2.1. Participants and Projects
Using purposive sampling, participants were adult attendees to one of six wellbeing projects in the UK that focused on interaction with nature, between January 2010 and September 2017 (n = 318; 125 males, 177 females, 16 did not report their sex; aged 43.17 ± 15.34 years, 37 participants did not report their age). All participants provided informed consent for their providing of data for research purposes; and where required, ethical approval was granted by the University of Exeter ethics committee prior to the commencement of each of the projects (dates of approval: 18/5/2010; 3/5/2012; 22/8/2012; 18/11/2014; 13/1/2015). Whereas some projects were attended by both the general public and individuals with defined needs, other projects were attended only by individuals with defined needs. All projects combined nature-based activities with other therapeutic approaches such as counselling and were free to attend.
Project A (n = 13; 10 males, 3 females; aged 37.5 ± 11.8 years) comprised a variety of conservation, ecotherapy, and craft-focused interventions across the UK, that were attended by the general public and individuals with defined needs, such as learning disabilities, mental ill-health, recovery from hospital stays, low confidence, anxiety, psychiatric disorders, depression and risk behaviours. Project A ran between February 2016 and February 2017 and was 12 weeks in duration.
Project B (n = 20; 7 males; 13 females; aged 39.9 ± 7.9 years) aimed to improve the physical and mental wellbeing of vulnerable adults through community gardening and food growing activities within a UK city. It comprised multiple interventions across the city that were attended by the general public and individuals with defined needs such as autism, learning disabilities, mental ill-health, physical ill-health or impairments, homelessness, alcohol or substance misuse. Project B ran between January 2015 and September 2017 and was 12 weeks in duration.
Project C (n = 7; 3 males; 4 females; aged 20.9 ± 1.8 years) was a wilderness therapy-based project that sought to engage and positively change the lives of young people who typically face multiple barriers to success. These barriers included low self-esteem and mental wellbeing, poverty, abusive or ineffective families, drug and alcohol abuse, school failure, and youth offending orders. Many participants had previously experienced significant social, psychological or physical trauma, and some entered the programme at a stage where they were self-harming. Project C focused particularly on participants’ self-esteem and self-worth, because of their strong influences upon many facets of their lives. The project blended wilderness therapy with other counselling techniques, centered around weekend and weeklong wilderness expeditions. Project C ran between April 2012 and December 2016 and was 26 weeks induration.
Project D (n = 12; 8 males; 3 females; 1 did not say; aged 38.5 ± 11.2 years) was a community-based project for adults experiencing mental health difficulties that aimed to enable participants to increase their aspirations, personal responsibility and ability to undertake challenges in their lives. The project combined elements of wilderness therapy and walking programmes, in the form of weekly, facilitated walks whereby participants walked together in a group, exploring and learning about countryside and coastal environments and wildlife. Project D also incorporated opportunities to camp in scenic and remote natural environments around the UK. Project D ran between January and December 2010 and was 26 weeks induration.
Project E (n = 208; 75 males; 118 females; 15 did not say; aged 44.6 ± 16.0 years) was a city-based community-wide project that comprised a variety of community activities such as community food growing, helping vulnerable groups to access nature, reducing the carbon footprint, tree planting, developing community and therapeutic gardens and helping the homeless. Its aims included supporting communities experiencing low wellbeing and PA levels and reducing inequalities in wellbeing across the city. Project E ran between January and November 2015 and was 26 weeks in duration.
Project F (n = 58; 22 males; 36 females; aged 45.0 ± 15.1 years) comprised eight different health and wellbeing interventions across two cities and other locations in the southwest UK. Interventions were partnerships between health staff working in primary care, local organisations owning and/or managing natural assets, and practitioners; and involved groups of four to ten participants. Each weekly session included walking and conservation or tasks with silent or meditative elements, and were based in woodland areas, coastal zones, countryside dominated by agriculture and greenspace in and around urban settlements. Project E ran between March 2015 and October 2016 and was 12 weeks induration.
2.2. Design, Measures and Data Processing
Participants reported their wellbeing via questionnaires completed at the start (first week) and end (final week) of the project. Wellbeing was measured using the Warwick–Edinburgh Mental Well-Being Scale (WEMWBS), which comprises a global wellbeing measure including affective-emotional aspects, cognitive-evaluative dimensions and psychological functioning. In Projects A, C, D and F, wellbeing data were collected using the full version of the WEMWBS, which consists of 14 positively worded items that address positive aspects of mental health [26
]. It is scored by summing responses to each item, which are scored on a five-point Likert scale from 1 (none of the time) to 5 (all of the time). Overall scores range from 14 to 70, with higher scores indicating better wellbeing. The scale is validated for use in both adults and adolescents in the UK. The original scale validation study reported a Cronbach’s alpha of 0.91 for a UK sample [27
], and more recently a value of 0.92 has been reported for England population-level data [28
]. WEMWBS scores correlate with indexes of happiness and general health, and low scores can be predictive of depression [28
]. In Projects B and E, wellbeing data were reported via the Short Warwick–Edinburgh Mental Well-Being Scale (SWEMWBS), which consists of seven items from the full scale. Overall scores range from 7 to 35, with higher scores indicating better wellbeing. A Cronbach’s alpha of 0.84 has been reported for SWEMWBS using recent England population-level data [28
], and correlation between the WEMWBS and the SWEMWBS has been calculated to be 0.954 [29
]. Raw SWEMWBS scores were converted to metric scores in accordance with Stewart-Brown et al. [29
] prior to further data processing and analyses.
