Health impacts of environmental and social interventions designed to increase deprived communities' access to urban woodlands: a mixed-methods study

Phase 1: environmental interventions

Phase 1 involved physical changes to the woodland environment, designed to facilitate better access to, and use of, the woods. The interventions took place simultaneously across the three intervention sites over a period of 6 months, between October 2013 and March 2014. As with all WIAT schemes, the interventions were responsive to local conditions and community needs and, therefore, followed WIAT principles but varied in the detail of their design and implementation. The WIAT interventions followed principles defined by FCS in its grant-aid schemes of that time (2010). 33 The FCS guide to developing interventions to enhance the woodland user experience59 also informed the interventions to help ensure that a consistent approach was taken.

Researchers documented the physical state of the sites before and after the first intervention phase of the WIAT programme, undertaking regular checks of the environment in the field. In addition, FCS kept a systematic record of the plans, progress and completion of the works and noted feedback from the rangers involved in the projects as they progressed.

The physical changes to each intervention site were:

• Intervention site A – upgrading main entrance with stone setts (paving blocks) and bollards (wooden posts to exclude vehicular traffic), surfacing a loop path (i.e. a path taking a more or less circular route to return to its starting point) (approximately 650 m), treeworks (trimming and clearing overgrown vegetation and tree branches), fence repair, installing two picnic benches, installing two benches and litter picking.

• Intervention site B – upgrading three entrances, surfacing a loop path (approximately 1900 m), treeworks (thinning trees and trimming and clearing overgrown vegetation and tree branches), installing two picnic benches; installing six benches, new signage, restoring pond and creating dipping platform and litter picking.

• Intervention site C – upgrading one existing entrance, creating one new entrance with stone setts, surfacing a loop path (approximately 990 m), non-surfaced path improvements, treeworks (felling and thinning trees and clearing overgrown vegetation and tree branches), installing one bench and litter picking.

Forestry Commission Scotland produced plans of the interventions and the researchers photographed the physical state of the sites before and after this phase of the WIAT programme. Figures 4–6 show images of typical physical changes to each intervention site in this phase of intervention, all taken in winter months for ease of comparison. See Appendix 1 for plans of the environmental interventions.

FIGURE 4.

Intervention site A before (above) and after (below) environmental interventions. Reproduced with permission from OPENspace.

Reproduced with permission from OPENspace

FIGURE 5.

Intervention site B before (above) and after (below) environmental interventions. Reproduced with permission from OPENspace.

Reproduced with permission from OPENspace

FIGURE 6.

Intervention site C before (above) and after (below) environmental interventions. Reproduced with permission from OPENspace and FCS.

Reproduced with permission from OPENspace and FCS

Phase 2: social interventions

Phase 2 involved social interventions via facilitated community engagement activities to advertise and promote woodland use. The social intervention started in June 2014 (4 months after the physical intervention was completed) and took place in the three intervention sites over a total period of 9 months, until March 2015. They were planned around typical interventions used in the WIAT programme and, as with the physical interventions, were responsive to local woodland and community characteristics and aimed to target a variety of different segments, including different age groups, within the community. Typically, they involved one or two forest rangers and up to 1 day of preparation.

Community members were invited to these activities by FCS using a variety of means to engage different groups via both targeted events [e.g. photography workshops, a continuing professional development (CPD) event for nursery teachers and assistants] and more general events aimed at the whole community. Public events were freely open to all and widely advertised, for example through posters on site, in local shops, libraries, health centres, schools, community council Facebook (Facebook, Inc., Menlo Park, CA, USA; www.facebook.com) pages and letter drops to people’s homes.

The FCS rangers completed a form devised by the researchers for each activity undertaken to ensure that a systematic record was kept of the social intervention delivery and the participants engaged. This included a record of:

• the number of people involved

• the estimated age group of participants

• the participants’ genders.

Table 4 summarises the 62 social interventions undertaken and the numbers of participants. For some public events the attendance was so high that only an estimate of numbers was possible. It should also be noted that those participating in some events might be the same people attending repeated sessions, so the numbers reported are not necessarily unique individuals but the sum of those attending any one event. Most activities attracted a reasonable balance of male and female participants, although school and nursery teachers and assistants were predominantly female and health walk participants were predominantly male.

Public events and those with school or nursery children included activities to introduce children and adults to the natural environment (e.g. ‘The Secret Woodland’, ‘Magic in Your Woodlands’, ‘Winter Woodland Wonders’, ‘Meet the Critters’, family fun day, a ‘Scavenger Hunt’, bulb planting). Art works were also created as part of the activities celebrated in public events (e.g. a wooden carving of a dragon at intervention site A and a woven willow sculpture of a deer at intervention site B).

Recording progress and monitoring of the Forestry Commission Scotland interventions

Forestry Commission Scotland recorded its plans and progress, including budgets and resource allocation for the study interventions, as part of normal WIAT procedures. Feedback included numbers of attendees and community engagement events, as described above, and comments from participants. For example, a participant at an intervention site B public event in September 2014 called ‘Magic in Your Woodlands’ noted ‘Just wanted to say what a wonderful day the boys had. We stayed for over 4 hours’.

In addition, regular progress meetings took place between the researchers and FCS staff managing the interventions to discuss plans for the interventions, progress, any problems or delays, any additional engagement with the local community and current or potential site users and any feedback on the process of intervention implementation. The meetings also ensured that detailed records of the interventions for research purposes were maintained. In total, 12 meetings were held between May 2012 and November 2015, with meetings most frequent during the process of physical and social interventions. At the end of the interventions, a reflective workshop was held in May 2016 by an external researcher from Forest Research with the FCS project management staff to explore the FCS experience of undertaking the interventions and identify good practice and lessons learnt. 60

The records of progress meetings with FCS show that the interventions were planned in accordance with standard WIAT procedures and both physical interventions (see Appendix 1) and social interventions were discussed and agreed by FCS staff and notified to the researchers in advance of implementation. It was, in particular, agreed that, so far as possible, the interventions would be typical of the WIAT programme, drawing on a relatively modest budget rather than being singled out for additional levels of funding or other resources (minutes of progress meeting on 11 June 2013).

The implementation of the physical interventions involved both Forest Enterprise staff and contractors to undertake the construction work, as is typical of WIAT projects. The progress meetings record that the phase 1 physical interventions were carried out as planned, without any significant divergence from the agreed programme of activities (minutes of meetings on 24 January 2014). Minor adaptations to physical interventions were responsive to local conditions and community responses; this is typical of the WIAT programme and would be expected of any such project.

The implementation of the social interventions was so compressed in timing that it was not possible for FCS staff to undertake all the activities, as would normally be the case. An experienced external ranger was therefore contracted by FCS to undertake the social interventions at intervention sites A and B (minutes of meetings on 20 May and 24 July 2014). This proved successful and received positive feedback from the project manager. However, it did mean that the interventions at site C were less numerous and not delivered by the same people as those in sites A and B (minutes of meetings on 29 January 2015).

The desirability of deterring vandalism and maintaining the quality of the site experience is always an issue in WIAT projects. During the two phases of interventions, between 2013 and early 2015, FCS ensured that any items suffering major vandalism were repaired or replaced and the sites were kept at reasonable levels of maintenance. However, once the intervention phases were completed, no further maintenance was carried out by FCS.

In summary, the interventions were not adapted in major ways, other than in the hiring of an external contractor as a ranger to undertake the social interventions at two of the sites. There was a close fit between what was planned and what was delivered under the interventions.

Other contextual factors

As part of the site selection process in 2011, contact (by e-mail or telephone) with the relevant local authority planning departments was necessary to establish that there were no plans for change in the woodland sites under consideration other than the WIAT interventions. We also enquired about any changes planned for the survey communities or the wider neighbourhood context during the course of the study that might influence the study results, but did not identify any factors that appeared likely to do so.

When preparing for community-led audits (see Environmental audits) and focus groups (see Community focus groups and interviews), we again contacted the local authority planning departments and community planning officers to assist in making contact with local community members and facilitators. We also enquired whether or not any important changes had been experienced in any of the communities, such as new building developments or urban renewal projects. In the case of intervention site B, wider community developments were identified that had taken place during the course of the study and may have had an influence on outcomes. The evidence for these comes from personal communications with local authority planners and from local newspaper reports (The Herald, Paisley Daily Express, The Sun) between 2012 and 2014.

In 2012, plans for major redevelopment of the town centre in which intervention site B is located, were approved. Proposals included a new supermarket, town hall, community centre and car park. In March 2013, a new multimillion-pound leisure centre opened next to the intervention site B woodland. In 2014 there was a housing redevelopment programme that required many people to move house within the intervention site B community during the study period, as well as people moving into and out of the area. The extensive town centre redevelopment programme also started in 2014. In May–June 2015 (and coinciding with the 2015 wave 3 core survey; see Core survey of community residents), access paths from nearby car parks to the edge of the intervention site B woodlands were upgraded by the local council. There were concerns about the possible impact of this work on the intervention and its operation, although these paths were outside the intervention woodland under study. As the path upgrade by the council took place only as the final wave survey was being undertaken, it is unlikely to have affected community use of the woodlands in the previous year but may have disrupted or enhanced use during the few weeks prior to some respondents completing the survey.

Questionnaire design

Sampling strategy

We were interested in community-level change resulting from the same programme of interventions in three different sites with three matched control sites, so our analysis involved comparison of the population sample from the communities that received the intervention with those that did not. We initially selected a repeat cross-sectional survey design rather than a cohort design for pragmatic reasons: a pilot study with two communities that met the WIAT site criteria undertaken in 2006–934 indicated that achieving a longitudinal cohort in these settings was unlikely to be successful owing to the unwillingness of participants to be recontacted and high levels of attrition over time. Recruitment and retention of a cohort of several hundred participants in each community over a multiyear study in this context appeared unfeasible. However, following the first survey wave, at baseline, the researchers revisited the decision to reject a cohort design. Considering again the potential value gained by the inclusion of a longitudinal cohort, it was decided to attempt to establish such a cohort. A longitudinal cohort of participants was therefore targeted, nested within the cross-sectional surveys.

In determining the sample size, the literature suggested that there were likely to be gender differences in the observed effects. 26 To answer the primary research question, our sample size needed to be large enough to (1) detect an effect of the WIAT programme in the intervention group compared with the control group at each post-intervention wave, and (2) allow us to detect a gender difference in that effect. Based on data from Stigsdotter et al. ,27 to detect a male/female difference in means of 1.2 in each group (intervention and control), with a common standard deviation (SD) of 6.2 based on a two-sided, two-sample test with a 5% level of significance and 80% power, would require a minimum of 420 males and 420 females in each arm of the study. Therefore, a total sample size of 1680, comprising 840 intervention group and 840 control group participants, was required, with an equal split of male and female participants in each group. We did not power the study for further subgroup analysis (the added cost to power for different age groups was not considered justifiable). However, we did consider other demographic and personal variables in analysis of the data, the sequential nature of the intervention and confounders, such as life events (see Chapter 3). We could not completely rule out a clustering effect and did not have data available to enable us to precisely calculate the design effect caused by clustering. To take account of this we allowed for a 25% increase in our sample size beyond that based on the above power calculations. Thus, we sought a total sample size at each survey wave of 2100 (1050 per intervention or control group).

The survey was administered in three repeat cross-sectional waves at the same time of year in each case: wave 1 (baseline pre-interventions, late April to June 2013), wave 2 (follow-up, minimum of 2 months post physical environment interventions at each site, May to June 2014) and wave 3 (follow-up, minimum of 2 months post woodland promotion interventions at each site, late April to July 2015).

Recruitment

Face-to-face surveys are a robust method of data collection that maximise the level of response. Other methods, such as telephone and postal surveys that rely on self-completion, have shown declining response rates in recent years, especially in disadvantaged areas. 74 Our questionnaire was designed to be administered in a 25-minute, face-to-face, computer-assisted personal interview (CAPI) completed in the respondent’s own home. The questionnaire was piloted in July 2012 to assess its time burden and participant comprehension of its questions and procedure.

A survey company with experience of recruitment in communities similar to those of the study collected the data. Fieldworkers employed by the survey company were given full training on administering the questionnaire. Surveys were completed in the participant’s own home with fieldworkers moving from door to door, recruiting participants and conducting the survey. A quadruple call-back approach was adopted with fieldworkers making a minimum of four attempts to contact an address before moving on to the next randomly assigned address.

Inclusion and exclusion criteria

Individuals aged ≥ 16 years living within the intervention and control communities and within 1.5 km of a woodland site were eligible for the study. Individuals were not eligible to participate if they primarily resided outside the study sites. Individuals recruited to the linked community-led audit and community focus group branches of the study (see Environmental audits and Community focus groups and interviews) were excluded from subsequent waves of survey data collection to avoid contamination of response by their experience of in-depth involvement in the research.

