Machines Research 1
Mapping the social and economic characteristics of
high density gambling machine locations
Prepared by Geofutures and NatCen for
The Responsible Gambling Fund /
The Responsible Gambling Strategy Board
Authors: Heather Wardle, Ruth Keily, Mark Thurstain-Goodwin and Gaynor Astbury
35 Northampton Square 108 Walcot Street
London EC1V 0AX
Bath BA1 5BG
1 Research objectives
2 Research evidence
5 Interpretation and conclusions
6 Access to interactive online maps
This study undertaken for RGF/RGSB by Geofutures Ltd and NatCen aims to address
the lack of empirical data for Great Britain on the spatial distribution of gambling
machines and the socio-economic characteristics of the neighbourhoods in which the
highest density clusters of gambling machines are found. The underlying objective was
to obtain baseline evidence on which to base further research in support of a strategy
that aims to help prevent vulnerable individuals from gambling-related harm.
While location data for the majority of licensed gambling machine venues were
available, for most venue types, machine numbers by venue were not. Various
research methods, including a fieldwork validation exercise, were employed to obtain
sufficiently accurate estimates of machine numbers by venue type to allow a machine
density analysis to be undertaken. Estimates and assumptions employed were based
upon a rapid evidence assessment of relevant recent literature as well as expert
Mapping, spatial statistical analysis and distribution analysis techniques were used to
identify ‘machine zones’ surrounding all gambling machine venues, and a subset of
‘high density machine zones’ within them. Key Census-based social and economic
characteristics of these zones were then analysed and tested for statistical significance.
High density machine zones were found to be positively correlated with areas of lower
income, economic activity and employment status compared with the respective
national averages, but the spatial distribution of high density zones is more complex
than this alone suggests.
Suburban, secondary urban, satellite and coastal locations were most notable among
the highest density areas, and even within these groups significant differentiation was
detected. A notable number of New Towns emerge within the highest density zones,
especially those with higher numbers of low-income neighbourhoods.
Income is clearly a factor in defining where highest machine density is found, but other
potentially significant relationships suggest themselves, including access, local
economic and leisure offer/diversity and the presence of key age groups.
Fruitful routes for deeper investigation were identified, including testing potential
circularity in results through regression analysis. There is also a need for further
validation of assumptions and more detailed case studies to reveal the differential
demand for machines among resident and visiting populations and the influence and
availability of alternative forms of entertainment on machine density. Finally,
behavioural research should take these findings forward to understand whether and
how machine density impacts propensity for gambling and/or gambling-related harm.
1. Research objectives
The decision to commission a series of research projects on terrestrial machine
gambling builds upon the publication of the Responsible Gambling Strategy Board’s
(RGSB’s) strategy paper in autumn 2009 and extends an initial programme of
exploratory work led by the Gambling Commission.
In its 2009 paper, the RGSB recommended that the Responsible Gambling Fund (RGF)
should support research focussed on gambling activities where there was a perceived
risk of gambling-related harm, such as high-prize and high-stakes gambling machines. In
addition, the RGSB clearly stated that much of their research agenda is driven by a focus
on the Gambling Commission’s third licensing objective: to protect children and
vulnerable people from harm.
To this end, the RGF has designed a programme of research to investigate issues relating
to terrestrial machine gambling, including the collation of empirical evidence about the
geographic location of gambling machines and the correlation of these data1 with
jurisdictional, regional, sub-regional and local socio-economic and demographic
1.2 Aims and desired outcomes
Understanding the micro and macro environments in which gambling is available is a
key component of any public health-based approach to gambling research, since the
geographical and social environments in which machines are located may interact with
individual gambling behaviour.
To be able to examine these interactions, it is important to understand objectively what
relationships, if any, are evident between the physical location of gambling machines
and the socio-demographic and economic environment in which they are situated.
Ultimately it may be valuable to examine whether any patterns are evident which might
suggest that machines are more or less clustered in areas where there may be a greater
propensity for gambling-related harm, as correlated with other relevant datasets, such
as the British Gambling Prevalence Survey (BGPS). To do this, we first need reliable and
systematic baseline information on which to ground further study. This is the primary
objective of this study.
To our knowledge, no study has previously attempted to map the location of gambling
machines in Britain either in aggregate, by venue, or by machine type. This study
therefore sets out to provide important empirical evidence in this area and allow for key
questions about the relationship between machine location, gambling access and
1 Note that throughout this report, the term ‘data’ is used in its grammatically correct form i.e. as the plural of
availability and broader local environments to be explored.
The study area was defined as Great Britain, not including Northern Ireland, conforming
to The Gambling Commission’s area of jurisdiction (with the exception of some small
and detailed areas of legislative control that apply across the UK).
The focus of this study was the physical location of terrestrial gambling machines and
not an examination of provision of machine-style games in online environments
(provided by sites including www.jackpotjoy.com, www.slotmine.com or
www.888games.com). At the outset it was acknowledged that this study could not in
any meaningful way take online gambling into account.
The streetscape aspect is relevant to recent British media coverage of gambling.2
Bookmakers’ shops have been perceived as increasing in a number in town centres,
especially where other retailers have ceased trading and left premises vacant, a feature
of the wider recent economic downturn. In areas such as London’s Chinatown and the
horseracing centre of Newmarket, local opposition has sought to preserve the trading
character of the high street and to voice concerns over lower-income and vulnerable
groups having increased access to gambling opportunities overall, not exclusively
The ability to interpret these stories in light of accurate data is another objective of this
1.3 Supplier credentials
RGF/RGSB commissioned a partnership of two independent suppliers to undertake this
baseline research, who between them have a strong track record of expertise in the
different aspects of the task.
Geofutures Ltd is a consultancy specialising in advanced spatial data analysis and
mapping. It was established in 2002 as a spin-out from UCL, among the pioneering
centres of geographic information science (GIS), and the company maintains academic
rigour in its methodologies while applying these to commercial needs.
Geofutures has much experience of undertaking complex research and data modelling
for government, corporate, and third-sector clients including the Department for
Communities and Local Government, the Office for National Statistics, Ipsos MORI, The
Local Data Company, Regen SW, and the Transition Network.
The company was commissioned to undertake the gathering and processing of data, the
mapping and spatial statistical analysis of gambling machine venue density and the
social and economic characteristics of the locations where the highest densities were
The National Centre for Social Research (NatCen) is an independent research institute
specialising in applied social research for policy purposes, with a commitment to
methodological rigour. NatCen conducted all three of the British Gambling Prevalence
2 E.g. ‘Give people the freedom to curb high-street gambling’ The Guardian
, 22.02.11; ‘Calls for government to limit ‘clustering’
of betting shops’ Racing Post
Survey studies and has broad ranging substantive expertise in the field of gambling
studies. The partnership of NatCen and Geofutures for this project therefore combined
substantive expertise in gambling research with methodological expertise in GIS,
underpinned by a commitment to rigour and transparency in research.
1.4 Understanding spatial relationships
At the outset, RGF/RGSB identified the significance of mapping data in revealing
important differences across space. Maps instantly reveal spatial differentiation, even
down to fine spatial scales, which together with our understanding of the differences
between locations – their physical geography, their accessibility or the social and
economic characteristics of their populations – can reveal previously unknown insights
into phenomena. With additional analysis, data correlations observed for some studies
may suggest causal factors.
The findings of this study are partly based upon machine density analysis of gambling
locations. At a more basic level, the visual interpretation of spatial patterns on the maps
provided also reveals aspects which were not previously known, such as the proportion
of the highest-density machine clusters which are in seaside towns relative to other
locations or a broad pattern by which satellite town and New Towns have higher density
of gambling machines relative to other areas, including inner-city areas. To interpret the
findings appropriately, it is important to understand some principles of spatial data
By giving each data point a spatial location on a map (a map co-
ordinate, or often an estimate of one based on postcode) we can
create a simple visualisation of a dataset across space.
As map (a) shows, together with physical features such as roads
and administrative boundaries, we can start to see the
distribution of points and identify high densities and patterns.
We then need to deal with the issue that data points are often
too concentrated to see detailed patterns clearly.
Map a: property point data
is visualised for Stoke-on-
Trent. The clusters are
clear, but where we have
the most data points we
can no longer discern fine-
Using recognised statistical methods which ascribe values to
locations relative to their proximity to known point data, we can
create continuous data ‘surfaces’ illustrating data values and
densities in a way we can easily see and understand.
These techniques also overcome another significant problem:
mis-representation of data in zonal maps.
In the example of map (b), a zonal map shows point data which
has been summarised according to the administrative boundary
Map b: data visualised
in which it falls, and the whole zone is given a single average
according to administrative
In map (c), the exact same point data is shown as a data ‘density
surface’, revealing hotspots and the true distribution of data
values across the whole area.
The zonal map obscures significant patterns that cross
boundaries, and averages out variations within each zone. This
means that for a given point on the map, we can only ascribe a
data value which is the average for a zone which is considerably
larger and which has an arbitrary shape. The hotspot map
reveals differential values at a much finer scale, based more
closely on the actual locations of point data, but it is still very
Map c: the same data
easy to understand.
visualised as surface
Several related methods are used in this study, as set out in the methodology below. The
data point locations of gambling machines are modelled to show spatial density surfaces,
which, in turn, are used to develop an appropriate method for identifying those locations
which can be statistically identified as ‘high’ density.
The zones around gambling machine locations are also identified simply as geo-located
points with a spatial ‘buffer’ around them, creating boundaries for these areas. Their
economic and social characteristics are then identified and aggregated based on the
resident populations living within and in close proximity to them.
Some social and economic data lend themselves to being spatially interpolated as
surfaces, and perceiving patterns in data such as income scores is far easier when
visualised this way. Other datasets used in this study categorise data into groups which
are not so readily visually interpreted. For example, ‘occupation group’ assigns a
percentage of population into one of nine categories; these multi-dimensional datasets
can technically be re-interpreted as surfaces but the time investment involved has to be
weighed against the ultimate accuracy of interpretation, and the data can still be
analysed without fine-scale data mapping and included in the analysis.
Chi-square tests for statistical significance are then applied to the findings to measure the
likelihood that patterns, where identified, are significant and could not have arisen
In all statistical analyses, spatial and otherwise, methods will by definition generalise and
allow for margins of error. Similarly all studies of this kind are limited by the accuracy and
availability of data. We aim in this report to be as open and objective as possible where
these limitations occur whilst making use of the best available information and methods
to derive results that can be safely relied upon, and clarifying where results may be
relatively less sound.
2. Research evidence
We conducted a Rapid Evidence Assessment to assess (a) what evidence currently exists
about the relationship between gambling machines and area characteristics, and (b)
what methods have been used to consider issues relating to the geo-spatial distribution
of gambling machines. The Rapid Evidence Assessment is a compressed and delineated
version of a Systematic Review, recommended by the Government Social Research
Department, which uses the same principles to identify, review and evaluate evidence.
However, it is conducted in a shorter time frame and accordingly discusses fewer
publications in less detail. Our approach was built on the following principles:
The identification of clear Our assessment focused on finding what evidence exists
questions that the review
seeks to answer.
a) the geo-spatial distribution of gambling machines
b) the impact of this distribution; and
c) the broader availability of gambling machines.
Our aim was to identify the central themes discussed in the
research literature relating to these questions and to
examine the methods used.
The specification of
We used various publicly-available databases to search for
search terms and
both peer-reviewed and ‘grey literature’ in this area.
Specifically, we used the London School of Economics’
cross-searcher database which searches over 100 different
databases, including PsycMed, PubInfo, Web of Science,
International Bibliography of Social Sciences, media reports,
communications, various unpublished reports and theses.
