likert scale technique (e.g. “very poor” to “very good”). Both results.
types of measures were used in the econometric analysis based on the The health indicator in each study measures the change in the
recognition that they are equally important in defining the relationship health effects due to an increase in exposure to green areas. In the re-
between exposure and individual health status. There is some evidence viewed studies, the health indicator may refer to objective indexes, such
that self-reported, subjective measures of health may underreport the as mortality incidence rate, five years' survival rate, life expectancy,
prevalence of certain conditions, including cardiovascular diseases, and annual prevalence/incidence of diseases, hospital admissions, mea-
that such measures may mask socioeconomic gradients in disease risk sured from estimated coefficients in epidemiological functions.
(Mosca et al., 2013). Controlling for the impact of the type of health Alternatively, it may also refer to subjective indexes, such as the general
measures used is hence important. health perception measured on a Likert scale. All these indexes taken
Defining exposure to green areas is another major issue when it from the different studies, once collected, were transformed into a
comes to analysing their effects on health. The studies in the literature standardized percent variation rate referring to different health outcomes,
review generally refer to increased availability of green spaces within a which defines our standardized indicator “health risk reduction” (HRR).
certain distance from people's living environment and use different As regards the explanatory variable of exposure to green areas, the
metrics for this purpose (e.g. spatial land cover data, Normalized indicators used in the reviewed studies may refer to the distance of the
Differences Vegetation Index – NDVI). Accessibility, usability and respondents' home to the nearest park, or percentage of green spaces in
quality of the green space, on the other side, are associated with a the surroundings of respondents' living environment, or normalized
number of factors such as promotional activities, provision of footpaths difference vegetation index (NDVI) in the living environment which
and exercise facilities, appropriate lighting, enhanced aesthetics and identifies if a target space contains green vegetation or not. In order to
mixed land-use, good air quality, while it can be hindered by factors create a common standardized indicator for exposure, we constructed a
such as low path connectivity, heavy traffic and contamination. There qualitative variable taking three values for exposure: low, medium and
are not many studies in literature with quantitative analysis in regard to high. For each study, we created three intervals based on the cumula-
the wider quality and accessibility metrics which could be used in our tive distribution function of the specific indicator of exposure used in
modelling exercise Greenspace is heterogeneous in nature, and though the corresponding analysis (size or distance from the homes of parti-
studies are starting to consider these factors (e.g. Wheeler et al., 2015), cipants). In each study, the lowest level of exposure is taken as the
they are as yet few in number. As a result, these factors have not been baseline, the second tercile is taken as a medium exposure level, while
contemplated in our analysis. We consider “increased availability” of the third tercile group represented a high exposure. The baseline acts as
green spaces and use it as a proxy of “exposure” as referred in the re- reference, and refers to those groups of individuals who are less ex-
viewed literature. Future research might build on the basic model de- posed, if at all, to green areas. Further detail is given in Chiabai et al.
veloped in the current analysis to include more complex analysis based (submitted DiB).
on more refined indicators reflecting exposure. The full set of variables included in the database (Supplementary
data) is presented in Table 2 and detailed in Chiabai et al. (submitted
DiB). The rationale behind the selected explanatory variables rest on
2.2. Variables in the model and standardization process
their use in the two equations of the Heckman model (Section 2.3). A
first set of explanatory variables are assumed to be affecting the health
The two main variables to include in the model are the health