Participants were selected from a postcode address file [Address Point – the then definitive Ordnance Survey product that provided a precise grid reference for each address listed in the Postcode Address File (Royal Mail data set)]. Selection used a stratified random sampling approach, stratified in accordance with distance from the WIAT intervention woodlands. This address file lists all deliverable addresses in the UK and can distinguish business and domestic addresses; we focused on domestic addresses only. Each unit postcode has a grid reference and this was used to stratify the sample by distance from the local woodland. We considered stratification by distance necessary because previous research suggests that the use of woodlands for populations living nearby may decline with distance,30 but there is also evidence that the quality of the natural environment may moderate the effect of distance,31 so distance was necessary to consider because the WIAT intervention is aimed at improving woodland quality. We stratified the sample in accordance with five distance points from the six WIAT-eligible woodlands. These distance points were in the range of 150 m, 300 m, 500 m, 750 m and 1500 m. Letters were then sent to the selected households informing them of the research project and that participants were being sought within the communities. The letters also contained the contact details of the research team members and their office, offering participants the opportunity to receive further information about the project or to opt out (see Report Supplementary Materials 2–4). Addresses of those residents who decided to opt out were then removed from the sample. A door-to-door approach and a quadruple call-back system were used to recruit participants. Recruitment was by face-to-face request to the first adult that responded to the door-to-door approach adopted.

Obtaining a longitudinal cohort

To maintain contact with the respondents from previous surveys, a thank-you letter was sent to all the addresses where wave 1 interviews took place. Unless respondents chose to opt out after receiving this letter (the letter described how to do so), the same addresses were visited in subsequent survey waves. Interviewers were instructed to confirm whether or not the person who answered the door was the same person (name, gender, age, address) who was previously recorded in wave 1. When the original respondent could not be found or recruited, recruitment of the new respondent in the household was attempted, if eligibility criteria were satisfied. If new recruits agreed to take part in the survey, a new respondent identifier code was generated. New recruits from the same household were asked to establish the relationship to the person previously interviewed (spouse/partner, child, parent, sibling, other family member or other as specified).

The size of the cohort was determined by the extent to which we were able to obtain repeat responses by this method. The cohort consisted of respondents who had participated in at least two waves of the survey. Once data were collated, checks were undertaken for age, gender and other individual characteristics to confirm that the respondents were correctly matched at each wave. The inclusion of this cohort in the initial cross-section design required changes to the initial analysis plan (see Chapter 5).

Response levels

To ensure that the correct sampling and interview protocols were followed, researchers attended the company’s field workers’ training sessions and took part in some of the visits to interview participants. This also gave insights into the challenges of recruitment in these very deprived urban areas, where, for example, there was a relatively high rate of refusal to answer the door (despite letters being sent in advance to targeted households).

Based on the random sample of addresses taken up by the survey company from the supplied postcode files, the overall response level achieved for the three surveys was 53%, lower than originally targeted. We also calculated the level of co-operation, that is, the proportion of successful interviews achieved once personal contact with a household had been made. The definition of ‘personal contact’ for this calculation includes effective interviews, doorstep refusals, people who wrote to opt out of the survey after receiving the introductory letter (these were comparatively few: 181 in total in wave 1) and those unable to take part owing to language issues or incapacity. The level of co-operation excludes incorrect or unusable sample addresses (empty properties, business premises, etc.), households from which there was no reply on the doorstep (despite quadruple call-back) and households (in waves 2 and 3) at which the named respondent was not available. The overall level of co-operation was 70%.

The response levels by wave and type of site (intervention/control) are shown in Table 5.

Assembling a longitudinal cohort represented a considerable achievement for the study given the challenges in participant recruitment. The availability of repeated measures from the same individuals provided more statistical power for the study and gave greater confidence in interpreting cause-and-effect relationships.

The sample respondents (n = 6317) were classified in accordance with the wave(s) of the study they participated in. This resulted in five different types of respondents across the six sites:

1. respondents who completed waves 1 and 2 (not wave 3) – 217 participants

2. respondents who completed wave 1 (not wave 2) and wave 3 – 235 participants

3. respondents who did not complete wave 1 but completed waves 2 and 3 – 420 participants

4. respondents who completed waves 1, 2 and 3 (the complete cohort) – 277 participants

5. respondents who completed only one wave, whether wave 1, 2 or 3 (cross-sectional data) – 5168 participants.

Data cleaning

The data collected via the core survey were cleaned using range, consistency and logic checks to confirm their quality. These checks involved identifying the correct codification of the responses, examining missing values or abnormal patterns in the data, assessing the average interview length or the performance of the interviewers, etc.

As part of this quality control, abnormal patterns in the data were noted, principally in relation to the score of the primary outcome (PSS). It was detected that an unexpectedly large number of participants appeared to have a PSS score of 0 (indicating no stress at all) at waves 2 and 3. Furthermore (but in wave 2 of the survey only), an unexpectedly high number of participants reported a PSS score of 21. Checks on interviewer identities and process revealed that five interviewers were associated with these suspect cases throughout the follow-up household surveys. A total of 857 cases with apparently unreliable PSS scores were linked to these interviewers, of which 426 corresponded to wave 2 and 431 to wave 3.

Although no other suspicious response patterns were observed in the data sets created by these interviewers, the possibility of data fabrication could not be discounted. It was decided that, to ensure the reliability of the analysis, all data collected by the interviewers whose results for PSS were questionable needed to be excluded. The 857 problematic cases were deleted from the final sample, reducing the original data set from 6317 to 5460. The number of losses from wave 2 and the number of losses from wave 3 (the two follow-up surveys) were about equal. The analyses presented in this report use this reduced sample.

Approach to analysis

Two general approaches to quantitative analysis informed the detailed statistical analysis undertaken. First, an overall effect was estimated. This was essentially an intention-to-treat (ITT) approach. It tested whether or not living in an intervention site alone was sufficient to produce primary and secondary outcomes of interest, regardless of an individual’s exposure to the intervention (captured by their engagement with woodlands, green space and natural environments).

The ITT approach considered the magnitude of the interactions between living in an intervention site (or not) and the wave of the survey. This therefore captured the differential impact between the intervention and control groups, in relation to the effect of the WIAT programme. The models assessed the effect of the WIAT programme by comparing the differential impact after physical (phase 1) interventions (wave 2) and after both physical and social (phase 2) interventions (wave 3) with respect to the baseline (wave 1). For this, although each difference estimate (captured by the interaction terms) was statistically determined by its own p-values in the models, the differences between the two post-intervention phases were established on the basis of the p-values of a conventional Wald test.

Second, we augmented this analysis by providing a closer inspection of the intervention effect on our primary outcome (perceived stress). This was possible by examining the differential impact of the WIAT programme as a function of three main factors: (1) levels of engagement with the woods (physical and visual), (2) gender and (3) distance to the woods. This augmented approach was used to model our primary outcome and required a rather different analytical strategy to estimate the intervention effect. A three-way interaction term was added in the models, denoted by the binary variables of type of site (i.e. intervention or control) and wave of the survey plus the corresponding indicators on levels of engagement with the local woods, gender or distance to the woods. The ‘main effect’ within each level of these three indicators was then given by calculating a joint test of interaction terms.

The two sets of analyses involved a series of multilevel regression models. The use of a multilevel framework was required because our full sample, created by three repeated surveys sampling individuals within spatially defined communities, also included a proportion of individuals who participated at more than one wave. Therefore, the models needed to allow for repeated observations nested within individuals as well as spatial clustering. Our multilevel approach accounted for the fact that observations made on the same individual at two different waves were likely to be correlated. For this reason, we used individuals – the lowest level in the data – as the clustering variable in all the statistical specifications.

Given the continuous and binary forms of our different outcome measures (see Chapter 3 for details of how these were derived), the analysis involved a combination of linear and logistic regressions. All the models were adjusted for a substantial set of individual-level characteristics considered to be potential confounders for any intervention effect. The selection was made a priori and was based on existing literature describing relationships between sociodemographic variables and both access to and use of natural environments.

As 4410 out of 5460 observations had complete information (81%), imputation techniques were considered to handle missing data. Imputation was used only for data from a particular survey wave that were missing for an individual who had participated in that wave of the survey. In other words, no participant’s data were imputed other than for variables in the survey wave in which they had participated.

Specifically, we followed Rubin’s rules to perform multiple imputations via chained equations. This imputation technique was used owing to its greater flexibility to account for uncertainty in the missing data mechanism; we assumed that the data were ‘missing at random’, meaning that the ‘missingness’ could be determined by known variables. The use of chained equations also had the advantage of being able to include different types of variables in the process (e.g. continuous, categorical, nominal). Appendix 2 provides further information on the approach taken.

All analyses were conducted using the cross-sectional data set (termed panel A) and repeated with the longitudinal cohort (termed panel B). Although the former enabled us to use all individuals regardless of the number of waves in which they participated, the longitudinal cohort (panel B) allowed us to track the same participants across time. Although the cohort was not necessarily a representative subsample, it provided a form of sensitivity analysis to corroborate (or otherwise) findings from the cross-sectional data. All analyses were conducted using Stata^® version 14 (StataCorp LP, College Station, TX, USA).

Chapter 3 describes the derivation of variables for considering primary and secondary outcomes in the study and details the analytical methods used.

Recruitment of community site auditors

Participants for the community-led audits were recruited initially through the baseline survey, in which respondents could indicate their willingness to be recontacted to participate in group walks or focus groups to help with the research. This produced comparatively few positive responses and so additional individuals were recruited through contact with local community groups and facilitators and through local advertising. Participants who took part in the first audit were invited to take part in all subsequent environmental audits. We sought a balanced sample in terms of gender but diversity with regard to age and life stage.

Inclusion and exclusion criteria

As with the community core survey, individuals aged ≥ 16 years living within the intervention or control communities within 1.5 km of the study woodland site were eligible to take part in the community-led audit sections of the study.

Sample

We aimed to recruit 10 diverse members of the community to each community-led environmental audit. In practice, we found that attendance varied greatly between sites and over time. It proved very difficult to recruit individuals to the audits programmed for control site A. Despite the best efforts of the study team, no members of the community could be recruited to the winter 2013 community-led environmental audit at this site, and only one community member could be recruited to the summer 2015 audit. Table 6 shows the participant numbers for each site at each audit point.

Approach to analysis

The audit data collected from both expert and community-led audits consisted of Likert scale scores from 1 to 5 for each of the 25 items considered and any additional textual comments. For each audit completed at each site, an average was calculated for each of the seven domains into which these items fall (maximum score of 5). A final score, summed over these seven domains (maximum score of 35), represents the overall perceived quality of the woodland and its immediate surroundings. The means and SDs of total scores were calculated across all community group participants and, separately, across the two expert auditors for each site at each time point. Appendix 3 contains the detailed results of these audits.

The resulting audit scores were used to compare intervention and control sites at each audit time point and to compare any changes over time in perceptions of the woodlands at each site. This allowed an assessment of whether or not the interventions had resulted in perceptions of improved quality of the woodlands over time compared with control sites (which had no interventions). The community audit scores were also compared with the expert scores. The use of both winter and summer audits were important because each woodland site varied in appearance quite markedly between these seasons. For example, features as varied as distant views or local litter, which might be covered by abundant summer vegetation, could be more clearly seen in winter audits. Evidence of types of use were also different between summer and winter.

To better understand the perceptions leading to community audit scores, any textual comments were also reviewed under each of the seven domains of the tool. Finally, these were compared with the themes that arose from the community focus groups (see Community focus groups and interviews) to reveal any common themes or disparities in community perceptions.

Recruitment of community members to focus groups

As with the community-led audit participants, focus group participants were initially recruited through the baseline survey, which allowed individuals to indicate if they wished to take part in focus groups or group walks. Additional participants were recruited to these sections of the study through contact with local community groups and facilitators and through local advertising (see Report Supplementary Material 6). We sought a balanced sample in terms of gender but diversity regarding age and life stage (Table 7).

When we were recruiting participants to the focus groups, a small number of individuals (n = 4) expressed a wish to take part in the study but were unable or unwilling to take part in any of the planned focus groups. Consequently, to facilitate their involvement, these individuals were invited to take part in a one-to-one telephone interview with a member of the study team. The topic guide, as employed within the focus groups, guided the interviews. The interviews were audio-recorded, with participants’ consent, and transcribed. Each interview took approximately 30 minutes.

Inclusion and exclusion criteria

As with the community core survey, individuals aged ≥ 16 years living within the intervention communities within 1.5 km of the study woodland site were eligible to take part in the community focus groups for the study.

Sample

We aimed to recruit six members of the community to each focus group, giving a sample of 18 participants across each set of focus groups. Ultimately, 11 individuals took part in the first set of three focus groups (one per site), with a further individual engaged through a telephone interview. The second set of three focus groups (one per site) featured 19 participants with a further three individuals participating in telephone interviews (see Table 7).

Approach to analysis

A hybrid approach was used to analyse the focus groups and interview transcripts. This sought to identify themes that emerged from the participants in addition to those imposed by the structure of the topic guide shown in Report Supplementary Material 5. Thus, we structured coding of the transcripts using both predetermined and an inductive thematic analysis. 76 Multiple siftings of the data were used to clarify and refine the themes. All transcripts were double coded and any variations in coding agreed and reconciled prior to producing the final thematic framework (see Figure 20).

Analysis paid particular attention to discordant voices or differing opinions and indications of unanticipated outcomes from the interventions. Interpretation of the themes was also informed by the outcomes of the community audits (see Environmental audits) to triangulate findings when possible.

