DATA

FOR
GOOD
How big and open data can be
used for the common good
Edited by Peter Baeck
February 2015

2 SUMMARY: THE GEOGRAPHY OF THE UKS CREATIVE AND HIGHTECH ECONOMIES

Nesta is an innovation charity with a mission to help people and

organisations bring great ideas to life.
We are dedicated to supporting ideas that can help improve all our lives,
with activities ranging from earlystage investment to indepth research
and practical programmes.
Nesta is a registered charity in England and Wales with company number 7706036 and charity number 1144091.
Registered as a charity in Scotland number SCO42833. Registered office: 1 Plough Place, London, EC4A 1DE.

www.nesta.org.uk

Nesta 2015

DATA FOR
GOOD
How big and open data can be
used for the common good

Contents
Foreword
1.

Introduction 5

2. Citizens

Advice Civic Dashboard: how DataKind UK helped

Citizens Advice Bureau get more from our data

Emma Prest, Datakind and Laura Bunt, Citizens Advice Bureau

3.

Using datadriven methods to understand hidden social action

3.1.

Mining the grant makers

3.2. Listening in: using social monitoring tools to understand the

social economy

Soft facts and spontaneous community mobilisation:

the role of rumour after major crime events

Colin Roberts, Martin Innes, Alun Preece and Irena Spasc Cardiff University

3.4.

Social action on social media

Carl Miller, Centre for Analysis of Social Media, Demos

3.5.

Mapping below the radar organisations on crowdfunding platforms

Maria Botello and Noel Hatch, European Alternatives

3.3.

3.6.

4.

24
26
32

Rowan Conway, the RSA

David Kane, National Council for Voluntary Organisations (NCVO)

What we learned about the potential of using datadriven methods

to understand hidden social action

Where next for big and open data for the common good

Endnotes

37

44
49
57

59
60

DATA FOR GOOD

Foreword

very day across the UK, volunteers, community groups and small
charities work to make the world a better place. Their devotion
and altruism are one of Britains most valuable assets. But for the
government, they are easy to overlook. The nature of governments is that
they pay most attention to what they can observe, measure and count and
grassroots social action is hard for a state to see.

We think there is a modest but important role for data innovation here. Over the past year,
Nesta has been working with researchers, charities and civil society groups to look at new
insights that data science can provide into what is going on below the radar in communities.
Of course, new data sources and algorithms, no matter how promising, are not a substitute
for good judgement or understanding. But we think the projects described in this report show
that better data can help provide a better understanding of the often neglected but vitally
important role of social action.
We hope you find the report useful, and we welcome your feedback.

Stian Westlake,
Executive Director of Policy & Research, Nesta

DATA FOR GOOD

1. Introduction

ew ways of capturing, sharing and analysing data have the potential to

transform how community and voluntary sector organisations work and
how social action happens.

However, while analysing and using data is core to how some of the worlds fastest growing
businesses understand their customers and develop new products and services, civil society
organisations are still some way off from making the most of this potential.
Over the last 12 months Nesta has grant funded a number of research projects that explore
two dimensions of how big and open data can be used for the common good. Firstly, how it
can be used by charities to develop better products and services and secondly, how it can
help those interested in civil society better understand social action and civil society activity.
Citizens Advice Bureau (CAB) and Datakind, a global community of data scientists
interested in how data can be used for a social purpose, were grant funded to explore how a
datadriven approach to mining the rich data that CAB holds on social issues in the UK could
be used to develop a realtime dashboard to identify emerging social issues. The project
also explored how datadriven methods could better help other charities such as St Mungos
and Buttle UK, and how data could be shared more effectively between charities as part of
this process, to create collaborative datadriven projects.
Five organisations (the RSA, Cardiff University, The Demos Centre for Analysis of Social
Media, NCVO and European Alternatives) were grant funded to explore how datadriven
methods, such as open data analysis and social media analysis, can help us understand
informal social action, often referred to as below the radar activity in new ways.
This paper is not the definitive story of the opportunities in using big and open data for the
common good, but it can hopefully provide insights on what can be done and lessons for
others interested in exploring the opportunities in these methods.

DATA FOR GOOD

2. Citizens

Advice Civic
Dashboard: how DataKind UK
helped Citizens Advice Bureau
get more from our data
Emma Prest, Datakind and Laura Bunt, Citizens Advice Bureau

Introduction

hrough its network of local community bureaux, digital advice services

and national phone lines, Citizens Advice sees firsthand the issues
that people are worrying about and the problems the country as a
whole is facing. As a charity we help millions of people every year to solve
problems, through offering advice, advocacy and coming together to
campaign for change. We help people work through issues with debt, welfare,
housing, employment, relationships, justice, discrimination and many others,
responding to individual and community needs.

The data Citizens Advice collects on these issues is rich and complex, but comes in a variety
of formats and does not directly match up. As an organisation already invested in using our
data to achieve change and improve services, we wanted to explore whether data science
methods could help us to understand more about Britain today. We are well versed in
identifying trends, but could this data help us to spot new or emerging problems? What could
we learn about social and economic issues if we analysed it in new ways?
This chapter documents a yearlong partnership between Citizens Advice and DataKind UK to
learn how data science, collaboration and hard work helped Citizens Advice do more with its
data. The outcome from this project was a realtime, responsive Civic Dashboard that reveals
the dynamics of social and economic hardship in Britain today.

Our starting point

On average, about 5,700 people walk into their local Citizens Advice Bureau every single
day; one of over 3,000 local Citizens Advice offices in communities and high streets across
England and Wales. Everyday 69,000 people look for help through Advice Guide, our digital
advice service, and 4,300 more reach us on the phone. You might have come to us yourself,
to seek advice, to find out information or access other services. Right now, at 7.06pm on a
Monday evening in January, 106 people were on Advice Guide in the last minute, and basic
rights at work was the most popular content searched for.
Citizens Advice helps people to solve problems. We do this by offering confidential, impartial
advice and support with any number of daytoday and difficult issues; anything from home
repairs, faulty goods, dealing with debt, finding work, claiming benefits, registering to vote,
coping with relationships or even taking people to court. Our network of volunteer advisors

DATA FOR GOOD

work with millions of people every year to help them find a way forward, and we come
together to influence policy, campaign for change, and devise local solutions to problems to
make society fairer.
The scale and extent of these daily interactions mean that Citizens Advice sees first hand the
issues that people are worrying about and what the country as a whole is facing. We know
the most common concerns for people in Leicester, or Doncaster, or Merthyr Tydfil. We see
questions people are most frequently asking on a Sunday night before work, or at coffee
time on a Monday, or in the darkest point in the night when everything feels much worse than
in the daytime. Through our services we have a window into the messy, complex, changing
pressures of modern life. And it is our responsibility to use this insight to make things better
for people.
DataKind UK was an obvious ally to help Citizens Advice get more out of this data. DataKind
UKs mission is to help third sector organisations benefit from advances in data science. Using
private sector volunteer data scientists, DataKind designs and manages data projects to help
charities understand what is in their data and use it for better decision making and greater
social impact. Unlike many of the third sector organisations that DataKind UK tends to work
with, Citizens Advice is fairly mature in its use and handling of data.
Citizens Advice has an inhouse team of data analysts that crunches, analyses and maps
our data, and regularly reviews the trends in issues from across the service. This team works
with our researchers, campaigners and communications teams to turn this data into action;
to showcase trends, insight and evidence to policymakers, service providers, regulators and
other stakeholders to help address the issues facing Britain. But though we already stretch
ourselves through our data work, we knew that there was an opportunity for it to do more.
This is where we started. What more could we learn about social and economic issues from
Citizens Advice data if we analysed it in different ways? For Citizens Advice, we wanted
to undertake an indepth analysis to better understand how our different data streams fit
together, to better identify and analyse problems and therefore improve our (and others)
response. For DataKind UK, this was an opportunity to work with an organisation with a vast,
rich and varied dataset (and quite a mature approach to working with data), and to use it to
help us work better. And so a natural partnership formed.

Setting the challenge

The project started with a big, bold idea: could data science enable Citizens Advice to
anticipate or even predict changes in the issues affecting people everyday, to act sooner to
prevent problems escalating?
In 2012 Citizens Advice spotted a problem with payday loans which was causing too many
people getting into debt. By 2014 we had worked with the Financial Conduct Authority to
establish new rules that aim to stamp out bad practice from the industry, rules that came into
force in January 2015. But who are the next payday lenders? We know that many people are
having problems with insecure work, set on zerohours contracts or without full employment
rights. How can we make sure we are spotting other examples of sharp practice and act
before they get worse? If we know the signs for someone facing eviction, could we identify
people at risk of homelessness and target resources to their support?
Our volunteer and staff advisors see these sorts of problems daily. With payday loans, lots
of people within our network were talking about this as a problem and suggesting that
something should be done. What we wanted to do was find more ways to make this insight
visible to others, and turn stories that could come across as anecdotal into analysis that was
hard to avoid as fact.

DATA FOR GOOD

Fundamental to making this happen was understanding the correlations between issues, the
timing of problems and how the datasets fit together. By doing so, could we build an early
warning system to show emerging structural problems across England and Wales?

The DataKind project

Supported by Nesta, Citizens Advice and DataKind UK joined forces to deliver a project that
had three goals:
1. Design a tool to harness Citizen Advices data so that the organisation could better identify
and react to emerging social issues in the UK.
2. Build awareness among Citizens Advice staff of new methods for mining and using data, and
encourage them to open up their data for others to use.
3. Develop lessons for other charities on how to better manage and use data.

The design of the project

DataKind UK uses several different formats to support partners. Given that Citizens Advice
had so much data in varying formats, both an iterative, exploratory process to understand the
data and test ideas, and structured time to focus on building and refining the final product
was needed. Therefore the project included two DataDives and one DataCorps.

WHAT IS A DATADIVE?

WHAT IS THE DATACORPS?

A DataDive is a twoday event that

brings the data science community
together with three or four selected
charities to tackle their data problems.
These events are free for both the
data scientists and the charities. They
provide direct data support to charitable
organisations, and make the social sector
aware of the power of using data in their
work.

DataCorps are longer projects where a

team of data science volunteers work
with a charity to tackle a tough data
problem. The DataCorps team work with
a charity for three to nine months to
clean, analyse, visualise, and otherwise
make use of data to make the world a
better place.

This twintrack approach allowed for:

Indepth exploration of the data.
Quick insights into what questions the data might be capable of answering.
Inspiration for the charity to make them aware of what is possible.
Relationship development between the Data Ambassadors, Citizens Advice and DataKind UK.
Based on the trust established, a slow move within Citizens Advice to open up their data.
Longerterm project planning possibilities through the DataCorps.

DATA FOR GOOD

An additional benefit of the DataDive model is that it is designed to incorporate numerous

charity projects. At the first DataDive three other charities took part Buttle UK, The Access
Project and Shooting Star Chase. At the second DataDive, two charities, St Mungos Broadway
and the North East Child Poverty Commission, were able to tap into and access Citizens
Advice data. This meant the project helped multiple charities to develop new methods to
mine their data, identify hidden issues and become more data savvy.

The Data
At the heart of this project is the rich and complex data collected by Citizens Advice. We
worked with three separate datasets throughout the project.
Bureaux issue data: this is a record of the issues that people bring to Citizens Advice. Each
issue is recorded at three levels of detail tier one being the most general (e.g. benefits),
tier two is more specific (e.g. employment support allowance), tier three is the most granular
(e.g. employment support allowance appeals). This information is recorded whenever
contact is made with Citizens Advice, whether facetoface in a bureau or over the phone.
Electronic Bureaux Evidence Forms: called EBEFs. These are forms completed by bureaux
staff when they want to flag an issue. These records are detailed writeups of an individuals
problem. They are also categorised as tier one, two and three.
Web data: this data comes from Google Analytics and shows which pages people visit on
the Advice Guide site and what search terms they use.

Citizens Advice at the First DataDive: understanding the data

At the first DataDive in June 2014, over 100 volunteer data scientists descended on the Mozilla
space in central London to donate their time and skills to help charities tackle complex data
problems.
The DataDive participants come from all walks of life and are brought together by a desire to
do good with data and work on charity projects. Almost all of them have fulltime jobs in the
private sector and enjoy donating their data expertise to good causes in their spare time.
There were three other charity projects involved in the DataDive The Access Project, Buttle
UK and Shooting Star Chase. Their findings are summarised later in this chapter.
Each charity was assigned three volunteer data scientists, called Data Ambassadors, to help
them prepare their data and hone their project idea ahead of the DataDive. This minimises the
time during the DataDive that needs to be spent on data preparation in analytical projects
this can be as much as 70 per cent of the work. Instead the volunteers who come to the
DataDive can focus on performing the analysis, and discovering new insights.
The Citizens Advice project was led by three volunteer Data Ambassadors1 who used the
event to develop a deep understanding of the Citizens Advices data, whilst generating usable
insight.
One of the main pieces of work over the course of the weekend was to begin matching up the
datasets.

10

DATA FOR GOOD

The web data shows which pages visitors land on, such as a page on Council Tax Benefit.
The graph below shows web pages that have been matched to the Citizens Advice issues
classifications and how they vary over time. The DataDive was a great way of flagging up
when the web data did and did not match up with the tiered classifications.