We used combinations of the following search terms to
identify literature: gambling; machines; access/accessibility;
availability; geography; GIS and/or density listed in the title
or abstract. All returned articles were shortlisted based on
(likely) relevance to the research questions; 21 articles and
book chapters were shortlisted for review.
Each article was summarised based on the abstract
information and assigned to an article type (empirical,
discussion, meta-analysis) and a broad theme (e.g. machine
location and area characteristics; relationship between
density and behaviour; other accessibility features). Based
on this process, 15 articles were then shortlisted for fuller
The basis for the
Empirical studies were assessed based on accepted
assessment of study
methodological standards and the transparency evidenced
by authors in presenting results and discussing limitations.
The basis for synthesising
In drawing conclusions, we have given more weight to
evidence from empirical studies with strong methodology.
2.2 Summary of relevant studies
Research evidence can be articles and studies that focus on the relationship
broadly categorised into
between machine density and area characteristics
two key areas:
articles and studies that focus on the relationship
between gambling and/or machine availability and
A number of studies in Canada, Australia and New Zealand have analysed the
relationship between machine location and area characteristics. The literature relating
to each of these themes is summarised in the following sections.
2.2.2 Machine density and area characteristics – evidence from Australia and New Zealand
Studies by McMillen and Doran (2006), Stubbs and Storer (2003) and Marshall and Baker
(2001) demonstrated that in certain Australian states, areas of greater deprivation or
lower resources had a greater density of Electronic Gambling Machines (EGMs). In
Victoria, this geo-spatial pattern led to the implementation of a cap on the number of
EGMs in 19 disadvantaged areas (Productivity Commission, 2010).
However, both McMillen and Doran (2006) and Marshall and Baker (2001) also
highlighted that the broad association between machine density and disadvantaged
areas needed further and more localised consideration. Marshall and Baker (2001)
focused their research on the metropolitan areas of Melbourne and concluded that
patterns observed at a national level were also observed at a more local level, whereby
more disadvantaged neighbourhoods, specifically those with lower economic resources,
were more likely to have greater numbers of EGMs.
McMillan and Doran (2006) compared three localities in Victoria, Australia. They
concluded that the association between machine density and disadvantaged areas
varied between them and that simply comparing machine density per 1000 people with
indices of deprivation (the broad methodology used to determine which areas would be
subject to EGM caps) was insufficient to fully explain spatial patterns at a local level, and
that broader domains of accessibility should be considered. They suggested that these
domains should include types and combinations of gambling machines,
technological innovation, the proximity of venues to community facilities,
consumer preferences, venue marketing strategies, convenient travel routes and
parking facilities, and other externalities (e.g. localised pockets of affluence and
disadvantage, changing urban and economic conditions, policy impacts, etc.)
In a further article, Young, Lamb and Doran (2009) argued that the geo-spatial
distribution of machines in relatively remote urban centres of Australia were the
product of various supply-side mechanisms. For example, they noted that pre-existing
spatial infrastructure largely determined the location of machines. This included
features such as the pre-existence of certain venues where machines could be housed,
proximity to central business areas (or other entertainment areas), transport corridors,
or community congregations.
They argued that these infrastructure aspects determined the basic spatial structure of
EGM supply in these regions. This provides further support for taking a more localised
perspective when assessing the geo-spatial distribution of gambling machines, and the
need to consider both supply side and demand side factors.
2.2.3 Geo-spatial distribution of machines – evidence from Canada
A body of research into the geo-spatial distribution of machines has also been
conducted in Canada. Robitaille and Herjean (2008) used GIS techniques to analyse the
location of venues with a license to operate Video Lottery Terminals (VLTs) in Montreal.
They used network analysis to calculate the time taken to walk to the nearest venue
with a VLT. Based on this, the authors concluded that there was a strong relationship
between VLT access and area vulnerabilities (defined using a composite index similar to
Great Britain’s Index of Multiple Deprivation), noting that the spatial distribution of
neighbourhood vulnerability was closely aligned to areas of greater accessibility of VLT
However, Robitaille and Herjean also noted that access to VLTs was concentrated in
central and pericentral districts of Montreal, and aligned along the major thoroughfares
of the city. This supports the importance of considering spatial infrastructure alongside
more standard measures of machine access (such as number of machines per 1000
people or distance to access).
Gilliand and Ross (2005) also noted the relationship between machine location and
regional vulnerabilities. In addition to investigating the density and prevalence of
machines, they also looked at adoption rates (the proportion of venues eligible to have
VLTs that actually held a license for these machines). They found a positive association
between VLT adoption and borough distress (an index which takes into account
unemployment, low educational attainment and lone parenthood); VLT adoption rates
(as defined above) increased as borough distress increased.
However, Gilliand and Ross also noted that other infrastructure reasons may affect the
distribution of VLTs, such as land zoning regulation, and variations in municipal liquor
licensing laws (as VLTs licenses are limited to venues with liquor licenses).
Finally, Wilson et al (2006) performed a similar analysis but instead focused on the
proximity of VLTs to secondary schools. This study used a number of measures to
examine the geo-spatial pattern of VLT distribution. These included measures of (a)
concentration (number of machines within a given radius) and (b) an accessibility index
(comprising the proportion of eligible venues with VLTs multiplied by distance from the
The results demonstrated that there were greater opportunities for VLT gambling in
economically-disadvantaged school areas and that the distribution of, and access to,
VLTs surrounding Montreal’s schools reflected local geographies of socio-economic
disadvantage. The authors did not specifically note any other spatial structural factors
that contributed to this observed relationship. However, they did state that other
normative and temporal factors may affect accessibility, such as parental influence and
supervision outside of school hours (social and personal accessibility).3
2.2.4 Relationship between machine availability and gambling behaviour
A number of studies have considered the relationship between machine availability and
gambling behaviour. Before considering the specifics of these studies, it is important to
outline the basic theories that underpin such research.
Broadly speaking, this
a) what is the nature of the relationship between exposure
body of research is driven to gambling and gambling behaviour?
by two main
b) what processes or mechanisms may propagate or
mediate this relationship?
Many researchers have argued that increased exposure to gambling opportunities will
lead to increases in gambling participation, expenditure and, ultimately, the experience
of gambling-related harm (e.g., Orford 2011, Cox, 1997). However, some scholars have
questioned this assumption, arguing that the shape and nature of the relationship
between exposure and behaviour is more complex than previously assumed and that
the relationship may be non-linear, as a range of adaptive processes by individuals and
the broader community may mitigate this relationship (e.g. Shaffer, LaBrie & LaPlante
(2004; LaPlante & Shaffer, 2007). The focus of this study, aimed at understanding the
geo-spatial distribution of gambling machines in Great Britain, speaks exactly to this
In this project, we are firstly focusing on understanding the core patterns of this
distribution. In the longer term, questions about how (any) differential exposure is
related to behaviour may be considered. Although not directly related to the core
objectives of this study, we have summarised some of the main themes emerging from
the literature below. In particular, these authors have raised some important issues to
be borne in mind when considering the relationship between machine density and
behaviour, such as the need to consider both accessibility and availability.
A meta-analysis conducted by Storer, Abbott and Stubbs (2009) concluded that the
prevalence of problem gambling increased as machine density increased, with little
evidence of a plateau with increasing density. This supports exposure theory. However,
they also found that problem gambling prevalence decreased over time which they
argued also provide some support for adaptation theory.
Of particular interest was the observed pattern whereby low-density EGM areas had the
greatest variance in problem gambling rates compared with high density EGM areas. In
3 The 2010 Australian Productivity Commission Inquiry into Gambling refers to research that suggests that
‘geographic and temporal aspects of accessibility’ are significantly and positively related to severity of gambling
behaviour whereas ‘social and personal aspects of accessibility’ are at best only weakly related (Moore et al. 2008;
Thomas et al. 2009 quoted in PC, 2010)
short, the relationship in low-density EGM areas was not uniform. The authors argue
that geographic clustering of venues in low density EGM areas could be driving this
association and that such clusters could have differential impact. They also note that the
type of venue in which EGMs are situated may play a role in mediating the risk of harm.
This detailed study, therefore, provides support for considering local community
contexts, spatial infrastructure, differential regulatory environments, and other
accessibility issues at a more fine-grained level when looking at the relationship
between gambling access and gambling behaviour.
Further studies have demonstrated a link between machine density and gambling
expenditure. The evidence for Australia has been summarised in the most recent
Australian Productivity Commission Report on Gambling (2010).
Among others, this report cites work by Delfabbro (2003), who found a strong
correlation between gambling machine density and net revenue, and also Stubbs and
Storer (2007), whose analysis showed that EGM density accounted for 77% of the
variation in gambling expenditure per adult. However, the Productivity Commission
report does not include Delfabbro’s 2008 article, which demonstrated that the
reduction in gambling machines in South Australia did not lead to a commensurate
reduction in gambling revenue or behavioural change among pre-existing patrons.
Delfabbro outlined various measures undertaken by industry to circumvent the
potential impact, again highlighting the need to consider this issue in a broader context.
Finally, most studies reviewed noted that machine density is only one aspect of
exposure. Based on the studies considered, exposure to gambling appears to comprise
two interlinked factors: availability and accessibility.
Measures of availability used in the studies reviewed included proximity to the nearest
venue, either using GIS analysis or self-report; density of machines per n
machine adoption rates (i.e., of those venues eligible, what proportion contain
machines) and venue/machine clustering at low level geographies.
All these aspects can potentially influence the availability of machines at different local
levels. Accessibility is generally considered to comprise broader aspects that may affect
behaviour, such as opening hours of venues, conditions of entry, ease of use, venue
type, layout and general environment, convenience and the spatial distribution within a
local area (such as whether the venue is close to transport links, other businesses etc).
All of these features may interact with broader machine availability to mediate the
relationship between machine exposure and gambling behaviour. Whilst many
researchers note this possibility, to our knowledge, there is no research evidence that
assesses both availability and accessibility issues, and interactions between the two, in a
This research gap is likely assessing both issues would require very specific local
to be evident because:
level analysis, whereas much of the research conducted
in this area to date focuses on larger, aggregate,
geographic areas combined with survey data; and
untangling the impact of these different aspects in an
empirically robust way is particularly challenging.
2.3 Key findings from Rapid Evidence Assessment
Research evidence from Canada, Australia, and New Zealand showed that the
geo-spatial distribution of gambling machines in these countries is associated
with underlying area vulnerabilities, such as borough distress or deprivation.
Localised studies in these countries have also demonstrated the importance of
considering both supply-side and demand-side factors in understanding this
distribution, such as consideration of local infrastructure and land zoning/uses.
No evidence base currently exists in Great Britain relating to this and thus
debates about relative exposure to gambling machines are not based on
Exposure to gambling contains two component parts: availability (measured by
metrics such as density and distribution of opportunities) and accessibility
(measured by metrics such as ease of access, venue environment and so on)
The relative contribution of availability and/or accessibility to levels of gambling
‘exposure’ levels in Great Britain is unknown.
2.4 Interpretation and application of evidence to Great Britain
As demonstrated above, a body of work, although in its relative infancy, has been
developed in Canada and Australia which examines the geo-spatial distribution of
gambling machines, and considers the impact of this distribution upon behaviour. This
has led to certain advances in thinking, such as the importance of looking at machine
distributions at a local level in order to fully consider how supply-side factors (such as
infrastructure and zoning) may influence this distribution. However, in Britain, no such
body of work exists and we do not have any empirical data that tell us:
a) how the location and density of gambling machines varies across Britain
b) if clusters of high-density machine areas exist and, if so,
c) what the underlying associations are with broader socio-demographic and
economic area characteristics.