Costs of interventions

In addition to health outcomes data collected via the core community survey, the economic evaluation drew on resource data collected directly from FCS. This analysis costed the time commitment of members of the FCS team involved in supporting the physical intervention, including the time spent administering the contracting process and monitoring compliance with the successful contractor. It also included time spent administering and delivering the programme of social interventions. An assessment of the costs of the WIAT programme was developed by using a top-down approach. 77,78 Units costs were determined by the FCS internal and external costing model: internal unit costs in the form of pay rates per day for various grades were applied to various activities related to the delivery of the interventions. To carry out the physical intervention, the work was contracted to third parties and, therefore, the cost equated to the contract value plus the additional monitoring/administration time of FCS using internal pay rates per day for a given grade. Staff involved in the WIAT programme at FCS regularly completed a costing model in Microsoft Excel^® version 16 (Microsoft Corporation, Redmond, WA, USA) to estimate this time commitment for the length of the study. The model recorded percentage time commitment in days on a monthly basis for different grades of staff. This time commitment was costed at an agreed unit rate that covered staff salaries and overheads and FCS support for both the physical and social interventions. The costs (see Table 27 for detailed costs relating to 2013/14) represent the fully costed input for the delivery of the interventions.

Assessment of health-related quality of life

The primary focus of the economic evaluation was the core survey response on the five-item EQ-5D scale that captures a description of five dimensions of health state. This allows the derivation of utilities to calculate quality-adjusted life-years (QALYs).

The EQ-5D data collection used both the old three-level version (EQ-5D-3L) and the new five-level version (EQ-5D-5L) of the EQ-5D questionnaire. The EQ-5D-3L was used in wave 1 and the EQ-5D-5L was used in waves 2 and 3. In both versions the health states are reported as an index on the questionnaire responses, with 11111 for full health for both versions and 33333 and 55555 for worst health for the EQ-5D-3L and EQ-5D-5L respectively. For the EQ-5D-3L, this index was used to derive utilities from the predetermined values sets obtained from the UK general public. 79 At the time of analysis, there were no utility value sets for the EQ-5D-5L for the wider UK population, only for England. 80 Furthermore, the EQ-5D-5L is not recommended for use by the National Institute for Health and Care Excellence (NICE) until the impact of its adoption is fully explored, as it has been found that improvements in HRQoL are valued less with the EQ-5D-5L than the EQ-5D-3L, with implications for economic evaluation. 81

When both the EQ-5D-3L and EQ-5D-5L have been used, NICE guidelines recommend the mapping approach, also known as a crosswalk approach, to ensure consistency of HRQoL utilities. 82 We used the crosswalk approach to calculate HRQoL utilities from responses from the EQ-5D-5L index profiles in waves 2 and 3 that were consistent with the HRQoL utilities responses from the EQ-5D-3L index profiles in wave 1. This approach is based on the distribution similarities of the two versions of the EQ-5D questionnaires. 83 The EQ-5D-5L health states provide 3125 indices that are distributed on a scale of –0.594 to 1 (index profile 55555 = –0.594 and 11111 = 1)83 and the EQ-5D-3L version provides 243 indices that are distributed on the same scale of –0.594 to 1 (index profile 33333 = –0.594 and 11111 = 1). 79 These distribution similarities provide comparability of the two versions of the EQ-5D. A Microsoft Excel tool known as the ‘EQ-5D-5L Crosswalk Index Value Calculator’ developed by the EuroQol group was used to calculate the crosswalk index values for the EQ-5D-5L dimension scores.

Cost–consequences analysis

Cost–consequences analysis (CCA) was used to present the total cost of the intervention and the primary and secondary outcomes of the WIAT interventions in a balance sheet format. 84 This approach presents policy- or decision-makers with a comprehensively wide range of outcomes to judge the impact of the intervention, including both health effects and non-health effects. However, it stops short of a full evaluation as it does not present the values placed on the outcomes that are affected by the intervention.

Cost–utility analysis

A cost–utility analysis (CUA) is the most common form of evaluative method used in health economic evaluation. It compares the total costs of the intervention with estimated QALYs gained from the intervention based on the estimated impact on HRQoL utilities of the sort provided by the EQ-5D descriptive system responses. In an exploratory analysis, a CUA was conducted from the EQ-5D responses for the WIAT intervention over the time scale of the study.

Sensitivity analysis

Uncertainty surrounding the costs and EQ-5D HRQoL utilities was quantified through a probabilistic sensitivity analysis (PSA). 85,86 This was undertaken using 5000 Monte Carlo simulations that repeatedly created random data through bootstrapping. A gamma probability distribution was assigned to the cost parameter, computed from the variation of individual costs in the costing model. 87,88 The EQ-5D HRQoL results from the adjusted ITT models were assigned a normal distribution with point estimates and standard errors (SEs) taken directly from the regression results. The PSA allowed the estimation of 95% confidence intervals (CIs) from the bootstrapped replicates of data using the percentile approach with a lower and upper percentile of 0.025 and 0.975 respectively. 88

All the bootstrapping was performed in Microsoft Excel and implemented using a Microsoft Excel macro. The results of all simulations were combined to give overall incremental cost-effectiveness ratio (ICER) results. The bootstrapped pairs of incremental cost and incremental QALYs are presented using the cost-effectiveness plane. The cost-effectiveness plane depicts the point in the quadrant where each bootstrapped pair of the incremental cost and incremental QALY is positioned. The vertical and horizontal axes represent incremental costs and incremental QALYs respectively.

Baseline characteristics of participants

Tables 8 and 9 draw on imputed data to give baseline descriptive characteristics of participants for the cross-sectional data set (panel A) and cohort sample (panel B) respectively. Because all the covariates take a categorical form, we report proportions, stratified by study arm. To inspect the extent of imbalance in the covariates at baseline, a p-value for test of differences is also reported.

At baseline, panel A comprises a total of 2117 participants (see Table 8). Of these participants, 609 correspond to the cohort sample (see Table 9), panel B, nested within the cross-sectional data set.

In the following commentary we note only significant differences based on p < 0.05.

At baseline, panel A (see Table 8) had differences between intervention and control samples in two of the age groups and more ‘much worse than normal’ life events in the control group. There were more lower social class participants in the intervention group and more higher social class participants in the control group. The difference is more significant with regard to the educational levels of the two groups, with the intervention group having more participants with no qualifications and the control having more with higher and further educational levels. This difference is also reflected in the larger numbers finding it difficult to cope on their income in the intervention group and more in the control group living comfortably on their income. Despite attempts to match sites in accordance with physical characteristics, including woodlands, the intervention group had fewer people living close to their local woodlands and more living at a distance of ≥ 751 m than the control group. The intervention group also had lower numbers of participants with access to a car. In health terms, there were more smokers in the intervention group. The control group had more dog owners than the intervention group. There were no other significant differences between the two groups in the sample.

There were some differences in profile between panel A and panel B. As with panel A at baseline, in the panel B sample (see Table 9) many individual characteristics were not significantly different between the intervention and control groups but there were a number of significant differences. Unlike panel A, there were no differences in the age profile between intervention and control groups and the control group life events ratings included more ‘better than normal’. The pattern of significant differences for social class, educational qualifications, income coping, proximity to the local woodlands, access to a car and smoking status matched that of panel A, with the intervention group showing a disadvantaged profile in these characteristics compared with the control group. Unlike in panel A, there are also significant differences in the proportions of the panel B sample from intervention or control sites in each of the site pairs.

Follow-up characteristics of participants in the wave 2 and 3 sample

The characteristics of participants in the two follow-up waves of data, both for the cross-sectional data set and for the cohort sample, differ from baseline in some important ways (see Appendix 4 for details).

There was a relative reduction in the sample size owing to the exclusion of problematic cases from the final data set as described earlier. The repeat cross-sectional data set used for analysis from waves 2 and 3 had 1672 and 1671 cases respectively. Likewise, the number of cohort participants was 350 at wave 2 and 402 at wave 3.

Panel A at wave 2 showed no significant differences between intervention and control in terms of age, whereas there were differences in two of the age groups in wave 3. Life events patterns were more varied across the two groups (more in the control group were ‘better than normal’ in both waves 2 and 3). There were a number of differences in social class (wave 2 only) and in educational qualification (waves 2 and 3), matching those found in wave 1, in which the intervention group mostly showed lower social class and educational qualifications than the control group. Unlike in wave 1, there were more not in work in the intervention group in waves 2 and 3. As in wave 1, more in the intervention group were finding it hard to cope on their income and more in the control group were living comfortably on their income (waves 2 and 3). As with wave 1, the intervention group had fewer people living close to woodlands and more living further away than the control group. As in wave 1, the intervention group in waves 2 and 3 had comparatively lower levels of access to a car, higher numbers smoking and lower dog ownership. In wave 3 only, there were more in the intervention group whose health was limited a little and more in the control group with no health limitations. There was a notable difference in sample size between the site pairs: there were significantly fewer participants in intervention site A (in both waves 2 and 3) and in control site B (wave 2 only).

In the panel B sample for waves 2 and 3 there was only one significant difference in one age group (more participants aged 55–64 years in the wave 2 control sample) between intervention and control samples. As with wave 1, wave 2 control participants had more life events in the ‘better than normal’ category than the intervention group. Social class and educational qualifications patterns in waves 2 and 3 largely matched those of wave 1, as might be expected, with differences between intervention and control groups more significant in wave 3 than in wave 2. As with wave 1, there were significantly more in the intervention group finding it difficult to cope on their income in waves 2 and 3, with the control group having significantly more living comfortably on their income by wave 3. Patterns of proximity to local woodlands and access to a car matched those of wave 1 in waves 2 and 3, with differences between intervention and control being more significant by wave 3. Differences in smoking were not significant in wave 2 but were significant in wave 3, with the control group having fewer smokers, as in wave 1, and the intervention group having significantly more who smoked in the past in wave 3, compared with significantly more currently smoking in wave 1. The only other significant differences were between site pairs, with a comparatively larger sample size in intervention site B and in control site C in waves 2 and 3 (as in wave 1), whereas the comparatively larger sample size in intervention site A (significant in waves 1 and 2) was no longer significant by wave 3.

Primary outcome variable: Perceived Stress Scale

The primary outcome of the study was perceived stress, measured through the 10-item PSS. We followed standard rules to derive our PSS score. The raw measure included four positively stated items and six negatively stated items. The items were based on Likert scale responses of ‘never’, ‘almost never’, ’sometimes’, ‘fairly often’ and ‘very often’. The final score was then calculated by reversing responses to the four positively stated items and then summing across all 10 scale items. The resulting variable was treated as a continuous variable ranging from 0 to 40, with lower values denoting better mental health outcomes.

Secondary outcome variables

The study considered a set of secondary outcomes, summarised in Table 10 (see Report Supplementary Material 1 for the full questionnaire and Appendix 5 for full reporting of all secondary outcome data). These secondary outcomes are related to the different pathways hypothesised to underpin the relationship between natural environments and mental health, particularly perceived stress, as shown in our logic model (see Figure 1).

The derivation of secondary outcomes proceeded as described in the following sections.

Health-related outcomes

Behaviour outcomes

Engagement with woods and their use

The social (phase 2) WIAT interventions were aimed at encouraging the local community to participate in a range of activities in the woods and to find out more about the woods.

Enhanced environment

Unadjusted patterns of stress

Table 14 reports the unadjusted patterns of stress levels of participants by study arm in the cross-sectional data set (panel A) and longitudinal cohort sample (panel B). As described earlier, because our self-reported measure of PSS is a continuous variable, we report means in stress (where a higher number indicates a greater level of stress).

Overall, there are similar unadjusted patterns of stress in the cross-sectional and cohort samples. On average, participants living in the intervention sites had lower stress levels than those in the control sites at wave 1 in both panel A and panel B data sets. These differences were statistically significant, as reflected by the test of equality of means reported in final column of Table 14.

By the post-physical intervention survey (wave 2), the unadjusted patterns of stress between the two groups had moved in opposite directions. Intervention site participants became, unexpectedly, more stressed on average. This pattern was more pronounced in panel A than in panel B. For example, although panel A showed a rise of 1.3 in the measure of perceived stress for the intervention participants, panel B showed an increase of about 0.4. There was a moderate reduction in stress for the control sites: a pattern that was observed in both and panel B, with a decrease of stress of approximately 0.4 and 0.7 respectively.

Finally, in the post-social-intervention survey (wave 3), we observed the effect of a similar trend in stress levels as at wave 2. Stress levels for the intervention participants had again risen, whereas those in our control participants had again decreased. Among the intervention site panel B participants, there was a rise of 1.7 in the measure of stress (PSS) by wave 3 over baseline levels. A greater PSS score increase of 2.7 over baseline by wave 3 was detected in panel A (see Table 14). In contrast, mean stress levels in the control sites for both panels were lower post social intervention (wave 3).

Unadjusted patterns of stress based on levels of physical and visual engagement with natural environments

Unadjusted patterns of stress by levels of physical and visual engagement with natural environments were considered for both panel A and panel B, comparing intervention and control groups. Physical engagement with natural environments was defined by nature visits as described in Preparing outcome variables for analysis, that is, whether or not participants had visited the local woods or other green spaces in the previous 12 months at each survey wave. Visual engagement with natural environments was defined by nature views as described in Preparing outcome variables for analysis, that is, whether or not participants were aware of views to natural environments while walking in the neighbourhood.