4
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

risk reduction in the outcome equation (see Section 2.3), and these in- sample which would lead to inconsistent and biased parameter esti-
clude increased exposure, mortality, disease cluster, a dummy variable mates (Copas, 2013). At the same time, the presence of more than one
to reflect whether the study used subjective measures or not, age, a estimated coefficient reported per study would give an excess weight to
dummy variable for gender, income per capita and hospital beds den- studies with many estimates (Stanley, 2001).
sity. Previous studies introduced a dummy variable for each study that
In order to differentiate the health impacts, the following variables provided more than one observation for the meta-analysis (Jarrell and
were constructed, discriminating among (a) mortality versus morbidity Stanley, 1990). Other solutions (Jeppesen et al., 2002) try to derive
effects (dummy “mortality”), (b) objective versus subjective studies estimates from meta-analyses combining a probit model and an un-
(dummy “subjective”), and (c) type of illness (categorical variable balanced panel data model to take into account the random researcher
“disease type”). Five dummies were derived from the categorical vari- effect and to assess the impact due to the commonality within a study
able: mental health, cardiovascular diseases, respiratory diseases, other and assuming that reporting a significant result in a study is separate
health impacts not included in previous categories (e.g. musculoske- from the amount observed. Further studies (Rolfe and Brouwer, 2012),
letal, neurological, digestive, diabetes, cancer), and a universal cate- used a mixed-effects Tobit model to take into account the censored
gory “general” (all-cause and general health). The latter is used in the nature of the data and the intra-study effects, assuming structural si-
literature as a comprehensive classification to refer the general in- milarity restrictions on coefficients for censored and non-censored ob-
dividual health status. servations. One way to take into account some of the limitations
The variable “subjective” is related to the differentiation between mentioned, would be to estimate with panel data a model selection
indicators used in “objective” and “subjective” studies respectively, (Wooldridge, 1995; Semikyna and Wooldridge, 2010), but this proce-
which might affect in a different way the relationship health-exposure. dure would not be feasible due to the nature of our data, as it presents
The same applies for the “mortality” and “disease type”. For example, neither the proper rationale (we could not compare the observations
we are interested in investigating possible differences in impacts among among the different papers), nor enough degrees of freedom to ade-
the groups of illnesses, and between mortality and morbidity. We also quately select the cohorts for the pseudo-panel needed to estimate the j-
include a number of demographic and socio-economic variables as Probit models in the first step, as proposed in those articles. Therefore,
control factors. Some of them were available in the studies reviewed in in this context of uncertainty, we tested the Heckman selection model
the database (“female” and “age”), while others were taken from sec- as a way to deal with the unobserved selection factors and correct for
ondary sources, such as “hospital beds density” and “income per capita” the bias in estimating the outcome equation, and we introduced vari-
at the country level. Hospital beds density (defined as the number of ables related with the study to control for the researcher effect. In our
hospital beds per 1000 people) represents a proxy for the access to the analysis, we name this effect as the i-Study Effect, as it is explained in
health care system, under the assumptions that higher access to health the next sections.
care services would guarantee better population health status. The Our objective is to gather the quantitative results available in the
variable “income per capita” is assumed to negatively affect the health literature about the relationship health and green areas in a meta-
risk reduction, in line with previous studies in the literature according analysis in order to model quantitative impacts on health associated
to which poorer groups are benefitting more from exposure to green with exposure to green areas in a context of study heterogeneity.
areas. The Heckman selection model is usually expressed in terms of latent
A second set of explanatory variables, some of the them in common variable models and relies on two equations, an outcome equation which
with the first set, are capturing the i-Study Effect in the selection equation includes factors affecting the outcome variable, and a selection equation
(see Section 2.3) designating: which considers the part of the sample which is observed and the fac-
tors influencing the selection process.
- Socio-economic aspects on the country where the study was done, In our case, the outcome equation relates the health risk reduction
captured by “income per capita”, “urbanization”, “literacy” at the with a set of explanatory variables such as increased exposure level,
country level, taken from secondary sources. “Urbanization” refers income per capita, type of disease and so on.
to the percentage of people living in urban areas and reflects a proxy In its general form, the outcome equation Ri can be expressed as:
for urban lifestyle. “Literacy” refers to the percentage of people
Ri = Xi β + εi (1)
literate aged 15–24 and reflects the effect of knowledge.
- Characteristics of the study, captured by “subjective” (meaning where Xi are the explanatory variables determining the health risk re-
subjective versus objective nature of the study) and “mortality” duction Ri; β is a vector of parameters to be estimated; andεi is the error
(meaning that the study focus on mortality versus morbidity out- term. In our analysis Eq. (1) takes the following form:
comes), taken from the reviewed studies.
Ri
We assume that these variables can affect the significance of the
= β0 + β1 morti + β2 subi + β3 cari + β4 resi + β5 meni + β6 geni + β7
results obtained in the reviewed studies. Some of these variables are in
both equations, as they capture both the effect on the health risk re- expm, i + β8exph, i + β9 f emi + β10 oldi + β11 adulti + β12 logGDPi + β13
duction as well as the i-Study Effect. This is the case of “income per log bedi + εi (2)
capita”, “subjective” and “mortality”. Their interpretation in light of the
results obtained is discussed in Section 3. The explanatory variables are those reported in Table 2, though
some of them have been further transformed in dummies, as specified
2.3. The Heckman model hereby. morti is the dummy variable “mortality” when mortality is
measured in study i. subi is the dummy variable “subjective” indicating
Though most of the studies reviewed support the idea that green if the observation is a subjective health perception derived from sur-
areas can have beneficial effects on human health, this relationship is veys. The four variables cari, respi, meni, geni are dummies derived from
influenced by multiple factors (environmental, socio-demographic and the categorical variable “disease type” in Table 2, and they are inter-
economic) and is therefore characterised by high levels of complexity preted in comparison with the category “others” (diabetes, cancer, etc.).
and uncertainty. Indeed, many of the studies found in the review, show cari, is the dummy variable for cardiovascular diseases, resi for re-
non-significant results. This implies unclear evidence for health benefits spiratory diseases, meni for mental health and neurologic diseases, and
from green areas at the current stage. In such cases, considering only geni for other diseases (digestive, muscular, etc.).Exposure is measured
the studies providing significant results would generate a censored with two dummies, expm and exph, derived from the categorical variable

5
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

Table 3 the health risk reduction and to identify the key variables in this re-
Descriptive statistics of principal variables. lationship.
Variable Whole sample (all Subsample (significant
observations) results only) 2.4. Marginal effects within the Heckman model