This basic understanding is needed before more nuanced perspectives can be applied.
This report aims to supply this missing evidence, and to develop a basis upon which
more localised research can be undertaken.
3.1 Available machine venue location data
3.1.1 Regulated gambling premises
The Gambling Commission (GC) regulates licensed gambling premises, with the
exception of pubs and bars. The Commission maintains a database of premises that have
a licence for gambling machines, based on statutory returns from every local authority
(LA), providing address details to the unit postcode4, status of licence (e.g. granted,
pending application, refused etc), and licence type (Adult Gaming centre [AGC], Family
Entertainment Centre [FEC] etc). This database was made available for this study in
Any such database is by definition a snapshot in time and will be subject to inaccuracies.
The database includes some missing data on licence status where LAs have not updated
the GC when pending applications are granted; lapsed licences are also updated
periodically. There will also be some typographic errors in the recording of some
licences, including errors in the recorded postcode, which is particularly pertinent to this
Subsets of the data based on the status of the application were created for inputs for
analysis. Given the known update pattern for the data, it was decided to include all
licences with the status of 'granted' or 'application'. There may exist, in a small number
of cases, more than one application for one potential site, (e.g. casino locations with
more than one company submitting an application) but the effect of this on results will
be minimal and resultant analysis assumes this level of accuracy. Records with no status
stated, or those refused licences etc have been omitted, giving us a better indication of
what existed 'on the ground' at the time the data were supplied.
With over 14,000 premises listed, the database was judged to have sufficient size to
provide a robust basis for venue location, even allowing for the flow of new and closing
3.1.2 Pubs, bars and restaurants
Pubs, bars and restaurants (given the collective term ‘pubs’ below for brevity) are
automatically allowed a limited number of gambling machines within the terms of their
alcohol licence, and can apply for a licence for additional machines. In both cases the LA
is the licensing body and no national body regulates these venues in equivalent terms.
We therefore recommended to the RGF that location data for this kind of premise
4 The UK Royal Mail defines 124 Postcode Areas by the first one or two letters in the standard postcode format,
divided into approx 2900 Postcode Districts by the number that immediately follows them. In the second half of
the postcode the district is divided into Postcode Sectors (of which there are approx 9,650) by the number, and
then into the Unit Postcode by the final two letters. There are approx 1.71m unit postcodes in the UK, each on
average representing 15 delivery points.
should be obtained from a commercial source.
There is no equivalent regulatory dataset of all pubs, let alone one which identifies the
presence or otherwise of gambling machines. We therefore recommended the use of a
publication-based list of approx 30,000 premises from which we were able to identify
16,000 venues where the publican had indicated the presence of one or more category C
or D gambling machines. This is a smaller population than the total 50-60,000 licensed
premises estimated by the British Beer and Pub Association, but one which we can feel
confident includes only relevant locations i.e. will not skew results with additional pub
locations where we do not know there are gambling machines present. Overall the
contribution of pubs to machine density is relatively low and evenly distributed, which
gives us greater confidence in the results of using these data.
These data were supplied by publisher William Reed5, including address details and unit
This dataset was merged with the GC data as a final input points dataset, defining the
location and attributes of the selected pubs for further analysis.
3.1.3 Gap analysis for pubs
To ensure that we could safely assume the pub venue datasets were sufficiently accurate
to provide a basis for research, we conducted a comparison analysis of the pubs location
data against relevant sector data from the Office for National Statistics’ Annual Business
Inquiry (ABI). The methodology used is outlined in section 3.2.4 below.
3.2 Machine location data collection
3.2.1 Data availability issues
In accordance with its aims and objectives as set out by RGF, this study deliberately
steered away from any attempt to ascribe behaviours to gamblers, aiming only to
provide a systematic view of where gambling machines were located.
However, in making the decision to seek data on number of machines in order to
calculate total machine density, rather than relying simply on venue distribution alone,
the view was taken that access to gambling machines in one type of venue is not
objectively the same as another. Access to several hundred gambling machines among a
series of neighbouring adult gambling centres cannot be judged equivalent to two or
three machines each across the same number of pubs. Therefore the number of
machines per venue is relevant when seeking to obtain an accurate view of gambling
In seeking ‘machine numbers per venue’ data, a number of key issues presented
1) LAs request the number of machines in the venue in a gambling licence application.
5 The UK Pubs & Licensees Database supplied by William Reed Business Media Ltd, based on the requested
circulation of the publicans’ publications The Morning Advertiser, Guide to Pubs, Bars and Nightclubs
and On Trade
They are not required to include these figures in statutory returns to the GC, however,
and research among LAs found none which captured this information in any way other
than by internal filing of paper copies of applications. Within the time and cost resources
available, it was not judged practical to examine data held by individual LAs in this
2) Some types of gambling venue are required to submit machine numbers information
to GC, by operator rather than by individual licensed premise. Where the GC holds this,
however, they were unable to release it for the purposes of this study, even in an
anonymised form which would allow us to ascribe an average number of machines per
venue across an operator’s estate.
We were advised by GC
The GC is prevented from providing any alternative
that the key reasons for
figures to those provided in official publications under the
this prohibition fall into
Statistics Act, and were not in a position to provide
historical data as an alternative since the database is
maintained as a current live snapshot of the estate.
Premises data and machines data by operator (which may
represent one or many premises) are provided to the
Commission as part of operators’ regulatory returns. The
Commission therefore considers that this information is
provided in confidence and should not be further
Gambling operators and GC experts agreed that within certain types of venue, such as
bookmakers, broad uniformity of operation would mean that machine numbers per
venue would be highly predictable relative to the legal maximum allowed within each
licence type, while in others this would not be so. Bookmakers are dominated by a small
number of chains including William Hill, Ladbrokes, Coral and others, and in most cases
they enumerate their machine operation openly in annual reporting.
Adult gambling centres rapidly emerged as the venue sector where accurate
generalisations would be least reliable. Not only do they vary considerably in size, but
multiple licences per address exist for a considerable number of AGC premises on the GC
database (see 3.2.4 below), and some venues are known to have additional machines
but only have their current legal limit of 20 category B machines switched on in rotation
at any one time.
In effect, the unavailability of comprehensive machine numbers data by venue makes it
necessary to create a proxy for these data, in the form of a weighted venue density. The
method by which this was done is outlined in 3.5 below.
The accuracy of this density surface depends upon the accuracy of the machine numbers
estimates by venue type, and the methods used to reach the best possible estimates are
outlined below (from 3.2.3). Provided that these estimates achieve a level of accuracy
which is sufficient for our purposes, a weighted venue density is fit for purpose.
The ‘weighting’ of a geocoded address point is calculated by the number of machines a
venue will have, so a series of small venues close together will have an equivalent
density to a single venue with a higher assumed number of machines.
While actual machine numbers per venue, if available, would remove the margin of error
that assumed machine numbers must create, an actual machine density analysis would
still reveal hotspots which are the result of high numbers of machines in close proximity,
whether they are found in one venue or many.
3.2.2 Machine categories
Just as not all gambling machine venues are objectively the same, nor are categories of
gambling machine. Without steering into behavioural waters, a potential gambler may
not view a fixed-odds betting terminal (Category B2) as equivalent to a category C
machine in a pub (see machine category summary below).
Given the challenges encountered in relation to accessing full information on numbers
for all machines (i.e. as an aggregated figure), let alone by an individual category, a
decision was made to set this difference to one side in order to undertake the analysis
within the scope of time and budget allowed. All machines are treated as ‘equal’ in this
analysis, and research into the relative densities of different machine types remains an
area for future study.
For ease of reference, the Gambling Commission’s published industry statistics on the
types and numbers of machines by category for 2009-10 are reproduced below.
Maximum stakes and prizes
Previously classed as
as of 31
Fixed odds betting
Amusement with prizes
Amusement with prizes
10p (cash prize
* Figures taken from regulatory returns. As The Gambling Commission does not regulate pubs, clubs,
working men’s clubs or FECs without adult areas, data from those sectors are not included in this table.
** This summarises information for different types of category D machines. See the Gambling
Commission website for exact details.
3.2.3 Direct data collection
Data collection for machines numbers therefore took a multi-pronged approach,
comparing industry estimates for machines per venue type from the GC and trade
bodies such as the British Amusement Catering Trade Association (BACTA) with primary
data collected from operators (see below).
To obtain reliable data on AGCs, the least homogeneous group of venues, a focused
fieldwork exercise was carried out by NatCen to visit premises and perform a simple
machine count (see section 3.3).
A number of gambling operators and machine suppliers were approached by the GC and
agreed to confirm their machine numbers to us for the purposes of this study. In some
cases data were offered per venue, and in others an average number of machines per
venue was supplied. Data were gathered in this way for a self-selected sample of bingo
operators, bookmakers, casinos and AGCs, including motorway service areas, and
compared with published industry estimates.
Major chains are also well-represented among pubs and bars in Great Britain. A small-
scale telephone exercise was carried out to request average or actual machine numbers
per outlet among the major chains, which yielded some results.
The aim was to derive a robustly-based average number of machines per venue type to
include in a density model. While industry estimates exist, it was necessary to obtain
some additional data verification to ensure that the results of the study were as
accurate and meaningful as the data available could possibly allow. The numbers for
bingo clubs, AGCs and casinos are likely to be under-estimates since their licensing does
not limit the number of category C and D machines.
Clearly, ascribing an average number of machines per venue is significantly less reliable
than using an actual number, but these more accurate data were not available for most
gambling sectors. In the absence of such information, the process described above
aimed to maximise the robustness of the machine numbers data included in our
The final figures used are detailed in the following table:
6 Based on validation sub-study results detailed below.
3.2.4 Data validation
A two-pronged data validation exercise was undertaken. The first was a gap analysis,
which was carried out to determine whether the William Reed data, which was known
to be a subset of all pubs, provided a representative dispersed spatial sample, displaying
spatial heterogeneity. By comparing the number of point locations of businesses in the
dataset to the spatially aggregated locations in the ABI, which are based on VAT returns
and which list businesses by standard industrial classification (SIC, the code which
identifies the precise industry sector to which they belong), we can identify any obvious
locations where pub data may be missing, and by how much.
The ABI 2008 workplace analysis (data units) was used. Data from the ONS Business
Register and Employment Survey (BRES) 2009 are more up to date, but workplace data
are no longer available at the fine spatial scale required. The chosen data include
SIC2007 4-digit codes, and the business type used was 5630 'Beverage serving activities'.
We implemented a spatial join query of counts of point locations of premises from the
Reed list within Census 2001 lower super output areas (LSOAs)7, representing 'data
units' used by the Annual Business Inquiry (ABI).
Some testing errors are assumed, since the SIC category may include non-relevant
businesses. A degree of spatial error may also be introduced since ABI data are supplied
by individual building location and then aggregated to LSOA, while the Reed data are
geocoded to the centroid8 of the unit postcode, which may fall within another LSOA, yet
these errors are considered to be minimal.
ABI data have to be suppressed since individual company records may enable
identification of a single company’s data, and breach commercial confidentiality, hence
we do not reproduce them here. However, the results of the gap analysis did not
indicate any skew in the data: LSOAs were identified where more SIC 5630 businesses
exist than the Reed data indicated, as expected, but no cluster or other spatial pattern
The second validation exercise related to AGCs and split licences (those AGCs where
more than one licence exists for an address).
Where multiple licences exist, these were included in the analysis, since it is assumed
that an additional licence is only present in order to increase the number of machines
available, and hence the concentration of machines increases for that location.
