Introduction {#sec-1}
============

The accompanying package contains an updated version of the **Unadjusted
Means-Tested Benefits Rate** (UMBR) data set. UMBR is a proxy measure of
income poverty for small geographic areas in England, Scotland and
Wales. It provides a single-number household poverty rate for somewhat
over 40,000 small area units annually from 2001 to 2013.

UMBR is produced from public data sources by the [Centre for Analysis of
Social Exclusion](http://sticerd.lse.ac.uk/case/) at the London School
of Economics, as part of the Social Policy in a Cold Climate (SPCC)
research programme[^1]. UMBR is suitable for a variety of purposes,
including analysis of the local distribution of poverty over time, and
the coding of other individual or area data-sets with an income poverty
indicator.

Key differences between UMBR-12 and UMBR-14 {#sec-1-1}
-------------------------------------------

The first edition of UMBR was published in 2012[^2]. The design of
UMBR-14 is substantially the same the as the previous edition, except
that:

-   UMBR-14 extends annual coverage to 2013. The previous version,
    UMBR-12 covered the calendar years 2001 to 2011.

-   UMBR-14 uses Census 2011 data, and associated updates to population
    estimates, to revise estimates from 2001 to 2011.

-   UMBR-14 is measured to the latest 2011 Census geographic boundaries
    (LSOAs) in England and Wales

Getting started {#sec-1-2}
---------------

The dataset is provided as a comma-separated (CSV) file, called
`umbr14-esw.csv`. This can be loaded in any statistical or numerical
software, including R, Stata or SPSS. Each row contains information for
one area (a 2011 LSOA in England and Wales, a 2001 Datazone in Scotland)
in one year. The poverty proxy rate (proportion of households that are
poor) is in the field `UMBR.HH`. The rest of this document describes the
design and potential use of UMBR in more detail.

Conceptualisation and design of UMBR {#sec-2}
====================================

The UMBR rate for a given area in a given year is the calendar-year
average **number of claimants of major means-tested benefits** in that
area, **divided by the mid-year estimate of households**.

UMBR as a poverty proxy {#sec-2-1}
-----------------------

UMBR is conceived of as a proxy for poverty according to the common
European and international standard which is widely used in major social
surveys, such as *Households Below Average Income*: poverty as an
equivalised household income falling below a threshold set as a
proportion, 60%, of the population median.

In addition, UMBR as a proxy for this rate was required to provide a
single number figure, covering all of Great Britain for 2001 to 2013 and
easily aggregable to higher level spatial units, which was reproducible
in a timely manner from publicly available data sources.

Source data selection and design {#sec-2-2}
--------------------------------

The selection of data for the numerator was informed by analysis of the
*Family Resources Survey*, looking at the validity and coverage of a
variety of administrative data proxies for income poverty[^3]. On this
basis, the receipt of four major means-tested social security benefits
(Job Seekers Allowance, Income Support, Employment and Support Allowance
and Pension Credit-Guarantee Element) is the numerator. Households are
used as a denominator to control as best as possible for inter-area
variations in family size and structure, as it is very rare for
households to contain more than one primary recipient of these benefits.

UMBR is not only a partial *count* of poor households, but a strong
*correlate* of the spatial distribution of all poor households. The
observed presence of low-income households who *do* claim benefits is a
strong predictor of the presence of low-income households who *do not*
claim benefits. This analysis indicates that for most areas, UMBR and
measures of underlying income poverty fall in a linear relationship.
Difference of scale in UMBR are thus typically equivalent to differences
of the same scale in the “true” poverty rate.

Similarities and differences to other poverty and deprivation indices {#sec-2-3}
---------------------------------------------------------------------

In its sources and design, UMBR is similar to several official measures
of local income poverty and deprivation available in the UK, such as the
various country-specific *Indices of Multiple Deprivation*, the
*Economic Deprivation Indices* (England) and HMRC’s *Child Poverty
Estimates* (Great Britain). In comparison with these sources, it:

-   measures only income poverty, not other aspects of deprivation;

-   gives a poverty proxy measured as a single real rate, rather than a
    rank;

-   offers annual coverage from 2001 to 2013;

-   is measured consistently across England, Scotland and Wales;

-   is reproducible only from public sources and using open-source
    software;

For some uses, these other sources may be more appropriate. Since they
are able to draw on source data that is not publicly available, they may
offer more accurate measurement than that provided by UMBR for the
equivalent years. In practice, UMBR is very closely correlated with the
income domains of the various IMDs.

Using UMBR {#sec-3}
==========

The UMBR dataset {#sec-3-1}
----------------

The UMBR-14 dataset is presented as a CSV (comma-separated values) file.
This should be suitable for use in any statistical application,
including R, Stata and SPSS. The dataset contains 536,289 rows, each of
which represent one of 41,253 small areas in one of 13 years. Each case
contains the following fields:

Geogcode
:   The ONS geographic code for the small area (2011 LSOA in England and
    Wales, 2001 Datazone in Scotland)

Year
:   The calendar year to which the values refer

All.MTB
:   The average number of claimants of means-tested benefits in this
    area in this year

Hholds
:   The mid-year estimated count of households in this area

UMBR.HH
:   The household UMBR rate (=All.MTB/Hholds) for this area

Geog.Indic
:   One-letter code indicating whether and how this area is affected by
    changes to statistical boundaries

Geog.XRef
:   For areas affected by major boundary changes, indicates where this
    area’s count data are held

Every area has a value for `UMBR.HH` for every year. For certain areas,
the `All.MTB` and `Hholds` fields are missing; these are marked with
“NA”. For further information on this and on the `Geog.Indic` and
`Geog.XRef` fields, see the section on below.

