The determinants of local police spending

The determinants of local police spending∗
Rowena Crawford †, Richard Disney‡and Polly Simpson§
July 8, 2016
Abstract
Since police forces in England and Wales obtained the ability to raise
revenue locally two decades ago, they have diverged significantly in the extent to which they use this power to fund police services. In this paper
we seek to explain the variation in locally raised police revenues over the
2000s, unpicking the role of local differences in preferences, central government funding, the production of public safety given police inputs, and
certain political economy features of the decision making process. We find
that around three-quarters of the variation in local revenues per capita can
be explained by differences in incomes, prices and preferences. Though
this leaves much local variation unexplained, we find little evidence that
inefficiency or a lack of accountability has consistently led to higher local
taxation.
JEL: H41, H71
1
Introduction
In this paper we seek to explain the wide variation in the extent of local funding of
police services across England and Wales. Police forces receive revenue from three
main sources: grants from central government, grants from local government
and a locally-levied property tax (a component of council tax) known as the
police precept. While government grants are allocated according to centrallydetermined formulae, local decision makers have discretion over the level of the
police precept, and consequently over the total budget available to the local
police force.
Over time sizeable differences have arisen in the precept rates levied by individual police forces. Figure 1 illustrates the level of police precept revenues
per capita across the 41 non-London police forces of England and Wales in 2010.
Locally-raised revenues range from £28 per capita in Northumbria to £95 per
capita in Surrey. For context, total spending on the police in 2010 ranged from
£178 in Lincolnshire to £291 in Manchester. In Northumbria total spending was
£243, and in Surrey it was £207.
∗
The authors would like to thank Daves Innes for invaluable research assistance during his
time at the Institute for Fiscal Studies.
†
Institute for Fiscal Studies
‡
University of Sussex and Institute for Fiscal Studies
§
Institute for Fiscal Studies
1
Figure 1: Precept revenues per capita (2010)
There are many possible reasons for this variation in local tax rates. It could
be driven by differences in local demand for public safety (for example, due to
income or price effects or other differences in local preferences); it could be due
to differences in local area characteristics which affect the level of public safety
that is provided by a particular level of police services; it could be a response
to the way grants are allocated by central or local government; or it could be
the result of other political economy considerations, such as the accountability
or motivations of the local decision makers.
Understanding the relative contribution of these different factors is particularly important in the context of two current policy debates surrounding police funding in England and Wales. The first area of recent debate concerns
the accountability of those responsible for setting the local tax rate. In 2012
the independent bodies responsible for setting the local police precept (‘Police
Authorities’) were replaced by elected individuals (‘Police and Crime Commissioners’) because the government believed that local decision makers were not
acting in the interests of local residents. Relatively high police precepts were in
effect taken as an indication of undesirable behaviour by the Police Authorities,
rather than the result of different local demand or supply factors. However,
in the absence of a comprehensive empirical examination of the factors driving
variation in police precepts, it is unclear to what extent the motivations behind
this policy reform were correct.
The second area of active policy debate is around the formula used to distribute central and local government grant funding between police forces. This
2
formula intends to compensate for relative funding ‘needs’ in different areas that is the extent to which different levels of resources are required to provide
a given level of public safety in different areas. However, the formula has come
under criticism for being opaque and for not reflecting the demands placed on
modern police services, and so the government is planning to reform it. In this
context, understanding the drivers of differences in locally-raised revenue would
be informative. In particular, if higher precept revenues appear to reflect ‘needs’
not being adequately compensated for by the existing funding formula, or if
higher precept revenues are related to some element of ‘need’ not currently captured by the formula, then such analysis could suggest appropriate directions for
reform.
This paper builds on a considerable body of literature that has examined demand for public goods. We follow in the tradition pioneered by Borcherding &
Deacon (1972) and Bergstrom & Goodman (1973) of using individual preference
theory to put structure on correlations between spending and local characteristics, but paying particular attention to the distinction between the public good
provided and what is of concern to individuals (as emphasised by Bradford et al.
(1969)). This is an important complication in our context, since government
grants to police forces are allocated according to the relationship between police
spending and public safety. In context our work is closely related to that of Preston & Ridge (1995), who examine demand for local public spending in Britain,
and find evidence of price elastic and probably income inelastic demand. There
are also close parallels with the literature that has sought evidence for political
influences on the local tax burden (for example, Borge (1995) in Norway and
Allers et al. (2001) in the Netherlands).
In this paper we bring together data from many different sources to examine the association between locally-raised police revenues and demand factors
(incomes, prices and preferences), supply factors (local ‘needs’ characteristics
that affect the level of public safety that is provided for a given level of police
services), and indicators of efficiency and motivation of the police force and the
local tax setting body. We find that around three-quarters of the variation in
local revenues per capita can be explained by differences in incomes, prices and
preferences. This suggests the government may have been somewhat overconcerned by the degree of difference in local tax funding of police forces. The
residual variation may be caused by differences unobservable local ‘needs’, or
by undesirable political economy factors. However, it is notable that the potential indicators of inefficiency or a lack of accountability that we identify are not
associated with higher local taxation.
The rest of this paper proceeds as follows. In section 2 we briefly describe
the institutional context and funding arrangements for the police in England
and Wales. In section 3 we present a theoretical model of local demand for
police services, illustrating the factors expected to play a role in determining
local police funding. In section 4 we discuss our empirical strategy and data,
and in section 5 we present our results. We conclude in section 6.
3
2
Police financing in England and Wales
Law enforcement in England and Wales is undertaken by 43 territorial police
forces operating at the county or metropolitan level.1 A map of these forces is
provided in figure 2. Each police force is an autonomous organisation, with its
own budget and responsibility for financing police services in its area. In this
paper we focus on the 41 police forces outside of London - in other words we
exclude from all analysis the City of London Police and the Metropolitan Police
Service (whose financing arrangements are slightly different to those in the rest
of the country, due to both national policing responsibilities and the different
local government structure in London).