Mean Std. dev Mean Std. dev To estimate the model coefficients, we used the full information
Health risk reduction 0.848 1.799 1.755 2.263 maximum likelihood estimation method. The estimation involves
Mortality 0.297 0.458 0.398 0.492 forming the joint distribution of the two random variables [εi, vi] and
Subjective 0.247 0.433 0.295 0.495 then maximizing the full log-likelihood function. The marginal risk
Exposure to green reduction induced by the model determinants was then calculated on
areas
• High 0.297 0.458 0.34 0.477
the basis of the estimated model considering the non-linear effects and
• Medium 0.445 0.498 0.307 0.464 for the mean values in the quantitative variables and the median values
• Low
Disease type
0.26 0.35 in the dummy variables.
The interpretation of the results from the model requires the
• Cardiovascular 0.220 0.415 0.08 0.272
•
transformation of the coefficients obtained in order to avoid selectivity
Respiratory 0.044 0.206 0.045 0.209
• Mental 0.198 0.399 0.216 0.414 bias. Vance (Vance, 2009) proposes marginal effects and significance
• General 0.324 0.469 0.477 0.502 testing following the equation:
• Others
Urbanization
0.214
81.128
0.411
6.458
0.182
82.447
0.388
4.103
∂E (Ri | Si∗ > 0, X )
= βk − αk ρσε δi (−Zα )
Hospital beds 6.131 3.993 5.476 3.79 ∂Xki (5)
Age
• Young < 16 14.285 7.854 14.28 8.638 where the inverse of the Mills ratio is denoted as δ(−Zα), and it is to
• Adults 16–65 59.724 26.489 60.72 27.511 control for potential bias emerging from sample selectivity and it is
• Elderly > 65
Female
25.991
51.966
31.236
17.226
25.0
48.673
31.731
14.448
calculated from the linear predictions (−Zα) of the selection equation.
In general, the marginal effect of a variable Xk will be different for each
Literacy 99.139 0.259 99.05 0.154
Income per capita 29,842.61 10,815.74 30,994.29 8190.77
observation (individual). As usual in such situations, we compute the
value of the marginal effect for Z, a mean or median vector of variables
Note: number of observations = 182. (for quantitative or qualitative variables, respectively). The marginal
effect estimated represents the variation in the health risk reduction
“exposure to green areas” in Table 2 and representing medium and high (HRR) associated with a variation in the explicative variable once
respectively compared with low exposure. femi is the proportion of fe- corrected for the selection bias.
males in each observation. oldi and adulti denote the proportion of po-
pulation over 65 and between 16 and 65 respectively, taken from the 3. Results of the model
variable “age” in Table 2. GDP is the “income per capita” expressed in
2005 USD. bedi is the number of “hospital beds” in the country per 1000 3.1. Descriptive statistics
inhabitants.
The selection equation is the probability that the health risk reduc- Table 3 shows the mean values and standard deviations of the
tion due to exposure is significant (probability of significance being created latent variable health risk reduction and its determinants. The
observed, Si), which can be expressed as: first two columns refer to the whole sample of observations created
from the results extracted from the literature review. Average values
Si∗ = Zi α + vi (3)
were taken for numeric variables and proportions in the case of dummy
where Zi are the explanatory variables assumed to capture the i-Study variables. The last two columns analyse the subsample created by se-
Effect; α is a vector of parameters to be estimated; and vi is the error lecting just those observations coming from significant results reported
term. Eq. (1) is observed if Si = 1, meaning that Si∗ shows significant in the reviewed studies. The two variables HRR and increased exposure
effects on risk reduction from exposure, and Si = 0 otherwise. are measured with the standardized indicators as defined in Section 2.2.
In our analysis the selection equation takes the following form: Overall, socioeconomic variables do not show great differences between
the whole sample and the subsample. However, for HRR and exposure
Si∗ = α 0 + α1 morti + α2 subi + α3 urbi + α4 log GDPi + α5 liti + vi (4)
to green areas the difference between the two samples is higher which
where urbi is the variable “urbanization” (percentage people living in justifies the use of the Heckman selection model to take into account
urban areas, per country) and liti is the percentage of literate people both significant and no significant results.
aged 15–24 in the country. Fig. 1 shows how the relation between per capita income and health
This is the latent variable model. If Si∗ shows significant effects of risk reduction is approximately flat when medium exposure is analysed
exposure on risk reduction then the observed latent function equals to while the slope becomes negative for a higher level of exposure. This
1, otherwise Ri = 0. The regression equation observes the value of Ri if relation has been built using the expected health risk reductions for
Si = 1. εi and vi are the error terms of the two equations which are each of the observations included database used for the analysis. This
distributed according to a bivariate normal with mean zero, εi~N result implies that targeting the inequality through development of
(0, σε2), vi~N(0, 1) and covariance ρ = Corr(εi, vi). The error terms are green spaces may require important developments in neighbourhoods
independent of both sets of explanatory variables. The model allows for in terms of green infrastructures in order to guarantee high exposure of
correlation between unobservable information of the two equations. As citizens. The apparent trend linking higher income with lower potential
it is well known, if ρ = 0, the standard regression model applied to Eq. for health improvements is nevertheless a relevant point. Higher po-
(1) provides consistent and asymptotically efficient estimators for all tential for health improvements in lower income areas would imply an
model parameters. When ρ ≠ 0, the standard regression model applied alleviation effect over health inequalities.
to Eq. (1) provides biased results, while the Heckman model with
sample selection provides consistent and asymptotically efficient esti- 3.2. Estimate of a systematized function for health risk reduction as a
mators for all model parameters. response to green areas exposure
The application of Heckman model in our context allows differ-
entiating among those factors affecting the significance of exposure on Table 4 shows the results of the Heckman Selection model