Of the 13,897 records on the GC database with valid postcodes, 3,450 were AGCs. An
7 ONS aggregates Census output areas (the finest scale of neighbourhood aggregation, typically 125 households or
300 residents) into a series of larger units. The next largest is lower-super output area (LSOA) which typically
represents 1,500 residents.
8 In this case, a centroid is the geographic point whose co-ordinates are the mean values of the co-ordinates within
the chosen zone. Centroids can in other cases be population-weighted, representing the ‘centre of mass’ of the
population values and locations within the zone.
initial data query found 503 addresses with more than one licence, based on an exact
match between premises street number, street name and postcode. This represents
14.6% of listed AGC premises.
Additional analysis e.g. running processes which would match addresses where details
are organised differently9 or with different spellings might find additional duplicates, but
as background insight this level of validation was sufficient to confirm that we should
use the number of licences at an address in the analysis, rather than the number of
3.3 Field validation for Adult Gaming Centres
As discussed in section 3.2, we were able to estimate the number of machines per
venue type for the majority of different gambling venues based on discussions with the
GC and various operators. However, with respect to Adult Gaming Centres (AGCs)
available data were much more limited. In particular, AGC venues vary considerably in
size along with the number of machines licensed at each premise.
In order to test our assumptions about the average number of machines per AGC, a field
validation study was conducted between May and June 2011. The purpose of this study
was to visit a number of different types of AGCs and to count the number of machines
located at each premise. To achieve this, we selected a sample of AGCs with different
characteristics and in varying geographic locations to ensure that our validation exercise
reflected the diversity which exists within the market. This was achieved by review of
the AGC venue database and by grouping AGC into sub-types. The sub-types developed
Urban 1 areas - these are AGCs located in the top ten urban areas in Great
Britain (as defined by the Office of National Statistics; Pointer, 2005).
Urban 2 areas – these are AGCs located in the top 25 (but not top ten) urban
areas in Great Britain.
Seaside areas – these are AGCs located in major seaside resorts in the UK; there
is some overlap with between seaside and urban areas – for example
Portsmouth is both a seaside and Urban 2 area.
Large towns – these are AGCs located in towns with a population greater than
20,000 people but are not within one the top 25 urban areas in Great Britain.
Small towns – these are AGCs located in towns with a population less that
20,000 people and which are not within one the top 25 urban areas.
Motorway and other travel location AGCs – this includes AGCs located in
motorway service stations, railway stations and airports.
The table below shows how many AGCs per category were selected for inclusion in the
validation study. Preliminary investigation of the venue database showed that AGCs
9 E.g. matching ‘Address field 1= 22-23 / Address field 2= High Street’ with another record ‘Address field 1=22-23
High Street / Address field 2=*no data+’
tend to be predominately located in urban areas. Therefore, slightly more AGCs in urban
areas were selected for inclusion. However, we also sought to gain a good balance
between area types and over 30% of the sample was located outside of the top 25
urban areas in GB (large towns, small towns and motorway areas).
NatCen interviewers were used to conduct the field validation. They were given detailed
project instructions and were asked to record full details of their attempt to count the
number of machines in each venue. Where interviewers were visiting seaside resorts,
they were given instructions to help them identify the AGC area within the venue and a
description of machine features to help them identify the correct machines to count.
3.4 Field validation results
Interviewers visited 65 venues in total. The main reasons that interviewers were unable
to visit some targeted venues were that the AGC either no longer existed at that address
or had never existed at that address (though the former was more common).
The table below summarises the results from productive venues.
Urban 2 Seaside
Overall, the average number of machines at each venue visited was 38.2. However, the
standard deviation was +/- 26.6, meaning that there is a great deal of variation around
this figure. This is largely driven by two observed facts: ( a) the size of seaside venues
and the number of machines they contain varies considerably, and (b) venues in
motorway service stations and other transport facilities only have a very small number
of machines, around 8-9. The median number of machines per venue was 34.
With the exception of seaside AGCs, the number of machines per venue was relatively
normally distributed, despite the large variance in observed numbers of machines per
venue, and therefore we have used the average number of machines (38.2) from the
field validation study in our calculations. The proportion of seaside AGCs in our achieved
sample is slightly lower than the proportion of seaside AGCs in Britain, meaning that this
‘average’ number may be an underestimate. In the premise register provided by the GC,
just under 30% of AGCs were in seaside locations. Our field validation study included a
slightly lower proportion (26%). This should be borne in mind when reviewing results.
3.5 Spatial analysis of venue and machine densities
The GC premises database was geocoded firstly to the unit postcode where possible,
and secondly to the street centroid where an exact match on the street and town name
exists. A record was generated of the positional accuracy of premise licence, and this
point file was joined with the Reed data pub locations, which was also geocoded to the
The unit postcode for the majority of records and street centroid for a minority of
records is therefore assumed as the input level of accuracy for resultant spatial analysis.
Some limited work was done to improve the positional accuracy of incomplete and
inaccurate records by hand. 864 records in the original premises register could not be
located geographically and have therefore been omitted from the analysis. Any errors
residing in the premises register provided, such as incorrectly-recorded postcodes, are
The average number of machines per premise licence type (see 3.2) was assigned to
each premise by its type. The average number of machines by the spatial location of
each premise licence was then aggregated by the OA in which the premise is found. This
enabled us also to calculate the average number of machines per person based on 2001
population falling within that output area.
Weighted kernel density estimate (KDE) techniques were then used to produce surfaces
indicating areas with high densities of activity, using ArcGIS Spatial Analyst. For surfaces
at the GB scale, the GB boundary was used as the extent and mask for analysis.
Surfaces were produced to
all the relevant registered premises licence locations
examine the density of:
the individual licence types locations e.g. density of
the average number of machines per premise licence
the average number of machines per head of
To explore how absolute machine density is related to population density, we calculated
the average number of machines per person.
The average number of machines by premise licence location (as calculated previously)
was aggregated by each 2001 Census output area based on its spatial location. The
average number of machines per person was calculated by the average number of
machines divided by all people in the output area. The population-weighted centroid of
the output area was then used as the input to a density surface (see section 1.4),
weighted by the number of machines per person.
A note on population data
For all population data in this study we have used 2001 Census of Population Usual
Resident Population (All People Count) at Output Area for both England and Wales and
Scotland as the underlying population dataset. As noted above, an Output Area is the
finest resolution zone for Census data, normally equating to approx 125 addresses, and
we required data at this level of resolution for comparison with the social and economic
data in the analysis.
More current Mid-Year Population Estimates are available from ONS, but these are
aggregated to Middle-Super Output Areas (MSOAs), much larger zones with a mean
population of 7200. Income is a key economic statistic for our analysis, and it is not
possible to accurately correlate income figures available by Output Area (OA) with data
aggregated to such large spatial areas.
Further, the Mid-Year Estimates are based on modelled data which aggregate age bands
either into very broad categories by LSOA, or 5-year age intervals at MSOA. Neither
offers the ability to select age bands, which is desirable when providing age-analysis
data comparable to the age bands used in the British Gambling Prevalence Survey (see
These constraints were weighed against the lack of currency in Census population data;
ONS estimates that the total GB population has increased by over 2 million since the
2001 Census. On balance, it was judged more important to be able to characterise areas
surrounding high densities of gambling machines at the neighbourhood scale and by
chosen age band than to reflect this growth over the last 10 years.
Clearly as soon as 2011 Census population figures are released at OA level, there will be
an opportunity to revalidate those findings which are based on population counts.
3.6 Defining Machine Zones
There is no single accepted convention in the gambling research literature for the spatial
extent of an area which is affected by the location of machines. US and Canadian studies
include assumptions for gamblers driving 50 miles to a gambling destination, which does
not seem appropriate for more closely-packed UK town centres and resorts, although
some new, i.e., Gambling Act 2005-permitted, casino destinations may be shown to
have a reach of this size.
Turning instead to a broader retail/leisure model, UK Planning Policy Statement 6
(UKPPS6) defined ‘easy walking distance’ for an average town centre visitor between
amenities as 300 metres. This allows for the visitor having arrived in the town centre by
a non-ambulatory means of transport, but assumes pedestrian activity once the
individual has arrived.
Separately in our definition process, we wanted to include social and economic data,
which are limited in their resolution to Lower-Super Output Area (LSOA). To include this
in a nearest-neighbour analysis, by which data from adjoining or zones are included
within a count defined by another zone such as a radius boundary, we needed a buffer
zone radius of 400 metres.
Taken together with the UK PPS6 guideline, we judged that a boundary around any
licensed gambling venue of 400m radius was a reasonable definition10. These created
areas we have called Machine Zones.
Buffer zones of 400m radius were created around every gambling machine venue.
Where these were contiguous with the buffer surrounding a neighbouring venue,
individual buffer geometries were dissolved into a single polygonal feature. The total
number of machines for each zone were aggregated and joined as attributes to the
machine zone (MZ) polygons.
The total area for each MZ was calculated, in order to calculate the average number of
machines per hectare11. The average number of machines attributed to each location
(see 3.2) was aggregated into the corresponding MZ and the sum of the average number
of machines per hectare calculated. This created our density estimate within the
assumed area of 400m proximity.
Individual areas with an average number of machines greater than 1 per hectare were
selected and exported as a sub-set dataset and termed High Density Machine Zones (see
3.6.3 Defining high density machine zones
We then needed to determine the best method for defining what constitutes ‘high
density’ and what spatial area around locations defined thus can be usefully said to be
affected by high machine density.
Since ‘high’ density encompasses both spatial distribution and relative density of one
10 There is no universal definition of distance used in the examination of gambling venue proximity. Wilson et al
(2006) used a 500 metre radius in their examination of the proximity of gambling venues to schools. Robitaille and
Herjean (2008) used a slightly different measurement, counting venues within a three minute walk from various
areas. They assumed a 6km per hour walking speed, essentially meaning that their threshold was a 300m radius.
Finally Welte et al (2006) in their American study used the threshold of 16km. Because Britain is not as
geographically dispersed as America and because of the likely location of machines, it therefore is sensible to use a
smaller radius in our analysis which represents a distance that people may be likely to walk.
11 A hectare is an area of 100m x 100m.
location to another, we needed to define a spatial density which was ‘high’ relative to
the total distribution.
By quantifying the number of machines per hectare for our Machine Zones and plotting
the result over a normal distribution, we found that one standard deviation from the
mean gave us a density of approximately one machine per hectare. As a working
assumption, we therefore determined that a value greater than one machine per
hectare robustly constitutes ‘high’ density.
3.7 Social and economic spatial data analysis
Once the Machine Zones (MZs), and within them the High Density Machine Zones,
(HDMZs) were identified, their characteristics could be measured according to a number
of key social and economic indicators.
Where the data allowed, we could then generate headline statistical values for income,
economic inactivity, occupation group, ethnicity and age for OAs in the relevant nation
as a whole against those for the MZs and HDMZs within that nation, to provide an
instant comparison of the average characteristics of each.
Social datasets were aggregated to these individual features/geometries by the
OA/LSOA population-weighted centroids falling within their boundaries. The data join
was carried out in ArcGIS based on the ID field.
This process was
the number of machines falling within the zone
implemented for each
average % of economically inactive population
MZ/HDMZ to enumerate:
average rank of IMD Income Domain
% population segmented into ethnic groups from the
total resident population
% in 10-year age bands from the total resident
% of population in occupation groups from the total
3.7.1 Known methodological issues
Those MZs/HDMZs zones with no OAs or LSOA centroids falling within them have been
omitted from the resulting machine zones dataset, and any further headline statistics
and statistical significance tests assume these areas are not included. Given the scale of
an OA (approx 125 households according to ONS) the great majority of the larger
merged areas will incorporate at least one of these centroids.
