Browse Catapults
Transport System Catapult - Logo Home

Mapping and Modelling Social Isolation with Buckinghamshire County Council

Historically, social isolation has been considered an issue that is faced predominantly by the elderly; there is, however, a growing body of literature which identifies it as a problem which touches a much broader range of society, starting in childhood, through the working-age period up to and including old-age. This is reflected in the Marmot Review report. Social isolation has a negative impact on quality of life and life expectancy in a number of ways, ranging from increased risk of heart disease and stroke to increased risk of the development of mental health issues; furthermore, it is associated with significant increases in public health spending.

It is, therefore, unsurprising that tackling this problem has become a key priority for a number of local authorities and charities, including Buckinghamshire County Council. Ongoing work at the council recognises this, and seeks to develop a shared approach to prevention when tackling issues around health and well-being.

The aim of this Data Science Fellowship, tying in with these efforts, was to develop a methodology for modelling all-age social isolation across Buckinghamshire, critically appraising the existing research and identifying the required datasets. Breaking this down, the project aims were :

  1. Undertake a substantive literature review which would allow Buckinghamshire County Council to establish a set of academically grounded definitions on social isolation, as well as providing an understanding of the presently existing work in the field.
  2. Collate data sources from around the council and beyond with a view to providing a central point of inquiry for those interested in tackling social isolation.
  3. Develop a robust methodology for mapping and modelling all-age social isolation across Buckinghamshire.

Of the three, the development of a tool for mapping and modelling social isolation would appear to have the most tangible immediate impacts. Following consultation with stakeholders both inside Buckinghamshire County Council and outside, it became apparent that the immediate use for such a tool would be to identify areas within the county which are at an elevated risk of suffering from social isolation, and subsequently establishing what the most appropriate interventions might be.

“Working with the Connected Places Catapult team and Buckinghamshire County Council has given me the opportunity to undertake a research project that has been both challenging and rewarding, and one that I feel will have significant social impact.”

Fellow, Buckinghamshire County Council

This may, for example. tie in with services such as Prevention Matters or those offered by the Citizens Advice Bureau which aim to provide guidance and support to individuals; it may allow them to better tailor guidance towards interventions, or help them to identify vulnerable groups who are not presently reaching out and consequently not benefiting from the existing support services. It may also be of use for the Joint Strategic Needs Assessment in describing the health and well-being needs of the population.

 

Beyond this primary change, the project may also help to drive some secondary organisational changes around knowledge- and data-sharing, and around collaborative approaches to tackling social problems. As has been highlighted in the previous section, social isolation is now recognised as an issue which can touch a broad range of society, and consequently is a concern for a number of different branches of government. The drive to deal with this challenge collaboratively fits within a broader organisational change process around unifying the districts comprising Buckinghamshire (i.e. Aylesbury Vale, Chiltern, South Bucks and Wycombe) into a single local authority, and also around breaking down barriers between departments to allow for greater intra-organisation collaboration. The project may therefore act as an example of what can be achieved through the new more open, shared approach.

 

In order to achieve the project aims, the workflow for this investigation was split between initial desk research (which would form the foundation of the project) and more technical research and development. The desk research for this project consisted predominantly of reviewing literature to answer three questions:

  1. How do we define social isolation?
  2. What data-driven studies have previously been undertaken on social isolation?
  3. How do we measure social isolation?

A review of the literature found the definition of social isolation to be relatively consistent, and so for the purposes of this investigation, social isolation was defined as an objective measure which applies to a state of having a lack of social relationships with respect to quantity and/or quality; this is to be contrasted with loneliness, which is defined as an individual’s subjective experience of whether their current relationships are adequate. This review led to an exploration of the factors associated with social isolation. These were split into individual factors, community factors and societal factors, reflecting the scale at which these factors act.

Following on from this initial exploration of contributory factors, a review of the literature revealed that social isolation is becoming an increasingly studied topic by both local authorities and charities (e.g. Age UK). A number of local authorities (e.g. Essex County Council, Lancashire County Council, Northamptonshire County Council) have sought to explore social isolation by constructing indices of isolation based on factors identified in the literature.