6
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

White et al. (2013) in the reviewed studies. These determinants describe the i-Study Effect
and include variables characterising the study and socio-economic
0- 5 Km 5- 50 Km > 50 Km
factors in the country under analysis.
The results arising from the selection equation show that the prob-
ability of seeing significant results in the health risk reduction from
increased exposure to green areas is significantly higher in studies
conducted in urbanized countries, with lower income per capita and
literacy rate, as well as in those studies looking at mortality outcome
and subjective health indicators. The negative effect identified for in-
17.15 44.34 38.51 come in determining the likelihood of a significant result may reflect
the potential publication bias in the publication of negative results.
Low Medium High Research in the medical sciences on clinical trials suggests that the odds
ratio for the publication of significant results in higher income countries
Maas et al. (2009)
relative to other countries was 0.41 in 2003 (Yousefi-Nooraie et al.,
Baseline +10% 2006), implying that studies in richer countries are more likely to re-
port negative results.
As it can be seen in Table 4, the Wald test shows that the covariance
between errors in the two equations is significantly different from zero,
so that the two equations have to be jointly estimated. Also, we have
tested the adequacy of our specification and our conclusions are: (i) we
reject the null hypothesis of non-global significance of the outcome
equation (χ2(13) = 143.38***), (ii) we reject the adequacy of the Tobit
Low Medium High specification for the structural similarity restrictions on coefficients for
censored and non-censored observations (χ2(15) = 207.15***) and (iii)
Fig. 1. Correlation between income per capita and health risk reduction for all- we observe problems of collinearity if we introduce a dummy variable
cause morbidity in the study sample. Marginal effects are calculated for mean for each study.
values of quantitative variables and median values of the dummy variables of In order to assess the magnitude of health risk reduction and its
the sample. Note: number of observations = 182. determinants, however, we need to look at Eq. (5) which estimates the
marginal effects (Section 2.4) from the system of equations. In other
Table 4 words, in order to explain the results on the HRR we need to jointly
Heckman Selection model results. estimate the outcome and selection equations, and interpret the re-
Explanatory Outcome equation (Ri): Selection equation (Si):
sulting marginal effects in light of both equations. Eq. (5) measures the
variables HRR probability that HRR significant marginal values for the health risk reduction as a response to changes in
the determinants—dy/dx for quantitative variables and discrete change
Mortality [mort] 0.5716 (0.897) 1.7911 (0.435)⁎⁎⁎ of dummy variables from 0 to 1 (Table 2). Results are reported in Fig. 2
Subjective [sub] −0.0523 (0.849) 1.1635 (0.334)⁎⁎⁎
Cardiovascular [car] −0.0875 (0.388)
and show that changes from baseline to medium exposure levels are
Respiratory [res] −0.0309 (0.282) expected to generate reductions in health risks of about 2.6% on
Mental health [men] 0.3941 (0.579) average in the study population. This impact increases to a 3.5% for
General health [gen] −1.7318 (0.672)⁎⁎⁎ high exposure levels compared to the baseline, though diminishing
Medium exposure 2.5682 (0.367)⁎⁎⁎
returns to scale can be intuited from the data, consistent with the lit-
[expm]
High exposure [exph] 3.4530 (0.591)⁎⁎⁎ erature (Pampalon et al., 2006). This implies that, all values held
Female [fem] 0.0051 (0.016) constant at the average, policies that increase availability of natural or
Elderly [old] 0.0593 (0.035)⁎ semi-natural spaces for the citizens may generate health benefits up to
Adults [adult] 0.0599 (0.038)⁎ 3.5% risk reduction.
log income per −2.1062 (1.310)⁎ −0.8408 (0.473)⁎
capita [log
Higher risk reductions are estimated for mortality compared to
(GDP)] morbidity (+1.4%). As regards the type of illnesses, mental health has
log hospital beds per 2.6754 (0.715)⁎⁎⁎ the largest impact on risk reduction (+0.39%) compared with the ca-
capita [log(bed)] tegory “other diseases” (encompassing many diseases, such as cancer,
Urbanization [urb] 0.0793 (0.028)⁎⁎⁎
diabetes, etc.). Though the coefficient is not significant, it shows a
Literacy [lit] −1.3205 (0.615)⁎⁎
Constant 13.0054 (12.947) 132.1250 (62.209)⁎⁎ tendency of the importance of green areas on mental health in the
Wald test of indep. eqns. (rho = 0): chi2(1) 3.27⁎ current context where mental disorders are strongly contributing to the
world disease burden (Burton and Rogerson, 2017). The broad and
Note 1: Figures are the estimated coefficients of the model and figures in comprehensive category “general health” shows lower risk reductions
brackets are standard errors. (−1.7%) compared to “other diseases” addressing specific health con-
Note 2: GDP and beds per capita have been transformed into log to consider the
ditions from exposure to green areas.
non-linearity effects.
As for the demographic variables, gender does not affect sig-
Note 3: number of observations = 182.
⁎ nificantly the impact, while adults and old people are those gaining
p < .1.
⁎⁎
p < .05. slightly more from increased exposure to green spaces, compared to
⁎⁎⁎
p < .01. young people (< 16 years old), though the magnitude of the effect is
small.
separately for each equation. The outcome equation (Ri) explains the Socio-economic variables have also an impact on risk reductions.
health risk reduction associated with exposure to green areas with a set Income per capita was found to be moderators of the improvement in
of explanatory variables identifying different determinants. The selec- health. In studies conducted in poorer countries, increased exposure to
tion equation, on the other hand, reveals the determinants affecting the green areas could lead to higher reduction in health risks, taking into
probability of finding significant results in the risk reduction estimated account the publication bias (−2.5%). From a different methodological
approach, other authors such as Wright Wendel et al. (2012) or

7
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

Fig. 2. Marginal effects for the Heckman model. Note:

marginal effects calculated on the basis of the estimated
model considering the non-linear effects and for mean
values of quantitative variables and median values of
dummy variables. Mortality (mort), Subjective (sub),
Cardiovascular (car), respiratory (res), mental health
(men), general health (gen), medium exposure (expm),
high exposure (exph), female (fem), elderly (old), adults
(adult), log income per capita (log(gdp)), log hospital
beds per capita (log(bed)). Note: number of observa-
tions = 182.