Figure 2: Map of police forces in England and Wales
Broadly speaking police forces receive funding from four sources: a general
grant from central government, a general grant from local government, a locallylevied property tax (a component of council tax) known as the police precept,
and specific grants from central government for spending on particular activities.
Police forces may also charge for special services, such as policing at sporting
events, but they are not allowed to make a profit from these services.
Figure 3 illustrates the relative importance of these funding streams, and
how the composition of real revenues per capita has changed over time. In 2000,
the locally-raised precept accounted for 19% of total revenues on average, while
the central government general grant accounted for 46%, the local government
grant for 34% and specific grants (only introduced that year) for 1%. By 2010
precept revenues had increased to account for 29%, while the central and local
government general grants had fallen to account for 37% and 28% respectively,
and specific grants accounted for 6%.
1
Scotland and Northern Ireland each have a unified police force
4
Figure 3: Composition of revenues over time
Notes: Aggregate composition of revenues across forces outside of London.
2.1
The allocation of grant funding between forces
Grant funding from central government is allocated to individual police forces on
the basis of a relative ‘needs’ assessment. This allocation attempts to take into
account the different level of police services required to provide a given level of
public safety in different areas, and to a lesser extent the different cost of providing those services in different areas. For example, areas with higher proportions
of young males on unemployment benefits would get a greater grant than an
otherwise similar area with a lower proportion of young males on unemployment
benefits on the basis that their population has a higher propensity to commit
crime; areas with a greater distance of trunk roads would get greater grant than
an otherwise similar area with a lower distance of trunk roads on the basis that
there will be more road traffic incidents to police.
Grant funding from local government is allocated to individual police forces
on the basis of the same needs assessment, but it also takes into account the
local tax raising capacity of each police force. Forces with a larger taxbase - who
would therefore receive greater precept revenues if all forces set the same local
tax rate - receive a lower grant than forces with a smaller taxbase (irrespective
of the tax rate that is actually set).
The exact formula employed to calculate relative needs has changed somewhat over time, as has the weight that the local government funding allocation
formula places on relative needs versus relative tax raising capacity. (More detail
on the various funding regimes that have operated over the period we consider
is provided in Appendix A). However, despite the increasing complexity of the
5
needs assessment, the formula has come under increasing criticism for being
“more and more detached from the real demands of policing”.2 Furthermore,
since 2002 there have been various ‘damping’ procedures employed to smooth
the year-on-year changes in forces’ grants. This has made the grant allocation
process considerably more opaque, and has reduced the extent to which grants
from central and local government reflect variations in policing needs or local
revenue raising capacity.
2.2
The police precept
The police precept is a component of council tax, a locally-levied property based
tax. For the purposes of council tax, domestic residences are banded according
to an assessment of their market value in 1991 (or, in Wales since 2005, their
market value in 2005). The police precept is set locally for a band D property,
and then the precept for properties in other bands is determined using a ratio
that is fixed across the country. For example, in all police force areas the precept
on a band A property is 2/3 the precept on a band D property, and the precept
on a band H property twice the precept on a band D property.
Until 2012 each police force had an associated Police Authority who was
tasked with deciding the forces’ budget and setting the precept level. Police
Authorities were independent of the police force itself, and had 17 members
- 9 from the local authority (or authorities) and 8 non-political members (of
whom at least 3 were local magistrates). In the 2010 general election campaign
Police Authorities came under fire for being unaccountable to local communities,
and in 2012 they were replaced by a single locally-elected Police and Crime
Commissioner for each force, who is tasked with setting the budget and strategic
priorities for the force.
While Police Authorities had the power to set precept levels according to
their local budgetary objectives, central government did place some constraints
on local precepts. Up until 1997 a system of ‘universal capping’ was in place,
whereby police authorities were informed before they formed their budgets of the
increase in the precept that would be permitted. In practice, the vast majority of
authorities simply set their budget at this maximum permitted level (Emmerson
et al., 1998b). In 1997, the new Labour government moved to a system of ‘selective capping’, whereby police authorities would first draw up their budgets and
then the government could designate forces for capping if it judged the planned
increase in the precept to be excessive. However, in practice it was not until
2004-05 that any force was threatened with capping (when three police forces
- Cumbria, Northamptonshire and West Mercia - were threatened with having
the increase in their precept capped, although in practice their precepts were
not actually capped). From 2011-12 onwards the government has discouraged
increases in local precepts by offering additional grant funding to those forces
who freeze their precept levels in nominal terms, and by requiring areas that
want to increase their precept by more than 2% (in nominal terms) hold a local
referendum to approve the decision.
2
Permanent Secretary to the Home Office, Mard Sedwill, Public Accounts Committee (2015).
6
Figure 4 shows how the distribution of the precept levied per band D property
has changed over time. In 1995, when the funding system described in this
section was introduction, virtually all forces set the same precept level.3 After
the move to ‘selective capping’ in 1997 there was a striking increase in average
precept levels, and a widening of the distribution over the time. Figure 5 shows
the year on year growth in the band D precept. This illustrates considerable
variation in changes in police precepts over the first half of the 2000s when
Police Authorities had most independent discretion over their local tax rate.
Figure 4: Distribution of precept rates over time
Notes: Boxes indicate the 25th to 75th percentiles, the horizontal bars indicate the median,
and the tails indicate the range.
2.3
Variation in revenues across police forces
In this paper we seek to explain the variation in police precept revenues among
the 41 English and Welsh police forces outside London over the 11 year period
2000-01 to 2010-11 (inclusive), since this is the period when Police Authorities
had the most discretion over the level of their local precept.
Figure 6 illustrates how total police revenues, and the contribution from
locally-raised precept and grants, varied across forces in 2000-01 and 2010-11.
In 2000-01 precept revenues per capita varied by £28 (from £23 to £51), while
in 2010-11 precept revenues varied by £67 (from £28 to £95). Looking at the relationship between precept revenues and grant revenues, areas with lower grants
tended to raise more local revenue: over the 2000-01 to 2010-11 period the correlation coefficient between precept revenues per capita and grant revenues per
capita is -0.345. However, overall, precept revenues per capita and total rev3
Prior to 1995 police services were funded via an open-ended grant based on actual expenditure.