Post intervention (waves 2 and 3), there was an increase in stress within the intervention groups, regardless of whether or not they had physical or visual engagement with the natural environment. However, when stratified by nature visits, these PSS score increases were greater in the intervention participants who did not undertake nature visits throughout the study (non-nature visits group) than in those who did undertake nature visits. When considering nature views, at wave 2 there was a marked increase in stress in the nature views group at intervention sites in both panels, which was not apparent in the non-nature views group for intervention sites. By contrast, at wave 3 the non-nature views group in the intervention sites showed sharp increases in PSS scores, whereas the nature views groups saw PSS scores fall in comparison with wave 2. The full unadjusted data for these variables are reported in Appendix 6.

These initial analyses suggested that the detrimental unadjusted pattern in stress established in Patterns of stress in the study sample for intervention sites may have been driven more by participants who were not affected by the intervention. For instance, in the cohort sample (panel B) there was an increase of 2.7 in mean PSS score within the non-nature visits group between wave 1 and wave 3, whereas there was only an increase of 0.7 in mean PSS score within the nature visits group. However, these unadjusted analyses do not take into account either individual-level characteristics or between-site differences in sample composition, and so the more detailed reporting of outcomes, as set out below (Is the intervention associated with changes in the primary outcome of perceived stress? to Sensitivity analysis), is based on models adjusted to take such variables into account.

Results of adjusted models

The multilevel models presented for the ITT approach were adjusted for participants’ age, gender, life events, social class, education, working status, income coping, access to car, smoking status, disability, health status, dog ownership, children, distance bands (to local woods: ≤ 150, 151–300, 301–500, 501–750 and 751–1500 m; i.e. five distance bands), site pair, type of site (intervention or control) and wave of survey. The augmented models included all these covariates plus binary variables for nature visits in the past year (‘yes’/’no’) and gender.

Overall effect: intention-to-treat approach

The results of the adjusted models using the ITT approach are presented in Table 15. We report only the coefficients for the interaction terms of type of site, that is, intervention or control (reference group: control arm) and wave of survey (reference group: baseline, wave 1). As described before, these interaction terms captured the differential impact of the intervention during the follow-ups. Dichotomised variables for type of site and survey wave were also included in all the models. Results based on the cross-sectional data set can be seen under panel A, whereas panel B shows the results using the longitudinal cohort.

TABLE 15 - The WIAT intervention effect on perceived stress (ITT approach)

B, unstandardised coefficient; n/a, not applicable.

Note

Multilevel models adjusted for age, gender, life events, social class, education, working status, income coping, access to car, smoking status, disability, health status, dog ownership, children, distance bands (to local woods: ≤ 150, 151–300, 301–500, 501–750 and 751–1500 m; i.e. five distance bands), site pair, type of site (intervention or control) and wave of survey.

Variable	B	p-value	95% CI	p-value for Wald test
Panel A: cross-sectional sample
Intervention × wave 2	1.52	< 0.001	0.78 to 2.27	n/a
Intervention × wave 3	3.58	< 0.001	2.85 to 4.31	< 0.001
Panel B: cohort sample
Intervention × wave 2	1.11	0.16	–0.46 to 2.68	n/a
Intervention × wave 3	3.03	< 0.001	1.54 to 4.52	0.004

Based on the cross-sectional analysis (panel A; see Table 15), stress levels in participants from intervention sites increased during the follow-ups (waves 2 and 3) compared with those in the control groups. The magnitude of the intervention effect was different at each wave, with larger observed changes at wave 3 [B (unstandardised coefficient) 3.58, 95% CI 2.85 to 4.31; p < 0.001] than at wave 2 (B 1.52, 95% CI 0.78 to 2.27; p < 0.001). These differences between intervention and control sites across the two post-intervention phases were statistically significant [as measured by the p-value (< 0.001) using the Wald test].

The adjusted predicted means of the PSS scores for the two groups, intervention and control, are shown in Figure 7. The size of the intervention effect is noteworthy. With regard to the baseline (predicted mean of PSS score = 12.09), by wave 2 (predicted mean of PSS score = 13.5) the predicted mean of the PSS scores within the intervention sites increased by approximately 1.5, and by wave 3 (predicted mean of PSS score = 14.5) a gain of 2.4 was achieved. On the other hand, the predicted means of the PSS scores within the control groups underwent almost no change by wave 2 and a moderate decrease of 1 by wave 3.

FIGURE 7.

Predicted means of PSS scores for panel A: cross-sectional sample. A higher PSS score denotes higher stress, controlling for different individual-level characteristics.

The results using the longitudinal cohort (panel B; see Table 15) echoed those from the cross-sectional sample. For example, post social intervention (wave 3) a detrimental effect of similar size for the intervention site participants was again established (B 3.03, 95% CI 1.54 to 4.52; p < 0.001). One important difference between the two sets of results is that the differential impact within the intervention sites post physical intervention (wave 2) was not statistically significant (B 1.11, 95% CI –0.46 to 2.68; p = 0.16). Overall, the p-value (0.004) of the Wald test confirmed the significance of the apparent detrimental effect. The adverse effect was smaller in the longitudinal cohort. Based on the predicted means of the PSS scores (Figure 8), from a baseline predicted mean of 13.8, stress levels within the intervention sites increased to 14.1 at wave 2 and 14.9 at wave 3.

FIGURE 8.

Predicted means of PSS scores for panel B: cohort sample. A higher PSS score denotes higher stress, controlling for different individual-level characteristics.

Results for the secondary outcome: mental well-being

To confirm the findings, analyses were repeated with an alternative measure of mental health: the SWEMWBS. SWEMWBS results range from scores of 7 to 35, with higher values denoting more positive mental well-being. The results can be seen in Table 16. The signs of all the coefficients for the interactions in both data sets (panels A and B) are negative, thus indicating a similar pattern of change over time in intervention as in the control (i.e. a detrimental intervention effect). Figure 9 shows predicted means of SWEMWBS using the panel A (cross-sectional) data.

FIGURE 9.

Predicted means of SWEMWBS scores for panel A: cross-sectional sample. A higher SWEMWBS score denotes better mental well-being, controlling for different individual-level characteristics.

Do changes in the primary outcome differ in accordance with levels of physical and visual engagement with natural environments?

In Tables 17 and 18 we refine the analysis to differentiate between participants in accordance with their levels of physical and visual engagement with natural environments. The ‘main’ intervention effect on stress levels of intervention site participants within each engagement group is reported using both the cross-sectional (panel A) and longitudinal cohort (panel B) samples.

Intervention effect on stress via levels of physical engagement – nature visits

Table 17 supports the notion that the detrimental effect on stress levels within the intervention group seems to be largely driven by participants who were not physically exposed to nature. The cross-sectional analysis (panel A; see Table 17) suggested a detrimental intervention effect on intervention site participants who were within the non-nature visits group: their interaction coefficients were positive and highly statistically significant at each point in time (wave 2: B 3.04, 95% CI 2.00 to 4.07; p < 0.001; wave 3: B 4.97, 95% CI 3.95 to 5.99; p < 0.001). By contrast, the intervention site participants within the nature visits group saw a very much smaller increase (B 1.9, 95% CI 0.8 to 3.0; p < 0.001). Results from the cohort sample (panel B; see Table 17) support this. There was no intervention effect on the intervention site participants who were within the nature visits group (wave 2: B –0.23, 95% CI –2.42 to 1.95; p = 0.83; wave 3: B 0.64, 95% CI –1.6 to 2.88; p = 0.57) but there was a significant increase in stress levels among the intervention site participants within the non-nature visits group at wave 2 (B 2.57, 95% CI 0.3 to 4.83; p = 0.03) as well as at wave 3 (B 4.69, 95% CI 2.66 to 6.71; p < 0.001).

Figure 10 shows the adjusted predicted means of the PSS scores computed using the difference estimates from the model of the cohort sample in panel B of Table 17. In the intervention group, from a baseline predicted mean PSS score of 13.6, the non-nature visits group had a predicted mean PSS score of 15.5 by wave 3, whereas those within the nature visits group had a difference of only 0.3 between the predicted mean PSS score at baseline (13.9) and at wave 3 (14.2). The non-nature visits group within the control groups, on the other hand, showed a decline in stress levels from a baseline mean PSS score of 15.4 to 12.6 at wave 3 and, for the nature visits group, a smaller difference in predicted mean PSS from a score of 15.1 at baseline to 14.6 at wave 3. It is evident that, although the differential effect of the intervention across the study arms (intervention and control) is not significant in the nature visits group, the effect on the non-nature visits group diverged considerably between intervention and control. This suggests that factors other than those related to visits to natural environments may lie behind the apparent intervention effect on stress levels.

FIGURE 10.

Predicted means of PSS scores by nature visits for panel B: cohort sample. (a) Intervention group; and (b) control group. A higher PSS score denotes higher stress, controlling for different individual-level characteristics.

Intervention effect on stress via levels of visual engagement – nature views

The main findings in relation to the intervention effect on stress by levels of visual engagement with natural environments are reported in Table 18. There is no simple picture apparent. In panel A, a strong detrimental effect on intervention site participants was found across both the nature view and non-nature view groups at all follow-up surveys (nature view group: wave 2 – B 1.54, 95% CI 0.53 to 2.54; p = 0.003; wave 3 – B 2.95, 95% CI 2.00 to 3.91; p < 0.001; and non-nature view group: wave 2 – B 2.06, 95% CI 0.95 to 3.17; p < 0.001; wave 3 – B 3.75, 95% CI 2.54 to 4.97; p < 0.001). However, results using panel B were quite different: a significant detrimental intervention effect for the nature view group was found only at wave 3 (B 3.19, 95% CI 1.30 to 5.09; p = 0.001).

Adjusted predicted means of the PSS scores from regressions in panel B of Table 18 are presented in Figure 11, which shows increases in the PSS score for the intervention site participants. Within the nature views group, predicted mean stress levels among intervention participants were 13.1 at baseline, 14.4 at wave 2 and 14.4 at wave 3. Participants within the control sites, on the other hand, were on average less stressed during the follow-ups, regardless of their degree of visual engagement with natural environments. For example, from baseline to post social intervention (wave 3), those from control sites and in the nature views group were about 2 PSS points less stressed. Results for the non-nature views group were similar.

FIGURE 11.

Predicted means of PSS scores by nature views for panel B: cohort sample. (a) Intervention group; and (b) control group. A higher PSS score denotes higher stress, controlling for different individual-level characteristics.

Analytical approach

Secondary outcomes were assessed mainly via the ITT approach. As noted before, an ITT approach enabled us to estimate change associated with the WIAT intervention within the intervention arm, relative to the control arm. Furthermore, as with our primary outcome, all secondary outcome analyses were undertaken based on both panel A and panel B. Likewise, when it is deemed relevant, the effect size of the intervention is presented using adjusted predicted means.

Results

Health-related outcomes: physical activity, connectedness to nature and social cohesion

Results can be seen in Table 19. They show the significant changes in participants’ health behaviours associated with the WIAT intervention programme.

TABLE 19 - The WIAT intervention and health-related outcomes: PA, connectedness to nature and social cohesion

*p < 0.5, **p < 0.01, ***p < 0.001.

Model estimates shown in terms of MET-minutes per week.

Model estimates shown based on a visual scale ranging between 0 and 7, measuring participants’ sense of connectedness to nature.

Model estimates shown based on a scale ranging between 3 and 12, capturing participants’ community collective strength.

Note

Each row in panels A and B reports interaction coefficients of type of site and wave for separate adjusted models.

Multilevel models adjusted for age, gender, life events, social class, education, working status, income coping, access to car, smoking status, disability, health status, dog ownership, children, distance bands (to local woods: ≤ 150, 151–300, 301–500, 501–750 and 751–1500 m; i.e. five distance bands), site pair, type of site (intervention or control) and wave of survey.

Outcomes	Wave, B (95% CI)		Wald test p-value
	2	3
Panel A: cross-sectional sample
(1) Vigorous activitya	–152.9 (–422.6 to 116.8)	221.20 (–43.46 to 485.90)	0.03
(2) Moderate activitya	–215.40** (–409.40 to –21.39)	249.20** (58.25 to 440.10)	< 0.001
(3) Walking activitya	203.3** (36.81 to 369.8)	–40.87 (–204.50 to 122.80)	0.01
(4) Overall PAa	–282.4 (–732.1 to 167.3)	275.2 (–163.2 to 713.5)	0.07
(5) Connectedness to natureb	–0.19* (–0.38 to –0.01)	0.39*** (0.20 to 0.57)	< 0.001
(6) Social cohesionc	0.44*** (0.22 to 0.65)	0.50*** (0.29 to 0.70)	< 0.001
Panel B: cohort sample
(1) Vigorous activitya	41.24 (–457.4 to 539.9)	382.3 (–87.75 to 852.4)	0.26
(2) Moderate activitya	–103.4 (–470.6 to 263.8)	559.3*** (211.3 to 907.2)	0.001
(3) Walking activitya	–11.57 (–346 to 322.8)	144.1 (–170.8 to 459)	0.61
(4) Overall PAa	–379.8 (–1185.9 to 426.3)	861.5** (106.5 to 1616.4)	0.02
(5) Connectedness to natureb	–0.29 (–0.674 to 0.102)	0.15 (–0.22 to 0.51)	0.13
(6) Social cohesionc	0.01 (–0.39 to 0.41)	0.02 (–0.36 to 0.39)	0.99

In most of the PA outcomes (e.g. vigorous PA) there was some evidence of an intervention effect. For example, compared with those living in control sites, levels of moderate activity among intervention site participants increased by wave 3. This was observed in both the cross-sectional sample (B 249.2, 95% CI 58.25 to 440.10; p = 0.01) and cohort sample (B 559.3, 95% CI 211.3 to 907.2; p = 0.002). Taking the three PA levels together (vigorous, moderate and walking), intervention site participants appeared, overall, to be more physically active post social intervention (wave 3), as reported by the cohort sample in adjusted model (4) of panel B (B 861.5, 95% CI 106.5 to 1616.4; p = 0.025).