The input socio-economic datasets, particularly derived from census data, are often
non-comparable in terms of date and measurement and from different sources across
the GB study area. Statistical sources such as the Office for National Statistics (ONS) and
Scottish National Statistics (SNS), create different categories and groupings of themed
datasets, making them not directly comparable.
3.8 Correlating Machine Zones and social and economic data
In each case the differences between the 3 location types (MZs/HDMZs/all) were
subjected to a chi-squared test to ensure that every difference can be safely assumed to
be statistically significant and not the result of a random coincidence.
For England we took the Index of Multiple Deprivation 2007 average income domain
score at LSOA scale (the higher the score the more deprived the area). For Wales, an
income domain score for 2008 is equivalent but differently calculated, i.e. the numbers
look very different but indicate the same differences between LSOAs. Scotland does not
publish a score for income, only a ranking, which is not possible to average in a
A total count of population from the 2001 Census for each category is broken down into
the age groups shown for all of Britain. Following the British Gambling Prevalence
Survey, the population is broken down into groupings of 11-15 years, then 16-24, and
the equivalent 10-year groups until ‘over 75’.
2001 Census data describing ethnicity according to ONS categories were analysed for
England and Wales. Data are collected in categories that are similar but not exactly
comparable for Scotland so these were enumerated separately. The results are shown
side by side below but this data collection difference should be assumed.
The average rate of economic inactivity for the population within HDMZs, MZs and GB
was analysed as a whole. Economic inactivity is defined by ONS as “people without a job
who have not actively sought work in the last four weeks and/or are not available to
start work in the next two weeks”. Students and retired adults fall into this category as
well as working-age unemployed individuals, so this group is not synonymous with
unemployment or necessarily with low income.12
Data for all employed people aged 16-74 are broken down into nine categories,
separating them by seniority level, and analysed for each location type.
3.9 Statistical significance tests
12 This distinction is important. Results from the British Gambling Prevalence Survey 2010 showed that those in full
time education (comprising mainly youths and students) and those who were unemployed were both more likely
than other groups to play slot machines and used fixed odd betting terminals. Those who were retired were less
likely to participate in these activities. Estimates for past year slot machine use were 23% among the unemployed,
17% among those in full time education and 3% among those who were retired. Therefore, understanding the
composition of economic inactivity in different regions is important when considering the distribution of gambling
We needed to explore the relationship between machine density and the socio-
economic datasets listed in section 4.4, to establish if the higher recorded values in
HDMZs were statistically significant.
We undertook a series of chi-square tests on the 7,243 machine zones (at GB level) in
which we were able to aggregate socio-economic indicators. The chi-square test was
performed on variable counts within a 5 x 5 matrix, defined by quintiles, of zone density
against the various indicators. We approached the analysis in this way to stabilise the
chi-square test as far as possible. (The only exception to this was the analysis for
Scottish data where the total count of machines zones was smaller and thus the matrix
In all the tests (as defined by the tables in section 4.4) there was a significant correlation
between higher density zones and high indicator scores (where the correlation is
positive) and higher density scores and lower indicator scores (the correlation was
However, we should be mindful that while the statistical correlation is significant, the
practical correlation may be less sure. Not only was the sample size large, which would
tend to emphasize any discrepancy between observed and expected values in the 1st
and 5th quintiles of both distributions, but also we need to caution against the
identification of correlation as signifying cause and effect, not least since we have simply
analysed the indicators singly, and not together. (We would expect there to be a
number of interactions and correlations between the indictors, such as between income
levels and age group, for example, which would need to be considered).
To understand the causal factors fully would require more detailed analysis, using multi-
level analysis or Geographically Weighted Regression for example.
Whilst we cannot at this point give a strong indication of what might be driving the
correlation, the fact that we find a strong statistical relationship that supports other
research findings is significant nonetheless.
4.1 Venue density
All venues: GC-licensed venues, together with pubs having gambling machines, visualised as
a density per square km. Centres of population stand out, as expected, but so do regional
centres and seaside resorts, reflecting the geographic spread of all the leisure sectors
associated with gambling machines.
Density of AGC licensed venues per square
Density of FEC licenses per sq km. Red = up
km. Red = up to 5.12 per square km
to 5.12 per sq km
Relative to the distribution of all gambling machines premises, AGCs show much greater
concentration in major urban centres and secondary town centres, with less
representation in suburban locations. The strong coastal presence of AGCs is clear, but
this is even more pronounced in the distribution of FECs, for which the major urban peaks
Density of bookmakers venues per square
Density of licensed bingo venues per
km. Red = up to 5.12 per square km
square km. Red = up to 5.12 per square km
Bookmakers show strong urban clustering in their distribution, with secondary centres
also evident (also see zoom below). Bingo venues are among the most scattered
geographically, found in primary, secondary and smaller urban centres. They follow
centres of population too, with particular clustering around the post-industrial cities
surrounding the Peak District e.g. Sheffield, Stockport, Stoke on Trent and Nottingham.
The proliferation of bookmakers in GB
high street locations has generated
media headlines; the distribution above
makes clear their strongly urban
nature. The zoom (left) visualises the
same data at a finer scale, and we see
some evidence of secondary clusters
close to an Army presence in Aldershot
and also Navy locations (Southampton,
Portsmouth). Concentrations around
well-known race courses are less
evident, although they are in some
cases within larger urban
concentrations (Ascot, Epsom,
Density of licensed casinos per square km.
Density of betting (track) licences per sq
Red = up to 5.12 per square km
km. Note that these are venues where
gambling machines are licenses within
track betting locations. Red = up to 5.12 per
Casinos are urban and resort amenities as this distribution above left shows. The track
venues (above right) are premises with gambling machines at racetracks and are
associated with dog racing as well as horses, hence the cluster around suburban north
4.2 Machines density
Density surface of average number of
Average number of machines per person,
machines per premise: red = up to 49
density per sq km. Red = 0.20 machines
machines per premise per sq km
per head resident population per sq km
On the left, above, we visualise the overall density of gambling machines according to their
average numbers per venue category. Since AGCs and FECs represent greater numbers of
machines, the presence around the coast of areas of high absolute machine numbers is
revealed in this analysis. While bookmakers have higher-stakes machines and their high
street visibility has generated media interest and some public concern their contribution is
relatively lower, in terms of numbers of machines.
Normalising this distribution by expressing machine numbers as a value per head of
population (the map above right) makes less difference than might be anticipated. Only in
the area surrounding Glasgow is a significant variation discernible, i.e. the cluster is
relatively more pronounced for the number of machines per head than absolute number.
One reason why the clusters barely change is the relatively low resident population in the
very centre of cities; this makes the numbers per head high. Whilst alternative
explanations may determine the location of machines in these urban centres (such as high
numbers of workers commuting into the area on a daily basis), this illustrates that in some
urban areas there are clusters where the number of machines per resident is
disproportionately higher than average. The same pattern is true for some coastal regions,
though, evidently, the ratio of machines to resident population in smaller suburban towns
is more proportional.
It is useful to view these national distribution patterns as context and to test basic
assumptions, but always difficult to identify detailed patterns at such a coarse spatial scale.
Maps and visualised data allow us to conduct exploratory spatial data analysis (ESDA),
which often reveals phenomena that have not previously come to light. ESDA is the process
by which we visually examine the mapped results of the methods explained above and
simply observe the patterns we see, and those which we may have expected but do not
Below we explore finer-scale distribution for the area between south London and Brighton
to demonstrate this. In each case the locations of HDMZs has been included (in orange);
these are explored further in section 4.3.
Density of average number of
machines per head resident
The cluster in Crawley
is a phenomenon revealed by this
mapping: to the north is a single
HDMZ at Gatwick airport itself but
the cluster in the town centre
relative to other towns of similar
size is notable.
Density of AGC licences.
As well as
Crawley’s cluster we see the coastal
towns’ concentration of AGCs and
the density created by a motorway
service area. Satellite towns
Aldershot, Maidstone and Ashford
also emerge as high density areas.
Eastbourne’s reputation as home to
high proportions of older people
may be reinforced by its lower
concentration of AGCs often
associated with younger users. In
fact Eastbourne’s visiting
population is more diverse than
this implies, comprising families
and younger groups as well as older
Density of bookmakers.
national map also shows,
bookmakers are a more strongly
urban phenomenon with the
greatest densities in outer London
and Brighton. Despite its HDMZs,
Eastbourne does not have a strong
density of bookmakers, perhaps
reflecting the ownership of the
seafront area by the Duchy of
Devonshire which is able to impose
development controls on the area.
Even in bingo venues
shows a marked cluster, equivalent
to that for Brighton where a
tourist-driven bingo sector might
be more expected. Commuter belt
residential towns including
Maidstone, Aldershot and Slough
also emerge strongly. The density in
south London is more patchy,
hinting at the suburban / regional
centre nature of bingo venues in
this part of the country.
Density of FEC licensed premises
As well as seaside family
amusements evident here, FECs are
found in car-accessible out-of-town
locations, together with cinemas,
bowling alleys and shopping malls,
indicated with the clusters here. In
Crawley, the FEC hotspot is north of
the town centre, closer to the
airport and associated mall-type
developments. FEC density is not
strongly spatially correlated with
HDMZs except on the coast,
suggesting they are not a major
contributor to the areas of highest
machine density within inland cities
and larger towns.
Density of pubs with category C&D
machines, with HDMZs overlaid.
Pubs with fruit machines are far
more evenly distributed than other
venue types, which is as
anticipated. Among the darker-
shaded areas of higher pub density
it is clear that HDMZs are not
uniformly clustered; this venue
type is unlikely to contribute
greatly to overall high machine
Surface illustrating % economically
inactive (higher inactivity in darker
shades), with HDMZs overlaid.
There is no immediate evidence of
a linear relationship between
highest economic inactivity and
highest machine density in this
region, although nationally some
spatial differentiation is revealed
(see 4.4). The higher proportion of
economically inactive adults is
evident in the coastal towns.
Surface illustrating IMD 2007
income domain score, lower
income in darker shades.
This is a relatively wealthy region,
with lower income neighbourhoods
mainly evident on the periphery of
larger settlements including some
of the coastal towns. There is some
evidence of spatial proximity
between some of the HDMZ
clusters and lower income
neighbourhoods, but not direct
correlation. This is true in Brighton,
for example, where
neighbourhoods characterised by
HDMZs are not those with lower
incomes, perhaps thanks to a more
diverse economy in the centre of
town and/or the cost of living
centrally in Brighton itself.
4.3 Machine zones (MZs) and high density machine zones (HDMZs)
Our analysis as outlined above identified 8861 machine zones (some become contiguous
with others; hence there is not a single machine zone for every gambling machine venue).
Of these, 383 are in our ‘high’ density category.
Of the 383, 126 HDMZs are within 1 mile of the Ordnance Survey coastal boundary.
Bearing in mind that this analysis counts multiple small zones clustered together as a
single contiguous zone, it is still noteworthy that fewer than half are in this coastal zone,
perhaps reflecting the more spatially clustered nature of HDMZs in seaside resorts but also
that many are found in non-coastal areas by this spatial definition.
Visualising zones with a radius of only 400m legibly on a national-scale map is not possible;
their visible scale is simply too small. In the map below we greatly exaggerate their spatial
extent and make the background mapping less dominant in order to show the national
distribution of this phenomenon. It is important to view this map in conjunction with the
finer scale mapping which follows, since the detail of spatial differentiation particularly in
urban areas is significant to the conclusions we can draw.