7
Figure 5: Distribution of annual increase in precept rates over time
Notes: Boxes indicate the 25th to 75th percentiles, the horizontal bars indicate the median,
and the tails indicate the range.
enues per capita are slightly positively correlated (with a correlation coefficient
of 0.150).
In this paper we examine what drives these differences in precept revenues
and overall revenues (equivalent to overall spending), explicitly taking into account grant income alongside other factors. In the next section we set out a
theoretical model of public good demand and supply, to illustrate which factors
would be expected to play a role and why, while our empirical strategy and data
are described in section 4.
3
A model of demand for local police spending
Here we set out a theoretical model that serves to illustrate the factors that one
would expect to play a role in determining local police funding, and the channels
through which these factors would be expected to operate. First we consider the
production of public safety, before turning to individual preferences, the grant
allocation mechanism, and how individuals’ preferences are represented in the
choices made by local police authorities.
3.1
Production of public safety
The output of interest for individuals is not police spending per se, but the level of
public safety in the local area. Public safety is a function of local police spending
per capita sF , the local cost of police services PF and local environmental factors
DF (for example, demographic, socio-economic or geographical characteristics
that may affect the underlying propensity for crime in an area, or the ability of
8
the police to deal with crime that occurs).
HF = h(sF , PF , DF )
3.2
(1)
Individuals’ demand for police services
Individuals’ utility is taken to depend on public safety Hi and other (composite)
consumption Ci . We assume that all individuals in a given police force area
enjoy the same level of public safety (Hi = HF ).
Ui = U (HF , Ci )
(2)
The individual faces a budget constraint, whereby their private income Yi
must cover both their private consumption Ci (with prices normalised to unity)
and their contribution to the funding of police services Πi .
Yi = Ci + Πi
(3)
Local police services are funded through government grants GF (expressed
in per capita terms) and the revenue raised from local residents. Individuals’
contribution to the funding of police services is therefore some proportion πi
of the difference between grant income and local spending on the police. The
individual’s budget constraint can therefore be rewritten as:
Yi = Ci + πi (sF − GF )
(4)
Individuals face the maximisation problem:
max U (HF , Ci ) s.t. Yi = Ci + πi (sF − GF )
si
(5)
HF = h(sF , PF , DF )
The solution to this is individuals’ demand for police spending per capita:
s∗i = f (Yi , πi , PF , DF , GF )
(6)
or in terms of locally-raised precept per capita:
t∗i = f (Yi , πi , PF , DF , GF ) − GF
3.3
(7)
Grant allocation
Government grants to police forces are allocated on the basis of observable relative ‘needs’ - that is, on an observable subset of the factors that affect the
production of public safety given local police spending (DˆF ⊂ DF ) - and on on
local revenue raising capacity tbF .
GF = g(DˆF , tbF )
9
(8)
3.4
The public choice mechanism
As set out by Borcherding & Deacon (1972), to get from a model of individual
preferences to a public choice over police spending we need to consider three
further factors: the mechanism for aggregating individual preferences, the preferences of the police authority and the costs to the police authority. With respect
to the first, we assume that the police authority chooses police spending with
reference to the optimal position of the median voter. This would imply police
spending per capita s∗m where m is the individual with the median income (and
median contribution to the locally raised revenue) in the local area. The level of
police spending per capita in police force area F is therefore:
sF = g(s∗m,F , IF , EF )
(9)
where IF indicates the ideology of the police authority, and EF their level of
efficiency. 4
3.5
Summary
This simple theoretical model illustrates that locally-raised police revenues would
be expected to vary across police force areas for a number of reasons. Demand
for police spending varies as a result of differences in private incomes, the tax
price of police spending, the price of police inputs, the factors that feature in
the production function for public safety, and grant income. Furthermore, the
level of police spending may depend on the preferences of, and costs faced by,
the decision making police authority.
4
Empirical estimation and data
Empirical application of this theoretical model requires making functional form
assumptions. Similar to Allers et al. (2001) we adopt a simple reduced form approach in which locally-raised precept per capita is modelled as a linear function
of the explanatory factors set out in equation 7.
Specifically our empirical estimation takes the form:
P receptF,t = α + β1 Ym,t + β2 πm,t + β3 Pt + β4 GF + δ 0 DF,t + γ 0 PrefsF,t + θ0 EF + (10)
where P receptF,t is real precept revenue per capita, Ym,t is real private income, πm,t is the local tax price of additional police spending, Pt is the real price
of police services, GF is real grant revenues per capita, DF are indicators of
‘needs’, PrefsF,t are indicators of local preferences for public safety, and EF are
indicators of efficiency. These data, and in particular the indicators of ‘needs’,
preferences and efficiency used, are described in more detail below. We do not
4
We assume that there is no feedback of the ideology or efficiency of the police authority to
the demand for spending by voters - despite the implications that these have for the tax price
of police services provided.
10
control for local decision maker ideology in our empirical analysis, as with the
data available it is not possible to identify this separately from the ideology of
the local population (which we count in local preferences).
Police revenues Data on precept revenues per capita and government grants
per capita are available for each police force from the Chartered Institute of
Public Finance and Accountancy (CIPFA).5 In some of our empirical analysis
grant revenues are disaggregated into general revenues from central and local
government, and revenue from specific grants.
Private income The measure of private income used is household gross disposable income per capita, aggregated to the police force area level from local
authority level data published by the Office for National Statistics (ONS). This
is a measure of mean income, as there is no available data on median income at
a sufficiently local geographical level.
Median individuals’ tax price of additional spending This is a function
of the size of the taxbase relative to the population, and the council tax band
of the median voter. Across our sample π ranges from 1.0 to 1.3, with a median
of 1.2. In other words, to raise police spending per capita by £1 would on
average require an increase in council tax for the median individual of £1.2
(assuming the median household contains two individuals). The tax price is
greater than one because some individuals are exempt from council tax. Data
on the distribution of properties across council tax bands was provided by the
Valuation Office Agency. Figures for population and the taxbase - the number
of band D equivalent properties, taking into account the proportion of people
entitled to reduced council tax payments - are provided by CIPFA.
