The results of the landscape mapping interviews were used to identify an initial sample of four middle managers at each study site. We selected them on the basis that they covered key areas of the local landscape between them; for example, we ensured that PCTs and local authorities were included, and that they were reported as working across organisations and/or professional groups.
We used structured network interviews: the data collection form is inAppendix 6. Other commonly used data collection techniques such as postal questionnaires were considered but we judged that structured, face-to-face interviews were most likely to ensure data quality and participation. Face-to-face interviews have also been used successfully in study designs similar to ours.21,22Interviews combined free recall
andfixed-choice strategies;16,23that is, we asked interviewees to name people in response to our
two questions, but allowed them to list only up tofive people. Our reasoning was that although we had defined a loose boundary for our network, we were most interested in identifying the core of the network, which would best be revealed by asking people to nominate their most important relationships rather than all relationships.
Although we focused on the‘core’of each network, we were also aware of the need to gain sufficient coverage of the landscape. This posed a practical challenge as the potential size of our networks
(as revealed through the landscape mapping exercise) was large. Owing to resource constraints, we would not be able to interview all of the managers identified on the landscape maps. We had tofind an
economic approach to collecting data about relationships across the landscapes at our three sites. Our solution was to gather‘secondary data’by asking interviewees to name people who their nominees talked with or went to. This method has been validated in the context of criminal networks, where research showed that the relational knowledge of a few informants compensated for the lack of objective
knowledge arising from low response rates.24If we subsequently interviewed someone for whom we had
secondary data from other respondents, we used their actual responses when constructing our data set. We took the view that two types of relationships would be of particular interest in relation to knowledge mobilisation. Thefirst was‘talks with’, which involved asking participants,‘Who do you talk with about [the pressing local topic/problem]?’The second was‘goes to’, which involved asking participants,‘Who do you go to get things done about [the pressing local topic/problem]?’These questions broadly correspond to the sociological concepts of speech and action. We also asked interviewees to provide further details (organisation or role/job title) about each of the people they named.
Having completed these initial interviews, we needed a strategy for sampling further interviewees without resorting to snowball sampling (for the reasons discussed above). Our solution was to select further interviewees by modelling the data from thefirst sample (hence the term‘iterative sampling’for this section). Because we were particularly interested in the collective thoughts and actions of middle managers, we modelled these partial networks using a method specifically designed to reveal
clusters–latent cluster analysis. We judged that identifying clusters would help to identify any‘holes’in the networks–gaps between groups within a network. In order to identify clusters, we used an approach which uses the concepts of latent position network models25and cluster latent position models.26
Further details about our modelling methods can be found inNetwork analysis: latent clusters, below. NHS provider
services
NHS primary care trust Coronary heart disease Stroke Diabetes Older people Primary prevention Weight management (children/adults) Physical activity Smoking Alcohol Health inclusion Social marketing Local community association Workplace health project Family services Children’s services, nutrition, physical activity, schools Adult services Older people Vulnerable adults Homeless
Sports and active lifestyle Cardiac rehabilitation Exercise on referral Public health unit Workforce well-being Local area initiative Primary care contracting and commissioning Commissioning executive committee Finance
Food and health Stop smoking
Health trainers Childhood obesity
Specialist obesity service
Sports clubs
Healthy living
centres Local employers
GP consortia Local councillors Leisure services Local medical committee Cardiology Information analysts Local authority NHS hospital trust
Regional cardiac and stroke network
GPs
Voluntary and community
groups
We also referred back to our landscape maps, in order to ensure that our sampling achieved the most extensive coverage of the landscapes possible with the resources available to us. Our selection was made using the following criteria:
(a) Actors near the centres of the clusters are sufficiently well specified in terms of their network links and need not be selected for interview (i.e. we knew enough about them already).
(b) Those near the edges of clusters had network links which might be less well specified, and could, therefore, reveal links with parts of the landscape that were not yet covered. These actors should be selected for interview.
(c) Individuals‘outside’clusters may be isolates, on the periphery of the network, or beyond our loosely defined boundary, but they could still be a key link to other parts of the landscape. They should be selected for interview if other sources (e.g. documents, observations or local knowledge) suggested that this might be the case.
Having identified, and collected network data from, the second phase sample, we modelled the network again. At this stage, we found that saturation of some parts of the landscape was already apparent. Equally, it was sometimes difficult to identify middle managers who would link to areas of the local landscape not yet covered. For our third andfinal phase of sampling, therefore, we focused on sampling individuals who had already been identified as particularly well connected in order to collect primary data in place of secondary, and to establish whether or not the managers identified in the landscape maps were linked to one another in practice.
We were aware, throughout, that there are always questions about the location of the boundaries of a network. Network and system theorists both emphasise that social systems are open, in the sense that they do not have well-defined boundaries with their‘environments’, and are continuously subject to external influences.3,27,28As some of the preceding points suggest, we addressed the problem in two ways.
First, we felt that it was reasonable to assume that interviewees would name people who were the most important members of a network and, by implication, not located close to any boundaries. Second, some of our interviewees were selected on the basis that they spanned the landscape (in our landscape maps), and this selection strategy gave us some confidence that we were able to achieve reasonable coverage within boundaries that were defined by our interviewees.
The network interview process was repeated approximately 8 months after the initial round of network interviews. For the second round of interviews, we did not develop the sample iteratively, but rather reinterviewed all round 1 participants. Where this was not feasible (e.g. owing to interviewees having left their posts), interviewees were replaced with individuals who were currently performing the same or a similar role within the organisation.