Figure 12 shows the adjusted predicted means of the total PA score, computed from the difference estimates of the adjusted model (4) of the cohort sample in panel B of Table 19. The graph shows an increase among intervention site participants in the total PA score (measured in terms of MET-minutes per week) from baseline to the two follow-ups (waves 2 and 3). It is important to note that the predicted means in the intervention and control sites were almost identical at baseline but that a difference had emerged by wave 3. Compared with the baseline predicted mean of overall PA of 2081.8, intervention site participants’ mean PA score by wave 3 was 2727.2, whereas those in the control sites saw mean PA levels drop to 1886.6 by wave 3 from 2102.7 at baseline. At wave 3, the difference between the two groups was notable: on average, intervention site participants had a PA score 841 MET-minutes per week higher than that of the control participants.

FIGURE 12.

Predicted means of overall PA score for panel B: cohort sample. An increase in the total IPAQ-SF score denotes higher levels of PA. Controlled for different individual-level characteristics.

Evidence for an intervention effect on the other two health-related outcomes was more mixed. For the cross-sectional sample (panel A), compared with those in the control sites, intervention site participants felt relatively less connected to nature at wave 2 than at baseline (B –0.19, 95% CI –0.38 to –0.01; p = 0.044), but at wave 3 this was essentially reversed, with intervention site participants feeling more connected to nature than control participants (B 0.39, 95% CI 0.20 to 0.57; p < 0.001). The pattern was similar but not significant within the cohort sample (panel B).

With regard to social cohesion, based on the cross-sectional data set [adjusted model (6), panel A], the intervention site participants reported stronger levels of social cohesion by waves 2 and 3 than the control site participants (wave 2: B 0.44, 95% CI 0.22 to 0.65; p < 0.001; wave 3: B 0.5, 95% CI 0.29 to 0.70; p < 0.001). However, in the cohort sample (panel B) an intervention effect was not detected (wave 2: B 0.01, 95% CI –0.39 to 0.41; p = 0.98; wave 3: B 0.02, 95% CI –0.36 to 0.39; p = 0.93).

Behaviour outcomes: nature visits, length and frequency of visits to the local woods, and nature views

Table 20 presents the results for the set of secondary outcomes linked to the behaviour domain. Because all of these behaviour measures were set in binary form, logistic regressions were used. We therefore report odds ratios (ORs) along with their respective CIs and p-values. Furthermore, it should be noted that the adjusted models, in which the length and frequency of visits to the woods were estimated, are reported based solely on the cross-sectional data set. This is because these models used only participants who had visited the target woodland areas for the study, and this reduced the sample significantly. For this reason, we were unable to exploit the cohort sample (panel B), which yielded a total sample size of only 314 observations, as models were likely to be overfitted (i.e. with several parameters fitted to a relatively small sample).

TABLE 20 - The WIAT intervention effect on behaviour outcomes: nature visits, length and frequency of visits to the woods, and visual contact with natural environments

*p < 0.01, **p < 0.001.

Model uses a reduced sample size because only participants who had visited the target woodland areas for the study were included (n = 1393).

Note

Each row in panels A and B reports interaction coefficients of type of site and wave of separate adjusted models.

Multilevel models adjusted for age, gender, life events, social class, education, working status, income coping, access to car, smoking status, disability, health status, dog ownership, children, distance bands (to local woods: ≤ 150, 151–300, 301–500, 501–750 and 751–1500 m; i.e. five distance bands), site pair, type of site (intervention or control) and wave of survey.

Outcomes	Wave, OR (95% CI)		Wald test p-value
	2	3
Panel A: cross-sectional sample
(1) Nature visits	1.33 (0.94 to 1.88)	2.69** (1.9 to 3.81)	< 0.001
(2) Length of woodland visitsa	0.43 (0.18 to 1.02)	0.83 (0.36 to 1.90)	0.15
(3) Frequency of woodland visits (summer)a	0.79 (0.43 to 1.45)	1.07 (0.57 to 1.99)	0.63
(4) Frequency of woodland visits (winter)a	1.45 (0.66 to 3.18)	0.82 (0.38 to 1.77)	0.42
(5) Visual contact with nature	0.88 (0.65 to 1.20)	0.43** (0.31 to 0.59)	< 0.001
Panel B: cohort sample
(1) Nature visits	1.56 (0.80 to 3.05)	2.77* (1.45 to 5.29)	0.008
(2) Visual contact with nature	1.43 (0.77 to 2.67)	0.64 (0.35 to 1.18)	0.08

The intervention did not appear to have had significant impact on length and frequency of visits to the specified local woods. The panel A analysis showed lower odds of visual contact with natural environments (nature views) for the intervention site participants by wave 3 (OR 0.43, 95% CI 0.31 to 0.59; p < 0.001), but the cohort sample did not exhibit a significant effect (wave 3: OR 0.64, 95% CI 0.35 to 1.18; p = 0.15). However, there was robust evidence to suggest that intervention site participants were more likely to undertake nature visits more generally than those in the control sites, an effect identified in both panel A and panel B. In both samples this was found post social promotion intervention (wave 3) (panel A: OR 2.69, 95% CI 1.90 to 3.81; p < 0.001; panel B: OR 2.77, 95% CI 1.45 to 5.29; p < 0.001).

Figure 13 shows the predicted probabilities of undertaking visits to nature computed by using the difference estimates from the adjusted model (1) of the cohort sample in panel B of Table 20. A higher predicted probability indicates a higher expected number of individuals visiting nature. At baseline, both study arms yielded comparable predicted probabilities. However, differences emerged by wave 3. For instance, within the intervention site participants, predicted probability increased to 53% by wave 3, from about 47% at baseline, whereas the predicted probabilities for those in the control sites reduced to approximately 40%, from 51%, over the same period.

FIGURE 13.

Predicted probabilities of nature visits for panel B: cohort sample. A higher predicted probability denotes a higher expected number of individuals visiting nature. Controlled for different individual-level characteristics.

We calculated the total increase in the expected number of individuals visiting natural environments by using the intervention site (n = 20,472) and control site (n = 12,615) populations meeting inclusion criteria from which the random sample was obtained. The effect translated to an additional 1228 individuals within the intervention sites undertaking visits to nature from baseline (expected number = 9622) to wave 3 (expected number = 10,850). By contrast, within the control sites, 1388 fewer individuals were likely to be visiting nature visits by wave 3 (expected number at baseline = 6434; expected number at wave 3 = 5046).

Secondary outcomes related to social support for environment use: awareness of the woods and activities undertaken within the woods

The results for the secondary outcomes associated with social support for woodland use can be seen in Table 21. Again, because all of the outcomes used within this category were dichotomised, ORs are reported. Because of the small sample size of participants undertaking activities within the local woods (the total sample size for panel B, as noted above, was only 314 observations) and thus to avoid overfitting models, we present results for these variables based only on the cross-sectional sample (panel A) for estimating changes in the different activities undertaken.

TABLE 21 - The WIAT intervention effect on secondary outcomes related to social support for environmental use: awareness of the woods and activities undertaken within the woods

*p < 0.01, **p < 0.001.

Model uses a reduced sample size because only participants who had visited the target woodland areas for the study were included (n = 1393).

Note

Multilevel models adjusted for age, gender, life events, social class, education, working status, income coping, access to car, smoking status, disability, health status, dog ownership, children, distance bands (to local woods: ≤ 150, 151–300, 301–500, 501–750 and 751–1500 m; i.e. five distance bands), site pair, type of site (intervention or control) and wave of survey.

Outcomes	Wave, OR (95% CI)		Wald test p-value
	2	3
Panel A: cross-sectional sample
(1) Awareness of the woods	2.26** (1.58 to 3.22)	3.1** (2.15 to 4.46)	< 0.001
(2) Go for a walka	2.98* (1.55 to 5.74)	3.3** (1.73 to 6.29)	< 0.001
(3) Walk a doga	0.53 (0.22 to 1.28)	0.72 (0.30 to 1.70)	0.36
(4) Go out with familya	3.42* (1.37 to 8.56)	2.14 (0.91 to 5.03)	0.03
(5) Relaxa	3.35 (0.61 to 18.28)	1.21 (0.23 to 6.40)	0.30
Panel B: cohort sample
(1) Awareness of the woods	3.85** (1.92 to 7.72)	3.39** (1.72 to 6.67)	< 0.001

Intervention site participants increased awareness of local woods to a much greater degree than control site participants. The intervention effect was large, held throughout the study (at waves 2 and 3) and was evident in the two sample panels (cross-sectional and longitudinal cohort). For example, using the results from the cohort sample (panel B), the OR for intervention site participants was 3.85 (95% CI 1.92 to 7.72; p < 0.001) at wave 2, followed by an OR of a similar size at wave 3: 3.39 (95% CI 1.72 to 6.67; p < 0.001). Panel A yielded similar results. Figure 14 presents the predicted probabilities of awareness of the woods between the two study arms using the difference estimates from panel B (cohort sample). It is interesting to note that participants from control sites were far more aware of their local woodlands at baseline than those from intervention sites but that this declined over time.

FIGURE 14.

Predicted probabilities of awareness of the woods for panel B: cohort sample. A higher predicted probability indicates a higher number of individuals likely to be aware of the woods. Controlled for different individual-level characteristics.

Within panel A, an intervention effect on the series of activities undertaken by participants within the woods was evident for the measures of going for a walk and walking with family. With regard to those in the control arm, the odds of going for a walk in the woods for intervention participants were 2.98 (95% CI 1.55 to 5.74; p < 0.001) at wave 2 and 3.3 (95% CI 1.73 to 6.29; p < 0.001) at wave 3. Similarly, the odds of walking with family for intervention participants were three times that of those in the control arm post physical intervention (wave 2) (OR 3.42, 95% CI 1.37 to 8.56; p = 0.009), although this level had dropped by wave 3.

Approach to analysis

To assess the differences in gender and distance bands to the woods, the augmented approach was taken, using the three-way interaction in the models. This enabled us to capture the differential impact of the WIAT intervention on perceived stress as a function of gender and distance to the woods. The main effects of the intervention on stress levels of intervention participants within each subgroup are presented based on the cross-sectional data set (panel A) and cohort sample (panel B).

Gender differences

Table 23 shows the difference estimates for the main effect of the WIAT intervention on stress levels by gender. The difference estimates in the cross-sectional data (panel A) show that, compared with those in the control arm, both male and female participants in the intervention group saw an increase in their stress levels during the two follow-ups (wave 2, female: B 1.63, 95% CI 0.67 to 2.58; p < 0.001; wave 3, female: B 3.42, 95% CI 2.49 to 4.34; p < 0.001; and wave 2, male: B 1.4, 95% CI 0.24 to 2.57; p = 0.018; wave 3, male: B 3.8, 95% CI 2.63 to 4.99; p < 0.001). This detrimental effect was confirmed in the cohort sample (panel B) for female participants only (wave 3: B 3.6, 95% CI 1.7 to 5.4; p < 0.001).

To add further insight about the effect size of the intervention, the predicted means of PSS scores are plotted by gender and study arm for the cohort sample (panel B) in Figure 15. It is notable that stress levels for women and men overlapped at each point in time, suggesting no significant differences by gender.

FIGURE 15.

Predicted means of PSS scores by gender for panel B: cohort sample. (a) Intervention group; and (b) control group. A higher PSS score denotes higher stress. Controlled for different individual-level characteristics.

Distance bands to the local woods

The difference estimates for the sample stratified within the different distance bands to the local woods are presented in Table 24. Based on the cross-sectional data set (panel A), we found that intervention participants within the lower distance bands had a lesser increase in their stress levels relative to those within the upper distance bands than control participants. This pattern was evident after each phase of intervention. For instance, at wave 2, the difference estimate for the lowest distance band (≤ 150 m) was a PSS score of 2.66 (95% CI 0.30 to 5.02; p < 0.001) compared with the estimate of 6.47 for the highest distance band (751–1500 m) (95% CI 4.4 to 8.5; p > 0.001). The gap between estimates for the lower and upper bands widened further by wave 3 with a difference score of 2.43 for the lowest distance (≤ 150 m) (95% CI 0.32 to 4.53; p = 0.024) compared with 9.71 for the highest distance band (751–1500 m) (95% CI 7.28 to 12.13; p < 0.001).

The intervention effect was less apparent in the cohort sample as few of the estimates for distance bands were statistically different from zero. However, it is notable that the estimates of difference in PSS score within the upper distance bands (501–750 and 751–1500 m) were some of the highest in magnitude and better in their precision as reflected by their p-values and CI, but only at wave 3.