Figure 1. Unique page views by classification - benefits

30000

20000

Unique

10000

July

October

January

April

Time
Benefit cap

Personal independence payment

Council tax benefits

State retirement pension

Employment support allowance

Unclassified

Housing benefits

Universal credit

Immigration

Welfare reform benefit loss

Income support

Working child tax credit

Job seekers allowance

The EBEF text data contained 40 fields with a variety of demographic information about the
client and six free text fields. Citizens Advice advisors fill these in to provide detailed insight
into what they consider to be critical cases.
Using natural language processing an advanced technique that gives insights from textual
data a group of the volunteers analysed the EBEF data and discovered a new term
appearing frequently in the Other category: blue badge. This refers to the scheme used for
parking by people with a disability. As a direct result of this analysis, Citizens Advice has set
up a new category that allows this issue to be appropriately and distinctly tagged, rather than
grouped with Other.
Though these insights were not new to Citizens Advice, the fact that they confirmed what
we knew already gave the team confidence in the method. Volunteers had the opportunity
to explore the data and gain a deep understanding of what was in it, and began to see what
could be achieved by matching disparate datasets.

DATA FOR GOOD

11

Citizens Advice at the second DataDive: comparing datasets

At the October DataDive held at the headquarters of the Royal Statistical Society Citizens
Advice opened up their data for the other charities to use.
This marks a potential sea change in the way that third sector organisations treat data
recognising for the first time the additional social value that can be had by linking data held
by more than one organisation. In addition to a continuation of the Citizens Advice project,
two other organisations participated North East Child Poverty Commission and St Mungos
Broadway.
Citizens Advice Data Ambassadors2 began to combine the data, and to test different options
for an analytical model that could predict new advice trends. They had made all data available
in Elasticsearch, an open source search engine based on Lucene (an open source information
retrieval software library) with an accessible and flexible interface. For this project,
Elasticsearch was a perfect choice because of the combination of data types we needed to be
able to search across and discover the patterns of new topics that might emerge.
The DataDive participants also explored the web data to show Citizens Advice the most
frequent issues on the website. They found that tier one issues yielded few insights Citizens
Advice already knew that benefits and employment were the most popular. However, looking
at the second tier of issues provided new insights. The graph below shows that requests for
advice on higher education are rising, while searches related to housing benefit have notably
declined.

Figure 2. How the relative importance of issues changes over time

Frequency of level 2 codes

0.06

0.04

Percent of
total issues
0.02

0.00

April
2013

July
2013

October
2013

October
2013

January
2014

April
2014

July
2014

October
2014

Date
2 Fuel (gas, electricy, oil, coal etc.)
7 housing benefit

3 Self employment/business

10 environment and neighbour issues

5 higher education

12

DATA FOR GOOD

As the weekend progressed, the DataDive participants began joining up the datasets and
comparing them. This was the first time Citizens Advice had been able to see the two streams
of data and compare the numbers of people accessing the two channels for different issues.
Below is a graph showing the web data and issue data for Universal Credit.

Figure 3.
Universal Credit
5000

4000

3000

Visits
2000

1000

February
2013

May
2013

August
2013

November
2013

February
2014

May
2014

August
2014

Date
Web

Bureau

A key element of the DataDive was when one volunteer3 created an interactive Kibana
dashboard based on 44,000 EBEFs. Kibana is a webbased analytics and search tool that can
be used with Elasticsearch.
This dashboard brought the otherwise impenetrable text data to life, making it searchable
and understandable for the first time. Citizens Advice and the volunteers immediately saw
the potential for this approach, and it became the inspiration for the final dashboard the team
created.

DATA FOR GOOD

Figure 4.

The DataCorps
In parallel to the DataDives, the core team of volunteer Data Ambassadors worked closely
with Citizens Advice to understand and refine what they wanted to achieve, and to extend the
work done at the DataDives.
The initial idea was to create a predictive model to identify new trends and issues. The first
step in that model was to match up the datasets and load them into a single repository
Elasticsearch to get a clear picture of all client interactions across the different channels. The
Bureau issue data and the EBEF data was already categorised using Citizens Advices issues
tiers. The web data, however, was not.

Matching up the web data

The Data Ambassadors needed to join the web activity data with the Bureaux issues data and
EBEF data something that had never been attempted before. Volunteers from DataShaka a
company that specialises in unifying datasets donated time and expertise to match specific
web addresses to their corresponding Citizens Advice issue classification codes, which were
to be the common data fields linking the three datasets.
Having considered several approaches, the team settled on the idea of using configurable
Google Analytics segments to split up the web traffic. An advantage of this is that
the segments can easily be reconfigured in future when Citizens Advice update their
classification codes. The web data consisted of two datasets: daily visits to the Advice Guide
website and Google natural search phrases leading users to the Advice Guide site.
The team used the DataShaka platform to configure a series of transformations on the data
(also known as a data pipeline), to be executed automatically every night. After acquiring the
datasets for the previous day from Google analytics, they converted the data into DataShakas

13

14

DATA FOR GOOD

data fabric for easier manipulation. They then added in the relevant tier one and two codes to
the data and uploaded the data to a remote Elasticsearch cluster.
The result was that whenever someone went to the Citizens Advice website and searched for
benefits and Employment Support Allowance (ESA), this could be matched up to someone
who phoned up asking about ESA, or visited a Bureau with an ESA inquiry. This enabled the
team to understand and compare the issues people were searching for online with the issues
dealt with in person or over the phone at the Bureaux.

Tackling the EBEF data

Citizens Advice staff knew they were not getting the most from the tens of thousands of
EBEFs submitted each year due to the very large volume of text and the manpower needed to
examine them. But understanding what is said in these text boxes can be key to spotting new
emerging issues as was the case with payday loans. Topic analysis of the EBEF data was
done to discover which words were appearing most frequently in the free text boxes using
natural language processing an advanced technique that gives insights from textual data.
To understand what was present in the EBEFs, the Data Ambassadors collected all the text
from each text fields in every form and analysed the text data to identify the most significant
one to three word topics. This resulted in a set of 300 topics that commonly appear, for
example under occupancy charge or legal aid.

From prediction to a dashboard

Once the data was more or less matched up for the second DataDive, and the DataDive
participants had tested out various ideas, one idea rose to the surface: an interactive
dashboard to understand social issues in near realtime.
This shifted the focus from the original concept of a predictive model to something potentially
much more useful the ability to understand Citizens Advice data across all channels, explore
hunches, test ideas and spot new topics.
This was extremely liberating for Citizens Advice staff as the present system is highly
dependent on the informational management team, with staff asking questions of the data
and the analysts pulling out spreadsheets.
The DataCorps team ran a user meeting with stakeholders at Citizens Advice to determine
what they wanted to be able to do with the data and which bits of data were most useful to
them. The communications and policy teams were especially excited to get a better idea of
how many people were raising which issues, what their stories were, and being able to explore
more uptodate and accurate data on which to base policy research work.
The work of cleaning and normalising the data continued so that the datasets matched up
seamlessly. The team also recruited some new expertise4 to develop the dashboard.
Building on the success from the second DataDive, the team choose Kibana as the tool for the
dashboard. The main reason for this that it is designed to be integrated with Elasticsearch,
where the data was stored, is open source, flexible and easy to use. The core functionality
was extended through the use of D3.js (a visualisation tool that uses a JavaScript library and
framework) for some of the custom visualisations. Elasticsearch staff gave the DataCorps
team additional probono support along the way this was very helpful as the dashboard was
built in a beta version of the software.

DATA FOR GOOD

The team presented the end result to Citizens Advice staff, who immediately saw the
possibilities that a more responsive, visual analytical tool would bring. The volunteers
continued to adapt the tool based on feedback and ran training with users, before finally
handing over the product to Citizens Advice to form a part of their information infrastructure.
This is the Civic Dashboard.

What is the Civic Dashboard?

One of the primary outcomes from the collaboration between DataKind and Citizens Advice
is the Civic Dashboard. The Dashboard brings together all the data from the three channels
the bureaux issue data, the web data and the EBEF data in a searchable interface for the
first time. While some trends may immediately spring out, it will be most useful for provoking
new thoughts and ideas, rather than providing all the answers. It allows for exploratory
analysis and the ability to drill down to see which issues are most prevalent in any particular
bureau or time period.
The Dashboard will give Citizens Advice staff an immediate overview of how issues are
changing across the UK. Though not an open or a public tool given the confidentiality of
the data and granularity of the dashboard, it gives teams the ability to quickly visualise and
analyse social and economic trends and use this to inform and shape the public debate and
strengthen the impact of service delivery. This will raise awareness of what problems are
happening where, and where Citizens Advice will need to target or improve interventions to
make society fairer.
The following paragraphs describe the Civic Dashboard in more detail.
The Dashboard is made up of a series of separate panels. The first one gives an overview of all
three datasets; the second one shows the web data; the third, the Bureaux issue data; and the
fourth the EBEF data. Within each panel the graphs are all linked, so that when you change an
element on one, such as the issue or the time period, the other charts respond.

Combining the three data channels

Figue 5 brings together the three streams of data so you can compare which issues are more
prevalent on which channels. The image on the left shows three striped bars: the left bar
shows the web data, the middle bar the Bureaux issue data and right bar the EBEF data. This
graphic shows how benefits (green) are the most common issue across all three channels,
but more so in the bureaux issue and EBEF data. While it is difficult to directly compare the
datasets on the same scale as one click on a website is not the same as one visit to a Bureau,
this graphic shows each channel as 100 per cent and the stripes show the percentage that an
issue is raised in that dataset.
The three smaller graphs on the right show the changes over time for all the data streams by
week. Web data at the top, bureaux issue data is in the middle, and EBEF is at the bottom.
You can therefore see whether there has been a surge of requests on one channel for an issue,
or equally a drop.

15

16

DATA FOR GOOD

Figure 5.

A window into the web data

Figure 6 shows an overview of web traffic to the Advice Guide site. This panel has been
filtered to show employment issues. The donut graphic shows Terms and Conditions of
Employment and dispute resolution are the most popular pages. In the right hand stacked bar
graph we see top five tier two codes over time on the website. The line graph in the bottom
right shows the number of users on the site, filtered for employment. The overlapped area
chart in the bottom right shows the most common Google search terms that lead to the
Advice Guide.

Figure 6.

DATA FOR GOOD

Summarising the Bureaux issue data

Figure 7 showing the Bureaux issue data is made up of three donut charts showing tier one,
tier two and tier three issues across the top. The graph on the bottom left shows the most
common level one issues by region. The bar graph on the bottom right shows uncommonly
common issues this means that there has been a change in the amount that the issue is
appearing in the dataset compared to the average. For example the amount that people are
asking about benefits and tax (dark blue) has deviated from the normal amount recently
either a lot more or a lot less reports.

Figure 7.

Exploring the EBEFs data

To understand what was being said in the EBEFs, the team examined all of the text word by
word using natural language processing, finding what the significant one to three word topics
were across the entire set. This resulted in a set of 300 highly significant topics that can be
explored using any combination of time, frequency, location, demographic fields and issue
codes. This means you can very quickly drill down to the information you are looking for.
In Figure 8 the map shows all the EBEFs related to unfair dismissal and the other charts have
been filtered for employment issues. On the top right is a list of the 300 two to three word
topics that most commonly appear in the employment EBEFs. We can see that the most
common topics are housing benefit, food voucher, poor administration. Where we are likely
to find the emerging issues is in the idle of the list of 300 topics. The most common ones are
likely to be what Citizens Advice staff already know about. The least common rarely appear,
but in the middle will be hints of problems that are just below the radar. The bottom graph
shows the most common 25 topics by month that appear in the free text fields for EBEFs
categorised as employment.

17

18

DATA FOR GOOD

Figure 8.

So whats changed?
When we started this project, Citizens Advice already took data seriously as a route to
influencing change and improving performance. But this project was a catalyst for a
completely different way of thinking about and approaching data as an organisation and
as individual teams. Primarily, this happened through the teams collaboration with data
scientists who offered new skills, experience and ideas that raised our expectations about
what was possible with data. It gave the team energy, confidence and status in a way that was
visible to other parts of the organisation, which in itself drove interest and engagement with
the project. There are three particular shifts that have come as a consequence of this project:

DATA FOR GOOD

1. Embracing openness
Citizens Advice takes client confidentiality very seriously. It is one of the pillars of the
service, and earns trust from clients. Whilst open to the idea of exploring the data in
more depth, at first the teams were naturally and rightly sceptical to the idea of inviting
external parties to dive into the data.
However, though we remained clear that no action would breach our commitments to
anonymity, confidentiality and privacy, the teams gradually became more comfortable
with sharing data as long as the right protocols were in place. By the second DataDive,
we worked hard to devise a safe and responsible way of sharing and linking data with St
Mungos Broadway to test if this would help us learn more about routes into and out of
homelessness (and ultimately how to prevent people becoming homeless).5
2. Democratising access to data
Perhaps the biggest impact of this project for the data analyst teams has been how it has
shifted the strategy for managing information and data across Citizens Advice. They have
moved from understanding their role as a team which provides regular analysis products
such as monthly stats breakdowns, maps, static dashboards and trends as well as
providing expert, adhoc guidance and consultancy to different teams to providing
the tools, platforms and methods that will allow others to interrogate data on their own
terms.
The dashboard from the DataCorps is core to this, as the visualisation tools and cloud
based access make it much easier for anyone in the organisation to access and analyse
data in realtime. This is combined with the team offering coaching and running
workshops about what it is possible to do with data, and even hosting their own internal
DataDive inviting teams across the whole organisation to pitch ideas.
3. A greater emphasis on questions and exploration
Ultimately, the dashboard makes Citizens Advice data more visible. This is prompting
questions, and inviting more people to make connections between issues and areas;
clusters on a map are much clearer than points in a data table. This greater visibility is
encouraging more people to explore the data and ask questions, rather than being reliant
on the data analysts team to come forward with their own analysis. Collectively, we are
more alert to what is going on and better equipped to explore more.