Prices All financial figures are converted into real terms (2015-16 prices) using the GDP deflator - a measure of economy wide inflation. To control for
geographical variation in prices we include the ‘area cost adjustments’ that the
government uses in the grant allocation formula to reflect the fact that some
areas face higher costs than others. However, it is worth pointing out that there
is relatively little geographical variation in the cost of police services in England
and Wales because the cost of the main input, police officer wages, is nationally
regulated. Police officers of an equivalent rank are paid the same across England
and Wales. We control for the change in the real price of police officer labour
over time using an index of the real change in the police pay scale relative to
2000-01.
5
Adjustments have been made so that pensions contributions are treated consistently over
time. (From 2006 onwards, the funding arrangements for police officer pensions (an unfunded,
PAYG system) were changed from one in which all pension outgoings (net of current employee
contributions) were funded explicitly in grant allocations, to a new arrangement where police
forces were required to make employee contributions for current employees, and government
provided a pension ’top-up’ when employee and employer contributions for current employees
were insufficient to meet outgoings for retired officers.)
11
Preferences In some of our empirical specifications we include controls for demographic characteristics that might be expected to be related to preferences for
public safety. Emmerson et al. (1998a) found that older individuals, and those
who supported the Conservative party, had higher preferences for spending on
police services. As controls for preferences we therefore include: the proportion
of the local population aged 65 and over, the proportion of local authority elected
seats held by members of the Conservative party, and the proportion of local authority elected seats held by members of the Labour party (with the residual of
these being the proportion of seats held by independent individuals or members
of other parties). The demographic structure of the population is estimated by
the ONS, and the data on the political persuasion of local authorities is available from The Elections Centre (http://www.electionscentre.co.uk/). We also
examine association between precept per capita and net internal immigration,
as this could be an indication of households moving local area as a result of their
preferences over local taxation and public spending.
Efficiency We wish to include indicators of the efficiency of the police force
and accountability of the police authority in order to test the hypothesis that
relatively high precept rates are the result of undesirable behaviour by the local
decision makers.
We use two indicators of accountability of the police authority. The first is
the number of billing authorities covered by a police force area - that is, the
number of local authorities who are responsible for sending out council tax bills
to residents in a police force area. The hypothesis being that it is more difficult
for local residents to coordinate and hold the police authority to account if they
are distributed across many local authorities, than if they all receive council tax
demands from the same local authority. The second indicator of accountability
we use is a measure of local election turnout rates.6 While police authorities were
not themselves directly elected, they included members from local authorities,
and it is likely that they would have felt more constrained to act in line with
perceived local preferences in areas with greater local political engagement.
We also include two controls that are potential indicators of the efficiency of
the police force. The first is the workforce exit rate. This may be an indicator
of forces in which inefficiency results in higher worker turnover. However, it
could also be argued that higher worker turnover is exogenous to the decisions
made by the police force, and therefore instead indicates that a particular force
faces higher costs due to higher hiring and training costs. The second potential
indicator of inefficiency we include is the ratio of support staff to uniformed staff
(police officers and police community support officers). However, there is debate
as to whether a higher or lower ratio is an indicator of police force inefficiency.
On the one hand a higher ratio could be indicative of inefficiently large back
6
Our election turnout measure is calculated using data from The Elections Centre. We
adjust for variation in the timing of local elections across England and Wales, strip out the
positive effect on turnout rates of general elections coinciding with local elections (as this is
unlikely to reflect an increase in local accountability in these years), and take into account the
different structure of local government in different parts of the country.
12
office operations, on the other more support staff could be being employed to
enable the more expensive uniformed staff to spend less time on paperwork and
more time on policing activities, in which case a higher support staff ratio would
result from a more efficient labour mix.
Needs variables The theoretical model presented in section 3 illustrated that
local ‘needs’ characteristics, that affect how much public safety can be provided
for a given level of police service inputs, can have an important role in determining police precept levels (and total police spending). Given the current policy
debate, in our empirical analysis we also seek to distinguish between needs characteristics that are included in the current grant allocation formula, and those
that are not. If any of the latter appear to have a significant association with
locally raised revenues then it could indicate that these factors ought to be taken
into account in a reformed funding formula.
College of Policing (2015) amongst others has argued that there are new demands on police officers outside what is considered normal crime-related duties,
such as incidents related to mental-health or child abuse. Unfortunately consistent time series data on local characteristics related to these activities is not
readily available, making empirical analysis of their association with local tax
revenues difficult. (Arguably this is exactly why these characteristics are not
currently included in the grant allocation formula). We include just two indicators of potential needs characteristics that are not included in the grant formula.
The first is the unweighted mean grant of neighbouring police authorities, which
aims to capture spill-over effects of need in neighbouring areas. The second is the
proportion of the local population who are black and minority ethnic.7 Areas
with higher proportions of minority groups may have less social cohesion and
therefore require greater police services than other areas, but this may not be
reflected in the current grant formula due to political sensitivities.
In addition we are able to control for the majority of local characteristics
that the government believes to be associated with the cost of producing public
safety (i.e. those characteristics included in the grant allocation formula). This
includes:
• The proportion of residents in routine occupations, semi-routine occupations, who have never worked or who are long-term unemployed (measured
in 2001 only).
• The proportion of households that i) live in rented accommodation, ii) are
occupied exclusively by students, iii) are overcrowded (by the ONS definition which adjusts for expected bedroom occupancy), iv) live in terraced
accommodation, v) are lone parent families (measured in 2001 only).
• The proportion of the population unemployed (defined as those claiming
unemployment benefits) and the proportion of the population on income7
Data on the proportion of the local population who are black and minority ethnic is taken
from the 1991, 2001 and 2011 censuses, and linearly interpolated for intermediate years.