Some of the close distance bands (151–300 and 301–500 m) did not show the trend noted above (i.e. the closer to the woodland area, the lesser the detrimental effect). Although the number of observations within each group was well balanced for both panel samples as a whole, there was a smaller sample size within the intervention sites for the group located ≤ 150 m to the woods (n = 191, panel A at baseline) than within the control sites (n = 809, panel A at baseline). To address this imbalance in sample numbers, two strategies were then implemented. First, we merged the lowest two distance groups (≤ 150 and 151–300 m), so that four groups were formed for analysis (≤ 300, 301–500, 501–750 and 751–1500 m). Second, distance groups 151–300 and 301–500 m were merged to create a different set of four groups for analysis (≤ 150, 151–500, 501–750 and 751–1500 m). Results from both of these analyses reinforced the trends reported earlier, with lower bands showing a lesser increase in stress levels and the opposite within the upper bands. The full results are shown in Appendix 6.

Seasonality

The winter audits in 2014 and 2015 were undertaken during the implementation of phase 1 and 2 interventions respectively, whereas the summer audits in 2014 were post phase 1 intervention and in 2015 they were post phase 2 intervention. The 2013 audits constitute the baseline, pre-intervention condition in both winter and summer.

Overall, audit scores were not significantly different (Mann–Whitney U-test) between summer and winter, for either the community or the expert audits, but it is notable that, for 2013 and 2014 community audits, summer scores were higher than winter scores in all cases but intervention site C. However, by 2015 community auditors scored all sites higher in winter than in summer, with the exception of intervention site C. This pattern is not found in the expert audits and is difficult to explain, other than that (for intervention sites A and B only) it might reflect a lack of maintenance post intervention.

Change over time

There was a significant difference (Mann–Whitney U-test) found between mean scores for intervention and control site scores across all time points, for both community (p < 0.001) and expert audits (p < 0.001). Figure 16 shows the differences in mean audit scores at each time point for the community audits and Figure 17 shows the pattern for expert audits.

FIGURE 16.

Difference between community audit scores for intervention and control sites over time. Error bars represent 95% CIs.

FIGURE 17.

Difference between expert audit scores for intervention and control sites over time. Error bars represent 95% CIs.

Figure 16 shows that intervention sites on average achieved higher community audit scores than the control sites at baseline and at all subsequent audits. However, there was a fairly steady increase in mean scores for the intervention sites once the interventions had begun in 2014, whereas the control sites have greater fluctuation in mean scores, having a mean score in winter 2015 of > 8 points higher than in winter 2013 but then dropping considerably in score in summer 2015. By summer 2015, the mean score for the intervention sites had increased by 4.53 points compared with summer 2013, whereas the control sites’ mean score had marginally decreased over the same period. The error bars illustrate significant differences between intervention and control in winter 2013, summer 2013, winter 2014 and summer 2015. The pattern shows that, although the quality of all woodland sites was perceived to improve in summer 2014 and winter 2015, by summer 2015 the quality of control sites had declined markedly compared with that of the intervention sites. The intervention sites were also considered to be significantly better quality than at baseline from summer 2014 onwards.

Figure 17 shows that the experts scored intervention and control sites very similarly, on average, at baseline in 2013 and it is only the intervention sites that show any substantial change over time, with a steady increase to a mean score of 9.48 points higher in summer 2015 than in summer 2013. By comparison, the control sites’ expert mean score increased by only 1.33 points over the same period. The control sites were scored significantly lower than the intervention sites in all 2014 and 2015 audits by experts, in contrast to the scores from community auditors.

In all but two audits, the community auditors were more positive than expert auditors about both baseline conditions and post-intervention conditions at the intervention sites. The lack of community auditors for one of the audit sessions in control site A may exaggerate the difference between the community and the experts in perceptions of quality over time.

Community audit scores for different environmental domains

Figure 18 shows the mean community audit scores for each environmental domain of the woodland audit averaged across all intervention sites. Winter and summer audit scores have been averaged for each of the 3 years of the study.

FIGURE 18.

Difference in community perceptions of seven domains of woodland quality, 2013–15: intervention sites.

Figure 18 shows an increase in quality for the intervention sites across all six domains focused on the woodland, with only the overall neighbourhood quality score showing little change. Facilities show the lowest score at each time point, despite improving over time, whereas overall woodland quality shows the smallest increase over time, followed by woodland safety/security. The greatest increase by 2014 is in woodland management and maintenance, although this had declined somewhat by 2015. The greatest increase in quality between 2013 and 2015 is in woodland access.

Figure 19 shows an increase in quality for the control sites across all seven domains in 2014 but a decline to near baseline level by 2015 in all domains apart from overall neighbourhood quality and woodland safety and security. This last domain is of interest as it might be considered an important aspect influencing people’s use of woodlands.

FIGURE 19.

Difference in community perceptions of seven domains of woodland quality, 2013–15: control sites.

Perceived environmental change in each intervention site

Appendix 3 gives the scores for each site at each audit point for both community and expert audits. It also summarises the comments made by auditors at each time point in relation to the different domains of the audit tool. In intervention site C, the difference between environmental quality pre and post interventions was evident only in summer audits for community auditors, who judged baseline conditions in winter more positively than the expert auditors. Over time, the pattern and direction of change in environmental audit scores was similar between the community and expert audits in intervention sites A and B, but the pattern varied more for intervention site C. More detailed information on audit scores for each of the intervention sites is given below.

Summary

In summary, the site audits show that the interventions were perceived as significantly enhancing the quality of the intervention woodlands compared with baseline, and that this was true regardless of seasonality. They also show that, by summer 2015, after both phases of intervention were completed, the intervention sites were, on average, considered of significantly higher quality than the control sites. However, for the community auditors (not the expert auditors) this had also been true at baseline.

The differences in scores between intervention sites and the commentary on what lay behind scores for different domains give some insight into the aspects of the woodlands that are attractive or a deterrent to community use (see Appendix 3). The focus groups and interviews undertaken post intervention give further insight into these concerns for the communities involved in the interventions.

Community focus group and interview findings

The topic guide used for the focus groups and interviews imposed an initial set of themes on the discussions: familiarity with the woodlands, use and perceptions of the woodlands pre and post intervention, awareness and perceptions of the interventions, engagement with any aspect of the interventions (especially the social interventions), impacts of the interventions including any behaviour and/or attitude change, access to the woodlands and reasons for not visiting the woodlands. In this context, it is worth noting that, in wave 3 of the core survey, respondents were asked whether or not they had taken part in an organised activity in the local woodlands in the previous year. Only 2% of those in intervention sites (and 3% in control sites) said ‘yes’, most of whom had done so with others (e.g. family or friends) and a few with their dog.

Subsequent coding of the focus group and interview transcripts using multiple siftings (as described in Community focus groups and interviews) produced a final, overarching thematic framework as set out here (Figure 20), which reflects community participants’ priorities and preoccupations in relation to the topics discussed. The main themes were awareness of natural open space for public use, barriers to access and use of the woods before the intervention, awareness and experience of the interventions, woodland behaviour and experience post intervention, barriers to use of the woods post intervention and wider neighbourhood issues. These main themes, and subthemes within them, are discussed in turn below in relation to each of the intervention sites. As the focus groups and interviews were undertaken only at the intervention sites, all sites referred to in this section are intervention sites. The respondents are identified as R1, R2 and so on, and noted as male (m) or female (f).

FIGURE 20.

Themes arising from focus groups and interviews.

Awareness of natural open space for public use

Barriers to access and use of the woods before the interventions

Access to site A was less apparent than it used to be owing to recent residential development blocking former access points and hiding views of entry points into the woodlands:

There’s a lot of new houses. There’s only one wee path now, that sort of takes you over the back road.

R1, f, site A, 2015

You can’t really get into it [. . .] The houses are all roundabout it [. . .] You used to go from [the park] just across the road, and that was you in [the woodland].

R4, f, site A, 2015

Several mentions of the quality of site A prior to the interventions noted that antisocial behaviour, including littering, fly tipping and drinking alcohol, were deterrents to use. One of the participants (a local councillor, although he was asked to participate in his personal capacity) had been invited to visit the site prior to the interventions because of concerns about activities going on there:

There was a real sense of, kind of, danger, almost, that you didn’t know what was lurking round the corner, or in the bushes, and things like that [. . .] It didn’t feel like the kind of place that people were particularly welcome in, to be honest [. . .] There was a pretty well established kind of den of, like, garden furniture, and all sorts of things, that had been put in the woodlands, and were protected with shopping trolleys, and all sorts of things [. . .] it was, basically, for gang fights, and under-age drinking, and all those kind of things, rather than pleasant woodland activities.

R2, m, site A, 2017

Site B was more accessible in terms of having a number of entrances that participants knew about from before the intervention. Deterrents identified before the intervention included the poor quality of paths on site:

At one time the path was just bogs [. . .] The mucky paths [. . .] it was almost impossible to get them [the children] from one place to another, without either you or them getting ankle deep.

R1, m, site B, 2015

At site C, antisocial behaviour, such as fly tipping, litter and fires, as well as the poor quality of the paths and vegetation (especially owing to the site’s steep gradient) prior to the intervention all acted as deterrents to use:

It was a bit diabolical, the woods were dirty, one section was really made a dumping ground, and again I would probably say there were more gangs, the kids were lighting fires.

R6, m, site C, 2015

[B]roken glass, nettles, it was overgrown, it was just an eyesore. Yes, it was in a really bad state.

R10, m, site C, 2017

[T]he woodland was all rough, it was dense [. . .] you couldn’t actually walk properly in the woodland [. . .] the kids [. . .] made dens in where they do their drinking and their drugs and everything else, so it wasn’t a nice place to go to.

R9, m, site C, 2017

Awareness and experience of the interventions

In 2017, many of the focus group and interview participants across all the intervention sites said they visited the woodlands at least once per week and in this respect they were not typical of the wider community as revealed in the core survey data. Awareness and use of the social interventions was more varied. Individuals had learnt about the interventions through direct observation of the physical changes in the woodlands, social media, posters and community networks. Lack of time and lack of awareness were factors alluded to by respondents for non-participation.

The majority of site A participants were unaware of the interventions and in 2015 the site was still not known as a woodland available for local community use. For this reason, they had little to offer on their direct experience of the interventions:

My son jogs quite a lot, and he jogs for miles [. . .] so he jogs all roundabout, but I don’t even think he knows about that [the woodland and their improvements].

R4, f, site A, 2015

When shown images of the paths and the upgraded entrance to the woodlands that the intervention in site A produced, participants in 2015 and 2017 were interested and some were positive about the images, but many had not noticed the changes:

I’ve never seen where that one [path] is [. . .] I didn’t know they had changed it all. I’m definitely going out looking for that.

R1, f, site A, 2015

I’ve never seen it [the new woodland entrance] myself [. . .] The road’s a wee bit better, but bet you it’s still got potholes in it. You need your wellies going along that one.

R3, f, site A, 2015

I think it looks alright, it looks like somewhere that you could go for a walk now.

R4, f, site A, 2015

This [new entrance] is much more open, much more welcoming, they’ve given you an idea of where you’re supposed to go. I know, from when I had gone before [. . .] you weren’t really sure if there was an official route that you were supposed to go. But this is much more clear, I think, for people.

R2, m, site A, 2017

However, at least one 2017 participant was aware of the community engagement activities associated with the intervention:

The only reason I was aware of it as a woodland, was because of some of the early things, some of the early communications that came out from the Forestry Commission, about the projects that you had been doing.

R2, m, site A, 2017

Participants from sites B and C showed greater awareness in 2015 of the interventions and some had been involved in activities undertaken as part of the phase 2 community engagement events. Participants had learnt about the interventions through direct observation of the physical changes in the woodlands, social media, posters and community networks, and some people who had not previously visited were attracted by the community events:

I personally didn’t [visit the woods before the changes]. It was a forestry-organised day that attracted me to the woodlands, where they had wood carving and stone carving and stuff, and that was the first time we had ever really noticed it. I knew it was there, but I wouldn’t have gone walking on it myself previous to that, I didn’t exactly know what it was [. . .] I think it was social media that I first saw something on through the [local community] development trust page, I think it was, and we went along for a nosey to see exactly what was happening, and they’ve held two or three events.

R2, f, site B, 2015

The people who turned up [. . .] there was definitely a couple of people that I would say probably would never appear in the wood, turning up in their flip-flops and pushing the pram, type thing, and they had a look about and they had fun, whether or not they would normally go.

R2, f, site B, 2015

Participants at site C in both 2015 and 2017 had engaged in a number of activities that they mentioned, including a photography workshop, a litter pick-up, bulb planting, a scavenger hunt and other family activities, which they also said had increased enjoyment and social interaction with family and neighbours. Participants also noted that people from further afield, beyond their community, had travelled to take part in the activities, suggesting that the reach of the interventions was wider than that captured by the core survey:

The whole family went down [for the scavenger hunt].

R6, m, site C, 2015

It was a positive change because it encouraged us to go into the woodland and we met different people and it’s a different area to go to, so it’s not the same place . . . It’s handy, you don’t need to take the car, no travelling really.

R10, m, site C, 2017

It’s good because it’s accessible now.

R7, f, site C, 2017

It’s usable, it’s owned, people will use it now.