And what was difficult?

Privacy
Throughout the project we were acutely aware of the sensitive and personal nature of the
data we were working with. We were often dealing with data about vulnerable individuals
whether that was sick children in the Shooting Star Chase data or homeless people in St
Mungos Broadways case the work required both an awareness of data protection laws, as
well as an emotional sensitivity to the stories in the data.
The risk for the charities involved was that someone could potentially identify an individual
in a dataset and misuse that information. DataKind UK has established structures to support
charities that take part in DataDives. During the DataDive weekends the charities worked with
their Data Ambassadors to anonymise the datasets before opening them up to the room of
data scientists. We asked the DataDive participants to sign nondisclosure protocols. We also
asked that they not publish or share the data with anyone outside of the venue and to delete
it once the event was over.

19

20

DATA FOR GOOD

Citizens Advice was particularly worried about sharing the EBEF data as it had small
amounts of profile information, along with free text fields which are harder to clean and fully
anonymise. However, Citizens Advice mitigated the risk by sharing that data in a strictly
regulated environment (participants could only access the data using an SSH tunnel) and
decided that the benefits being able to learn what was in the free text fields outweighed
the potential risks.
The St Mungos Broadway project presented a particularly challenging data management
problem. Linking up St Mungos Broadway data with Citizens Advice data meant sharing
their data in its raw form to match up individual records by name, data of birth and national
insurance number. St Mungos Broadways information management team was concerned
about the riskiness of exposing an individual. In order to minimise the risk they established
a protocol for how to share the data, and ultimately they decided that the results would be
worth it. To begin with, St Mungos Broadway asked Citizens Advice to sign a datasharing
agreement. St Mungos Broadway then shared the relevant data on a secured pen drive with
one Citizens Advice staff member who matched up the data, reanonymised it and deleted St
Mungos Broadways original raw data. There is little guidance on how to link data in this way
between organisations or what best practice in this area looks like.
Getting the most out of data without compromising confidentiality and privacy is tricky. It
presents some genuinely difficult judgments as organisations weigh the risks and benefits,
and is an area where society needs to catch up with changing technology.

Consent
As organisations collect individual level data they ask people to complete a data consent
form. The form outlines the specific uses to which personal information will be put, one of
which is usually analysis and service improvement. With St Mungos Broadway the implicit
assumption is that the analysis they conduct using individual level data is done internally. This
was the first time that they had interpreted consent in a different way, and raises questions
about the correct wording of consent for future projects.

Practical problems
Parts of the project were slow to move along, in part due to the heavy cogs of working in a
large organisation. Asking two large organisations to share and link data, something neither
had done before, proved to be a time consuming process and in some ways, perhaps it is
surprising that it happened at all.
Other problems arose around sustaining momentum in the volunteer team to continue to
work on the nine month project. The odd volunteer dropped out and others were recruited.
All the volunteers had fulltime jobs, and we would like to thank their employers, many of
whom were flexible and allowed them to use work time to complete aspects of the project.
We would also like to mention the challenges we did not face. There was buyin from all the
partners, from the senior levels of all the charities, and importantly an incredibly dedicated
and engaged information management team at Citizens Advice. Internally, Citizens Advice
had also started using Elasticsearch so there were some happy coincidences in terms of tool
consistency.

DATA FOR GOOD

How did the DataDives help the other charities?

The two DataDive weekends were not only run for Citizens Advice. Three charities took part
in the June event Shooting Star Chase, Buttle UK and The Access Project and two more
in the October one The North East Child Poverty Commission and St Mungos Broadway.
Below are the summaries of their findings and insights.
For all the charities, the DataDives were an eyeopening experience into how they could
be making more of their data. Some of them took away interactive visualisations that they
continue to use with stakeholders (Shooting Star Chase and The North East Child Poverty
Commission), others used it to refine their processes and systems (Buttle UK and the Access
Project). The most ambitious project involved linking St Mungos Broadway data to Citizens
Advice data on an individual level. This was pioneering for the sector in that it showed that it
was possible to match up datasets across organisations.
Many of the data scientists who attended the DataDives and the Data Ambassadors continue
to work with the charities. The DataDives were not ends in themselves, they were the start of
many of these organisations data journeys.

CASE STUDY: SHOOTING STAR CHASE

Shooting Star Chase, a leading childrens hospice, was keen to discover where thousands of
children in need of support from childrens hospices around the country might actually be
a real concern as many hospices are sited for historical rather than needsbased reasons.
During the investigation, the Shooting Star Chase volunteers streamlined the referral paths
of how children come to be at the hospices saving up to 90,000 for childrens hospices
around the country by refining the referral system. These savings are the equivalent of
hiring three nurses a significant discovery for such an underfunded sector.
A group of volunteers continued working with the charity after the DataDive to extend
the work by building interactive visualisations. These enable Shooting Star Chase to
better understand where the children with lifelimiting conditions are and which health
facilities they visit with which sicknesses. They now use this dashboard to persuade clinical
commissioners of the high demand for childrens endoflife services and the need for
more funding for the sector.6

CASE STUDY: BUTTLE UK

Buttle UK is a charity that provides grants to some of the most vulnerable children and
young people in the UK. At the DataDive they used text analysis to unearth useful insights
about the most common shared experiences among the families they support. They are
now using the results from the DataDive to guide them in a new organisational strategy, to
show which issues their clients face and how they are interconnected, and what kinds of
longerterm support Buttle UK should provide.7

21

22

DATA FOR GOOD

CASE STUDY: THE ACCESS PROJECT

During the DataDive, The Access Project, an organisation that provides tutors for young
people mapped the lifecycles of their volunteer tutors to better understand what makes
for a successful pairing between tutor and student. They were able to predict when
volunteers were likely to drop out of the programme and which factors did, and did not,
affect their involvement.
However, the main takeaway was the realisation that they were not collecting the right
data. The volunteer data scientists quickly refined the reporting process so that The Access
Project could collect more useful data and get to the root of the problem, something they
immediately implemented.8

CASE STUDY: THE NORTH EAST CHILD POVERTY COMMISSION

The Commission seeks to reduce child poverty in the north east of England. At the Autumn
DataDive they wanted to explore how they could gain a more realtime understanding of
the issue and communicate the data in a more actionable way to encourage immediate
responses. The team of volunteers first looked at whether the Citizens Advice data could
be used as an indicator of child poverty. However, they found that it was not statistically
significant. The team had a lot more success with creating an interactive visualisation to
help the public understand the data around child poverty. The volunteers had an epiphany
when they discovered that a DataKind project in the US was remarkably similar. They
repurposed code from the DC Action for Kids project9 and, within a matter of hours, had
produced an interactive map,10 which lets the viewer choose any one of the loaded datasets
and see how it plays out across the map.11

CASE STUDY: ST MUNGOS BROADWAY

St Mungos Broadway wanted to look at what kinds of issues people went to Citizens
Advice for help with, before they ended up homeless in St Mungos Broadways system.
The team of 30 volunteers who worked on the St Mungos Broadway project said that
St Mungos Broadways clients were most likely to have approached Citizens Advice for
help with employment support allowance. They also began creating a predictive model to
establish how likely it is that a client will return to St Mungos Broadway for help and if so
which type of service they are likely to require.

Looking ahead
The Civic Dashboard is up and running as a responsive, realtime tool to understand the
social issues people are facing in Britain. Citizens Advice teams are using this as a way to spot
trends, interrogate questions and identify emerging challenges.
Our data analyst teams are hosting workshops to walk people through the Dashboard, and
using this as a springboard for coming up with new ideas and opportunities for datadriven
innovation. This is only augmented by the incredible and inspiring generosity of the DataKind
UK volunteers who now act as an expert advisory group and informal mentor network for the
teams and individuals internally.

DATA FOR GOOD

But the work doesnt end there. Even since the Dashboard was first shown, we have explored
the possibility of new projects such as: linking Citizens Advice data with healthcare data to
more fully understand the relationship between health problems and social problems, and
therefore design more holistic interventions; supporting more civic action through making
data open as to whats happening where (e.g. plotting instances of scams, misselling or
poor advertising); allowing for more responsive monitoring and evaluation, tracking demand
and moving resources to where they are most in need; combining census data or other data
with Citizens Advice data, or using Citizens Advice data as a benchmark to track impact of
interventions in other areas.
Through exploring our data, we have laid the foundation for much more innovation in using
data to address social needs.

Conclusion
This chapter shares some insights and lessons from Citizens Advice, DataKind UK, and the
other charities that are using data science methods to understand more about social issues.
And we are not alone. There has been a real surge in interest in recent years, in the potential
of open data and data science to help charities make the most of their data, and open up their
data to others.
Much like any innovation project, we have learnt that making this work requires hard graft,
and lots of time and commitment from dedicated teams. Whilst technology has introduced
the possibility of new methods and faster processing, ultimately the magic of data science
comes down to people and organisations with the determination to make things work and
improve outcomes for the people we are here to support.

Acknowledgements
This project would not have gone anywhere without support from Nesta, TeraData, Citizens
Advice and DataKind UK, and the dedication of lots of fantastic volunteers and staff. In
particular: Peter Passaro, Iago Martinez, Arturo Sanchez Correa, Henry Simms, Billy Wong,
Emmanuel Lazaridis, Sam Leach, and all of the other volunteers who participated throughout
the project; Peter Watson, Ian Ansell, Kevin Benson, Hugh Stickland and Mike Dixon at Citizens
Advice. From DataKind UK: Duncan Ross, Francine Bennett, Heather Underwood, Jake Porway
and Kaitlin Thaney. Wed also like to thank Mozilla and the Royal Statistical Society for hosting
the DataDives; DataShaka, Datanauts, Elasticsearch and Experian for their support; and all the
many data scientists who came to the DataDives. Thank you.

23

24

DATA FOR GOOD

3. Using

datadriven methods to
understand hidden social action

nderstanding the dynamics of social action is one of the key components

of Nestas work on supporting public and social innovation. In reports
such as the Compendium for the Civic Economy12 and People Helping
People13 and practical programmes such the Centre for Social Action Innovation
Fund14 and Innovation in Giving15 we have sought to understand where social
action happens, how it happens, who the people and organisations supporting
and enabling social action are and what can support more social action to
happen at scale. This combined with research from organisations such as the
Charity Commission,16 the Third Sector Research Centre (TSRC)17 and the NCVO
Civil Society Almanac,18 mean that we already have a very rich understanding of
social action and civil society activity in the UK.

Funding social action

through crowdfunding

Mobilising social action

through social media

Discovering hidden social

action through open data

DATA FOR GOOD

However, most of our understanding to date is based on studies and registers of established
civil society organisations such as registered charities or social enterprises.
From research by the TSRC and others we also know that looking at established social
economy organisations does not give us a full picture, as there is an enormous amount of
social action happening below the radar (BTR) in informal groups coming together to
identify and address a social need. This type of social action has traditionally been, and still is,
very hard and resource intensive to research.19 However, as more of what could be considered
BTR activity moves online and we develop more and better methods for capturing, opening
up and analysing data we are interested in understanding, the potential in developing data
driven methods to better understand this phenomenon. Examples of this type of activity
include citizens using crowdfunding platforms to finance social action projects, or using social
media such as Twitter to address social needs.
This led us to fund five research grants exploring different methods for answering some of
these questions:
Demos Centre for the Analysis of Social Media (CASM) looks at the role of Twitter in
mobilising social action, examining two case studies, the Somerset floods and the Step up to
Serve campaign.
The Royal Society for the Encouragement of Arts, Manufactures and Commerce (RSA)
Examines how traditional methods of mapping used in RSA Connected Communities
research in Hounslow compare to data driven methods when mapping local below the radar
activity.
Cardiff University Investigates how we can begin to analyse and understand new types
of community mobilisation and social action through social media analysis, with a specific
focus on the events surrounding the murder of Lee Rigby.
European Alternatives Study projects on civic crowdfunding platforms, to explore if this
can help identify new types of collaboration and funding for projects not connected to
established organisations.
The National Council of Voluntary Organisations (NCVO) Considers if mining open data
from grant funders such as the Big Lottery Fund and Wellcome Trust can identify grantee
organisations that do not appear on any existing registers of civil society organisations.
The following articles summarise the different projects, their methodologies and what they
found. The full length papers with more indepth descriptions of the methods and datasets
can be downloaded from www.nesta.org.uk/publications.
Finally, it is important to note that this research is focusing on a very small subset of BTR
activity, looking at people and activities that are mobilised online. This naturally excludes
interesting offline activity and people not online. We acknowledge this and hope these
methods can be used to complement other types of research looking at offline activity.

