We analyzed a population from Southern Sweden aged 18–80 years in a follow-up study (1999/2000 to 2005). The Swedish registration system provides a personal identification number for every individual. This number can be used to link data from different registers, and can be used to follow each individual during the entire study period. In this study we also used register data for geo-coding of each individual. All other data were extracted from the surveys.
A health survey was distributed as a mailed questionnaire in 33 municipalities in the Scania region (a Southern Swedish province). The total sample comprised 24945 persons. Three mailed reminders and one reminder by telephone were used. In the baseline survey (1999/2000) answers were obtained from 13 604 (54.5%) respondents and 10 485 (77%) responded to the follow-up (2005). In this study we excluded the individuals living in larger city centres, due to lack of detailed landscape data in the evaluation of green qualities in these areas. The final cohort included 9230 persons.
The sampling was conducted with individuals, not households, as sampling units. We do not have data on the number of persons in the sample that belong to the same households, but it is assumedly a negligible number.
The initial public health survey was stratified to constitute a representation of the total population in Scania regarding gender, age, and education level . At follow-up, women were slightly overrepresented (55.4% vs. 49.7% of non-responders) and fewer persons were born outside Sweden (9.2% vs. 14.8% among non-responders). There were also differences regarding unemployment (4.7% vs. 9.5% among non-responders), students (3.4% vs. 17.1% among non-responders), and low, middle, or high level non-manual workers (10.6%; 17.6% resp. 15.5% vs. 9.1%; 11.0% resp. 8.5% among non-responders). The responders were slightly more educated (38.1% had >13 years of education vs. 30.4% of non-responders). However, responders and non-responders had similar age (mean 49 years in both groups). Likewise, the level of physical activity was equivalent among responders and non-responders (low to moderate physical activity 78.8% resp. 77.8%).
There was a selective attrition based on mental health at baseline (21.7% of non-responders reported poor mental health at baseline compared to 17.6% of the responders).
Changes in residential addresses or in access to green qualities were not assessed for the non-responders.
In cases of extreme values (“outliers”) in data from 2005 those were controlled for and replaced with the values from the survey in 1999/2000. This was done for 60 cases concerning height, 10 cases concerning age (+ 5 years were added to 1999 value), and two cases concerning weight and ‘number of persons in the household’ respectively.
The survey and linking of register data were conducted in accordance with the Declaration of Helsinki and approved by the local committee of ethics (Regionala Etikprövningsnämnden i Lund, reference no. 2005–471).
The survey contained in total 106 questions on varied aspects of health. For the aim of this study we explored data on background variables – age, gender, economy, marital status, ethnicity, and education. Further on we extracted data on mental health (as measured by the General Health Questionnaire, GHQ-12) and data concerning habits of physical activity. Level of education was classified into four categories, close to the classification system of ISCED (International Standard Classification of Education) (UNESCO, 1997) : 1) < 10 years at school, 2) 10–12 years at school, 3) vocational training, 4) university.
Outcome variable, 2005
Mental health and general health questionnaire
There are several different versions of the self-administered General Health Questionnaire (GHQ), including GHQ-12 and GHQ-28. The GHQ-12 is a shortened 12-item version of the GHQ-28 , and is among the most widely used screening instruments for general mental health . Prevalence of poor mental health is defined as reporting a problem in three or more of 12 questions in the GHQ-12 . Each item (e.g. “Have you, during the past few weeks, felt unhappy and depressed”) is rated on a four-point Likert scale: 1) less than usual, 2) no more than usual, 3) rather more than usual, 4) much more than usual. Reporting a problem is defined as rating 3 or 4 on the item (scoring 0-0-1-1).
In general GHQ focuses on two main classes of phenomena: 1) inability to carry out one’s normal healthy functions; 2) emergence of new phenomena of distressing nature .
GHQ-12 has proven cross-cultural validity [42, 43] and reliability with an internal consistency between 0.82 and 0.86 (Cronbach’s alpha) [44, 45].