R5, m, site C, 2017

It’s more attractive to the eye, which makes you want to go in.

R4, f, site C, 2017

There was also a wish for community-based facilitation of activities to continue:

They were a great idea, but they should be done again now that the woodland’s being used more. It’s time for round two and get all those people that you didn’t have before, get involved.

R4, f, site C, 2017

However, across all sites, concern was expressed in 2015 about lack of adequate consultation on the interventions that happened, and some cynicism about how much difference it would have made to what was undertaken:

I’m actually on the Tenants and Residents Association for [nearby community], and I’ve never heard of anything on that [FCS community engagement activities].

R4, f, site A, 2015

I think it’s always nice to know about what’s going on on your own doorstep, and too many things are done underhanded, so even though I wouldn’t have objected to what’s been done or it didn’t affect me in any way at all, it still would have been nice to know.

R2, f, site B, 2015

There’s never been any consultation or anything like leaflets or anything. Not that I’m aware of.

R5, f, site C, 2015

A girl came round the door when the Forestry Commission was just going to go into [the woods].

R4, m, site C, 2015

Would it have mattered if we had had a view?

R2, f, site C, 2015

Would they have listened?

R1 f, site C, 2015

Woodland behaviour and experience post intervention

Barriers to use of the woods after the interventions

Participants at sites B and C identified a number of barriers to use after the interventions, particularly by 2017, once the activities and maintenance associated with the interventions had ceased to be undertaken by FCS in 2015. The main themes of complaint were about vandalism and uncontrolled dogs and dog faeces. Issues of safety were also raised and concerns about the lack of ongoing maintenance, which restricted some people’s use of the woods, but there were differences of opinion on whether or not all of these things had got worse again.

Wider neighbourhood issues

The final theme covered issues about whether or not there were aspects of the wider neighbourhood context, positive or negative, that might have had a bearing on people’s well-being and quality of life during the study period.

Participants at site A considered that theirs was a good neighbourhood in general, both for those who had lived there all their lives and for those who had recently moved there. However, the woodland site for the intervention was marred in terms of access by recent (prior to the study) housebuilding:

Like a community, so people know people.

R4, f, site A, 2015

It’s a good neighbourhood for me. Because I’ve stayed in areas that weren’t as good.

R2, f, site A, 2015

They built all they houses, they just built them alone, and they never consulted, and they sort of barricaded it off so you couldn’t get in [to the woodlands].

R4, f, site A, 2015

At site B, there had been local council-funded work on footpaths adjacent to the woodlands, linking them to the sports centre and other parts of the community, in the summer of 2015. A parkrun (parkrun Limited, Twickenham, UK) running group had been set up after the study was complete and attracted a good number of people each Saturday. Although there was appreciation for recent building developments, participants had mixed views on the sense of local community, after a period of regeneration following the decline of a retail centre built 20 years previously:

One of the tragedies was that this [the town hall where the focus group was held] wasn’t built 20 years ago.

R1, m, site B, 2015

I think when [the community] was smaller there was a stronger sense of community.

R2, f, site B, 2015

Apart from the community council and youth group, the likes of the development trust, other than that, I don’t think there is a great deal [of a sense of community].

R3, m, site B, 2015

I have lived in about eight different houses in [the community], family circumstances change, bigger houses needed, that kind of thing.

R2, f, site B, 2015

At site B in 2017 there was also mention of aeroplane noise from the flight path overhead:

Sometimes my house shakes [. . .] You hear a noise when it’s a big plane [. . .] you definitely notice an increase in noise.

R6, m, site B, 2017

I think you just get used to it. We are directly on the flight path and probably in the summer you’re more aware of it.

R1, f, site B, 2017

At site C, comments were made about the desire for more consultation over developments, including the WIAT interventions and new housing developments:

They should stop building more houses on the top, because the skyline is kind of spoiled. But, it is, it’s beautiful, I wouldn’t move anywhere else.

R5, f, site C, 2015

Objectives

To evaluate the cost-effectiveness of the WIAT interventions in terms of the health outcomes and undertake the overall CCA for the primary and secondary outcomes.

Methods

The cost of the WIAT interventions was assessed using a top-down approach based on resource use. The QALY benefit from the WIAT programme based on the adjusted regression models was combined with cost to give an indicative cost per QALY in a CUA that included an assessment of uncertainty. All analyses were undertaken from a societal perspective for the 2-year period. A CCA presented the cost of the interventions together with the primary and secondary outcomes in a balance sheet format.

Results

The total cost of the interventions was £241,667 across the estimated eligible population (n = 20,472). The average cost for the physical intervention in wave 2 was £7.68 (95% CI £7.67 to £7.69) and £11.80 (95% CI £11.79 to £11.82) for both physical and social interventions in wave 3. There was no evidence of a statistically significant association between the interventions and HRQoL. An illustrative CUA in panel A reveals an incremental ICER of £935 (95% CI £399 per QALY to dominated) in wave 2 for the physical intervention (thus higher cost and lower QALY than in the control) and an ICER of £662 (95% CI £206 per QALY to dominated) in wave 3 for both social and physical interventions. The cost per QALY in panel B is £361 (95% CI £160 per QALY to dominated) for wave 2 after the physical intervention and £165 (95% CI £71 per QALY to dominated) for wave 3 after both social and physical interventions. Overall, the CCA suggests significant effects for the secondary outcomes of moderate physical activities, walking activities, connectedness to nature and social cohesion in panel A, and moderate activity and overall PA for panel B.

Conclusion

The interventions were not associated with any statistically significant improvements in health, hence there is no basis for demonstrating cost-effectiveness. However, the cost per individual in the eligible population is highly significant but low in magnitude. The CUA suggests that even small benefits in HRQoL, if found perhaps after a longer-term evaluation, have the potential to be relatively cost-effective at willingness to pay (WTP) of £20,000. The CCA also shows that the WIAT interventions have the potential to provide other benefits, if causally attributed to the interventions and considered of sufficient magnitude, at low cost.

Unadjusted results

The observed mean EQ-5D HRQoL utilities in the intervention and control sites for all the waves are shown in Table 28 for panel A and panel B data sets.

These results are also presented in graphical form in Figure 21 for panel A and Figure 22 for panel B.

FIGURE 21.

Unadjusted predicted means of EQ-5D HRQoL utilities for panel A.

FIGURE 22.

Unadjusted predicted means of EQ-5D HRQoL utilities for panel B.

The results for panel A show an approximate balance across control and intervention sites at wave 1. After the physical intervention at wave 2, the results have not greatly changed. At wave 3, after the social intervention, whereas the control groups have a EQ-5D HRQoL utility comparable to baseline, there was a strong fall in EQ-5D HRQoL utility in the intervention group that was both significantly different from the baseline measure and significantly different from the wave 3 utility in the control sites.

The results for panel B show a significant imbalance at baseline, with worse HRQoL in the intervention group. By wave 2, the control sites have lower EQ-5D HRQoL utility than baseline, whereas EQ-5D HRQoL utility at the intervention sites has increased, with the net effect that the difference between the intervention and control sites is negligible. In wave 3 there is no apparent change in EQ-5D HRQoL utility for the control sites, but EQ-5D HRQoL utility for the intervention sites has fallen.

The overall effect of the WIAT interventions on HRQoL was estimated using an ITT approach as described in Core survey of community residents. This was captured by the interaction term of type of site (intervention and control sites; control sites as reference group) and wave of survey (waves 1, 2 and 3; baseline wave 1 as reference group). This has the advantage of adjusting for any imbalances in the EQ-5D HRQoL utilities at baseline. Table 29 presents the unadjusted results of panel A and panel B samples.

The overall results for both samples mirror the observed results but the overall effects are less pronounced after adjustment for baseline imbalances. There is a positive effect of the physical intervention in both panel A and panel B analyses and a negative effect of the physical and social intervention combined at wave 3. However, none of these results are significant and they are unadjusted for other imbalances between the intervention and control sites.

Adjusted results

In addition to adjusting for imbalances in EQ-5D HRQoL utilities at baseline, the model can also be adjusted for age, gender, social class, perceived income coping, distance bands to local woodlands, working status, education, car ownership, life events, smoking, disability and differences in site pairs (intervention and control sites). Table 30 presents the results of the effect of the WIAT interventions on HRQoL after adjustment and, as before, the best estimate of WIAT effect is from the intervention interactions with the wave of the data from panels A and B.

The results of the adjusted analysis show very similar results to the unadjusted analysis for panel A. As for panel B, the intervention effect for the physical and social intervention combined at wave 3 shows a reversal from very slightly negative to very slightly positive. However, all these results need to be interpreted in context: none of the interaction terms were statistically significant, so there remains substantial uncertainty as to the true effect of the WIAT interventions on HRQoL.

The impact of the Woods In and Around Towns programme of interventions on mental health in the community

In relation to perceived stress levels (PSS), our main outcome measure, the intervention was associated with an increase in stress levels over time, whereas the control group showed a marginal decline in stress over time. The difference in PSS between the intervention and the control was significant for both cross-sectional and cohort panels by the wave 3 survey (post physical and social interventions). This finding was the opposite of what was predicted in our logic model based on previous research. Measures of mental well-being (SWEMWBS) showed a similar pattern, with a negative effect on well-being associated with the intervention. Our data offer no definitive explanation for this pattern of increasingly poor mental health associated with the intervention sites, but the analyses show that participants who visited nature did not display the same pattern. Qualitative findings suggest that certain other factors may have contributed to changes in mental health over time. For example, in intervention site B, an increase in disturbance from aircraft noise – a recognised stressor – and the difficulties associated with housing churn owing to urban renewal, may have contributed to increases in stress. Although the community audits indicated that the interventions noticeably improved the quality of the physical environment, the focus group and interview findings suggest that some concerns remained about using the woodland sites, especially after FCS maintenance ceased, although these were not so marked in 2015, the year of the wave 3 survey, as in 2017.

Health-related quality of life (measured using EQ-5D) was also considered using the same adjusted data sets as for stress levels. In this case, we found no evidence of a statistically significant association between the intervention and HRQoL.

The main finding relating to the above is that the intervention was associated with an increase in perceived stress and a decrease in mental well-being. Any beneficial effect from the WIAT interventions, as evidenced by the woodland audits and qualitative data, was not sufficient to produce beneficial effects on community-level mental health. Evidence on participants’ engagement with the intervention woodlands and other natural environments, discussed in Perceived stress and mental well-being outcomes associated with levels of engagement with woodlands or natural environments after implementation of the Woods In and Around Towns intervention, suggests that the pattern of stress and mental well-being levels may reflect influences external to the intervention programme. In communities classified as areas of high multiple deprivation in accordance with the SIMD, there may well be other, area-level factors not captured by this study that have contributed to mental health outcomes in the study communities.

Perceived stress and mental well-being outcomes associated with levels of engagement with woodlands or natural environments after implementation of the Woods In and Around Towns intervention

In both panels, the increase in perceived stress and the decrease in mental well-being over time associated with the intervention was considerably greater for individuals who had not made nature visits (i.e. to woodlands and other nearby green spaces) in the previous year than for those who had made nature visits. In the panel B longitudinal cohort we did not find a significant increase in stress levels in the intervention sites for those who had made nature visits, whereas there was a highly significant increase in stress for those who had not visited nature. A similar pattern for mental well-being showed, in panel A, a significant increase in such well-being for those who visited nature, a significant decline for those who did not and a pattern for panel B mirroring that for perceived stress. This suggests that nature visits may have some stress buffering effects in this population, as other research has found. 94,95 It is also possible, however, that greater mental well-being is necessary as a precursor to people feeling able and willing to access and enjoy natural environments. We did not find a similar pattern of stress for participants who noticed views to woodlands or green spaces when walking about their neighbourhood, suggesting that it is nature visits, rather than views of nature, that are associated with any stress buffering effect. However, for mental well-being, views of nature were not associated with a significant decline, whereas those without nature views showed significant declines in mental well-being in both panels A and B, suggesting some beneficial effect from views of nature in support of mental well-being.

The effect of changes to the physical woodland environment alone compared with the combined effect of physical and social interventions

We explored whether physical interventions in the environment were sufficient to produce beneficial effects in health or awareness of the woodlands or whether promotional activities, such as led walks and other community initiatives, were also needed to produce significant change. The core survey results indicate that neither of the intervention phases was sufficient to produce a community-level benefit in mental health. However, we did find a significant effect with regard to awareness of the local woods. Compared with the control site participants, intervention site participants increased awareness of local woods to a much greater degree. The intervention effect was large, held throughout the study (at both wave 2 and wave 3) and significant in both sample panels.

Although the site audits and qualitative findings suggest a positive response by participants to both the physical and social interventions, these participants represented only a very small minority of the community populations under study, because the core survey found that most did not engage with the woods and the facilitated community activities. Nonetheless, in the wave 2 survey, 23% of those in intervention sites who were aware of changes to the woodlands had read about them or (more commonly) heard about the changes from others. This indicates that there might be a greater effect from the interventions than participation in community engagement activities would suggest, and that word of mouth is an additional means by which to engage community members in natural environments. The focus groups also included participants who had become aware of their woods because of the social interventions only, suggesting that the phase 2 interventions were effective in increasing awareness, even if this was for only a comparatively small number of individuals.