25

26

DATA FOR GOOD

3.1.

Mining the grant makers

David Kane, National Council for Voluntary Organisations (NCVO)

What did we set out to do?

This project set out to develop and test a methodology for identifying below the radar
organisations within data held by grant makers (both statutory and private) on who they
fund. By matching data about who these grant making organisations fund with data on
registered organisations, we hoped to identify the remainder as below the radar. At NCVO, we
believed this approach would enable us to pick up organisations outside the sphere of known,
registered organisations. However, we realised from the start that the method would only pick
up particular types of organisations namely, those that have an interest or an ability to seek
out grant funding. In this sense, we thought this method would lower the radar, rather than
bypass it entirely.

What do we mean by below the radar?

In the context of the project we defined below the radar organisations as associations
of people with a charitable or social aim, but which dont have a formal registration as a
charity, company or other legal form. This definition focuses on organisations that may not
be formally constituted and are usually unregistered due to their small size (as measured by
income, expenditure, assets or employees). However, it does not capture some below the
radar activity, particularly new forms of activity such as social media.

How did we do the research?

We identified below the radar organisations in three stages. First, we gathered the source
data from grant makers (either directly or through the 360 Giving initiative).20 Secondly, the
data was matched with official registers of organisations based on the name of organisations
a process known as reconciliation. Groups of registered organisations such as universities,
local authorities and schools were also found using keyword searches. Lastly, we tested the
process and produced our results. Our methodology was set up to be flexible and iterative,
with improvements and refinements identified as the project progressed.

How many below the radar organisations did we find?

After applying the matching process we identified over 111,000 grants out of our population
of 240,000 grants that were not matched to a registered organisation, or identified as a
registered organisation or other above the radar organisation (such as a local authority or
school) through keyword searches.
After deduplication, we were left with 125,000 organisations in the grants database. Of these
organisations, 63,000 had received one or more grants that had been previously identified
as registered organisations, indicating that 49 per cent of organisations in our dataset were
unregistered (around 62,000 organisations).

DATA FOR GOOD

27

A further check of the results was done by looking at the amount awarded. As the threshold
for charitable registration in England and Wales is an income of over 5,000, this can be
used as a proxy for the size of the organisation in the dataset. Of the 62,000 unmatched
organisations, 33,000 received no grant greater than 5,000. This would indicate that just
over half of the unmatched organisations could be classed as below the radar.

Figure 9. Matching results flowchart

Initial population: 240,012 grants (125,149 organisations)

Reconciliation

Filter by matching with

registered organisations
or identifying as registered
organisation

Registered charity

Keyword searches:
Companies

24,475 grants

Public sector

19,787 grants

Universities

13,903 grants

Registered charities

Person

Reminder: 111,054 grants

61,885 organisations

55,510 grants

(not matched)
10,839 grants

4,444 grants

Matched or identified as registered:

128,958 grants
(63,264 organisations)

Filter by size of
largest grant
Grant amount
over 5,000

Reminder:
33,217 organisations

Most likely to be below the radar

28,668 organisations

28

DATA FOR GOOD

What did we find out about these organisations?

Activity
The largest category of below the radar organisations is culture and recreation (including
arts and sport organisations) with 30 per cent of organisations, followed by social services
(24 per cent), which encompasses a wide range of social activities, including childrens clubs
and support for the elderly. When comparing the distribution of organisations in our below
the radar population with those for registered charities, we found that culture and recreation
organisations are not only the largest group, but also disproportionately represented in below
the radar organisations.

Figure 10. Distribution of organisations by ICNPO category

Culture and recreation

Social services

Religion

Education and research

Grant makers and infrastructure

Development and housing

Law, advocacy and politics

International

Health

Environment

0%
All funders

5%

10%

15%

20%

25%

30%

35%

Big Lottery Fund

Geography
The distribution of below the radar organisations by country largely reflects the distribution
of registered charities, with the exception of Wales, which has 12 per cent of below the radar
organisations but only 5 per cent of registered charities.

DATA FOR GOOD

29

Distribution of organisations by country

Country

Below the radar

Registered charity

England 71% 75%

Northern Ireland 3%

4%

Scotland 14% 15%

Wales

12% 5%

UKwide 0.1% 1.5%

International 0.0%

0.1%

Looking just at England, we found a more even distribution across the country than for
registered charities. Organisations identified as below the radar do not appear to have
the same headquarters effect that registered organisations do, whereby looking at the
registered address of an organisation shows overrepresentation of London and the South
East, where many larger organisations have their headquarters. The North West, North East
and East Midlands appear to have a greater proportion of BTR organisations when compared
to registered charities. This may provide some support for the idea that below the radar
organisations are more active in areas where charities are less common.

Figure 11. Distribution of English organisations by region

North East

North West

Yorkshire and Humber

East Midlands

West Midlands

Eastern

London

South East

South West

0%
BTR

2%

4%

Registered charity

6%

8%

10%

12%

14%

16%

18%

DATA FOR GOOD

Organisational size
The data used did not have a field that specified the size of the organisation. However, we
were able to use the size of the grant that organisations received as a guide to the size of the
organisation. In general, the organisations that were identified as registered received larger
grants than those that could not be matched. However, there were a number of large grants
(over 10,000) made to organisations that were not identified as a registered organisation
indicating that the matching process has likely missed these organisations (as organisations
receiving a grant of this size are likely to be registered).

Figure 12. Distribution of English organisations by largest award size

25%

20%

15%

Percentage
10%

5%

+
m
10

m
-1
0
1m

k1m

10
0

k10

10

9k
-1
0

8k
-9

7k
-8

6k
-7
k

5k
-6

4k
-5

3k
-4

Unmatched

2k
-3
k

1k
-2

to

1k

0%

U
p

30

Registered

So was our approach successful overall?

The methodology we used had mixed results:
The process of matching organisations that have received funding to official registers of
organisations (notably the Charity Commission register) appears to have gone well. But
the picture is complicated by the difficulties in finding other registered organisations in the
dataset. In particular, there is no comprehensive register of public sector organisations which
includes, for example, parish councils, schools and other smaller organisations. Even looking
for discrete groups of organisations such as universities or local authorities through keyword
searches is made difficult by the way these organisations are described.

DATA FOR GOOD

However, it was possible to identify organisations that could be considered as more

likely to be below the radar. To that extent we have been successful in lowering the radar.
These organisations that have been identified as likely to be below the radar do exhibit
characteristics that might be expected from smaller unregistered organisations:
They are likely to be smaller in size (as measured by the size of the grant they received) on
average than those identified as registered organisations.
They are also more likely to be arts and sports based, often running small arts projects or
sport clubs.
They show a much more even spread throughout the country.

What next?
This project has highlighted a number of possibilities for further work and development:
A comprehensive register of organisations and reconciliation services needs to be available.
Resources such as Opencorporates21 and Opencharities22 provide examples of what is
needed. These will help researchers and others repeat similar exercises using large lists of
organisations.
There is a particular gap around public sector organisations. There is no official list that
contains every public organisation, and no unique identifier or URI scheme for organisations.
This is vital when looking at a data resource that covers the boundaries between sectors as
this one does.
There are other gaps in available data on civil society organisations. Scottish and Northern
Irish charities are not currently included, although Scottish charities are now available as
open data from OSCR23 and the Northern Irish charities register has only recently been
set up. Other types of bodies could also be incorporated such as royal charter bodies and
Community Amateur Sports Clubs (CASCs).
The 360 Giving programme is providing useful data which is of value and offers insights
that cannot be found elsewhere. We would encourage the continuation of that programme,
and greater use of the data and improvements to the quality of data imported will make
it a more useful resource. Of particular use would be the inclusion of charity numbers and
company numbers which would remove the need for an imperfect matching process.

31

32

DATA FOR GOOD

3.2.

Listening in: using social

monitoring tools to understand
the social economy
Rowan Conway, the RSA, and Jimmy Tidey, the RCA, with thanks to Gaia Marcus

Going where the (social) action is

Whether it is local mums coming together to support each other or a group of neighbours
joining forces on a guerilla gardening campaign, there is below the radar social action
happening everywhere. As well as the official services delivered by charities and social
enterprises, it is the informal or semiformal activities led by community groups or individuals
that provide the renewable energy which powers the social economy
The transformative potential of below the radar social action is enormous, but it is also hard
to track. Understanding grass roots community activity outside of the established social
economy (such as registered charities or official databases) has the potential to redefine
community engagement methods; transform social investment models; or even challenge
established local governance systems by widening the pool of talent to include a range of
unusual suspects.
The Connected Communities research team at the RSA has been exploring how social
network analysis can be used to combat isolation at a community level since 2010. In a seven
site Big Lotteryfunded study called Building Inclusive Communities we have been using
traditional social research methods to gain a rich picture of the localities we work in across
the UK. Through social network analysis and asset mapping, our data has provided a snapshot
of each community at a hyperlocal level, providing a platform for coproduction and inspiring
social action in each locality.
The programme uses assetbased community research methods and surfaces below the radar
activities with tools such as surveys, interviews and focus groups. Local community volunteers
might survey a local area through door knocking to capture data about its resources, skills
and talents. Or community groups might come together in a participatory process to compile
an asset map and use the map to build collective plans for an area. Social network surveys
provide another lens through which we view community assets, building a picture of a
communitys relationships as a social graph. In using these research methods, the RSA has
observed that the process of research itself can be empowering, as the journey of discovery
of local assets sometimes inspires social action in itself, as well as building relationships and
thickening local social networks.

DATA FOR GOOD

While these research methods offer a multilayered picture of a community way beyond
the official stakeholder databases or registries of voluntary sector bodies getting this data
is not simple. The process of recruiting and training community researchers, managing local
volunteers, and processing the data can be very resource intensive. The data is also time
sensitive as it shows a communitys assets or an individuals social network at a particular
moment in time. These limitations are largely outweighed by the benefits, but the RSA was
interested in exploring alternative methods to see if they might yield similar results in a less
resource intensive and more dynamic way.

The Community Mirror

As more and more community activity moves online into hyperlocal blogs and onto social
media sites, the opportunity to apply digital research tools for local community data
gathering is becoming a reality. The RSA wanted to test digital research methods to map
community assets to see if they could deliver similar results to the traditional facetoface
methods that we deploy in the Connected Communities work. With Nesta funding, we took on
a microresearch project to see if social media analytics might provide an alternative way to
view a communitys assets and activities.
This collaborative research project, which we called Community Mirror, was carried out
to understand the ways in which online datagathering techniques compared to door
todoor research. We worked with the Royal College of Art (RCA) using their LocalNets.
org application to gather the online data and contrasted the findings with one of the RSAs
existing community research projects in the London Borough of Hounslow that used faceto
face methods.

Mining the Twittersphere

The LocalNets.org app mines the social web by aggregating Tweets and blog posts from
selected Twitter accounts and blogs. This activity is in effect a social listening platform as it
provides the capability to systematically collect online conversations about specific phrases or
words. Each item is coded with topic tags, geolocation where applicable, and any entities that
are relevant to it. Words that are captured include people, places, organisations and events
the same type of data that we seek through facetoface surveys.
As it is a microblogging platform, Twitter is a rich source of social data and is a good
starting point for social web mining because of its inherent openness for public consumption
(Facebook is a more closed network). Twitter data is particularly interesting because it is so
dynamic. The LocalNets.org web app was live over a four month period in 2014 and collected
data about the online activity in Hounslow in order to produce a visualisation to show the
following:
A list of the communitys assets, including people, places, organisations, and events.
A geographic map of places that the localitys community members are talking about online.
A network diagram of how the community assets are connected to one another.

33

34

DATA FOR GOOD

The resulting online community map looked like this:

Figure 13. Hounslow social network map of the community assets identified by the

LocalNets.org web app
Hounslow Green Party
Centre for Spirituality & Cultural Advancement
River Crane
Eid
Seema Malhotra MP
Corinna Smart Hounslow Mayor
Paul Robeson Theatre
Hounslow Culture

apblackwell

Feltham Band
Feltham Local

FelthamBand

LBofHounslow

Reach Academy Feltham

Get West London

Hounslow Bob

Hounslow Chronicle

Bridge Link Community Centre

Mary Macleod MP

Cranford Community College

Salina Patel

Sgt Andrew Pugh @MPSHounslowSth

Inwood Park

Feltham Pond

Hounslow Homes
Cranford College PE

Heathrow Airport
Bedfont Lakes

OsterleyNT

Holy Trinity Church

Mumsnet Hounslow

LB of Hounslow

Lampton Park

Crane Park

GreenFeltham
Ponyboy1234

Lampton Park FC

Treaty Centre Hounslow

Team Hounslow AC
Get Active Hounslow

Hounslow Volunteer Centre

Hounslow Heath

Hounslow Borough Police

Hounslow Highways

Isleworth Unofficial Traffic Warden

Green Feltham
Lampton Park Conference Centre

Osterley Park
Hounslow Cycling

Hanworth Park

TW13.net

Feltham Warriors
Feltham Community College

Hillingdon Borough Police

Hanworth Air Park Leisure Centre and Library

Sgt Nigel Treacy
Feltham Police
West London Mum

The offline research (doortodoor surveying) that happened concurrently in the Cranford
ward of Hounslow, captured the people or resources that local community members look
to for social support. Respondents were asked to identify people, locations, organisations
and events that help them feel part of the community. These community assets and the
connections between them and the respondents were then visualised on the social network
map below:

Figure 14. Social network map of the community assets mentioned

by survey respondents in the offline door-to-door research

Note: Different clusters of connected assets are shown in different colours for clarity.