In this study GHQ-12 was used in Swedish and all items were applied. According to the validated syntax for GHQ12 a binary value was calculated for each individual 1999/2000 and 2005 – considered as having good mental health (interval 0–2) or not (interval 3–12).
Exposure variables (place and person factors) and confounders, 1999
Green qualities (place factor)
Based on interview studies (focusing on how people perceive the landscape regarding preferences and habits), field studies, and inventories conducted in 1995–2005 in landscape architecture/environmental psychology, eight basic characteristics (or qualities) of the landscape were revealed (Serene, Wild, Lush, Space, the Common, the Pleasure garden, Festive and Culture) [46–50].
The green qualities have been suggested to be beneficial to health (hence they are sometimes denoted “recreational characters” or “recreational values”) and when used in previous epidemiological studies associations between access to these qualities and neighbourhood satisfaction as well as to physical activity have been demonstrated [26, 51]. The green qualities have been used as a gold standard in a recently published epidemiological study , where area-aggregated assessments of the qualities demonstrated convergent as well as concurrent validity. However, though developed by experts in landscape planning, the qualities as such are not yet considered validated constructs.
To grasp features considered as healthy, resources for recreation have been classified and analyzed with GIS in former Swedish projects . The National Land Survey of Sweden (Lantmäteriet) has within the European Union programme CORINE (Coordination of Information on the Environment) mapped the land and vegetation cover of Sweden into 58 classes, using 25 × 25 m grids . With this data it was possible to establish objective definitions of the qualities that could be implemented using the GIS technique for five of the eight green qualities (Serene, Wild, Lush, Spacious, and Culture). These qualities were described and defined in GIS as below:
Serene – a place of peace, silence, and care. Sounds of wind, water, birds, and insects. No rubbish, no weeds, no disturbing people.GIS-criteria: broad-leaved forest, mixed forest, pastures, inland marshes, wet mires, other mires, water courses, lakes and ponds.*
Wild – a place of fascination with wild nature. Plants seem self-sown. Lichen and moss-grown rocks, old paths. GIS-criteria: Slopes more than 10°. Forest, thickets, bare rock, inland marshes, wet mires, other mires, water courses, lakes and ponds. Each >15 ha if >1 km from the city. **
Lush – a place rich in species. A room offering a variety of wild species and animals and plants. GIS-criteria: Mixed forest, marshes and mires, beaches, dunes, sand plains, bare rock. All registered “key biotopes”. Pasture land of regional interest. Biodiversity areas, bird biotopes. National parks
Spacious – a place offering a restful feeling of “entering another world”, a coherent whole, like a beech forest. GIS-criteria: Beaches, dunes, sand plains, bare rock, sparsely vegetated areas, burnt areas, natural grassland, moors and heath land, forest > 25 ha. Slopes > 10°. Farmland pointed out in a national plan. Coastal zone preservation. ***
Culture – the essence of human culture. A historical place offering fascination with the course of time. GIS-criteria: Non-urban parks. Farmland pointed out in a national plan. National interests of cultural preservation. Nature reservation areas. * Excluded areas: noise > 30 dB, artillery ranges. **Excluded areas: noise > 40 dB, <800 m to wind power aggregates. ***Excluded areas: noise >40 dB.
Only persons from rural or suburban areas, or smaller towns were included in this study (n = 9230), since the assessment of the green qualities could not be made objectively for inner city areas with available data. Hence individuals from the larger inner city areas (Malmö, Lund, Kristianstad, and Helsingborg) were excluded (n = 1245).
Residential geocodes were obtained for the participants. With the aid of those geocodes in combination with the GIS database the green qualities were included in our analysis. We assessed for each respondent the presence/absence (regardless of amount/area) of each of the five qualities within 300 m from the centre of the property at the geocoded residential address. We assessed either amount of green qualities (zero to five), or access or not to each single quality respectively (i.e. access to serene or not, access to wild or not, etc.)