13
providing means-tested benefits (we calculate the ’lag’ of these variables
as the average proportion over the previous 3 years).
• Motorway lengths and urban B/C road lengths in km (time-varying from
2005 onwards only)
• The natural logarithm of the number of bars per hectare
• Population density (hectares per 1000 population)
However, unfortunately the data on many of these local area characteristics is
collected only periodically (for example, in the census every 10 years). This
means that we could understate the extent to which these factors vary across
police forces over the period we consider, and potentially the extent to which
they explain differences in local tax raising.
Summary statistics on the distribution of real precept revenues per capita,
and the explanatory variables described above, are set out in Table 1. These
distributions are described over the 410 police force-time observations - those
variables in italics are those were there is less, or no, time series variation due
to data limitations.
5
Results
Our main results are presented in Tables 2 and 3, where we incrementally add
additional categories of explanatory variables to our linear model of explaining
police precept revenues per capita.
In specification (1) we include as explanatory variables only incomes and
prices. Private income (Y ) is positively associated with the level of precept
revenues per capita (or analagously, the level of spending per capita) but the
magnitude is relatively small. A £1000 increase in local mean income is associated with a £2.37 increase in locally-raised precept. Given mean income across
our sample is around £17,000, while police spending per capita is around £155
per capita, this implies an average private income elasticity of police spending
that is positive but considerably less than proportional. (Around 0.25 on average, though since we have estimated a linear model rather than a log model the
estimation does not imply a constant income elasticity of demand.)
Grant revenues are disaggregated into general grant revenues (those allocated on the basis of an assessment of relative ‘needs’ and local revenue raising
capacity) and specific grant revenues (those ring-fenced for specific purposes).
General grant revenues are negatively associated with precept revenues, suggesting that they actually crowd out some locally raised spending on police services
and increase total spending by only around 80 pence for each extra £1 of grant
revenue. In contrast, special grant revenues are positively associated with precept revenues. This is likely due to the nature of special grants, which often
have to be actively applied for by police forces and which may involve matching
requirements on funding.
14
In terms of prices, the negative relationship between the tax price of additional spending (π) and precept revenues suggests that demand for police services
is price elastic. The level of police pay appears positively associated with precept revenues, but since police wages are set nationally this variable contains
no cross-sectional variation, and so one might be concerned about whether it is
truly identifying a relationship between precept and prices, or just time effects.
The area cost adjustment used by the government in the grant allocation formula
to account for differences in costs across police force does suggest higher costs
depress demand and spending, but it is not statistically significant.
In column 2 we expand our specification to also include the controls for
other potential indicators of local preferences for public spending. Contrary to
expectations the proportion of the population aged 65+ has little association
with the level of police spending. The proportion of local authority seats held
by members of the Labour party is negatively associated with police spending,
though the effect is small and barely significant. Police spending per capita is
significantly higher in Wales than in England, though this may capture other
institutional differences as well as preferences.
Interestingly a higher level of net immigration per capita does appear to be
significantly associated with precept revenues: a one standard deviation higher
net immigration rate is associated with around £3 lower precept revenues (and
police spending). Whether this is indicative of individuals sorting themselves
into areas depending on their preferences for police spending is an interesting
question. If so it suggests that either on average households prefer lower taxation
and lower police spending, or that individuals with preferences for lower police
spending being more likely to relocate than those with preferences for higher
police spending. Alternatively, this association may be driven by an omitted
third factor rather than being indicative of individuals’ preferences. For example,
it could be that a higher police precept is partly the result of greater need for
police services to ensure public safety, which could also reduce the attractiveness
of a given area and reduce net immigration.
The explanatory power of this simple linear model with only incomes, prices
and preferences is relatively high, with an R2 of 0.75 (even with just incomes and
prices it is 0.65). Running the model separately for each year suggests a similar
level of explanatory power in cross-section, though clearly the limited number of
observations then limits statistical power to precisely estimate coefficients.8 .
The role of efficiency and accountability
In column 3 we expand our specification again to include a number of potential
indicators of the efficiency of the police force and accountability of the police
authority. While many of these indicators appear significantly associated with
the level of the police precept, none are related in the direction that would be
expected if inefficiency and a lack of accountability were playing a role in driving
up the local revenue burden. The number of billing authorities that a police
force area covers has a negative association with the precept level, suggesting
8
Results not reported here but are available on request
15
that police authorities are not raising more revenues in areas where they can
shroud these tax changes behind the face of more billing authorities. Election
turnout rates appear positively associated with precept revenues - suggesting
that police authorities in areas where the local population is less engaged are
not taking advantage of this to raise more revenues. Higher officer leaving rates
are negatively associated with precept. This is contrary to the expectation that
forces with higher staff turnover (potentially driven by force inefficiency) may
pass on the higher costs associated with this to local residents. Finally, the
ratio of non-police officer staff to police officers is not significantly associated
with police precept (though, as previously discussed, in theory it is ambiguous
whether a higher or lower ratio would be indicative of inefficient labour structure
in any case). Including these variables also has a relatively minor impact on the
explanatory power of the model.
Taking account of ‘needs’
In Table 3 we add to our model various ‘needs’ factors - local area characteristics
that might be expected to feature in the production function for public safety. In
specification (4) we include two potential indicators of ‘needs’ that are not are not
factored into the grant distribution formula. The first is the mean formula grant
of neighbouring police forces, which aims to capture spill-overs effects between
areas.This variable is indeed positively correlated with precept revenues, though
it it is only weakly statistically significant. The second variable we include is
the proportion of the population who are black and minority ethnic, testing the
hypothesis that areas with a greater proportion of minority groups may have
less social cohesion and so require greater policing for a given level of public
safety. This variables does not, however, appear associated with police precept,
suggesting it is not associated with demand for police services.