Much of this work, however, seeks to explore social isolation in the elderly, and overlooks the impacts that it has on a broader range of society. Furthermore, much of the literature fails to draw a distinction between observables which are factors which contribute to the phenomenon of social isolation and observables which are indicators of it. The review of previous work also raises the question of how we go about measuring social isolation. A review of academic literature leads us to a set of measures of social isolation which should be applicable to a broad section of society:

  • Frequency of social contact:
  • Social network support
  • Reciprocity

The technical aspect of this project was made up of three processes:

  1. Data gathering
  2. Data mapping
  3. Modelling

Data was acquired from a number of different sources, both internal and external. Part of the data gathering process also involved basic profiling, fixing data issues and exploratory data analysis. As part of the exploratory data analysis aspect of the data collection phase, an interactive dashboard was developed which would go on to form the basis for one of the project outputs. This dashboard contains facilities for univariate analysis and mapping, comparative mapping and was also intended to encapsulate the results of the model.

 

Having undertaken exploratory analyses of individual datasets in isolation, the datasets were brought together for the data mapping phase. Part of the exercise of bringing them together was around developing an understanding of how each of the datasets contributed to the picture of social isolation, be it as a contributing factor or as an indicating metric. As part of the data mapping phase, a picture arose which divided variables from the datasets into contributing factors (which would go on to become inputs for the model) and indicators (which would go on to become outputs for the model).

The modelling process made use of two approaches – linear regression and geographically weighted regression.

 

Whilst the linear regression model performed reasonably well with regards to fitting the data, it suffered from two issues: firstly, it suffered from spatial autocorrelation in its residuals, ultimately meaning that the model residuals were not normal (a condition of using linear regression); furthermore, this approach only provides global parameter values and therefore does not offer us any insight into the way in which relationships between input and output variables might vary spatially. The latter approach allows us to model the spatial variation in how different factors contribute to the phenomenon of social isolation. The training of a geographically weighted regression resulted in a reduction in spatial autocorrelation, and allowed us to map the relationships between the input variables and the output variable using choropleths. These maps were then made more usable by inputting them into the dashboard framework with the aim that this would allow council staff to explore how the impact of different contributing factors varied across the county.

This work has laid the foundations for understanding all-age social isolation across Buckinghamshire. In three months, we have come a long way. We started with disconnected datasets managed in council silos, an absence of all-age social isolation research in the literature, and a desire to visualise, explore and understand social isolation across the county. The outputs of this project have started a journey – adding new ideas and methods to the all-age social isolation discourse, as well as delivering a prototype for a tool to model social isolation geospatially across Buckinghamshire. However, this work is just the start. The model developed in this project provides an introduction to the use of geographically weighted regression in modelling social isolation, but there are many other avenues to explore.

There are also countless novel datasets that could be included in the model that would leverage further insight. With new data at varying granularity, a multilevel modelling approach could be compared to the geographically weighted regression. Finally, the dashboard could be extended beyond a local application to a web service, to allow access across the council. Hopefully the work done in this Data Science Fellowship can be the catalyst for a new way of looking at social isolation and can have a lasting impact both in the council and in the community it serves.

 

Find out more about our Data Science fellowship programme here
“We participated in a Data Science Fellowship to further our understanding a complex and long-standing issue for the Council and wider partners – social isolation. Following previous work completed in this area with significant limitations, and a growing body of published research, we tasked a data scientist to review and critically appraise the existing evidence, develop a robust methodology to calculate risk of social isolation and model this across Buckinghamshire. The work was delivered to a high standard, in line with the specification required, and within expected timescales even with the disruption caused by Covid-19. The resultant outputs from this fellowship enhance our understanding of this issue, and will enable us to design and deliver meaningful interventions to improve outcomes for our residents”.

Buckinghamshire County Council