Below we examine the distribution of MZs (as a reminder, this is a 400m radius zone
around any venue with one or more gambling machines, irrespective of size or type) and
HDMZs (a 400m radius zone in which the modelled density of gambling machines is
greater than 1 machine per hectare; some HDMZs are contiguous with neighbouring
zones) at a regional scale, with emphasis on those areas where the greatest
concentrations of both zone types are found.
In southern Scotland, MZs (top) follow centres of population as we would expect. HDMZs
(immediately above) are found in secondary centres including Glenrothes, Airdrie,
Stirling, Wishaw, Kirkaldy, Kilmarnock, Dundee and Clydebank, together with a few
The same comparison for the Tyneside area reveals a similar pattern. MZs (top) are
evenly distributed taking population concentrations into account, but the areas of
highest density (above) are found in urban locations around Newcastle and Gateshead
but not in either centre: Sunderland, Ashington, Bedlington, Wallsend, Chester-le-Street,
Bishop Auckland, Spennymoor, Washington, Doxford Park and Peterlee, a number of
which are or are near to ex-colliery locations. There are also coastal clusters in Whitley
Bay, Seaton Carew, North and South Shields and Redcar.
Here we visualise MZs (top) and HDMZs (above) in North West England. The HDMZs in the
Blackpool and Southport areas are clearly evident. Other notable clusters are found in
centres including Halifax, Barnsley, Huddersfield, Wakefield, Doncaster, Rotherham,
Preston, St Helens, Blackburn, Hull and Castleford, with others evident in smaller centres
surrounding Manchester and Liverpool but not evident in their centres. York and Sheffield
have no HDMZs, perhaps suggesting more diversified local economies.
Finer scale mapping for London and the area south of it is provided in maps above (MZs
top, HDMZs above). In these regional views the lack of HDMZs in central London is evident
alongside the ‘ring’ of satellite centres where they are found (Watford, Romford, Bexley,
Lewisham, Sutton, Feltham) with Luton, Stevenage, Harlow, Woking and Slough among
the regional towns having the most obvious clusters. We also see HDMZs in the main
In the West Country the coastal resorts are revealed as HDMZs, as expected. In Bristol the
two clusters are edge of centre in highly road-accessible residential areas. Of smaller
inland centres only Taunton and Launceston have HDMZs. The two motorway service
areas also creating local clusters, though we should be cautious about these in light of
their known smaller average machine numbers (see 3.4).
Earlier analysis of machines density per head resident population suggested a higher
HDMZ result for the Welsh valleys than is evident according to our final spatial density
definition of these zones (see 3.6.3). While MZs overall are distributed throughout the
centres of population (top), HDMZs (below) are notably thin on the ground, only found
here in coastal towns Barry (site of the amusement centre, Barry Island), Porthcawl and
Swansea. Inland we find HDMZs in satellite town Neath, Cardiff Airport, and in the valleys
at Tredegar, Ebbw Vale and New Town Cwmbran.
In the next phase of our exploratory spatial data analysis, we zoom into major cities to
view the spatial distribution of HDMZs at this finer scale.
Above we visualise HDMZs overlaid on a data surface of the IMD Income domain score,
for which the darker shades indicate lower income areas. Here we see that the location
of HDMZs is by no means simply an inner-city phenomenon in the larger conurbation of
Liverpool and Birkenhead, though Wigan and Warrington both have central urban
concentrations of HDMZs. Whilst HDMZ are in lower income areas, they are not always
located in the poorest areas and there are some notable outliers also, see for example,
the HMDZs in Ashton-in-Makerfield or Bebington.
Similarly, the same data combination for the area around Manchester points to HDMZs
concentrated in secondary urban centres and, like that observed in the greater Liverpool
area, a pattern by which HMDZs tend to be in lower income areas, but not always
necessarily the lowest income area. Again, there are notable exceptions. See, for
example, Altrincham, which is an interesting case study. The town itself is relatively
affluent and in close proximity to similar areas such as Hale and Bowdon but also less
affluent areas such as Wythenshawe and Baguley. Altrincham is the largest retail town
centre in this district and therefore should be an economic and recreational focus for
these areas. However, in 2010, a survey by the Local Data Company found that
Altrincham town centre has the second worst retail vacancy rate in the country (within
its category of medium town centres), with nearly 1 in 3 retail units being vacant. The
presence of HDMZs in this area may well be related to this and it would be of interest to
Birmingham and surrounding towns show the same characteristics: HDMZs are not found
in the inner city but in suburban and fringe locations. The presence of diverse economic
activity in the inner cities relative to secondary locations may be a driver for this spatial
difference, creating hotspots of machine gambling in locations where rents are lower and
alternative forms of leisure are thinner on the ground.
Finally, the same data for Greater London follow similar patterns: neither the inner city
nor neighbourhoods showing a high concentration of lower income households (the
darkest shading) are where the areas of highest machine density are found. These are
instead in surrounding towns. When we come to analyse the social and economic
characteristics of HDMZ neighbourhoods, we are not therefore looking at locations that
are purely central-urban, nor is their distribution clearly correlated with urban areas with
the lowest income.
4.4 Socio-economic data analysis
For England, we took the Index of Multiple Deprivation 2007 average income domain
score, at LSOA scale (the higher the score the more deprived the area). For Wales, an
income domain score for 2008 which is equivalent but differently calculated, i.e. the
numbers look very different but indicate the same differences between LSOAs. Scotland
does not publish a score for income, only a ranking, which is not possible to average in a
England, 2007 IMD average income domain score
(higher score indicates higher deprivation i.e. lower income)
All machine zones
Wales, 2008 IMD income domain score
(higher score indicates higher deprivation i.e. lower income)
All machine zones
It is evident from the results above that English 2001 Census output areas that fall into
HDMZs are populated by, on average, consistently and significantly lower-income
households than the all-England average. MZs also show lower incomes on average
though the difference is smaller.
In Wales, both MZs and HDMZs fall into output areas with significantly lower income
characteristics than the all-Wales average, though MZs fall into areas of considerably
lower income than HDMZs. This is consistent with the spatial patterns seen in Section 4.3
for Wales, in which MZs were distributed throughout the urban areas of south Wales and
HDMZs are relatively rare.
A total count of population for each category is broken down into the age groups shown
for all of Great Britain. Following the British Gambling Prevalence Survey, the population
is broken into groupings of 11-15 years, then 16-24 and the equivalent 10-year groups
until ‘over 75’.
Key results: In all age groups the average proportion of the population in HDMZs and all
MZs was close to and below the national average except
for those aged 16-24, 25-34 and
Selected age profiles, all GB
All machine zones
% of population 16-24 13.63
% of population 25-34 16.28
% population over 75
We are seeing a pattern here which suggests that HDMZs are found in locations with
above-average proportions of younger people and the oldest generation. The higher
share of older people may well be accounted for by the coastal clusters: the rate of
economic inactivity is consistently greater around the coast, and retired people are a key
component of this statistic.
The greater concentration of under-35s may be consistent with the clustering of HDMZs
in secondary and edge of centre urban locations. These are locations associated with
lower accommodation costs, attractive to families, and also to economic migrants who
tend to fall into lower age categories13. The 2001 census occurred before peak migration
from the newer countries of the EU but the latter phenomenon may still have had some
effect on these numbers.
The ethnic makeup of the resident populations of MZs and HDMZs was compared to the
relevant nation as a whole to explore whether any differentiating characteristics were
evident. The results below present broadly equivalent English/Welsh and Scottish
statistics side by side for basic clarity, but Scottish statistics are collected separately and
only high-level comparisons should be made.
The categories are determined and named by ONS, with our additions for clarity in
13 The Royal Geographical Society recently noted that migrant workers tend to be younger than the domestic
workforce. In recent years some 80% of migrant workers have been under 35 years old, compared to only 42% of
the overall UK working-age population (Crawley, 2010).
Selected ethnic groups, England and Wales, % of total population of output area
All England / Wales
All white persons
All mixed race
All [south] Asian
and Asian British
All Chinese persons 0.46
includes all East
All black or black
The findings above are consistent with the generally urban clustering of all MZs having a
greater proportion of non-white population, while HDMZs are found in secondary urban
areas with populations more consistent with the national average. Very slightly higher
average proportions of mixed race, south Asian and East Asian residents are found in
HDMZs, while the share of black residents is significantly lower.
Selected ethnic groups, Scotland, % of total population of output area
All white persons
All [south] Asian
All black persons
and other black
all east Asian
Scotland’s ethnic population is considerably smaller than that for England and Wales and
will be concentrated in the urban areas where MZs are clustered. The slightly higher
average proportion of Black and east Asian residents in output areas surround HDMZs is
notable, however, suggesting a different ethnic pattern around these zones compared
with their non-Scottish counterparts.
Breaking down all employed people aged 16-74 into 9 categories, broadly separating
them by seniority level, we find patterns including the following:
Selected occupation groups, all GB
All machine zones
% Managers and
% Skilled trades
% Sales & customer 9.18
(3rd lowest group)
These statistics reveal a picture of HDMZs as areas with consistently fewer residents with
higher-status occupations and consistently more in lower categories than the GB
average. The ‘suburban’ locations of many HDMZs might at first sight suggest the leafy
commuter belt, but the data above suggest these are not neighbourhoods characterised
on average by higher-earning workers.
This pattern suggests that higher densities of gambling machines are found in locations
where economic opportunities are fewer. In our analysis to follow, the lack of economic
diversity emerges as a key potential factor fostering higher densities of gambling
machines, and this may have important policy implications too.
4.5 Headline social and economic characteristics of high density Machine Zones
4.5.1 National ‘averages’
Analysis by averages always risks obscuring differences between the locations under
study, but when we have clear patterns emerging from fine-scale statistics based on
more than 218,000 census output areas in Great Britain, averages offer us important
If an ‘average’ HDMZ exists, then, its characteristics would likely include the following –
which immediately have to take account of differences between Scotland and
A suburban or secondary urban location, or a coastal resort
The average income domain score (which rises as deprivation increases) is over
50% higher than the average if in England and is double the average if in Wales
More 16-34 year olds than average and more over-75s
More residents of south Asian descent and fewer Black residents than average in
England and Wales; slightly more Black and east Asian residents in Scotland
More people than average working in lower-status occupations and fewer in the
highest status jobs.
One way in which we can start to segment this ‘average’ picture is by using the
exploratory analysis outlined above to identify spatial patterns among the HDMZ
Two patterns emerge which warrant further initial study: coastal towns, which were
anticipated as locations of high gambling machine density, and New Towns, which were
4.5.2 Correlating GB seaside resorts with highest density Machine Zones (HDMZ)
Among seaside towns, we find some very large clusters of HDMZs in some resorts but
none in others. Further investigation suggests that the difference between the two
groups may be explained, in part, by the income of the resident population.
We took a listing of the 74 largest seaside resort towns in England14 by population from
two comparable benchmarking studies for CLG by Beatty, Fothergill & Wilson (2011;
2008). We investigated whether (a) they have a HDMZ within them and (b) whether
roughly more than half the output areas (OAs) in the settlement fall into the lowest 20%
band for income.
The results show that a correlation between these two characteristics occurs more
frequently than otherwise i.e. seaside resorts with a high proportion of low income
neighbourhoods are more likely to have a HDMZ.
Has no HDMZ and low % of low income OAs:
Has no HDMZ but high % of low income OAs:
Has HDMZ but low % of low income OAs:
Has HDMZ and high % of low income OAs:
14 Regrettably no equivalent listing of seaside resorts is available from a reputable source for the other GB nations.
This relationship is noteworthy, confirming an anecdotal expectation that ‘richer’ seaside
towns are not characterised by amusement arcades. Of the 5 which have relatively low
resident income levels but no HDMZ, three are in Cornwall, perhaps suggesting both the
low relative incomes of more remote coastal communities and a more diverse tourist
offer than in other regions.