Finally, in specification (5), we include many of the indicators of ‘need’ that
are included in the grant distribution formula. While theoretically appealing,
since we would like to understand whether ‘needs’ affect demand for police spending over and above their determination of grant revenues, this specification suffers
from the empirical problem that the level of general grants and these needs are
highly collinear. The grant distribution formula, while it has become increasingly complex over time, is as its core linear in its arguments. (Table 4 indicates
that a simple regression on the relevant ‘needs’ variables can explain 93% of the
variation in general grants per capita.) The coefficient on general grant revenue
per capita in specification (4) is therefore not robust to the inclusion of formula
‘needs’ variables in specification (5), and one suspects that the separate effects
of grant income and ‘needs’ factors on the level of locally-raised revenue is not
well identified in specification (5).
6
Summary and conclusions
Our empirical results suggest that demand for policing is income inelastic and
price elastic. This is in line with the findings of Preston & Ridge (1995) for
16
demand for local public spending in Britain.
Our simple linear model including just prices, incomes and some other indicators of local preferences can explain three-quarters of the variation in the level
of police precept per capita. While police revenues per capita in 2010 range by
£67 (with a standard deviation of £14), the residual unexplained portion of precept revenues per capita ranges by just £39 (with a standard deviation of £8).
This suggests that simply comparing differences in levels of precept per capita
across areas might lead one to be over-concerned about the differences between
police forces. Those forces with high precept rates relative to what our simple
model suggests are typically those forces with high precept rates. Concern levied
at forces with high precept rates may be therefore be qualitatively justified, but
the magnitude of the problem may not be as concerning as might first appear.
Furthermore, the evidence presented here finds little evidence that higher
precept revenues are driven by a lack of accountability on the part of the local
police authority, or by inefficiencies of the police force itself. One limitation of
the analysis in this paper, however, could be a lack of suitable indicators of these
factors.
We find some support for the idea that forces may suffer negative spillover
effects from being adjacent to forces with higher ‘needs’, and therefore raise more
revenue locally to increase police spending in response as the grant allocation
formula does not take these spillover effects into account. However, the role of
other ‘needs’ in driving variation in precept rates is difficult to identify either
because their impact on demand through the production function cannot be
separated from their impact on demand through grant funding, or simply because
they are unobservable. This is particularly the case for local characteristics
thought to be associated with ‘new needs’ for policing, such as dealing with
cases of child abuse or incidents related to mental health). A lack of consistent
time-series data on a range of local area characteristics presents a significant
challenge for the government as it faces trying to reform the allocation formula
for grant funding to police forces in England and Wales.
17
References
Allers, Maarten, De Haan, Jakob, & Sterks, Cees. 2001. Partisan influence on
the local tax burden in the Netherlands. Public Choice, 106(3-4), 351–363.
Bergstrom, Theodore C, & Goodman, Robert P. 1973. Private demands for
public goods. The American Economic Review, 63(3), 280–296.
Borcherding, Thomas E, & Deacon, Robert T. 1972. The demand for the services
of non-federal governments. The American economic review, 62(5), 891–901.
Borge, Lars-Erik. 1995. Economic and political determinants of fee income in
Norwegian local governments. Public Choice, 83(3-4), 353–373.
Bradford, David F, Malt, Richard A, & Oates, Wallace E. 1969. The rising cost
of local public services: Some evidence and reflections. National Tax Journal,
22(2), 185–202.
Emmerson, Carl, Hall, John, & Brook, Lindsay. 1998a. Attitudes to local tax and
spending. Institute for Fiscal Studies.
Emmerson, Carl, Hall, John, & Windmeijer, Frank. 1998b. Impact of capping
on local service provision. Institute for Fiscal Studies.
of Policing, College. 2015. Estimating demand on the police service.
Preston, Ian, & Ridge, Michael. 1995. Demand for local public spending: evidence from the British Social Attitudes Survey. The Economic Journal, 644–
660.
18
Appendix A - Grant allocation regimes
Prior to 1995-96 police funding in England and Wales was ’demand-led’, in that
actual police expenditure was covered by central government through an openended grant. In 1995-96 this changed to a new system with cash-limited grants
made to police forces, and the ability of forces to raise additional revenue through
local taxation.
Under the new arrangements, each police force received a grant allocation
from the Home Office (central government), and a grant allocation from local
government via (what is now called) the Department for Communities and Local Government (DCLG)9 . These grant allocations were intended to take into
account the relative funding ‘needs’ of different areas. Allocations from both
departments were based on the same assessment of needs, but DCLG made additional adjustments to take into account the revenue raising capacity of each
police force (i.e. the size of its taxbase).
Significant changes to local government funding happened twice during the
2000s, in 2003/04 and 2006/07, which affected the formula used to allocate the
local government grant between police forces (but not explicitly the Home Office
grant formula). In this section we describe in more detail the allocation formulae
in the three time periods: 1995-96 to 2002-03 (the so called ’Standard Spending
Assessment’ model), 2003-04 to 2005-06 (’Formula Spending Shares’) and 200607 to 2012-13 (’Four Block model’).
A.1 1995-96 to 2002-03
Home Office grant
In broad terms the formula for home office police grant was as follows:
HOgrant = (population ∗ needf actors ∗ areacost) ∗ policegrantrate
where the ’need factors’ term is an estimate cost of fighting crime per capita
in that police authority. To calculate this, the Home Office estimated a series of
crime regressions, using the observable characteristics of an area to predict the
number of crimes per capita (split by different crime types). For example:
M urderspercapita = β1 ∗ U nemploymentrate + β2 ∗ Strivingpopulation + ...
The local area characteristics uses in these regressions are primarily drawn from
the census and quarterly labour market data. Examples include the proportion
of residents in terraced accomodation, the proportion of households renting,
unemployment benefit claimant rates, and motorway lengths (kms.) The coefficients in these regressions were updated annually over this period, as was the
underlying labour market data when more up to date releases were available.