DATA FOR GOOD

Findings
In the four month period of data collection, the online dataset generated a list of 294
community assets, categorised as people, locations, organisations and events. These assets
were from across Hounslow, not specifically in the Cranford ward. It showed perhaps
unsurprisingly that when they were online, Hounslow residents often identified community
assets outside of their own locality. When it came to events and people, there were no direct
correlations between the offline and digital data. The online asset mapping discovered 25
community events and 72 people who might be seen as influencers including MPs and local
footballers.
It was with the local organisations and places where correlations could be found. The digital
asset mapping found 168 locations and organisations and the offline asset mapping found 51
locations and organisations. Twenty per cent of the assets mapped offline were picked up by
the digital tool, with ten clear matches between the two datasets. With a further six probable
matches, it is reasonable to say that 31 per cent of the place and location assets mapped
offline were discovered by LocalNets. This suggests that the LocalNets digital tool has validity
in surfacing relevant local data, while at the same time supplementing this with a large volume
of additional assets not mentioned in the offline research.
Whether these items can accurately be described as constituting the below the radar
social economy, however, is open to question. Most of the assets included in both datasets
hospitals, parks, churches are well known to authorities and are key parts of the formal
social economy rather than below the radar. What the LocalNets data can provide, however, is
a map of how these institutions engage with each other online and it is this relational aspect
of the social economy that is often below the radar, and one of the key benefits of LocalNets.
The tool can produce up to date and locally valid assets maps, with some insight into how
these institutions interact with each other or not online.
Overall, the study found that LocalNets.org software is an efficient way of collating hyper
local information about community activities and organisations in a given area. With further
user interface development, it is possible that is could become a reasonably inexpensive way
of gaining insight into the below the radar activity in a local area as part of a social media
analytics process.

Whose radar is it anyway?

While the study did prove that the LocalNets.org app can surface additional data about the
people and organisations that are supporting and enabling social action, what one does with
this intelligence is another matter. Digital methods like social listening can provide a relatively
inexpensive way to bring a new, more dynamic, dimension to a community asset map, but the
tricky questions arise when you ask who exactly will use these methods? Will it be the local
authority, local community groups themselves or even central government? Whose radar is it
that we are going under?
Mapping things that are below the radar sounds inherently covert, like somehow we are
smoking out a shadow economy. Certainly, there are ethics and privacy questions that need
to be explored when social media monitoring is applied to communities. Lessons can be
learned from the corporate experience where social intelligence tools such as Brandwatch
are routinely used by corporations to track social media and online conversations about their
brands in order to understand consumer behaviour. Companies have vast datasets about
consumers, but many still struggle with how to engage with customers about the data. The
process of interacting with users of social media for organisational purposes like customer

35

36

DATA FOR GOOD

service or sales brings up awkward questions such as: do Twitter users know you are
listening? Will they find it acceptable? Does regular social listening turn you into an online Big
Brother? These questions, when applied to a local community context, will require thorough
consideration.
In our conclusions, we find that digital methods do offer a promising approach to mapping
the below the radar social economy assets, but that traditional methods are key to
engagement with communities. It is human connections that take an asset mapping process
beyond a simple audit to a platform for social action. While the on the ground research
certainly requires significantly more resource than digital tools, our study did not suggest that
the digital tools could effectively replace offline methods rather they were complementary.
Combined, they can provide a powerful source of insight. Indeed, they can feed each
other, and digital asset mapping can be seeded by existing local knowledge and used to
complement offline community asset mapping approaches. LocalNets.org can be primed with
keywords that come from other asset mapping processes, and the output of LocalNets.org
can be used as a starting point for deeper investigation.
So digital tools that go beneath the radar can indeed produce useful and useable insights
about community. This study certainly informs the Connected Communities work going
forward, and you never know, it might even evolve into a new way of tackling that old
chestnut of finding the hard to reach. But that is another conversation entirely

DATA FOR GOOD

3.3.

Soft facts and spontaneous

community mobilisation: the role
of rumour after major crime events
Colin Roberts, Martin Innes, Alun Preece and Irena Spasic Cardiff University

This study examines how social media is transforming processes of community mobilisation
in the aftermath of major crime events. In so doing, it illuminates processes that are below
the radar, in terms of being frequently seen but unnoticed, owing to how public attention
fixates upon the response of the police and other security agencies. The analysis focuses
upon two specific issues: (1) how rumours disseminated via social media platforms work as
soft facts to influence patterns of collective reaction in such circumstances; and, (2) how
these communication platforms are being used to organise particular forms of spontaneous
community mobilisation.
The analysis pivots around four case studies of community mobilisation following the murder
of soldier Lee Rigby in Woolwich, London in 2013. It distils some more generalisable ideas
about how social media is transforming the ways in which contemporary collective action and
social reactions are organised. The key insights derived from the analysis are:
Social media communications function as both an engine and a camera they propel
social reactions, but simultaneously leave digital traces that can be used to develop a
picture of these reactions.
Rather than big data, the key quality of social media following major crimes and disasters is
its functioning as fast data, with information about what is happening travelling rapidly out
from the scene.
The velocity of data influences both public sensemaking and processes of collective
action, and comes close to outstripping the capacity of existing social networks to respond.
Consequently, rumours play an important role in shaping public sentiments on social media
following major incidents.
Concurrently, forms of community mobilisation spring up, often reflecting polarised political
and ideological interpretations of what has happened and what should be done about it.
The evidence suggests that effective spontaneous community mobilisation frequently has to
be scaffolded by existing social networks and institutional structures, which are repurposed
in moments of perceived crisis.
A key idea developed through the research is that of spontaneous community mobilisation.
This is intended to capture how, in the volatile wake of a major event, individuals, communities
and groups come together to engage in forms of urgent and improvised collective action and
sensemaking. This is spontaneous in that it was not (and often could not be) planned, and
social media, because of the networked communication it represents, plays an important role
in enabling these mobilisations to occur.

37

38

DATA FOR GOOD

Research design and data

The data analysed are drawn from a larger dataset of social media materials collected in the
wake of the killing of Lee Rigby. In excess of 34 million data points were collected using a
textmining platform developed by the authors and their research team. These stretched from
the first Tweet from a witness at the crime scene, through to the conclusion of the court case
where the two suspects were found guilty. The scale of the data is important, because of the
high resolution picture it affords of processes of social reaction. It enables us to track and
trace in finegrained detail what happens following a major crime, in ways that have not been
possible previously.
For this study, samples from this larger dataset were extracted for more detailed analysis,
on the grounds that a preliminary look had identified the presence of potential rumours and
community mobilisation. Framed in this way, the research engaged with three principal aims:
1. To use datadriven analysis to develop an understanding of how rumours about crime risks
and threats encourage forms of community mobilisation that are not dependent upon
formal organisations.
2. To bring forward new concepts that account for the role of social media data in seeding new
forms of social organisation.
3. To consider the policy and practice implications of these modes of spontaneous community
mobilisation for agencies involved in community impact management work following the
occurrence of major crime events.
The data were collected and analysed using a bespoke suite of software tools called
Sentinel.24 Sentinel can collect social media data from a variety of sources, however this study
focuses on material collected from Twitter. Analyses of these data were guided by application
of a 5W (who, what, when, where, why) framework:
Who: Identify participants in events, and relationships between them.
What: Characterise events including crimes and communities mobilisation.
When: Maintain a timeline of linked events.
Where: Determine locations by both geotags and place names.
Why: Apply social science models to uncover causal links between events.

DATA FOR GOOD

39

Figure 15 below provides a conceptual representation of the platforms architecture and

Figure 16 shows how data are displayed by the system:

Figure 15. An Overview of the Sentinel Platform

Customisable apps
Semantic APIs

who

what

when

where

why

Sentinel core services and models

Signal
crimes

Conflict

Extremist
narratives

Expression and term recognition

Data collection services

Figure 16. Example of a Sentinel What/When Data Visualisation

40

DATA FOR GOOD

Key findings from the four case studies

The first case study focused on processes of social reaction in the first 24 hours after the
killing of Lee Rigby, and how rightwing political groups rapidly tried to mobilise their
supporters for public protests using social media. In response to which, a coalition of groups
opposed to their political views also sought to rapidly mobilise a counterdemonstration.
Again, much of the work of contacting and trying to persuade people to get feet on the
streets was conducted using social media.
The key insight developed into how mobilisation happens is the interaction between the two
sides. Each was adapting and responding to messages broadcast by members of the groups
to whom they were opposed. These messages were soft facts in that it was difficult to
corroborate them and establish their validity given the fast moving nature of the situation. But
they were sufficient to act as indicators of the presence of a perceived threat needing to be
countered. This seems to be a key ingredient in terms of persuading supporters of the need to
physically mobilise.
Analysis of the Twitter data showed that it peaked on the first evening with nearly 900 Tweets
per minute relating to the Rigby case, as people flocked online to try and find out more about
what was happening. More finegrained analysis though revealed that embedded within these
large volumes of data, were defined thought communities actively using these platforms to
try and rapidly mobilise their political support.

Figure 17. Tweet Volumes Per Minute for Lee Rigby Murder During First 24 Hours
Total Tweet Volume - Sentinel - Woolwich Incident
1000

900

892

800

705

700

600

Tweets per
minute in
Woolwich
Channel

549

500

420

400

300

279

200

100

22/05/13
09:36

22/05/13
14:24

22/05/13
19:12

23/05/13
00:00
Temporal period

23/05/13
04:48

23/05/13
09:36

23/05/13
14:24

DATA FOR GOOD

This case exemplifies just how social media can be used to track and trace elements of
complex and unfolding patterns of social reaction. Initially the community of Twitter users
mobilised online, using the broadcast function to interrogate their social network about what
might be happening, and to share what they knew. Much of what was communicated in these
exchanges was soft facts in that the validity and reliability of the information was clearly
uncertain. But the interesting thing is that collectively some of the participants did a good
job in establishing an accurate definition of the situation. Their soft facts were progressively
hardened up into a coherent public narrative, often as a result of interventions by mainstream
media outlets reporting official sources. When LBC radio in London began to suggest that
the murder might be a terrorist incident, their statements were treated with a degree of
authority and their intervention was widely retweeted.
Following on from such sensemaking processes however, other voices became more active
on social media. Notably, the English Defence League, an established community of interest,
became a significant actor in the story once the incident was defined as Islamist extremism.
They used specific rhetorical techniques to try and spontaneously mobilise their support, as
did those opposed to their ideas. The analysis clearly evidences that both sides struggled in
this respect. Social media and the soft facts communicated by it were partisanly deployed,
but converting interest and online support into a physical presence and action proved difficult.
In the days and weeks following the murder of Lee Rigby, reports were received from
around the country of hate crimes being committed against people and symbolic buildings.
Community tensions were running high. One story though cut through this general climate
of unease and concern. It told of how tensions relating to an EDL march in York had been
diffused when members of a local Mosque had come out and talked with the marchers over
a cup of tea, and a friendly game of football had ensued. This story was picked up by many
national media outlets and was repeatedly recounted by journalists. It became almost iconic
in the narrative of the Lee Rigby murder, conveying a morality tale of how community impacts
were managed and a sense of order restored. For implicit in the narrative that was worked up
was how symbols of English identity (a cup of tea and biscuits, and football) were effective at
diffusing the potential for racial conflict.
However, our analysis of social media data around this event evidences that what actually
happened, was significantly different from the mythologised narrative propagated by
broadcast media and the press. Very few farright political supporters actually mobilised
despite calls for them to do so on Twitter, possibly because there was a major national event
elsewhere in the country the day before. A far more successful counterdemonstration was
organised though. From the point of view of the interests of this study, two particular aspects
stand out:
The coalition of left wing and antifascist groups that quickly mobilised in response to the
threat of a right wing antiMuslim march was scaffolded by several key local institutions that
provided an infrastructure and centre of gravity for the action.
On both sides, potential participants were actively using social media to try and test the
level of support that was actually going to show up so that they were not in danger of being
isolated.
In a study of below the radar activities the latter behaviour is especially intriguing as those
involved were effectively using social media platforms as a sonar radar pinging a signal
to see what response they received, which would tell them something about the social
environment they could expect. This was not about being above or below the radar, but about
how social media affords a network of lots of microradars.
The final case study in the empirical sections of the report shifts focus to attend to how a
particular online community mobilised its resources to intervene in the aftermath of the Rigby

41

42

DATA FOR GOOD

case. This is especially interesting because it is a purely online action, potentially pointing
to new directions that need to be taken in studies of what actually counts as community
mobilisation. On the 27 May 2013, the webpage of the English Defence League was hacked by
the online collective Anonymous UK. All data relating to personal information of EDL leaders
and supporters, including addresses and telephone numbers had been copied and over the
coming days was publicly posted on Pastebin. It is not clear how spontaneous this action was
from the available data, but it was certainly below the radar in terms of coming unexpectedly
from a nontraditional civil society organisation.