Germann-Chiari and Seeland (2004) have also considered the role of less effort. Investment on green areas may therefore be a strategy to
access to green space in low-income groups and areas. Finally, health alleviate health inequalities in poor areas. The findings suggest how-
risk reductions are expected to be higher in countries with higher access ever that interventions may require important increases in green space
to healthcare (measured as number of hospital beds per 1000 in- available to obtain a certain level of health benefits.
habitants in the country) (+2.7%).
Fig. 3 shows a simulation of expected HRR using OECD GDP per
capita. It shows the negative and logarithmic decrease in impact asso- 4. Discussion and conclusions
ciated to higher income levels as simulated using sample's average
values as reference. It can also be seen in the figure the difference in The existing literature on the impact of exposure to green areas
impacts between higher and medium exposure levels. The graph marks shows a high level of heterogeneity with respect to both the methods
average income as calculated from OECD countries, as well as the lower and indicators used for health and exposure to green areas. In this
decile from the sample of OECD country average income. Potential for context, this study argues that it is important to consider both sig-
improvement therefore depends on context in the model drawn from nificant and no-significant results in the literature in order to construct
this study. Richer countries require stronger improvements in their an overall framework to study the relationship between green spaces
environmental conditions in order to achieve health improvements in and health. For this purpose, we performed the following steps: (i)
their populations. While less developed countries can also benefit from literature review of studies with quantitative results on the health
stark environmental action, they can obtain these advancements with benefits associated with increased exposure to green areas, (i) con-
struction of a database with standardized indicators for health and

Fig. 3. Change in health risk reduction (HRR) in relation

to GDP per capita for all-cause morbidity, 10% quantile
(Q10) and average. Marginal effects are calculated for
mean values of quantitative variables and median values
of dummy variables. Source: own data and OECD Income
Distribution Database (via http://oe.cd/idd). Note:
number of observations = 182.