9
Previously local government funding was handled by the Department of Environment,
Transport and the Regions (1997-2001), Department for Transport, Local Government and the
Regions (2001-2002), and the Office of the Deputy Prime Minister (2002-2006), before coming
into its current incarnation in 2006
19
Using police activity data, the Home Office then combined the predicted
crime frequencies with the average cost-per-crime to the police, in order to predict the amount of money each police force area would need to spend dealing
with particular types of crime:
Expenditureonmurderspercapita = Costpermurder ∗ M urderspercapita
By adding up all the crime-specific per capita expenditures, and multiplying
them by population, this gives an estimate of the total spending need of a police
force. To adjust for variation in the average cost of policing across police forces
an area cost adjustment is included:
T otalexpenditureonpolicing =(expenditureonmurderspercapita
+ expenditureonrobberiespercapita + ...)
∗ population ∗ areacostadjustment
A proportion of the need predicted by this formula is paid to forces by the
Home Office (the ’police grant rate’). This proportion is the same for all police
forces but varies over time.
Local government grant
The formula for the local government police grant begins with the same formula
that the Home Office use, calculating an expenditure per capita which is scaled
up by population and adjusted for regional differences in costs. This estimated
total spending need of a police force was referred to as its ’Standard Spending
Assessment’ (SSA). This is then multiplied by 1 minus the police grant rate, to
identify the proportion of a police force’s spending needs that were not being
met by the Home Office grant.
Then, consistent with how funding was allocated to all different tiers of local
government (including district and county councils, unitary authorities and fire
authorities), additional adjustments are made which take into account the local
revenue raising ability of each police force. Specifically:
LGgrant =(population ∗ needsf actors ∗ areacostadj) ∗ (1 − policegrantrate)
− (assumedcounciltax ∗ taxbase)
where assumed council tax is the level of precept on a band D property
assumed to be levied by every council in the country, and taxbase is the number
of band D equivalent properties in a police force area. The level of assumed
precept was called the ’standard tax element’ and, as an example, was set at
£51.70 in 2000-01. Police forces who levied a precept greater than the assumed
rate of council tax would have a total budget greater than their SSA, while those
who levied a precept less than the assumed rate of council tax would have a total
budget of less than their SSA.
20
In 2002/03, a minimum increase in police forces’ funding (defined as all funding received from government, including special grants) was introduced, such
that a police force would receive at least a 2.3% increase on its previous year’s
funding allocation. This was paid for by proportionately scaling down the DCLG
allocations of all police authorities who were to receive an increase of more than
2.3%, and setting the maximum increase at 4
A.2 2003-04 to 2005-06
Home Office grant
Over this period, there were minor tweaks made to the Home Office funding
formula, and some new data was used where this became available.
Local government grant
In 2003-04, the local government finance arrangements were rebranded - moving
from the ’spending share assessment’ system to the ’formula spending shares’
model. The government rhetoric implied that this represented a shift from calculating absolute spending needs for local authorities (including police forces) to
assessing the relative needs of local authorities. In practice, very little changed
in terms of the structure of the funding formula used.
Local revenue raising ability was given a new name of ’assumed national
council tax’ with each type of authority given a ’share of assumed national
council tax’. For police forces the level at which this was set jumped up from
£57.63 in 2002-03 to £88.08 in 2003-04 (the assumed level of council tax generally
also jumped up at this point). This meant that for a police force to achieve a
given level of total revenues it was required to raise a greater proportion from
local taxation than had previously been the case.
Minor changes were also made to the minimum and maximum permitted
annual increases in funding.
A.3 2006-07 to 2012-13
Home Office grant
In 2006/07 the crime regressions were updated, and a number of new local area
characteristics were added as explanatory variables (for example, bar density).
A major change was also made to the way police pensions were funded. The
police pension scheme is an unfunded, pay-as-you-go scheme. Previously, all
pension outgoings (net of current employee contributions) were funded explicitly in grant allocations. However, from 2006/07 police forces were required
to start making employer contributions in respect of current employees, and the
government then provided a pension ’top-up’ grant when employee and employer
contributions for current employees were insufficient to meet outgoings for retired
officers.
21
Local government grant
In 2006/07 a new system of allocating local government funding, the ‘Four Block’
model, was introduced. Under the new system, the total DCLG funding is split
into 4 pots, or ‘blocks’. Funding from the first block, “relative needs”, is allocated
based on the same formula as the Home Office grant. Funding from the second
block, “relative resources”, is allocated based on the taxbase per capita of local
areas. Funding from the third block, “central allocation”, is allocated purely
on a per-capita basis. Finally, the fourth block “floor damping”, applies floors
and ceilings to changes in grant levels from year to year. The total budget for
local government is divided between these four blocks on a relatively ad hoc, and
allows DCLG the flexibility to change the relative priority it assigns over time
to the neediness of areas, or their ability to pay, or the stability of funding over
time.
This new funding allocation system therefore essentially depends on the same
elements as before, but is far more complicated and less transparent.
22
Figure 6: Distribution of annual increase in precept rates over time
23
Table 1: Summary statistics
VARIABLES
mean
sd
min
p25
p50
p75
max
Precept p.c.
Mean income p.c.
πi
GF (general)
GF (specif ic)
Pay index
Area cost adj.
% LA seats Labour
% LA seats Conservative
% pop. aged 65+
Net internal immig.