Conclusions
Taken together the case studies provide important insights into the role of social media in
how and why community mobilisation does and does not occur, as well as the wider concepts
involved in thinking about below the radar social action.
Social media provides fast data about emerging events and unfolding situations. But the
speed of communication often means that the provenance, and validity and reliability of the
data cannot be confirmed. As such, the information is often treated by potential users as
contingent soft facts in due course it may have to be updated but in the absence of more
authoritative sources it fills an information gap.
Soft facts can be especially influential when they provide a stimulus for collective action, in
terms of a perceived problem or threat that needs to be countered. In the aftermath of the
Lee Rigby killing, politically opposed groups sought to use social media to mobilise their
supporters. It was interesting that both sides struggled to convert soft support online into
actually mobilised support in real life. Community mobilisation efforts did become more
successful when they were supported by established institutions. This is in keeping with recent
research identifying the importance of institutional infrastructures in providing a platform for
social action.25 But in part, it may also reflect the particular challenges of the circumstances
being examined, in that in the aftermath of a dramatic crime where emotions were elevated,
direct involvement entailed personal risk, given the conflict dynamics unfolding.
These findings connect to some wider implications the analysis has for thinking about civic
action being below the radar and for the application of social media analytics to such issues
more generally. With the widespread adoption of social media, being below the radar is not
the same as being completely off the radar. In terms of our analytics, we found we had to
tune our radar to be able to detect community mobilisation that was taking place. Given the
volumes of data circulating through the social media ecology, the digital traces of how these
mobilisation events were being organised were difficult to see with standard instrumentation,
but they were there.
This reflects a more general point that many orthodox data science techniques for crunching
large volumes of social media data are probably insufficiently calibrated to sense the
kinds of complex social processes involved in community mobilisation. For example, our
analysis shows how polarised forms of social support developed around different ideological
standpoints, and these interacted in driving some individuals to physically mobilise. These
interactive processes would not have been picked up by standard sentiment analysis.

DATA FOR GOOD

How communities react and mobilise informal social control following major crimes is not
something that has been studied very often. These are processes that are below the radar
in the sense that they have been largely neglected by the authorities. Some attention is now
paid to conducting community impact assessments and consequence management of any
community tensions, but this is far less significant in terms of public service agendas than
finding the perpetrator. An interesting finding in this regard concerns the important role
played by some institutions in scaffolding responses performed by civil society organisations
and other networks. In this sense, an important question to ask when talking about being
below an organisations radar is whose radar are we talking about? The community
mobilisation actions may have been below the radar for the police, but they plainly werent
below the radar for the individuals and groups engaged in rapidly and urgently mobilising.
Developing this line of thought, possibly the most important thing about the new information
environment is how it has significantly widened the scope of the radar. Accepting the point
about needing to calibrate the radar properly given the issue of detecting small signals of
interest in the vast volumes of noise, it nevertheless remains the case that:
Social media is directly engaged in enabling rapid forms of community mobilisation to be
performed.
The digital traces of these mobilisations that social media cast allow us to study and
understand aspects of the social reaction to major events that were previously seen but
unnoticed.

43

44

DATA FOR GOOD

3.4.

Social action on social media

Carl Miller, Centre for Analysis of Social Media, Demos

Social media platforms are often seen to be nasty, even dangerous places. Misogynistic and
racist language, cyberbullying and hateful abuse all appear on social media platforms in large
quantities. Extremist groups from across the political spectrum, even terrorists, have found a
voice on these digital platforms, and use them to spread their message and find new recruits.
However, these new digital platforms are also used to help others. These are vibrant new
places which people use to volunteer, organise, share skills, mentor, fundraise and donate,
participate in local civic projects and work as activists to change laws and minds. Taken
together, this is digital social action a new class of social good and community resource.
Demos set out to understand the emerging contours of social action on one of the most
important social media platforms: Twitter. We wanted to learn how it happened and who
conducted it. We wanted to understand the contexts that inspired or provoked social action
and the problems it was directed towards. We were especially interested in social action that
usually sails below the radar; conducted outside of the formal structures of charities or social
enterprises, that runs on little or no money and because of this is often missed. Ultimately,
we were interested in how this emerging class of social action could be supported and
encouraged.
We looked at Twitters reaction to two events. The first was the Somerset Floods during
January and February 2014. Following the wettest weather on record, the UK suffered
widespread flooding. Thousands of homes and offices were flooded, causing hundreds
of millions of pounds of damage. The second was the launch of the Step Up To Serve
Campaign,26 launched on 21 November 2013. Supported by Prince Charles and the leaders
of all three main political parties, it aimed to increase the number of young people routinely
conducting social action in their local communities.
Both events prompted large reactions on Twitter, and we collected over 120,000 Tweets
in total. The sheer scale of data has become a perennial problem to those interested in
studying social media: these platforms now routinely create more data than any researcher
can hope to read themselves. We used new technology that CASM has developed to respond
to this problem, called method 51. It allows us to train algorithms to automatically analyse
Tweets. We give the algorithm examples of different kinds of Tweet, and it learns the kinds of
language use that most indicate each. The advantage: we can analyse many more Tweets than
we could ever read. The disadvantage: its never perfectly accurate, and there will always be
some noise in the analysis that we make using this technique.
We built eight algorithms overall for this research, and used them in a number of ways.
Some sorted Tweets relevant to the events in question from those that were not. Some
sought to find Tweets that represented social action in some way, and others to sort Tweets
representing social action into those that came from charities and institutions from those that
came from individuals. On average, the classifiers got it right about 78 per cent of the time.

DATA FOR GOOD

45

The Somerset floods

After a period of extremely wet weather, widespread floods hit the UK in January and
February 2014, causing millions of pounds of damage, and severe disruption to Britains
transport networks. This was selected as a largely unanticipated event that caused
widespread hardship and disruption to communities throughout the UK. In the shocked
aftermath of the floods, as affected communities across the country were reeling from the
impact, people took to Twitter for a number of reasons.
Of the 116,000 Tweets that we collected between 510 February, 84,000 were about the UK
floods and 39,000 were related to social action. These were sustained throughout the time of
the period studied.

Figure 18. Number of social action Tweets versus total number of Tweets relevant to

the flooding, over time
1400
1300
1200
1100
1000
900
800
700
600
500
400
300
200
100
0
5 February

Social action

6 February

7 February

8 February

9 February

10 February

Total Tweets about the flooding

The vast majority of these Tweets were online social action, where people were using Twitter
to try to help others by sharing information, updates, warnings and advice. These included
people were sharing information on which roads and rail services were disrupted by the
floods, weather reports, official warnings and advice and to report on the extent, damage or
disruption of the flooding firsthand, often linked to a photo of the scene.
A smaller body of Tweets (about 3,000) represented offers or mentions of offline help to
those affected by the floods. This included mentions of volunteers working to contain the
floods: pumping water, laying sandbags and transporting supplies. Others offered to rescue,
move or stable animals both livestock and pets. People also used Twitter to offer a range of
other help, such as sewage treatment, the provision of fodder or relief using 4X4 vehicles, and
donating money to relief organisations. The reverse was also true. Some people needed help,
and didnt know who could provide it. Strictly, of course, these Tweets did not represent social
action, but instead the opportunity for it to be done.

46

DATA FOR GOOD

Figure 19. Proportion of Tweets by category of offline social action

9%
6%
33%
10%

Animals
Volunteering
Specialists
Sympathy
Donations

42%

Of those that could be geographically located, a very broad pattern emerged. Tweets
reflecting offline social action of volunteering, donations and offers of help, tended to
come from areas affected by the flooding. Tweets reflecting online social action of sharing
information and advice tended to come from areas especially London and central England
that were not affected by the flooding, but were densely populated. Most of both kinds
were below the radar. These were not Tweets sent by large, organised charities. These were
individuals taking to Twitter, largely outside of any organisational context, trying to help, or
find help, however they could.

Figure 20. Tweets related to social action.

Orange = online social action.

Blue = offline social action.

Shaded = area affected by the

floods. Size of colour =

quantity of Tweets.
Overall, these Tweets represented a form of digital
community resilience. In the aftermath of a largely
unanticipated disaster, communities across the
UK reacted spontaneously to try to help one
another. This research suggests that there is now
an important digital dimension to this reaction.
During the floods, Twitter became an important
forum for important information, realtime
updates from the ground, and a key exchange to
make offers for help and to appeal for it.

DATA FOR GOOD

Step Up To Serve Campaign

The Step Up To Serve Campaign was a highprofile call to get over half of young people
regularly conducting social action by 2020. It was launched in Buckingham Palace by Prince
Charles, David Cameron, Ed Miliband and Nick Clegg, received major support from businesses
and third sector organisations and was widely covered in the media. The Campaign asked for
people to use Twitter to tell people how they planned to help young people volunteer using
the hashtag #iwill. It was selected because it was an opportunity to study an event that has
caused many people to speak about social action, their plans, views and priorities.
Over the first month of the Campaign, we collected 10,665 Tweets, of which 3,780 were
relevant to the Step up to Serve Campaign. People used Twitter to respond to the campaign
in a number of ways. Some Tweets were attempting to increase awareness of the campaign,
and sharing links to news stories covering it. Others spoke about the campaign, discussing its
objectives and intentions. Importantly, some people used Twitter to make pledges to conduct
social action or help others to do so.
The nature of this response was markedly different from the first case study. In responding
to the Step Up To Serve campaign, people did not use Twitter to look for, or offer, specific
help. Instead, people offered their longerterm views and plans related to social action. Of the
3,780 responses to the launch of the campaign, 1,884 Tweets contained a pledge to support
or conduct social action; 690 of these came from, as far as we can tell, normal people acting
outside of any kind of structure or organisation. The highest proportion of Tweets pledged
were a commitment to social action in general. They contained pledges that supported the
concept of social action and generally promised to help, but did not specify what this help
would be. Some Tweets contained offers to help young people to find work as a form of
social action, or to allow young people to improve their employability through volunteering
themselves. Others specifically mentioned social action to support the local community,
disadvantaged groups, the education system or to increase how social action is understood
and appreciated.
Overall, Twitter was used to talk about peoples experiences, what they have done that
day. It was also used to mention what people find inspiring and laudable. Hence, some
Tweets contained information about specific acts of volunteering that the person had done
themselves, or had seen happen. This shows that Twitter can be studied to gain insight into
attitudes towards social action its prestige and sense of importance, the kinds that people
think are a priority and how they intend to do it.

Conclusion
Two different case studies produced two very different reactions from Twitter, but overall
they tell us an important truth: Twitter is a significant new forum for social action. It reflects,
mobilises and unlocks new forms of mobilisation, of people helping others.
There are two important opportunities presented by the rise of Twitter, and social media more
broadly. We can use these new platforms to understand social action better, and also to help
encourage it.
First, theres a research opportunity. Social media platforms and especially Twitter produce
huge amounts of data, much of it in realtime. Alongside the rise of these valuable new
sources of social information are the emergence of powerful new analytical techniques that
are capable of crunching and making sense of all this information. Even within the scope of
this limited pilot study, 41,211 pieces of information relevant to social action were found. None

47

48

DATA FOR GOOD

of these new technologies and techniques are perfect, and many struggle to produce insight
as robust as that you would expect from conventional sociology. However, social media
research methods are improving all the time, and, in my view, this is a valuable new research
coalface to be worked on.
Second, theres an opportunity to proactively increase the extent and effect of social action.
People used Twitter to both offer help and also ask for it, but the process was ad hoc and
chaotic, and it wasnt clear whether suppliers and providers effectively linked up. Valuable
offers for help are buried within huge quantities of other, irrelevant Tweets, and its very
difficult to find them. We need to do more to better support the growth and increase the
effect of social action on social media.
I recommend an Ebay for social action a powerful application of digital platforms in other
areas especially commerce and the provision of services be developed to directly connect
people providing something with those who need it. When social action information is found,
it could be centralised onto a realtime online platform, information exchange or brokerage
hub, clearly related to a specific event and segmented either into the type of help that people
are offering, or where the help is being offered. This would allow people to find and help, and
also know what help is needed and who needs it. Its worked for commerce; it could work for
social action too.

DATA FOR GOOD

Mapping below the radar

organisations on crowdfunding
platforms

3.5.

Maria Botello and Noel Hatch, European Alternatives

1. Introduction
It provides a methodology to map below the radar (BTR) activity on two crowdfunding
platforms: Spacehive and Crowdfunder. Crowdfunding platforms seemed a suitable environment
to find data on BTR organisations as they can provide these organisations with the financial
resources needed to carry out their work. As BTR groups are characterised by their activity, the
crowdfunding projects were the unit of analysis selected to account for this type of organisation.
This methodology has been developed into a software prototype that uses webcrawling
techniques to navigate crowdfunding projects through hyperlinks and captures those projects
complying with certain criteria. Namely, projects are expected to be run by BTR organisations
if 1) they meet the condition upon which co-link analysis is based, 2) have reached their
funding target, and 3) are not run by registered organisations.
The application has been used to collect data in order to conduct a case study for each of the
platforms. This paper explains the methodological approach undertaken and the development
and evaluation of the methods, based on the results obtained in the case studies. The paper
will conclude with a summary of the lessons learnt across the two case studies and the next
steps to be taken in the development of datadriven methods for BTR research.