[The full table is included as an appendix to this report.]
4.5.3 Correlating GB New Towns with high density Machine Zones (HDMZs)
We undertook the same exploration for the 27 towns that were created under the New
Towns Act 1946 or rapidly expanded under provisions in the same Act or its replacement
1964 Act. Again, the full table is included as an appendix to this report.
Of the 27 towns in total, 18 (two-thirds) have gambling machine locations in the highest
machine density bracket. 14 of these are also towns where more than half of the output
areas are in the lowest 20% income band. New towns without an HDMZ were much
more evenly split for low income prevalence than the seaside resorts.
This suggests that average income is an influencing factor, but that the overall proportion
of new towns with HDMZs is perhaps worthy of further exploration in its own right. Do
new towns share physical characteristics such as road access, neighbourhood zoning or
diverse catchment areas from which demand for certain types of recreational and leisure
pursuits is generated which make them more likely to host an HDMZ? Or do they have in
common a relative lack of features such as heritage-led leisure amenities which might
provide alternative entertainment?
Has no HDMZ and low % of low income OAs:
Has no HDMZ but high % of low income OAs:
Has HDMZ but low % of low income OAs:
Has HDMZ and high % of low income OAs:
5. Interpretation and conclusions
5.1 Key findings
The results presented in section 4 highlight some key themes in relation to the geo-
spatial distribution of gambling machines in Great Britain.
Firstly, the locations of venues containing gambling machines are focused in main
population centres but also in regional centres and seaside resorts. When adjusted to
take into account the number of machines per venue type, the distribution changes,
with a greater concentration of gambling machines in urban centres and secondary
towns being evident.
This pattern is driven largely by the geo-spatial distribution of AGCs that have the
highest number of machines per venue and, to a secondary extent, by the distribution of
bingo clubs, which are found largely in primary, secondary and smaller urban centres.
Secondly, adjusting data to take into account population density where machines are
located has minimal impact upon the machine density distribution. In some areas, this is
likely to be due to the fact that many venues are located within urban centres, which
typically have a relatively low resident population, making the number of machines per
The observed distribution of high density machines when population size is taken into
account remains similar to the overall pattern observed when the blunter measure of
average number of machines per square kilometre is used. This suggests that total,
resident, population density is not a significant driving factor governing location and
clustering of machines.
Examination of machine zones (areas where any machines are located) shows that there
are a number of high density machine zones in Great Britain, with 1 or more machines
per hectare. Of all 8861 machines zones in Great Britain, 383 or 4.3% were defined as
High Density Machine Zones (HDMZ) and the distribution of these high density zones
The maps shown in section 4.3 demonstrate a broad pattern by which HDMZs tend to
be located either in seaside areas or in secondary/satellite towns surrounding major
cities. For example, in the North West there are, unsurprisingly, clusters of HDMZs in
Blackpool and Southport, but also in Warrington, Altrincham, Stockport and Barnsley.
Despite the proliferation of machines zones in central Manchester, the only significant
cluster of high density machine zones is observed in and around Salford, extending
westward towards Eccles.
This pattern is broadly replicated in the Greater London area, Glasgow and Liverpool.
For example, in Greater London, HDMZs cluster in satellite towns such as Feltham,
Woking, Slough, Crawley and Watford, with none being observed in Central London,
despite a high number of machine zones being in this area. In Glasgow HDMZs are
located in secondary towns such as Airdrie, Paisley and Clydebank
Examination of the socio-economic characteristics of residents in both machine zones
and HDMZs shows some interesting variations. Firstly, for income, economic inactivity
and lower socio-economic occupation, there is a broad pattern by which machine zone
areas are more likely than average to be lower income areas and have a higher
proportion of economically inactive people. Those who are in employment are
somewhat more likely to have ‘elementary’ occupations and less likely to be in skilled
trades or be managers/senior officials.
When looking in aggregate at the difference between HDMZs and average results for all
areas in Great Britain, the differences were even more pronounced, with HDMZs being
more likely to have an economically inactive population (comparative to all machine
zone areas and to all areas in Great Britain). They are also more likely to have a local
population in elementary occupations and less in managerial or professional
occupations. In England, those living in HDMZ areas are also much more likely to have
lower income levels (as measured by IMD income scores). Finally, HDMZs are also more
likely to be in areas where there is a slightly greater proportion of Asian/Asian British
residents in the local community.
Some of these associations may be related to the observed patterns by age group.
Notably, HDMZs have a greater proportion of residents aged between 16-34 and also
aged 75 and over. These age groups are more likely to be economically inactive (e.g.
students and retired people) and are also likely to have lower incomes than the
population at large.
Examination of sub-groups of HDMZs (those in seaside resorts and those in new towns)
also shows that a disproportionate number of seaside resorts with low income
neighbourhoods are more likely to have a HDMZ. A similar pattern was observed in ‘new
towns’ whereby those with a HDMZ also had a higher percentage of low income
neighbourhoods. This shows differential density within seaside resorts and new towns,
with those with a greater proportion of low income areas having the highest density of
However, these broad patterns also mask some localised nuances. As demonstrated
through our presentation of regional case studies, focusing on Manchester, London,
Liverpool, Birmingham and Newcastle, not all HDMZs are located in lower income areas
or even the poorest. Some are located in areas which, by comparison, have relatively
higher income levels; for example Altrincham, Bebington, Bexley and Sutton. None of
these areas are in the lowest tertile of area deprivation in England (as measured at ward
level by the Index of Multiple Deprivation 2007). Further investigation is needed at a
local level to understand what features contribute to the development of high density
machine zones in these areas and in particular to explore commonalities between them.
5.2 Potential implications for policy and planning
As noted in section 2.4, there is no pre-existing empirical evidence base in Britain which
looks at the geo-spatial distribution of gambling machines. To date, debates about
increases in availability and clustering of gambling venues and gambling machines have
not been based on empirical data. This report provides this information for the first time
and shows that:
The density of gambling machines in Great Britain varies and is centralised in
urban, secondary/satellite towns and seaside resorts.
Areas of high machine density tend to have poorer socio-economic indicators,
with a higher proportion being low income areas, a higher proportion of
residents being economically inactive and, of those who are economically
active, a greater number active being in the lowest socio-economic occupations.
Adjusting for population density does not vary the observed distribution and
clusters of high density zones are apparent even after population numbers have
been taken into account. Therefore, the assertion that high density machine
areas are only located in the high density population areas cannot be made with
any confidence. Further support is provided by examination of the spatial
distribution of high-density machine areas, which shows they are largely in
seaside towns or satellite or arterial towns to major city centres. Focus on
seaside towns and new towns shows that those with a higher proportion of low
income neighbourhoods are more likely to have HDMZs.
Review of similar work conducted in Australia and Canada emphasised the
importance of understanding supply side factors in understanding the
distribution of machines. This appears pertinent also to Britain. The location and
density of machines is likely to be governed by various factors, such as
availability of commercial property, local planning controls as in Eastbourne,
proximity to local infrastructure and transport links, proximity to central
business districts of various areas, integration of the gambling offer within and
the diversification of the local economies as well as local population demand.
Analysis of the density of machines in this report is largely influenced by the
location and number of AGCs, simply because they have a greater number of
machines per venue than pubs and bookmakers. One would expect that AGCs
are, naturally, more likely to be located in areas where there is good access to
the venue, in terms of a high street location and/or transport links to the area,
combined with a perceived market either indigenous to the area or a ‘passing’
population, or located in a regional centre with economic pull from a
surrounding catchment area. In terms of seaside resorts, there is clearly a
further factor, that of the tourist economy. Further investigation at a local level
is needed to understand the nature of these visiting populations and any other
possible features governing this distribution in more detail.
However, whilst an interaction of supply and demand features may determine
the spatial distribution of gambling machines, the effect on the local population
needs to be taken into account. As this report shows, high density machine
areas are more likely to be located in areas with some of the poorer socio-
economic indicators and the potential impact of this should be considered.
Where HDMZs are located in towns which act as local centres for recreational
and leisure economies, further consideration also needs to be given to the
catchment area from which consumers to the town centres are drawn.
Policy makers and planners need greater understanding of how market forces
and other externalities may promote clustering of machines in certain areas,
which may, in turn, potentially affect the local population. Such an assessment is
beyond the scope of this report, which was to explore the location of gambling
machines and the socio-economic characteristics of the areas in which they are
located. However, it points to the importance of understanding the local context
in which gambling opportunities exist, and the need to balance supply-side and
For example, supply-side and other externalities may mean that machines are
always likely to be greater in number in urban settings (population density
notwithstanding). This is potentially unavoidable; it will not be of commercial
interest to locate machines in areas where people can not access them (hence
the relative paucity of machines in rural areas). However, if these features mean
that machine density is elevated in areas where the indigenous local resident
population displays some of the poorest socio-economic outcomes, it raises
questions about the potential impact of this.
The implications of this for future policy and further research must therefore
include a deeper understanding of the economic diversity of those locations
showing the highest density of gambling machine venues. There is also a need
to understand who is using the machines in these areas. Whilst high density
machine zones may be in locations where the resident population experience
higher economic inactivity, lower incomes, and lower-status occupation, we
need to understand whether it is this population who is actually making use of
the machines and, if so, in what volume. This is particularly the case for HDMZs
in satellite and secondary towns, where a passing population in central business
districts or a non-resident population (i.e., those from surrounding
neighbourhoods who visit the town centre for leisure and recreation) may be
key consumers of machines in this area. Fundamentally, more information is
needed about who consumers of machines in HDMZs areas are, where they
come from and how and why they integrate this behaviour with other
(recreational) activities on offer in the local area.
If there is a perceived need to limit or exercise greater regulatory control over
the development and/or distribution of present or future venues, the ability of
the local economy to support other commercial activity needs to be fully
understood in order for such limits to have the desired effects.
Before any such conclusion could be reached, there is a clear need to
understand the detailed mechanisms which are governing this observed
5.3 Considerations and limitations
A number of considerations should be borne in mind when reviewing this study, its
findings, and their apparent implications.
Firstly, our analysis is based on average number of machines per venue rather than
actual totals. Whilst we have endeavoured to collect actual data where possible, a
number of practical limitations prevented us from obtaining actual number of machines.
Therefore all estimates presented are based on our best understanding of the actual
Secondly, we were not able to take into account different machine categories. There
may be a qualitative difference between the social and economic characteristics of areas
with a high density of B2 machines and those with a high density of category C and D
machines. The data were not available to allow us to explore this.
Thirdly, further analysis should be undertaken to explore the associations observed in
this report. Where possible, we have provided case study examples. However, the
resources were not available to enable us to provide detailed analysis of, for example,
even every local authority area in Britain, let alone super output areas. Evidence has
shown that spatial distribution does vary at different local levels and this should be
Fourthly, our remit was to map the location of machines and to examine the socio-
economic and demographic features of the areas in which the machines were located.
As such, we have focused on using nationally available indices of socio-economic and
demographic status. Further analysis at a local level may uncover differential spatial
distributions and associations which may be important, such as the diversity of the
economic offer within certain locations, or proximity of high density machine areas to
Finally, readers may note some circularity in the results presented. High density
machine zones have a disproportionately high percentage of people aged 16-34 in the
local population. Those aged 16-34 are most likely to report playing machines and are
most likely to have lower income levels. This age group may also be disproportionately
more likely to live in urban areas, although as we have seen, the highest density
machine clusters are not generally found in the inner neighbourhoods of primary cities.