8
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

exposure levels, and (iii) econometric analysis using the Heckman Yet, the most rapidly urbanizing areas are not located in such
Selection model to correct for the unobserved selection bias and analyse countries, and are often subject to social, economic and demographic
key emerging patterns from the literature. pressures that do not allow for such measures to be implemented. It is
Our results show that, while diverse, studies in the literature tend to precisely in these countries where quantitative studies are scarcer. The
find a positive correlation between green spaces and health benefits, model predicts an inverse relation between income and health impacts,
especially strong for high levels of exposure. One of the most significant though the absence of studies in developing countries poses a pathway
conclusion extracted from this analysis is the relevance of contextual for future research. Context is central to the associated health outcomes
factors. The notion that different contexts yield different interconnec- of green spaces and adaptation strategies associated with green spaces
tions is supported by the results obtained, which pointed towards in- should be tailored to the specificities of the area where they are applied.
come, education, and urbanization as possible factors affecting the re- This leads to another conclusion, that the effects of improving
sults of the different studies. health through higher access to green space could lead to direct eco-
A number of limitations have been identified in the approach pro- nomic benefits. These benefits could take the previously mentioned
posed in this study. First, the approach does not consider the pseudo forms of decreased medical expenditure, augmented productivity and
panel structure of the database. Due to the nature of our data (the less work absenteeism, which could be added to other benefits such as
observations responding to different health indicators), we do not have increase in property values, diminished flood risk, etc. Comprehensive
either the rationale (we cannot compare the observations among the economic valuations of green spaces that includes their impacts over
different papers), nor the numbers of observations and associated de- health should be expected in future analysis. Cost-benefit analyses may
grees of freedom to adequately select the cohorts for the pseudo-panel. otherwise underestimate benefits and lead to sub-optimal allocation of
Second, we are unable to consider the quality of green space as well resources to green spaces. It is important that we properly identify these
as accessibility and usability given the lack existing in the literature in values, drawing on the use of techniques such as the Heckman mod-
terms of quantitative analysis on health benefits. Third, heterogeneity elling approach, so that policy can be appropriately targeted to pro-
of literature in relation to exposure required us to construct a qualita- tecting green space and, in so doing, protecting the health of the po-
tive indicator for this metric and ideally this would be standardized pulation.
across the literature to allow for comparability in quantitative terms.
This study has been performed in a field where the literature is growing Acknowledgements
but heterogeneous. While its intention is precisely to help in the task of
having a general overview of the potential health benefits of green The authors would like to acknowledge three research projects for
spaces in health, it highlights the need for a more common approach to support: Horizon 2020 research project INHERIT (INter-sectoral Health
metrics used in such studies. and Environment Research for InnovaTion); ECOHEALTH (Adaptation
Strengths and novelties of the proposed approach include the fact to Climate Change in Spain: analysing Co-benefits among health,
that this is the first time that the Heckman model has been used in a tourism, ecosystem and food) Fundación Biodiversidad; and EU COST
meta-analysis study to our knowledge, which guarantees a better ap- Action IS1204: Tourism, Wellbeing and Ecosystem Services
proach compared to the Tobit model to synthetize the literature on (TObeWELL).
environmental exposure to human health. This is also among the first
studies that derive a marginal effect of exposure to green areas on Appendix A. Supplementary data
health from existing studies in the literature that could be used for
identifying health impacts in different contexts. Shanahan et al. (2016) Supplementary data to this article can be found online at https://
found that the health benefit can be affected by the “dose” of nature doi.org/10.1016/j.ecolecon.2019.106401.
experience. We find similar non-linear benefits.
Furthermore, this paper is based on the sound idea that the use of References
meta-analysis in reanalysing key but heterogeneous studies from the
literature, taking into account both their significant and insignificant Antrop, M., 2004. Landscape change and the urbanization process in Europe. Landsc.
results, can provide a better understanding of the relationship between Urban Plan. https://doi.org/10.1016/S0169-2046(03)00026-4.
Aschan-Leygonie, C., Baudet-Michel, S., Mathian, H., Sanders, L., 2013. Gaining a better
exposure to green spaces and human health. Finally, we highlight the understanding of respiratory health inequalities among cities: an ecological case
unbiased nature of the results which can lead to better informed policy. study on elderly males in the larger French cities. Int. J. Health Geogr. 12, 19.
Our study has relevant implications over several social aspects. https://doi.org/10.1186/1476-072X-12-19.
Bowler, D.E., Buyung-Ali, L., Knight, T.M., Pullin, A.S., 2010. Urban greening to cool
First, it opens a pathway for considering the co-benefits arising from towns and cities: a systematic review of the empirical evidence. Landsc. Urban Plan.
adaptation to climate change using green spaces (Chiabai et al., 2018). 97, 147–155. https://doi.org/10.1016/j.landurbplan.2010.05.006.
The increase in the amount of available green space in urban areas has Burton, J., Rogerson, M., 2017. The importance of greenspace for mental health. BJPsych
Int. 14 (4), 79–81 (Published online 2017 Nov 1).
been proposed in order to adapt to several impacts of climate change Chen, H., Jia, B., Lau, S.S.Y., 2008. Sustainable urban form for Chinese compact cities:
such as increasing temperatures (Bowler et al., 2010; Doick et al., 2014; challenges of a rapid urbanized economy. Habitat Int 32, 28–40. https://doi.org/10.
Harlan and Ruddell, 2011) or flood risks (Claessens et al., 2014; 1016/j.habitatint.2007.06.005.
Chiabai, A., Quiroga, S., Martinez-Juarez, P., Higgins, S., Taylor, T., 2018. The nexus
Opperman et al., 2009). Such measures are often referred to as Eco-
between climate change, ecosystem services and human health: towards a conceptual
system-based Adaptation (EbA). The potential for health improvement framework. Sci. Total Environ. 635, 1191–1204. https://doi.org/10.1016/j.scitotenv.
could arise as a positive side effect or co-benefit of EbA strategies. An 2018.03.323.
area where this could have implications is urban planning. The urban Chiabai, A., Quiroga, S., Martinez-Juarez, P., Suárez, C., García de Jalón, S., Taylor, T.
Quantifiable results from a literature review in health impacts of changes in exposure
areas in developed countries are increasingly taking an ecological to greenness. Ecological Economics: Data in Brief (Submitted).
perspective towards development and new built areas include public Claessens, J., Schram-Bijkerk, D., Dirven-van Breemen, L., Otte, P., van Wijnen, H., 2014.
open spaces including green areas. Literature suggests that green spaces The soil-water system as basis for a climate proof and healthy urban environment:
opportunities identified in a Dutch case-study. Sci. Total Environ. 485–486, 776–784.
are not optimally distributed among all citizens but that wealthier https://doi.org/10.1016/j.scitotenv.2014.02.120.
neighbourhoods dispose of higher amount of them (Germann-Chiari Copas, J.B., 2013. A likelihood-based sensitivity analysis for publication bias in meta-
and Seeland, 2004; Mitchell and Popham, 2008). Therefore, develop- analysis. Appl. Statist. 62, 47–66.
Doick, K.J., Peace, A., Hutchings, T.R., 2014. The role of one large greenspace in miti-
ment of green spaces in poorer neighbourhoods may decrease health gating London's nocturnal urban heat island. Sci. Total Environ. 493, 662–671.
inequalities within developed countries. Such reductions have direct https://doi.org/10.1016/j.scitotenv.2014.06.048.
economic impacts in the form of less medical expenditure, increased Dunstan, F., Fone, D.L., Glickman, M., Palmer, S., 2013. Objectively measured residential
environment and self-reported health: a multilevel analysis of UK census data. PLoS
productivity and lower work absenteeism.