2. Methodology to track BTR activity on crowdfunding platforms

What do we mean by below the radar organisations?
The first thing that becomes apparent when conducting any kind of research on BTR activity
is the need to establish a working definition for this concept. To that end, a very concrete
notion as to what can be considered a BTR organisation was selected from the multiple
definitions documented in third sector research literature.
The term BTR organisations is used in the context of this project to refer to those
organisations that, while carrying out voluntary activity with a social purpose, remain
unregistered and unregulated. That is to say, they do not appear in official registers such
as the Register of Charities by the Charity Commission. In summary, it describes those
organisations that are under the regulatory radar.27

Why are they likely to be found on crowdfunding platforms?

Crowdfunding is defined as the practice of funding a project or venture by raising many small
amounts of money from a large number of people, typically via the internet.28 The fact that
BTR organisations lack a legal structure implies that they might not be able to raise funds
for their activity through the conventional channels used by organisations with a registered
formal status. Therefore, crowdfunding platforms constitute an available alternative to BTR
organisations in terms of funding.

49

50

DATA FOR GOOD

As the scope of this project is limited to the study of BTR organisations within the UK,
CrowdingIn.com, a directory of crowdfunding platforms that operate in this country, was
used to select the platforms. The reason why the crowdfunding platforms Crowdfunder and
Spacehive were eventually selected is that they are oriented to projects with a social purpose,
often referred to as civic crowdfunding projects, and that presented an appropriate data
structure for the method devised to find new BTR groups.

Colink analysis as a way to identify thematically related projects

The main method used to identify new BTR groups in crowdfunding platforms is co-link
analysis, which was implemented for the first time in the Issue Crawler. The Issue Crawler,
a web application developed by the Digital Methods Initiative, deploys co-link analysis to
enable the location and analysis of issue networks on the web, defined as a set of web pages
dealing with a common theme that are connected by hyperlinks.29, 30
In the framework of this project, co-link analysis is used to identify a set of crowdfunding
projects thematically related, rather than web pages, departing from a single seed project.
Any project on those platforms could be used as a seed, that is to say, as the initial project
from where to start the search - or crawl - of new projects by following its hyperlinks.
As this implies shifting the focus from crawling the whole web to crawling specific platforms,
co-link had to be adapted to fit the particular data structure on the selected platforms. On
these platforms, projects do not show hyperlinks to other projects. Instead, each project has
hyperlinks to its backers profile pages, and these from pages hyperlinks to the pages of all
the projects they have pledged money to.

Figure 21. Colink analysis as deployed in this project

Seed project

Contributors

New projects

DATA FOR GOOD

As a result, co-link analysis is deployed to follow the connections or hyperlinks between two
different objects: the projects concerned and their contributors. Thus, given a seed project,
new projects can be obtained. The premise here is that only the projects that have at least
two contributors in common with the seed are captured for further analysis. That is to say,
projects that have been funded by at least two people who also pledged funds to the seed
project.
The new groups are expected to present a thematic relation with the seed and between
themselves. That is to say, if a seed project is selected that pursues a social aim that concerns
the environment, the projects obtained from that seed will all be expected to revolve around a
similar theme.

Complementary conditions
Meeting the premise above didnt prove to be a sufficient condition to ensure that the new
projects were the type of organisations sought for this research. Therefore, extra conditions
were established in order filter the projects resulting from the co-link. Namely:
Being worth studying. Only projects that have reached their funding target are deemed
worthy of attention, the reason being that only these projects will actually receive the
money to be realised. As projects are often developed over an extended period of time
during which the organisations behind them can be considered to be active, the successfully
funded project seemed a good indicator of sufficient and significant activity beyond the
purely digital domain.
Being run by BTR organisations. It is usual to find projects on these platforms that, while
aimed at a social purpose, were created by registered organisations, from charities to
different types of social enterprises, and therefore, do not count as being below the radar.

3. Deploying the methodology on Crowdfunder

Crowdfunder is currently the UKs leading reward-based crowdfunding platform. By April
2014, it was found to have more projects than all other UK platforms.31 As a consequence,
even if this platform is not exclusively aimed at projects with a social purpose, it presents a
significant number of this kind of project or, in other words, enough relevant data.

Development of methods
Both the co-link analysis and the conditions above were implemented into an application
intended to find projects by BTR organisations on Crowdfunder. With regards to the co-link
analysis, an extra iteration was developed in order to get more new projects from a single
seed project. That is to say, the program runs the crawler for the first time based on the seed
URL entered by the user, and then uses the URLs of the resulting projects as the new seeds
for a second iteration of the co-link analysis. The set of projects output by the app will include
those projects obtained from both the first and the second iteration.
The condition that checks for those projects worth studying was implemented so that the
app only returns those projects that had reached their funding target within the present or
previous year. This was done to ensure that the organisations behind projects are alive at the
moment data is collected.

51

52

DATA FOR GOOD

Figure 22. Crowdfunder application

The approach undertaken to filter out projects by BTR organisations was not the one
originally intended. This alternative solution was implemented based on the fact that
registered organisations tend to make explicit their legal status on the description of their
projects on the platform. In doing so, a set of keywords was selected that account for the
most common ways in which the different legal status is featured on Crowdfunder.32 While
running, the app checks whether the selected keywords are shown on the description of
the project, and if so, discards it, keeping only projects that do not contain those specific
keywords.
The app also allows the export of the results in the appropriate format for Gephi, an opensource network analysis and visualisation software package. When imported into Gephi, the
resulting data is used to create a network visualisation where projects are represented as
nodes and the connections between them represent their common contributors.

Case study
The application has been used to collect data for a case study aimed at testing the efficiency
of the methods in identifying BTR organisation on Crowdfunder and exploring what insights
can be drawn from the analysis of the data on BTR activity.
The selected seed project to run the application for the study was Grow a Future for
Families, as the number of projects it returned was the highest among those seed projects
used to test the app. Namely, a total of 16 new projects presumably created by BTR
organisations. Then, the application was run a second time using the same seed but leaving
unchecked the keywords filter, a feature enabled by the app. This resulted in up to 61 new
projects both by registered and BTR organisations, allowing for a more accurate evaluation of
the co-link analysis method.

DATA FOR GOOD

This brings us to the next step undertaken, visualising the results using Gephi. Two different
network visualisations were created based on the set of results above. On these networks,
project nodes are coloured based on their category on the platform as a way to render visible
thematic relations between them. Results on both networks were used in the evaluation of
methods.

Figure 23. Network of projects

In doing so, the following questions were formulated: do the resulting projects pursue a social
purpose? Are these projects thematically related? To what extent can platform categories
be used to determine that? With regard to those projects that are supposedly run by BTR
organisations, is this actually the case?
Based on the evidence provided by the biggest network, the method seems to work
successfully. The majority of the projects returned have a social purpose and are thematically
related to those other projects with which they share connections, even if they do not match
the same platform category. The reason is that projects often do not stick to a single theme
but rather have several thematic dimensions, which are not all reflected by the assigned
category.
Thus, for example, the purpose of two projects on the platform Snact and The Happy Pig
reflect strong environmental values, even though those projects were assigned categories
other than environment. And in fact, both projects, especially The Happy Pig, present a
significant number of connections to projects under that category.
For the purpose of checking whether the organisations behind the 16 projects on the BTR
network were indeed below the radar, a more detailed analysis of the pages of those projects
on the platform and their own websites had to be carried out. In doing so, it can be concluded
that even if the keywords filter works successfully, in that the selected keywords did not
appear on the description of the projects, this method presents significant limitations, such

53

54

DATA FOR GOOD

as the fact that it does not take into account the context of the keywords. Furthermore, it was
found that registered organisations do not always reveal their legal status on Crowdfunder.
As a result of the above, only six out of the 16 projects on this network, proved to be by BTR
organisations. Both the Enchanted Acres and The Happy Pig projects were intended to build
spaces where people can explore and learn about permaculture and sustainable ways of living
at no cost. Similarly A tipi for earth education, aimed at raising funds to build a tipi for school
visits where children can learn environmental values and connect with nature.
The purpose of the The Big Sun Flower project is to send out seeds to people at no cost, in
order to raise awareness of centronuclear and myotubular myopathy by growing sunflowers
all over UK and beyond. Anna Harbour Reopening was created with the goal of bringing
together the necessary resources to clear a local harbour of silt accumulated over 50 years
and turn it into a community asset. Ask Amy is an app prototype aimed at increasing the
number of young people accessing and getting involved in politics. The project on the
platform intended to raise money to develop the prototype into a fully functional application.

4. Deploying the methodology on Spacehive

Spacehive is said to be the worlds first crowdfunding platform for civic projects.33 It is
primarily aimed at infrastructural projects at the local level in order to help communities to
transform their public spaces. Thus, it seemed a suitable place to locate BTR activity.

Development of methods
Even though the data structure on this platform is suited to deploy co-link analysis, this
method was not eventually employed due to the lack of the relational data needed for that
purpose. That is to say, that at the time the method was being implemented it became
apparent that funders hardly ever contribute to more than one project. This might be due
to the local nature of Spacehive projects that results in funders contributing exclusively to
projects in their own area.
The reason why it was decided to go ahead with the development of a more simple
method to map BTR activity on this platform is that Spacehive is specifically addressed at
civic projects. As this type of project is presumed to have a social purpose, it seemed that
deploying a method based on the two conditions previously described could still fit the
purpose of this research.
In compliance with the first of those conditions, an application was developed that captures
all those projects on Spacehive that have succeeded in meeting their funding target,
regardless of when, as the platform does not provide this information.
Moreover, in order to keep only those projects run by BTR organisations, the app combines
two different filters: the keyword filter already described and another one that checks whether
the names of the organisations behind Spacehive projects appear on certain registers.34
Where this is the case, the project is discarded and otherwise captured for further analysis.
This later condition could be implemented because on Spacehive the username of the project
promoter usually matches the name of the organisation behind the project.35

DATA FOR GOOD

Figure 24. Crowfunder application

Case study
The app was also used to collect data on Spacehive. As the method to detect BTR activity on
this platform is not based on co-link analysis, and the relations between the resulting projects
based on common contributors were not tracked, there was not suitable data available to
build the network visualisation of thematically related projects.
As a consequence, other questions were formulated in order to check for the efficiency of this
method: Is the fact that the platform is aimed at civic projects reason enough to ensure that
the resulting projects pursue the type of social purpose that characterises BTR organisations?
Do the resulting projects actually belong to BTR organisations?
From the resulting 13 projects, nine were found to pursue a social purpose to some extent.
As for the second question, the combination of both restrictions seemed to work well for
a good number of projects, as it was observed that six out of the nine projects that have a
social purpose have been created by BTR organisations. However, the second restriction also
presents limitations, as it was observed that among the results there are a few instances of
projects whose creators on the platform do not match the name of the organisations behind
them for different reasons. Moreover, it should be noted that a proper copy of all concerned
registers would allow for more refined results. The purpose of the six BTR projects found in
the platform were:
Burghead Tennis was aimed at purchasing a portable tennis net and other equipment to be
used by a tennis club and local communities in a school yard.
The Northgate Herb and Fruit Beds project had goal was to convert a derelict eyesore into
small community garden where people can grow and harvest their own fruit and vegetables.
A Childs Dream sought funding to build a sensory garden for a young girl with autism as a
pilot for a community garden.

55

56

DATA FOR GOOD

The next three projects present a more artistic approach to their purpose.
The Porty Light Box was intended to raise funds to turn a decommissioned phonebox into
light box from which to display images by local artists, schoolchildren or other groups.
The project Drum Together Brum Street Party was about creating a one-day free event for
the local community featuring a programme of live music performances and workshops.
After the riots - Happiness in Tottenham raised money to set up an exhibition based on a
series of proposals intended to address the psychological and financial impact that the riots
of 2011 had on this area.

5. Lessons learned across the two case studies

While civic projects on Spacehive have by definition a social dimension, they not always have
the wider social purpose characteristic of voluntary activity. And in fact, when comparing the
results from both platforms it becomes apparent that co-link analysis, based on crawling the
common preferences of contributors, constitutes a more efficient way to check for projects
with the kind of social purpose sought.
The proportion of BTR projects returned with the Spacehive crawler was nonetheless greater
than for Crowdfunder. To some degree this might be due to the fact that two different filters,
the keywords and the registers restrictions, were used for the former platform. However, it
seems that, even if both filters present limitations at present, the filter using the registers
implies a more consistent approach to check for the legal status of organisations on
crowdfunding platforms.
Furthermore, the notion of BTR organisations as those being under the regulatory radar
seems appropriate for the purpose of developing automated data collection methods.
Nonetheless, in order to enable a richer understanding on BTR activity on crowdfunding
platforms, a more nuanced analysis of the projects and organisations identified from
deploying those methods is needed.
For example, when researching BTR organisations behind projects on both platforms it
became apparent that relatively often BTR activity appears combined in varied ways with
that of registered organisations and governmental bodies. This is especially the case for those
crowdfunding projects intended to have continuity over time, for which they receive support
from other organisations in the form of funding, material resources or spaces.