2.3. Categorising Wellbeing Scores
In order to contextualise the reported raw WEMWBS and metric-converted SWEMWBS scores, they were categorised in relation to UK population mean and standard deviation (SD) values. Scores within one SD of the mean were considered as ‘average’, scores more than one SD below the mean were categorised as ‘low’, and scores more than one SD above the mean were categorised as ‘high’ [28
Wellbeing scores were categorised in line with the Health Survey England (HSE) 2016 data (published Dec 2017) relating to the respective versions of the scale. As mean and SD SWEMWBS values were not published in HSE’s report, these were calculated using the full published dataset, including only data whereby participants completed all items with a 1–5 score (i.e., no missing values or answering ‘don’t know’). For WEMWBS, the categories were calculated as ‘Low’ 14–38; ‘Average’ 39–61; ‘High’ 61–70. For SWEMWBS, the categories were calculated as ‘Low’ 7.00–18.59; ‘Average’ 19.25–26.02; ‘High’ 27.03–35.00. For analyses, categorised (low; average; high) wellbeing scores were then dichotomised to create categories of ‘low wellbeing’ and ‘average to high wellbeing’.
Data for age were categorised in line with Barton and Pretty [2
], creating four age categories (≤30 years, 31–50, 51–70, and ≥70 years of age).
2.4. Creating a Single Variable for Analyses
In order to create and use a single, amalgamated variable for the analysis of changes in wellbeing across the projects, the data from the respective scales were ‘normalised’ to ‘percentage scores’ that represented scores as percentage of the scale on which they were reported, using the following formula:
Percentage Score = ((wellbeing score − minimum score for respective scale)/range of respective scale) × 100
Short form: Percentage Score = ((wellbeing score − 7)/28) × 100
Full form: Percentage Score = ((wellbeing score − 14)/56) × 100
2.5. Statistical Analysis
Whole sample. Standardised meta-analysis methodology was used to assess changes in reported wellbeing (Δ wellbeing) across the whole sample from start to end of engagement. Mean pre and post-intervention wellbeing scores were entered into Comprehensive Meta-Analysis Version 3 (Biostat, Englewood, NJ, USA), by intervention, for multi-study analysis. Data were pooled to calculate an overall intervention effect estimate. This represents the weighted average of the combined individual intervention effects. To reduce the imprecision of the pooled-effect estimate, the inverse-variance method was used to assign weights to each project, so that larger projects with smaller standard errors were given more weight than smaller projects with larger standard errors. As the various project interventions took different approaches, the combined intervention effect estimates were calculated using a random-effects model meta-analysis [30
]. 95% confidence intervals were calculated on the basis of the standard error of the pooled intervention effect. Statistical significance was set at p
< 0.05. In addition to the overall meta-analysis, random-effects model moderator analyses examined the influence of project duration on the Δ wellbeing.
Following multi-study analysis, in line with Barton and Pretty [2
], further analyses examined the effects of individual differences on the intervention-associated changes in wellbeing scores. The factors identified and available for this analysis were: sex (male, female); age group (in line with Barton and Pretty: ≤30 years, 31–50, 51–70, and ≥70 years of age); starting wellbeing status (low, average- high). Due to non-normality of data, non-parametric tests were used to examine the effects of these factors on the Δ wellbeing. A Kruskal–Wallis test was used to examine the effect of age group, and a Mann–Whitney U test was used to examine the effect of sex. As the distributions of data were significantly different between the dichotomised levels of wellbeing status, a bootstrapped (10,000 samples) independent samples t-test was used to examine effect of starting wellbeing status.
Low-wellbeing subsample. To investigate the efficacy of GE interventions in ‘low wellbeing’ adults, a multi-study analysis and follow-up analyses were conducted on data from participants who reported ‘low’ wellbeing pre-intervention. Due to a SD value of zero within the multi-study-analysis calculations, full Hedges’ g calculations were not possible when analysing pre- and post-intervention mean values. Therefore, the multi-study analysis examined the event rate of movement from reporting ‘low’ starting wellbeing status, to reporting ‘average-high’ end of intervention wellbeing status.
Similar to analyses of the whole sample, due to non-normality of data, non-parametric statistical tests were used to examine the effects of sex and age group on Δ wellbeing. A Kruskal–Wallis test was used to test the effect of age group, and a Mann–Whitney U test was used to examine the effect of sex.