This report shows broad correlations and associations. It is beyond its scope to explore
causality or the direction of these associations. However, whilst it is possible that factors
such as the age distribution of the local area may underpin some of these associations,
the broad patterns are relatively clear: the areas of highest machine density are more
likely to be in locations where younger people live, where people have lower incomes
and where more people are economically inactive.
5.4 Directions for future research
A number of directions for refining and building upon this research emerge.
A regression analysis, preferably geographically weighted to take account of the
proximity between data points and the statistical values potentially influencing
them, may significantly increase our understanding of the strength of the
relationship between the locations of HDMZs and the social and economic data
from their surrounding areas. For example, this would allow us to take into
account the differing age profile between areas and to see whether low income
or economic inactivity was independently associated with HDMZs. It would also
allow us to test the relative strength of these associations, to link them together
and ensure related factors (e.g. income and economic inactivity) are not double-
counted, and to identify which factors are the strongest predictor of this
distribution. This is beyond the scope of this study but could be readily
undertaken as a short follow-up based on the same data.
Following on from the above, additional research and analysis is needed that
would allow us to examine the relationships between proximity to machines
and probability of use. This would need to differentiate between the propensity
of local resident populations and those who are visitors to the location, to use
gambling machines. This should also explore the different characteristics of
visiting machine user groups in tourist destinations and elsewhere. In certain
HDMZs, it should also take account of potentially important differences in
behaviours between tourists/visitors and local, resident, populations, who may
be likely to have different levels of gambling involvement. Clearly, although local
populations are exposed to gambling venues and are the welfare responsibility
of the local authorities licensing those venues, they are not the only or even
main commercial target audience for operators. However, this report has shown
that HDMZs are in areas with some of the poorest socio-economic
characteristics. Exploring the potential impact of this increased availability upon
behaviour is the next logical step; a key component of which is to better
understand the local market forces which affect machine location, the
anticipated profile of consumers using these provisions compared with
understanding who is actually using these machines.
An important component of better understanding local market forces relates to
economic diversity. Some HDMZs may exist purely because the leisure and
recreation offer within an area is thin and is competing with other towns and
centres. Economic diversity is relatively readily analysed, and Geofutures used
such a measure based on industry sectors of companies in a previous study
delimiting UK town centre boundaries for the Department for Communities and
Local Government. Relating the distribution of MZs and HDMZs against this
measure would be likely to reveal relationships that the characteristics of the
resident population alone hint at but cannot confirm.
This report represents the first attempt to map the location and density of both
machines and gambling venues in general in the Great Britain. What this
endeavour has highlighted is the difficultly of doing so because of the current
way information is collected and stored. Questions about the relationship
between exposure to gambling opportunities and gambling behaviour form a
critical part of the debate about how gambling should be regulated and
monitored in Great Britain. There is very little robust empirical evidence relating
to these issues and administrative data, as used in this report, are vital to
ensuring this evidence base can be developed. Public policy decisions should be
based on robust, empirical data. To facilitate this, we would recommend
consideration be given to the development of a joined-up data strategy which
considers what information may be needed by policy makers, regulators,
researchers and industry stakeholders in better understanding the relationship
between gambling access/availability and behaviour.
For example, if further and better data became available on the actual
distribution of machine numbers, we could feel even greater confidence in the
findings presented in this report. Data on the categories of machines would add
a layer of understanding which might, if added to the analysis as a weighting
factor for example, reveal different density patterns based on some measure of
gambling ‘intensity’. In the absence of regulatory data becoming available,
additional ‘ground truthing’ fieldwork of the type employed for AGCs in this
study could be used to refine the accuracy of the assumed machine numbers for
other venue types and across a larger sample of locations.
It would also be valuable to test further the assumption that a density of 1
machine per hectare is the ideal threshold value for ‘high’ density. While the
method used to derive this measure stands up to statistical scrutiny, testing
lower threshold values of machines per hectare could yield results that alter the
spatial patterns we revealed, with different implications. Equally, it might still
place clusters of HDMZs in arterial, secondary and suburban centres and
reinforce this result further.
As patterns emerge they suggest further undiscovered insights that may await.
For example, why are Crawley, Luton and Stevenage such gambling machine
hotspots? Why Tamworth and not Burton upon Trent? Are there other
similarities between the seaside towns with HDMZs and those without? Case
studies exploring local data in greater depth may be valuable, perhaps involving
local groups and individuals, especially if future research aims to link access to
machines with gambling behaviours or to help shape public policy.
6. Access to interactive online data maps
6.1 Access details
These are currently hosted at:
6.2 Using the map viewer
This application gives instant access to data helping to visualise the locations of
gambling machines and the social and economic data for those locations.
As with all applications visualising semi-transparent surfaces, this tool cannot be
effectively viewed in Internet Explorer 6, and this browser is no longer supported by
Microsoft. All other major browsers (IE7 and above, Mozilla Firefox, Chrome and Safari)
should allow the application to be viewed correctly.
Pan around and zoom into / out of the map using the controls at the top left of the map,
or click, hold and drag the map with the mouse.
Some contextual data have been visualised as data surfaces. Data values are statistically
‘smoothed’ using spatial interpolation methods to allow the data to appear as a
Point data show specific locations within the different categories of care provision, and
each category can be viewed on top of contextual data and together with other
To select data to view on the map, check the relevant box on the right hand side. Please
note that not all data layers can be viewed at finer scale zoom levels. If your data are no
longer visible, zoom out until it reappears.
For assistance with any technical issues, please call Geofutures Ltd on 01225 320050
This reference list includes articles which were included in the Rapid Evidence Assessment but have not
been specifically referenced in this report. These articles are marked with an asterisk.
Australian Productivity Commission. (2010) Gambling Inquiry Report
. Available at
Beatty C., Fothergill S., Wilson I. (2011) England's smaller seaside towns: a benchmarking study
Department for Communities and Local Government. Available at
Beatty C., Fothergill S., Wilson I. (2008) England’s Seaside Towns. A ‘benchmarking’ study
Department of Communities and Local Government. Available at
*Clarke D., Samson T., Abbott M., Townsend S., Kingi P., Manaia W. (2006) Key indicators of the
transition from social to problem gambling. International Journal of Mental Health and Addiction
Crawley, H (2010). UK migration controversies – a simple guide
. Royal Geographical Society. Available
Cox, S., Lesieur, H.R., Rosenthal, R.J. & Volberg, R.A. (1997). Problem and Pathological Gambling in
America: The National Picture
. Columbia, MD: National Council on Problem Gambling.
Delfabbro P. (2008) Evaluating the effectiveness of a limited reduction in electronic gambling machine
availability on perceived gambling behaviour and objective expenditure. International Gambling
*Hing N., Haw J. (2009) The development of a multidimensional accessibility scale. Journal of
Gillilan JA., Ross NA. (2005) Opportunities for VLT gambling in Montreal: an environmental analysis. Canadian Journal of Public Health
; 95(1): 55
*Ladouceur R., Jacques C., Sevigny S., Cantinotti M. (2005) Impact of the format, arrangement and
availability of Electronic Gaming Machines outside casinos on gambling. International Gambling
; 5(2); 139-154
Shaffer HJ.(2007) Understanding the influence of gambling opportunities: expanding
exposure models to include adaptation. American Journal of Orthopsychiatry
*Lund, I. (2009) Gambling behaviour and the prevalence of gambling problems in adult Electronic
Gambling Machines gamblers when Electronic Gambling Machines are banned: a natural experiment. Journal of Gambling Studies
; 25(2): 215-225
Marshall DC., Baker RGV. (2001) Clubs, spades, diamonds and disadvantage: the geography of
Electronic Gaming Machines in Melbourne. Australian Geographic Studies
; 39(1): 17-33
McMillen J., Doran B. (2006) Problem gambling and gaming machine density: socio-spatial analysis of
three Victorian localities. International Gambling Studies
; 6(1): 5-29
*Moore SM., Thomas AC., Kyrios M., Bates G., Meredyth D.
(2011) Gambling accessibility: a scale to
measure gambler preferences. Journal of Gambling Studies
; 27(1): 129-43
Orford J. (2010) An unsafe bet: the dangerous rise of gambling and the debate we should be having.
*Pearce J., Mason K., Hiscock R., Day P. (2008) A national study of neighbourhood access to gambling
opportunities and individual gambling behaviour. Journal of Epidemiology and Community Health
Office of the Deputy Prime Minister. (2005) Planning Policy Statement 6: Planning for Town Centres,
Responsible Gambling Strategy Board. (2009) Research, Education and treatment: the next steps
Robitaille, E., Herjean P. (2008) An analysis of the accessibility of Video Lottery Terminals: the case on
Montreal Int. Journal of Health Geographics;
7(2) doi: 10.1186/1476-072X-7-2
Shaffer HJ, LaBrie RA, LaPlante D. (2004) Laying the foundation for quantifying regional exposure to
social phenomena: considering the case of legalized gambling as a public health toxin. Psychology of
Storer J., Abbott M., Stubbs J (2009) Access or adaptation? A meta analysis of surveys of problem
gambling prevalence is Australia and New Zealand with respect to the concentration of Electronic
Gaming Machines. International Gambling Studies
; 9 (3): 225-244
D, Jessop G. (2011) Gambling and the
multidimensionality of accessibility: More than proximity to venues. International Journal of Mental Health
; 9(1): 88-101
Welte JW., Barnes GM. (2007) Type of gambling and availability as risk factors for problem gambling:
a tobit regression analysis by age and gender. International Gambling Studies;
Welte JW (2004). The relationship of ecological and geographic factors to gambling behaviour and
pathology. Journal of Gambling Studies;
20 (4); 405
*Wickwire EM. (2007) Perceived availability, risks and benefits among college students. Journal of
Wilson D.H., Derevensky J., Gilliland J., Gupta R., Ross N. (2006) Video Lottery Terminal Access and
Gambling Among High School Students in Montréal. Canadian Journal of Public Health
Young M., Lamb D., Doran B. (2009) Mountains and Molehills: a spatiotemporal analysis of poker
machine expenditure in the Northern Territory of Australia. Australian Geographer
Over 50% OAs in
Isle of Wight
Saltburn by the Sea
Sutton on Sea
Grange over Sands
Dymchurch/St Marys Bay
Wells next the Sea
Chapel St Leonards
Over 50% OAs in
Hemel Hempstead, Herts
Welwyn Garden City, Herts
Newton Aycliffe, Co Durham
East Kilbride, Lanarkshire
Central Lancashire (Preston & Leyland)
Milton Keynes, Bucks
Peterlee, Co Durham
Washington, Co Durham
Contains Ordnance Survey data © Crown copyright and database right 2010. All rights reserved. Licence
number 100045515. Contains Royal Mail data © Royal Mail copyright and database right 2011.
© Geofutures Ltd and NatCen 2011
RGF retains all rights arising in any work carried out by the contractors,
For further information
including copyright and any other intellectual property rights. However, RGF
has granted to the contractors a royalty-free, non-exclusive licence to exercise
appropriate intellectual property rights in relation to project materials for the
purposes of wider research and dissemination activities.
Geofutures and NatCen shall not in any circumstances be liable in any way
whatsoever to any other person for any loss or damage arising in any way as a
result of reliance on this document.
Geofutures and NatCen warrant that all reasonable skill and care have been
108 Walcot Street
used in the preparation of this document. Beyond this, neither organisation
Bath BA1 5BG
shall be liable for any loss or damage (including consequential loss) sustained
by the client or his/her agents arising in any way, whether directly or
Tel: 01225 320050
indirectly, as a result of reliance on this information. Geofutures and NatCen
make no warranty, express or implied, as to the accuracy of any data compiled
by any other person and used by Geofutures and NatCen in preparing this