9
A. Chiabai, et al. Ecological Economics 167 (2020) 106401

One 8, e69045. https://doi.org/10.1371/journal.pone.0069045. Rolfe, J., Brouwer, R., 2012. Design effects in a meta-analysis of river health choice ex-
Gascon, M., Triguero-Mas, M., Martínez, D., Dadvand, P., Rojas-Rueda, D., Plasència, A., periments in Australia. J. Choice Model. 5, 81–97.
Nieuwenhuijsen, M.J., 2016. Residential green spaces and mortality: a systematic Rook, G.A.W., 2013. Regulation of the immune system by biodiversity from the natural
review. Environ. Int. 86, 60–67. https://doi.org/10.1016/j.envint.2015.10.013. environment: an ecosystem service essential to health. Proc. Natl. Acad. Sci. 110,
Germann-Chiari, C., Seeland, K., 2004. Are urban green spaces optimally distributed to 18360–18367. https://doi.org/10.1073/pnas.1313731110.
act as places for social integration? Results of a geographical information system Rook, G.A.W., Lowry, C.A., Raison, C.L., 2013. Microbial “Old Friends”, immunoregula-
(GIS) approach for urban forestry research. For. Policy Econ. 6, 3–13. https://doi. tion and stress resilience. Evol. Med. Public Heal. 2013, 46–64. https://doi.org/10.
org/10.1016/S1389-9341(02)00067-9. 1093/emph/eot004.
Harlan, S.L., Ruddell, D.M., 2011. Climate change and health in cities: impacts of heat and Semikyna, A., Wooldridge, J.M., 2010. Estimating panel data models in the presence of
air pollution and potential co-benefits from mitigation and adaptation. Curr. Opin. endogeneity and selection. J. Econom. 157, 375–380.
Environ. Sustain. 3, 126–134. https://doi.org/10.1016/j.cosust.2011.01.001. Shanahan, D.F., Bush, R., Gaston, K.J., Lin, B.B., Dean, J., Barber, E., Fuller, R.A., 2016.
Jarrell, S.B., Stanley, T.D., 1990. A meta-analysis of the union wage gap. Ind. and Labour Health benefits from nature experiences depend on dose. Sci. Rep. 6. https://doi.org/
Rels. Rev 44, 54–67. 10.1038/srep28551.
Jeppesen, T., List, J.A., Folmer, H., 2002. Environmental regulations, and new plant lo- Stanley, T.D., 2001. Wheat from chaff: meta-analysis as quantitative literature review. J.
cation decisions: evidence from a meta-analysis. J. Reg. Sci. 42, 19–49. Econ. Perspect. 15, 131–150.
Kerr, J.H., Fujiyama, H., Sugano, A., Okamura, T., Chang, M., Onouha, F., 2006. Stansfeld, S.A., Matheson, M.P., 2003. Noise pollution: non-auditory effects on health. Br.
Psychological responses to exercising in laboratory and natural environments. Med. Bull. 68, 243–257. https://doi.org/10.1093/bmb/ldg033.
Psychol. Sport Exerc. 7, 345–359. https://doi.org/10.1016/j.psychsport.2005.09. Takano, T., Nakamura, K., Watanabe, M., 2002. Urban residential environments and se-
002. nior citizens' longevity in megacity areas: the importance of walkable green spaces. J.
Lee, A.C.K., Maheswaran, R., 2010. The health benefits of urban green spaces: a review of Epidemiol. Community Health 56, 913–918. https://doi.org/10.1136/jech.56.12.
the evidence. J. Public Health (Bangkok). 33, 212–222. https://doi.org/10.1093/ 913.
pubmed/fdq068. Tamosiunas, A., Grazuleviciene, R., Luksiene, D., Dedele, A., Reklaitiene, R., Baceviciene,
Lovell, R., Wheeler, B.W., Higgins, S.L., Irvine, K.N., Depledge, M.H., 2014. A systematic M., Vencloviene, J., Bernotiene, G., Radisauskas, R., Malinauskiene, V.,
review of the health and well-being benefits of biodiverse environments. J. Toxicol. Milinaviciene, E., Bobak, M., Peasey, A., Nieuwenhuijsen, M.J., 2014. Accessibility
Environ. Health. B. Crit. Rev. 17, 1–20. https://doi.org/10.1080/10937404.2013. and use of urban green spaces, and cardiovascular health: findings from a Kaunas
856361. cohort study. Environ. Health 13, 20. https://doi.org/10.1186/1476-069X-13-20.
Maas, J., Verheij, R.A., Groenewegen, P.P., De Vries, S., Spreeuwenberg, P., 2006. Green Vance, C., 2009. Marginal effects and significance testing with Heckman's sample selec-
space, urbanity, and health: how strong is the relation? J. Epidemiol. Community tion model: a methodological note. Appl. Econ. Lett. 16, 1415–1419. https://doi.org/
Heal. 60, 587–592. https://doi.org/10.1136/jech.2005.043125. 10.1080/13504850701466049.
Maas, J., Verheij, R.A., de Vries, S., Spreeuwenberg, P., Schellevis, F.G., Groenewegen, Ward Thompson, C., Roe, J., Aspinall, P., Mitchell, R., Clow, A., Miller, D., 2012. More
P.P., 2009. Morbidity is related to a green living environment. J. Epidemiol. green space is linked to less stress in deprived communities: evidence from salivary
Community Heal. 63, 967–973. https://doi.org/10.1136/jech.2008.079038. cortisol patterns. Landsc. Urban Plan. 105, 221–229. https://doi.org/10.1016/j.
Marselle, M.R., Irvine, K.N., Warber, S.L., 2013. Walking for well-being: are group walks landurbplan.2011.12.015.