Concerning the chosen distance of 300 m it can be commented that in Scandinavia a common average distance to urban green areas is 300 m . In addition 300 m has previously been estimated as a crucial limit for people to exploit green spaces for recreational purposes and it is believed to represent rather well a person’s recreation area in his/her neighbourhood [17, 26]. A distance of 300–400 m is often reported as the threshold after which use starts to decline rapidly [17, 20, 55].
Physical activity (person factor)
There are varied approaches to measuring physical activity . In this study we dichotomized the population according to a single question concerned with leisure-time physical activity – “How often are you physically active of perform exercise during you leisure time? Excluding domestic work” (Response alternatives: 1) Sedentary 2) Moderate physical activity 3)Regular exercise 4) Regular advanced exercise). Low to moderate leisure-time physical activity was defined in this study as responding 1 or 2 (n = 6811; 78.8%), and regular leisure-time physical activity as responding 3 or 4 (n = 1838; 21.1%).
To study the potentially increased mental health effect of physical activity and access to nature, or any particular kind of nature, interaction-variables were created between low or regular physical activity respectively and either access to the green qualities or not, or access to each single quality (Wild, Lush, Serene, Culture, or Space) or not.
Financial stress and living conditions (person factors)
Classification in three groups of financial stress was based on data about having troubles paying bills (1)every month 2)every second month 3)it occurs rarely 4)never). Persons reporting troubles often (i.e. responding 1 or 2) were classified as financially stressed, reporting problems rarely (i.e. 3) as slightly financially stressed, and never troubled (4) as not financially stressed. These groups were used to study potential interaction effects between financial stress and access to the green qualities (in aspects of amount or particular quality).
Concerning living arrangements and form of housing the classification was constructed in accordance with living in detached houses or terrace-houses (group 1), living in a block of flats (group 2), or other living forms (group 3). Again those groups were used to study any interaction effect with the green qualities; those living in a block of flats were assumed to benefit the most of access to nature.
All persons born in a country other than Sweden were merged into one single category. Hence the categories of country of origin are “Sweden” or “other” (n = 810; 9.1%).
Marital status was classified into two groups (cohabiting or not) according to four response alternatives – 1)married or cohabitant 2)unmarried 3)divorced 4)widow/er. Hence any other response than 1 was merged into one group, considered as living alone (n = 2237; 25.3%).
Adjusted for baseline mental health, crude odds ratios (OR) and 95% confidence intervals (CI) were calculated in order to analyze associations between different exposure variables in 1999, and the outcome, mental health in 2005. Thereafter multivariate logistic analyses were performed. Apart from the exposure variables mentioned in the previous section, mental health 1999 and age were included as confounders.
In a similar survey conducted in the same population of Southern Sweden, multilevel analyses of green qualities aggregated to 1000 square metres did not change results compared to single level analyses, suggesting negligible clustering-effect even for much smaller areas than municipalities . The same negligible cluster-effect has also been found in other studies of relationship between behaviour, health outcome and green space . For the data from 1999/2000 the non-clustering-effect was also tested empirically. Thus in this study we fitted single level regression models to the data.
In the preliminary analyses we found no support for any interaction effects between financial stress and green qualities, nor for living arrangements and green qualities. Since a pattern appeared for physical activity and the green qualities we decided to focus on that in the following analyses. The effect of the interaction variable, constructed from physical activity and access to green qualities (both quantitatively and qualitatively), was explored by logistic regression analysis concerning the association and OR for mental health outcome. Any significance of positive departure from additivity of effects by the interaction variable was calculated by relative excess risk due to interaction (RERI) [58, 59].
Given the central focus on setting and living environment the analyses were restricted to those who did not experience any change in environment between base-line and follow up (n = 7549). This was to keep the access to green qualities constant between the occasions and hence reduce the potential effect on mental health that may be expected from a move, and that is not attributable to the environment itself.
All analyses were conducted using SPSS 18.0 for Windows (SPSS Inc, Chicago, Illinois, USA). The statistical significance level was set to p-value < 0.05 and 95% CI for mean differences and OR.