6. Next steps for improving the efficiency of BTR methods

In summary, co-link analysis seems a valid method to map new projects thematically related,
which, if combined with a suitable method to check for the type of BTR activity sought, can
serve to identify BTR groups in crowdfunding platforms. Having the data in a format that suits
methodological requirements would be crucial in making the necessary improvements to both
methods.
To that end, a form of partnership with the platforms concerned could be established. For
example, the data required to deploy co-link analysis could be enabled through an API, which
would allow for the development of a more robust application.36
With regard to the methods aimed at filtering out projects by BTR organisations, it seems
that if further developments were to be made, they should focus on improving the restriction
based on the registers. However, platform collaboration would still be needed for that
purpose. So, for instance, it would be extremely useful if platforms ask users to specify the
name of the organisation they work for whenever they are acting on its behalf.

DATA FOR GOOD

3.6.

What we learned about the

potential of using datadriven
methods to understand hidden
social action

So what do the five research projects tell us about the opportunity in datadriven methods?
Datadriven methods, particularly those looking at Twitter data, can help us understand new
types of hidden social action. The work by the RSA, Cardiff University and CASM show that
people do have conversations about local social issues online, and that datadriven methods
can indeed be used to understand these new types engagement and of hidden social action
that results from conversations. The study of the Somerset floods and the Lee Rigby incident
show how Twitter in particular is a powerful medium for producing and sharing knowledge
about a social issue that can lead to action.
What is unclear, however, is the extent to which activity and engagement online add value
and impact, either through the distribution of valuable knowledge between peers or resulting
in offline social action. The study of the Somerset floods did prove that this was the case in
some instances, with people offering to help others affected by the floods with housing, but
the extent to which there was an actual exchange is unclear.
As outlined in the introduction to this chapter, this research builds on the existing studies
of below the radar groups. As demonstrated by the RSA study in Hounslow, datadriven
methods can be used to create a local map of assets and resources that add to, and
complement, existing maps of assets developed using offline methods.
An emerging understanding of what constitutes BTR activity when it is happening online,
and methods for identifying this activity in large datasets: One of the big challenges
is how to definite the type of activity that people engage in online; how do we define
what constitutes a group or a BTR community and not just a few people having a
conversation online? This is easier when looking at examples of a common social project
that everyone participating in the project joins in with, such as a crowdfunding campaign,
but more challenging when looking at more unstructured and varied data, such as Twitter
conversations. However, the work on tools and methods such as the Sentinel, Localnets and
method 51 demonstrate how the analysis of big datasets can identify those incidents where
larger groups of people come together to discuss and potentially take action on a social issue.
The question remains, as alluded to earlier, as to what the effect of this is and the extent to
which the community forming online translates into social action either off or online.
However, the question that follows the ability to map this activity through new datadriven
methods is whether or not below the radar is the right label? Firstly, much of the activity
that the different methods have mapped can be described as being very different from the
traditional offline informal activity, often focused on knowledge exchange and mobilising
people around single events rather than ongoing longterm commitment to a social issue.
We still know very little about the below the radar experience and how is it different from
participating in above the radar activities, or the differences between off and online BTR
activity.

57

58

DATA FOR GOOD

There is often a relationship between those above and below the radar. While the purpose of
BTR research is to identify and understand activity not driven by established above the radar
organisations, such as charities and social enterprises, it is important to remember that this is
a classification developed by academics and government to define what is and isnt in public
registers, and not necessarily how people think about their own social action, or how charities
think about mobilising and managing volunteers.
Case studies such as the study of the Lee Rigby murder and the Somerset floods show how
key nodes play a strong role in mobilising online BTR activity. However, those nodes will often
be an above the radar organisation, such as local mosque, or a person affiliated with this.
While this challenges the terminology and makes it a more complex field to research, it adds
to the growing knowledge around the potential for civil society organisations in mobilising
people to take part in social action, via social media in particular.37
Following on from this, the NCVO study showed how open data from funders can help
lower the radar and develop a more refined understanding of civil society activity through
identifying those organisations that are established enough to receive a small grant but dont
feature in any official registers.
Increased access to data will enable more and better datadriven BTR research; the
challenge is to do this without invading property. All of the methods described in this paper
are driven by the increased social action taking place online, and the access to this data
through open data, webscraping APIs etc.
In this context it is important to note that the five projects described above are primarily
looking at data from Twitter and other social media platforms, crowdfunding platforms and
open datasets in their analysis. Focusing on these data sources means that there is lots of
online BTR activity and types of social action that is being mobilised through other platforms
that the research methods wont pick up.
As all the research projects have focused on either accessing twitter APIs, open datasets or
scraping information that is already in the public domain, the issue of privacy has been less
of an issue in this research that it was for the CAB and DataKind. However, as these methods
become more refined and widely accessible, the civil society research community needs
to continue the discussion on how to tread the fine line between using these methods and
developing a better understanding of civil society without invading privacy.

DATA FOR GOOD

4. Where

next for big and open

data for the common good

s demonstrated by the projects in this paper, there is real potential in

using datadriven methods to both help charities develop better services
and products, and understand civil society activity. Key lessons and
recommendations for future work in this space include:
Better investment in data skills in the third sector: Government and philanthropic
organisations should invest in programmes that build collaboration between data scientists
and organisations in the third sector, such as the collaboration between Datakind and CAB.
While the work by CAB and Datakind has focused on the opportunities for a bigger charity
such as CAB, these programmes should focus on how to support the sector as a whole,
including smaller community and voluntary sector organisations.
Data dives is one approach to this. Another example is the Open Data Challenge Series, a
collaboration between Nesta and the Open Data Institute. It runs seven challenge prizes
that invite businesses, startups and individuals to develop innovative solutions to address
social issues using open data in areas such as Crime and Justice, Energy and Environment,
Education, Housing and Food. This has already led new solutions such as MoveMaker, an app
which uses open housing data to help social tenants house swap.
Support the development of better and more open data: Much of the work on both
researching and developing new products is only possible because of access to highquality
data and the sharing of open data between organisations.
Open data from grant funders can lead to a better understanding of where money is spent
and the size of civil society, as illustrated by the experiment mixing data from St Mungos
Broadway and CAB to detect and address homelessness issues, and the work by NCVO.
However, to reap the benefits of this, potential organisations, from government and big grant
funders to big and small charities, need to continue experimenting with new ways of opening
up and collaborating around data, without putting privacy at risk. The 360 Giving programme
which helps grant funders open up their data has pioneered much work in this space, and a
continuation of this programme to support better collaboration around data is recommended.
Make better use of social media to engage with and build civil society capacity: The work
done by CASM, Cardiff University and the RSA has provided some early insights into how
social media, in most cases Twitter, can be used to better understand new types of civil
society activity and local assets. However, following on from this it would be interesting
to explore the real potential of social media to help charities mobilise volunteers and
communities to address issues and signpost people to local resources in more structured
ways. Furthermore, how can public services, and local government in particular, use these
approaches as a tool for understanding local conversations around issues and assets, to better
respond to public issues, such as the Somerset floods.
A new type of civil society organisation, a new way of understanding of civil society: While
the vast majority of civil society activity still happens through established organisations
such as registered charities, the below the radar grantees and Nestas other research in to
areas such as digital social innovation and crowdfunding show that people are constantly
finding new ways of mobilising time, resources and money to address social issues via online
platforms. As this trend continues, we need to keep working towards understanding what a
digital civil society looks like.

59

60

BIG DATA STORIES

Endnotes
1. The Data Ambassadors: Peter Passaro, Chief Data Officer at Datanauts and a natural language processing expert, Iago Martinez, VP of
Data Engineering at DataShaka, and Arturo Sanchez, who at the time was a data scientist at YouGov.
2. Arturo Sanchez Correa was replaced with Sam Leach, a data scientist at Inquiron.
3. Ian Huston, a data scientist at Pivotal.
4. Henry Simms, an Integration Technical Lead at Experian Data Quality UK, lead the design and development of the visualisations, and
Billy Wong Principal at Hedgehog & Fox, was recruited to analyse the connections and relationships between issues.
5. St Mungos Broadway is also open to the idea of sharing data and has approached Citizens Advice about continuing to collaborate on a
new research project.
6. For more information about the Shooting Star Chase project see http://www.datakind.org/finding-30000-missing-children/
7. For more information about the Buttle UK project, see: http://www.datakind.org/making-sense-of-text-data-to-help-disadvantagedfamilies/
8. For more information about the Access project, see: http://www.datakind.org/projects/improving-access-to-education-by-supportingtutors/
9. http://www.datakind.org/blog/dc-action-for-children-long-term-collaboration-for-long-term-impact/
10. The map can be viewed at: https://datakind-uk.github.io/child-poverty-commission-dashboard/
11. For more information about the project with The North East Child Poverty Commission, see: http://www.datakind.org/projects/
communicating-about-child-poverty/
12. http://www.nesta.org.uk/publications/compendium-civic-economy
13. http://www.nesta.org.uk/sites/default/files/people_helping_people_the_future_of_public_services_wv.pdf
14. http://www.nesta.org.uk/project/centre-social-action-innovation-fund
15. http://www.nesta.org.uk/project/innovation-giving-fund
16. http://apps.charitycommission.gov.uk/showcharity/registerofcharities/SectorData/SectorOverview.aspx
17. http://www.birmingham.ac.uk/generic/tsrc/index.aspx
18. http://data.ncvo.org.uk/
19. http://epapers.bham.ac.uk/1550/ and http://www.birmingham.ac.uk/generic/tsrc/documents/tsrc/working-papers/working-paper-29.
pdf
20. http://threesixtygiving.org/
21. https://opencorporates.com/
22. http://opencharities.org/
23. http://www.oscr.org.uk/charities/search-scottish-charity-register/charity-register-download
24. Development of Sentinel was supported by the TaRDiS project (Tackling Radicalisation in Dispersed Societies, funded by the European
Commission, 2012-2015) and the After Woolwich project (funded by the ESRC, 2014-2015).
25. Sampson, R. (2012) Great American City: Chicago and the Enduring Neighbourhood Effect. Chicago: Chicago University Press.
26. http://www.iwill.org.uk/
27. McCabe, A. and Phillimore, J. (2009) Exploring below the radar: issues of theme and focus. University of Birmingham, Birmingham:
Third Sector Research Centre. See: http://www.birmingham.ac.uk/generic/tsrc/documents/tsrc/working-papers/working-paper-8.pdf
28. http://www.oxforddictionaries.com/
29. Marres, N. (2012) The re-distribution of methods: On intervention in digital social research, broadly conceived. London University,
London: Goldsmiths Research Online. See: http://research.gold.ac.uk/7773/1/Marres_redistribution_of_methods.pdf
30. The Issue Crawler deploys co-link analysis in the following way: from a set of seeds or website URLs entered by the user, the crawler
returns only the URLs of those sites that receive links from at least two of the seeds. The resulting websites are expected to be relevant
actors on the Issue Network of the particular theme introduced by the seeds.
31. http://www.ukbusinessangelsassociation.org.uk/news/crowdfunder-smashes-through-500000-funding-target-just-over-3-hoursequity-crowdfund-campaign
32. The keywords used were: CIC, foundation, charity, social enterprise, cooperative, co-op, co-operative, ltd, association, company,
business, trust. It should be noted that these keywords were selected from the observation of a limited sample of projects. Thus, adding
other well-documented keywords to that list would help to achieve more accurate results.
33. http://www.bigsocietycapital.com/how-we-invest/spacehive
34. Up to three different registers were found susceptible to contain the types of formal organisations present on crowdfunding
platforms: the Register of Charities by the Charity Commission, the Register of Companies by Companies House, which includes
regular companies, CIC and other social enterprises, the Mutuals Public Register by the Financial Conduct Authority, which includes
organisations such as Co-operatives. Although it was not possible to get an official copy of those registers on a suitable format, I could
get through more informal channels two updated files with the names of all registered charities and community interest companies. In
order to get more accurate results, a proper version of all the above registers should be used instead.
35. This method was not eventually developed for Crowdfunder due to the fact that relatively often project owners do not register on the
platform with the name of the organisation they work for, but use instead their own name or a random username.
36. An API, or application programming interface, enables applications to access data on platforms in an easy and stable way. As a result,
apps that use APIs do not need to be modified every time the data structure on platform pages changes. This is, however, the case for
those applications using web crawling techniques such as the one developed for this project.
37. http://www.carnegieuktrust.org.uk/getattachment/ee2be1a3-8a94-4d8b-963f-94e29c8a9781/Making-the-Connection--The-Use-ofSocial-Technolog.aspx

Nesta
1 Plough Place
London EC4A 1DE
[email protected]
@nesta_uk
www.facebook.com/nesta.uk
www.nesta.org.uk
Nesta is a registered charity in England and Wales with company number 7706036 and charity number 1144091.
Registered as a charity in Scotland number SCO42833. Registered office: 1 Plough Place, London, EC4A 1DE.