in certain types of natural environments better for well-being than group walks in Ward Thompson, C., Aspinall, P., Roe, J., 2014. Access to green space in disadvantaged
urban environments? Int J Environ Res Public Health 10 (11), 5603–5628. urban communities: evidence of salutogenic effects based on biomarker and self-re-
Martinez-Juarez, P., Chiabai, A., Gómez, S.Q., Taylor, T., 2015. Ecosystems and human port measures of wellbeing. Procedia - Soc. Behav. Sci. 153, 10–22. https://doi.org/
health: towards a conceptual framework for assessing the co-benefits of climate 10.1016/j.sbspro.2014.10.036.
change adaptation (no. April 2015). In: BC3 Working Papers, (Bilbao). Wheeler, B.W., Lovell, R., Higgins, S.L., White, M.P., Alcock, I., Osborne, N.J., Husk, K.,
Mendis, S., 2014. Global status report on noncommunicable diseases 2014. Geneva, Sabel, C.E., Depledge, M.H., 2015. Beyond greenspace: an ecological study of po-
Switzerland. doi:ISBN 9789241564854. pulation general health and indicators of natural environment type and quality. Int. J.
Mitchell, R., Popham, F., 2008. Effect of exposure to natural environment on health in- Health Geogr. 14. https://doi.org/10.1186/s12942-015-0009-5.
equalities: an observational population study. Lancet 372, 1655–1660. https://doi. White, M.P., Alcock, I., Wheeler, B.W., Depledge, M.H., 2013. Coastal proximity, health
org/10.1016/S0140-6736(08)61689-X. and well-being: results from a longitudinal panel survey. Health Place 23, 97–103.
Mosca, I., Bhuachalla, B.N., Kenny, R.A., 2013. Explaining significant differences in https://doi.org/10.1016/j.healthplace.2013.05.006.
subjective and objective measures of cardiovascular health: evidence for the socio- Whitehead, M., Dahlgren, G., Evans, T., 2001. Equity and health sector reforms: can low-
economic gradient in a population-based study. BMC Cardiovasc. Disord. 13. https:// income countries escape the medical poverty trap? Lancet 358, 833–836. https://doi.
doi.org/10.1186/1471-2261-13-64. org/10.1016/S0140-6736(01)05975-X.
Nowak, D.J., Crane, D.E., Stevens, J.C., 2006. Air pollution removal by urban trees and World Health Organization, 2006. Preventing disease through healthy environments.
shrubs in the United States. Urban For. Urban Green. 4, 115–123. https://doi.org/10. Towards an estimate of the environmental burden of disease Available at: https://
1016/j.ufug.2006.01.007. www.who.int/quantifying_ehimpacts/publications/preventingdisease.pdf.
Opperman, J.J., Galloway, G.E., Fargione, J., Mount, J.F., Richter, B.D., Secchi, S., 2009. World Health Organization, 2011. Global status report on noncommunicable diseases.
Sustainable floodplains through large-scale reconnection to rivers. Science (80-.) 326, https://www.who.int/nmh/publications/ncd_report_full_en.pdf Available at:.
1487–1488. https://doi.org/10.1126/science.1178256. Wooldridge, J.M., 1995. Selection corrections for panel data models under conditional
Pampalon, R., Martinez, J., Hamel, D., 2006. Does living in rural areas make a difference mean independence assumptions. J. Econom. 68, 115–132.
for health in Québec? Health Place 12, 421–435. https://doi.org/10.1016/j. World Health Organization, 1986. Ottawa Charter for Health Promotion. World Health
healthplace.2005.04.002. Organization, Ottawa. https://doi.org/10.1093/heapro/1.4.405.
Pereira, G., Christian, H., Foster, S., Boruff, B.J., Bull, F., Knuiman, M., Giles-Corti, B., Wright Wendel, Heather E., 2011. An Examination of the Impacts of Urbanization on
Martin, K., Christian, H., Boruff, B.J., Knuiman, M., Giles-Corti, B., 2012. The asso- Green Space Access and Water Resources: A Developed and Developing World
ciation between neighborhood greenness and cardiovascular disease: an observa- Perspective. Graduate Theses and Dissertations. https://scholarcommons.usf.edu/etd/
tional study. BMC Public Health 12, 466. https://doi.org/10.1186/1471-2458-12- 3413.
466. Wright Wendel, H.E., Zarger, R.K., Mihelcic, J.R., 2012. Accessibility and usability: green
Pretty, J., Peacock, J., Sellens, M., Griffin, M., 2005. The mental and physical health space preferences, perceptions, and barriers in a rapidly urbanizing city in Latin
outcomes of green exercise. Int. J. Environ. Health Res. 15, 319–337. https://doi.org/ America. Landsc. Urban Plan. 107, 272–282. https://doi.org/10.1016/j.landurbplan.
10.1080/09603120500155963. 2012.06.003.
Roe, J., Thompson, C., Aspinall, P., Brewer, M., Duff, E., Miller, D., Mitchell, R., Clow, A., Yousefi-Nooraie, R., Shakiba, B., Mortaz-Hejri, S., 2006. Country development and
2013. Green space and stress: evidence from cortisol measures in deprived urban manuscript selection bias: A review of published studies. BMC Medical Research
communities. Int. J. Environ. Res. Public Health 10, 4086–4103. https://doi.org/10. Methodology 6 (1), 37.
3390/ijerph10094086.

10