Num. billing authorities
Election turnout rate
Workforce exit rate
Support staff ratio
Mean(G) of neighbours
% pop. BME
Population density
Log(bar density)
% pop. NSSEC 6,7,8
% households renting
% households student
% households overcrowded
% households terraced
% lone parent households
Lag. % pop. M-T benefits
Lag. % pop. unemployed
Km of motorways
Km of urban roads
51.50
16.93
1.209
139.9
15.47
102.5
1.020
29.34
38.01
17.00
0.199
8.237
34.80
5.535
0.524
145.0
5.853
418.7
-1.005
25.16
26.50
0.338
4.956
25.26
6.105
8.061
1.498
75.14
344.3
15.65
2.280
0.0692
29.39
8.916
0.830
0.0350
17.81
18.08
1.919
0.331
3.509
4.077
1.299
0.0910
17.91
5.049
404.4
0.736
3.430
3.677
0.208
1.062
5.858
1.189
2.299
0.629
58.76
162.5
21.77
13.01
1.016
93.20
0
100
1
0.871
0.402
13.21
-0.685
2
22.92
1.805
0.128
115.1
0.699
34
-2.625
15.15
20.45
0.0235
3.369
15.16
4.002
3.602
0.439
0
90.70
39.52
15.23
1.150
119.3
9.331
102.5
1
14.44
25.93
15.79
-0.0441
5
31.93
4.651
0.458
130.6
2.654
211.9
-1.517
23.06
24.08
0.215
4.022
20.85
5.117
6.460
1.039
29.20
226.2
51.30
16.31
1.217
129.3
14.85
102.8
1
25.36
41.00
16.70
0.210
7
34.92
5.349
0.523
139.6
4.301
271.9
-0.985
26.01
25.21
0.297
4.965
24.88
5.798
7.413
1.343
68.20
308
61.53
18.30
1.259
158.6
20.42
102.9
1.026
42.91
51.87
18.04
0.405
10
37.70
6.265
0.583
157.1
6.900
478.4
-0.658
27.59
27.78
0.402
5.499
29.52
6.952
9.676
1.778
108.3
473.8
96.12
26.06
1.342
239.7
61.69
103.0
1.159
76.96
72.14
22.34
1.372
17
45.82
12.73
0.865
192.0
28.89
2,300
0.870
30.69
38.00
0.919
7.364
38.88
9.669
15.46
4.142
231.2
752.7
24
Table 2: Estimation results - without controlling for needs (dependent variable
is precept revenues per capita)
(1) Demand
Y
π
G(general)
G(specif ic)
Pay index
Area cost adj.
(2) + Preferences
(3) + Efficiency
2.369
3.182
2.785
(1.002)**
(0.805)***
(0.742)***
-40.704
-42.377
-28.940
(18.816)**
(16.488)**
(16.510)*
-0.211
-0.146
-0.137
(0.044)***
(0.053)***
(0.050)***
0.658
0.383
0.346
(0.107)***
(0.066)***
(0.068)***
4.211
4.458
3.667
(0.681)***
(0.544)***
(0.536)***
-73.891
-121.519
-74.465
(54.993)
(47.494)**
(47.239)
% LA seats Labour
% LA seats Conservative
% pop. aged 65+
Wales
Net internal immig.
-0.135
-0.123
(0.076)*
(0.075)
0.072
0.041
(0.098)
(0.091)
0.760
1.104
(0.840)
(0.700)
16.793
13.159
(2.878)***
(3.240)***
-10.534
-9.566
(3.264)***
(2.820)***
Num. billing authorities
-0.678
(0.239)***
Election turnout rate
0.406
(0.155)**
Workforce exit rate
-1.141
(0.390)***
Support staff ratio
22.738
(17.527)
Constant
R2
F
-276.181
-280.773
-276.888
(80.674)***
(76.207)***
(72.946)***
0.65
61.09
0.75
58.62
0.79
44.39
* p < 0.1; ** p < 0.05; *** p < 0.01
N = 451. Standard errors are clustered at the police force level
25
Table 3: Estimation results - controlling for needs (dependent variable is precept
revenues per capita)
(4) + NF needs
Coeff.
Std.Err
Y
π
G(general)
G(specif ic)
Pay index
Area cost adj.
% LA seats Labour
% LA seats Conservative
% pop. aged 65+
Wales
Net internal immig.
Num. billing authorities
Election turnout rate
Workforce exit rate
Support staff ratio
Mean(G) of neighbours
% pop. BME
Constant
Population density
Log(bar density)
% pop. NSSEC 6,7,8
% households renting
% households student occupied
% households overcrowded
% households terraced
% lone parent households
lag IS
lag unemp
Km of motorways
Km of urban roads
R2
F
2.307
-27.622
-0.123
0.338
3.380
-44.093
-0.173
0.136
0.785
15.244
-9.390
-0.736
0.371
-0.957
16.642
0.103
-0.284
-280.150
0.636***
14.447*
0.049**
0.070***
0.589***
40.062
0.075**
0.092
0.710
3.651***
2.914***
0.250***
0.168**
0.365**
18.110
0.054*
0.260
73.638***
0.80
40.51
(5) + F needs
Coeff.
Std.Err
4.836
-19.960
-0.043
0.297
2.042
-10.098
-0.270
0.001
0.755
13.597
-8.799
0.001
0.468
-0.640
17.469
0.043
-0.167
-252.532
0.005
3.587
2.202
-0.887
4.694
-1.995
-0.198
-0.265
1.123
-2.022
0.004
-0.015
0.86
253.27
* p < 0.1; ** p < 0.05; *** p < 0.01
N = 451. Standard errors are clustered at the police force level
26
0.816***
9.724**
0.075
0.078***
0.608***
39.461
0.074***
0.084
0.651
4.221***
2.436***
0.230
0.166***
0.282**
12.913
0.065
0.306
78.950***
0.003
2.694
0.545***
0.393**
3.030
0.963**
0.152
2.363
1.116
2.937
0.019
0.007**
Table 4: Estimation results - dependent variable is general grant per capita
Coeff.
Std.Err
Taxbase p.c.
% households renting
% households overcrowded
% households student
% households terraced
% lone parent households
% pop. NSSEC 6,7,8
% pop. on MT benefits
% pop. unemployed
% unemp. who young male
% unemp. who LT unemp.
Log(bar density)
Population density
Paid staff in 1995
Km motorways
Km urban roads
Constant
R2
F
-61.948
2.881
2.725
-0.084
0.941
2.801
0.122
5.066
-7.602
-64.328
73.092
12.264
0.009
-0.002
0.073
0.003
18.330
0.93
367.93
28.248**
0.196***
1.452*
2.315
0.115***
1.167**
0.318
0.822***
2.188***
30.118**
16.845***
1.748***
0.002***
0.001***
0.012***
0.006
16.578
* p < 0.1; ** p < 0.05; *** p < 0.01
N = 451. Standard errors are clustered at the police force level
27