Other health and well-being outcomes associated with the intervention

There was some evidence of the intervention being associated with other health and well-being outcomes. Measures of PA (IPAQ) showed a positive and significant association with the intervention group, especially for moderate levels of activity (both panels) and overall PA (for panel B at wave 3). This was notable compared with the control group, in which levels of PA went down over time. Although we cannot discount the possibility that (as with the increase in stress levels) this outcome is attributable to external factors other than the intervention, it is possible that the reported increase in the activity of walking in the local woods in the intervention sample compared with the control sample (see Other behavioural, social and experiential outcomes associated with the intervention) accounts for some part of this effect.

With regard to our other health-related outcomes, there was evidence of an increase in connectedness to nature by wave 3 of the survey for the intervention group compared with the control, but this was significant only for panel A. We found a similar pattern of comparative increase in social cohesion in the intervention group, significant both in wave 2 and wave 3, but only for panel A. The focus group findings illustrate a number of ways in which the interventions are likely to have influenced these outcomes. For example, participants enjoyed views of nature and wildlife and many were positive about facilitated community events with rangers that increased knowledge of the natural environment as well as events that engaged children and families.

Other behavioural, social and experiential outcomes associated with the intervention

Other outcomes identified in our research questions include the impact of the intervention on length and frequency of visits to natural areas and local woods, experience of local woods, awareness of them, activities undertaken there and visual contact with woodland.

The intervention did not show a significant association with length and frequency of visits to the specified local woods; this was analysed only for panel A owing to low total numbers visiting their local woods. However, there was a significant association, found in both panel A and panel B, between the intervention and those undertaking nature visits in general (i.e. to woodlands and other nearby green spaces) by wave 3. The increase in expected number of visits to nature was about 6% of the intervention sample. There was also a significant increase in those who had visual contact with nature, that is, those who were aware of views to woodlands or green spaces when walking about their neighbourhood. This was significant at wave 3 for panel A only.

Awareness of the local woods also increased significantly for the intervention group in both panels at waves 2 and 3. Panel A showed an increase of 13% in awareness of the local woodlands in the intervention sample between wave 1 and wave 3 and, for those who were aware, an increase in assessment of the quality of the local woodlands as good or very good from 26% in wave 1 to 51% in wave 3. The control sample showed a decline in all these measures, by comparison. For those who visited their local woods (a minority of participants; hence this was analysed only for panel A) there was a significant increase in going for a walk in the woods (at waves 2 and 3) and for walking with family and/or friends (wave 2 only).

The environmental audits by community members showed that the intervention woods were perceived as significantly better in quality than the control woods in the summer of 2015, when the wave 3 survey was undertaken, and significantly better quality than at baseline from summer 2014 onwards. This perceived enhancement of the physical environment may have contributed to greater use and awareness of the woods and other natural environments, although it is notable that the community auditors (unlike the expert auditors) scored the intervention woods significantly better in quality than the control woods at baseline too, in 2013. Focus group participants, many of whom visited their local woods, illustrated their positive response to the interventions in relation to walking in the woods and visiting with family, although for some the lack of longer term maintenance after the interventions was a problem for continuing such enhanced use of the woods over a year later.

We considered measures of experience of woodlands associated with attention restoration theory. We found a significant association between the intervention and both being away and fascination, and this was true for both panels. Such findings support a considerable literature on the restorative effects of experiences in nature13,16 and point to potential psychophysiological responses that may have beneficial effects for health and well-being, including stress relief. 14,15 If woodlands need to be perceived as offering benefits for mental restoration or relief from stress before such benefits can be experienced, and perhaps before local woodlands are likely to be visited at all, then these findings point to a potential pathway by which mental health may improve over the longer term.

Any gender differences associated with impacts of the interventions

Although, on average, women were consistently associated with higher levels of stress than men, there was no significant difference in stress levels by gender. Furthermore, the audits and focus groups did not produce any notable differences in gender-based responses to the interventions.

Differences in impacts of the interventions in accordance with distance of woodlands from participants’ homes

We found an association between distance to the local woodlands from home and stress outcomes. Differences in stress between intervention and control were largest in those living furthest away from the woods, with significantly higher stress for the intervention group (in both panels A and B) for those living ≤ 500 m from their local woodlands. This was especially notable when four rather than five distance bands were used in the analysis: two bands ≤ 500 m and two > 500 m. In panel B, when using five distance bands for analysis, increases in stress levels were not significantly different between control and intervention site participants living within 500 m of their local woodlands. These support findings from many other studies that consider the effect of distance from green and natural environments in determining any beneficial effect from nature, with the threshold of 500 m one limit of influence often cited,96,97 although a closer limit of 300 m has recently been recommended for consideration by the WHO. 98 These findings suggest that proximity to woodlands may play a role in any stress buffering effects found from engagement with natural environments.

Cost consequences of the intervention in relation to the primary and secondary outcomes of the study

In terms of economic evaluation, there is no basis for demonstrating cost-effectiveness because the interventions were not associated with any significant enhancement of HRQoL. However, the CCA shows that the WIAT interventions are low cost and have the potential to provide benefits other than mental health benefits, such as improvements in moderate physical activities, walking activities, connectedness to nature and social cohesion. The overall CUA reveals an incremental ICER of £935 (95% CI £399 per QALY to dominated) in wave 2 for the physical (phase 1) intervention and £662 per QALY in wave 3 (95% CI £206 per QALY to dominated) for both physical and social (phase 2) interventions in panel A. The cost per QALY in panel B is £361 (95% CI £160 per QALY to dominated) for wave 2 after the physical (phase 1) intervention and £165 (95% CI £71 per QALY to dominated) for wave 3, after both physical and social (phase 2) interventions. This suggests that even small benefits in HRQoL, if found perhaps after a longer term evaluation of the WIAT interventions, have the potential to be relatively cost-effective at WTP of £20,000.

Choice of study sites

One major challenge for this study was identifying appropriate sites for intervention and control arms of the study and matching them in terms of community characteristics.

The choice of sites and implementation of interventions were typical of the WIAT programme in terms of community characteristics and nature of woodlands. The sites were well matched on the aggregate socioeconomic characteristics of the community and matched as well as possible (given the other constraints) on the physical arrangement of the woodlands and community. There were no better matching sites available to us that met all the inclusion criteria. However, as noted above, FCS report that the time scale for planning, preparation, delivery and evaluation of the interventions was shorter than is considered ideal for the WIAT programme. The choice of sites was based on those eligible for funding support under the FCS WIAT programme. However, there were additional criteria required for this study, which included the requirement to match each intervention site as closely as possible with a control site, and the need to manage the timing of interventions to suit the study schedule and programme of community surveys. It was necessary to choose sites that best matched the research requirements rather than those whose characteristics and communities appeared most ready for engaging with WIAT.

The FCS reflective workshop suggested that, ideally, WIAT sites have pre-existing community groups able to engage with the woodland, or evidence of existing community use of the sites, and that negotiations with local authorities and other stakeholders about permission to operate in sites have already been carried out. They also considered that woodlands larger than intervention sites A and C were generally more successful for WIAT projects. 60

In addition, the research time scale meant that the typical WIAT approach was compressed, with a short lead-in time to start the project at all three woodland sites, minimal local community and other stakeholder consultation prior to the project start and a compressed timetable (8–9 months) for delivery of each phase of the intervention. FCS staff reflected that, ideally, the social interventions in particular would be built over a longer period of time, perhaps 2–3 years. 60 FCS staff also considered that it would have been beneficial to have produced an outline longer term plan for the intervention sites at the start of the WIAT process for the study sites. 60 Thus, although every effort was made to implement the WIAT intervention in accordance with usual practice, it must be recognised that the requirements of the research study may have constrained the effectiveness of the interventions somewhat.

Study design

Although the study design drew on the existing relevant literature, both in terms of calculations of sample size and in choice of outcome measures, the lack of directly comparable studies based on natural experiments means that there may be limitations in terms of study power and in sensitivity of measures used. In particular, it is possible that the PSS, EQ-5D and SWEMWBS are not sensitive to any health and well-being benefits resulting from the WIAT interventions and that other measures better reflect any such outcomes.

There were challenges owing to the design of this study that are common to many environmental interventions. In general, it is not possible to randomise such studies, or the population who live in the intervention and comparison areas, meaning that it is always possible that there will be differences between the populations in intervention and comparison areas. We therefore have to rely on being able to control for fundamental differences through the regression models but for this we are restricted by the information collected in the survey. It is also possible that the results reflect interventions and external influences for which we have no data, which may have affected intervention and/or control sites in different ways.

As with many environmental intervention studies, this study had a limited number of sites: three intervention and three control sites. Because the site is the unit of intervention, we would expect some differences between sites in terms of the effectiveness of the intervention, yet the expense necessarily limits the number of sites that can receive the intervention or in which the surveys can be conducted. It is therefore difficult to account for variability in the success of the intervention between sites and to know whether or not the ‘typical’ effect of such an intervention has been seen.

It is also unfortunate that the study was not designed for further follow-up in subsequent years after the intervention, although this again would have had serious implications for the expense incurred for additional waves of the survey. It is possible that different PSS or other health effects might have been found after a longer period post intervention, although the findings suggest that this might be dependent on continuing investment in maintenance of the physical infrastructure and/or further social interventions to engage the local community.

Core survey sample

We faced considerable challenges in recruiting core survey participants. Although response rates were not as high as targeted, co-operation rates at each survey wave were at least 63% with a total co-operation rate of 70%. The literature confirms that response rates in epidemiological studies have in general been declining over recent years and highlights the particular challenges of recruitment in areas of deprivation. 74,100 Against this background, recruitment in a very deprived urban context proved extremely difficult. Although the response rates were not as high as we had initially targeted, we consider that the co-operation rates reflect a sample that is adequate for the study and realistic given these challenges.

However, the sample was reduced post recruitment, after data checks revealed concerns over some of the sample. This reduced the cross-sectional sample and led to a significantly smaller sample in some of the sites, in particular in intervention site A in both wave 2 and wave 3. We therefore had a less balanced sample in terms of sites than originally planned and this may have affected the results.

Despite our best attempts to match intervention and control sites based on environmental qualities and socioeconomic, area-level and health indicators, it was also notable that our cross-sectional (panel A) intervention sample showed greater levels of deprivation and poor health, indicated by significant negative differences compared with control participants in educational qualifications (waves 1 and 2) and in coping on income, smoking, access to a car and proximity to local woodlands in all three survey waves. There were also significantly lower numbers of intervention site dog owners in all three waves. All of these potentially have a bearing on woodland and natural environment use. In particular, higher affluence and social class have been shown to be associated with greater numbers of visits to outdoor environments for leisure or recreation in Scotland. 101 This imbalance between the intervention and control sample characteristics may have been further exacerbated by the difference in patterns of proximity to the local woods. Although we attempted to match intervention and control communities and woodland sites on a variety of physical characteristics, the intervention sample finally used for analysis lived significantly further from their local woods than the control sample.

Recruitment of a longitudinal cohort within the sample was, as predicted, much less successful than the cross-sectional survey recruitment and means that we can have less confidence in the representativeness of the cohort sample. In addition, we found imbalances in levels of recruitment for different sites. In intervention site B, 45% of respondents in wave 3 of the survey were not the same individuals who responded in previous waves because they were ‘not known’ at the address. This seemed very high and the quality of these data was queried. Information from the local authority planning department indicated that there was a lot of churn within this community and people moved house within the public housing stock, adding to the challenge of retaining a longitudinal sample and perhaps to levels of stress within this population.

Recruitment for community site audits and focus groups

Recruitment of audit participants and focus groups via the survey was very challenging, with particular difficulty experienced in recruiting from the community in control site A. Despite efforts to increase recruitment by engaging with community groups and facilitators as the study progressed, those who did participate may not be typical of the communities from which they were drawn and are unlikely to reflect the views of many subgroups within that community. After substantial effort to recruit more widely, there were higher recruitment numbers, and from different participants, in focus groups or interviews in February 2017 than in November 2015. However, the time difference between the two means that their findings are not directly comparable, with the latter reflecting attitudes to and experience of sites no longer maintained by FCS > 20 months after the core survey was undertaken and the interventions completed.

Economic analysis

In the CUA, the time horizon for the analysis was assumed to be up to 2 years for the costing of the WIAT interventions. The full cost of the physical intervention was incurred in the first year. In the second year, the social intervention was added to the cost of the physical intervention. QALY gains were estimated for the year in which they would be measured. For this reason, there is an overestimation of cost because of the way that the CUA was conducted. In reality, the physical intervention could last in perpetuity if additional resources are employed to maintain the physical intervention. A model could have been developed to estimate the annual equivalent cost of the physical intervention assuming a percentage maintenance cost appropriately discounted into the future. This would have had the effect of reducing the physical intervention cost. However, we saw little advantage of adding this extra step because (1) the cost of the WIAT interventions is already very low, so they effectively provide an upper bound in the CUA and (2) the main uncertainty is the lack of evidence of the effect on HRQoL utility. Refining the cost estimation does not change the conclusion that the WIAT interventions cost very little and that the main uncertainty is whether or not there is a measurable effect on HRQoL. Our main conclusion stands: given the modest cost of the interventions, WIAT would need to have only a small impact on HRQoL to show cost-effectiveness. The problem remains, however, in being able to show small changes in HRQoL in the context of a population sample.