Appropriate uses {#sec-3-2}
----------------

Uses to which UMBR can be put include:

-   To make statements about relative poverty rates in different parts
    of a wider area (a city, a region). For single LSOAs and Datazones,
    small differences of less than ±10% (certainly not less than ±5%) at
    a single point in time should not be considered as indicating
    significant differences in the underlying poverty rate;

-   To assess how poverty rates have changed over time in an area or a
    group of areas over time, including aggregating to higher spatial
    units (such as MSOAs or Wards) to make comparisons at that level;

-   To judge how much any such change might be attributed to falls in
    absolute numbers of poor households, or to increases in the
    household population;

-   To code other datasets, such as survey, census or administrative
    data, with a poverty indicator for further analysis.

Limitations and cautions {#sec-3-3}
------------------------

It is important to note that although UMBR is an interval measure, it is
not directly comparable to a poverty rate as normally conceived.
Furthermore, the measurement of poverty for small areas inherently
involves estimation of quantities that are not directly observed, as
well as compromises between the simplicity of the method and the
complexity of the underlying construct (poverty).

Some particular cautions apply to UMBR as a representation of poverty as
it is conventionally measured using a threshold of population median
income:

-   UMBR does not fully reflect the incidence of in-work poverty, or
    poverty resulting from high housing costs. Thus UMBR tends to
    indicate lower levels of poverty in areas of high housing costs
    (notably London), in comparison to survey or other sources.

-   It follows from this that caution should be exercised in directly
    comparing areas with very dissimilar housing and labour market
    characteristics.

-   UMBR also does not fully account for changes in the welfare system
    which affect the material welfare of claimants or their eligibility
    for means-tested benefits. Thus the effects of, for example, cuts to
    Housing Benefit and Council Tax Benefit on poverty rates are not
    reflected in UMBR, nor are changes to lone-parents’ claims to Income
    Support.

-   The source data from which UMBR is derived include multiple forms of
    estimation, rounding, and geographic resampling, and UMBR makes some
    interpolation and projection of data over time. As noted, small
    differences (less than ±5%) between single years or between single
    areas should not be interpreted as indicating a significant
    difference in the underlying poverty rate.

-   The average size of small areas and the algorithm by which they are
    constructed is different in Scotland than in England and Wales. Thus
    measures of dispersion (for example, standard deviation) cannot be
    directly compared between Scotland and England and Wales.

-   The limited information available on population change between the
    Censuses in 2001 and 2011, and changes to statistical boundaries,
    mean that estimates for areas where there were substantial
    population or housing changes are subject to substantially greater
    uncertainty.

-   In areas with large institutional populations (people living in
    communal establishments such as nursing homes), and especially in
    areas where such establishments closed or opened between the
    Censuses, UMBR may produce erroneous results, such as UMBR scores
    substantially greater than 1.

-   These cases cannot be systematically identified, so are left
    unaltered. Inferences should not be drawn from single outlying
    cases, and detailed analysis of a small number of specific areas
    should be supplemented with other sources.

Comparing UMBR-12 and UMBR-14 {#sec-4}
=============================

This edition of the dataset follows the same method as the first edition
of UMBR. In comparison to the previous edition it draws on additional
sources, which both extend the timeframe covered by the dataset, and
improve the accuracy of the estimates that were already in the previous
edition, by:

-   using administrative data on social security benefits to Q4 2013;

-   using the results of the 2011 Census of Population;

-   using revised population estimates from ONS (England and Wales) and
    GROS (Scotland);

-   providing results aligned to updated (in England and Wales) standard
    geographic units.

The effect on existing estimates for 2001 to 2011 follow from revisions
to official small-area population estimates and to the UMBR household
estimates as a result of these revisions and new Census data. The
effects of the availability of new data on the previous UMBR estimates
are described in the following sections.

Size of revisions between UMBR-12 and UMBR-14 {#sec-4-1}
---------------------------------------------

Comparisons can be made for the years and areas which appear in both
UMBR-12 and UMBR-14 to describe the scale of revisions made in the light
of newly available source data. For simplicity, only areas unaffected by
major boundary changes are compared (≈98% of areas in England and Wales,
100% in Scotland), and thus the effects of major housing developments
and local population changes are not included. Table 1 shows the
absolute change for these rates from the previous (UMBR-12) to current
edition, by year.

[ht]

           $<$1%   1%-5%   5%-10%   10%-25%   $>$25%
  ------ ------- ------- -------- --------- --------
    2001    92.7     4.6      0.0       0.0      0.0
    2002    87.9    10.0      0.2       0.0      0.0
    2003    84.0    14.1      0.4       0.1      0.0
    2004    80.7    17.4      0.7       0.1      0.0
    2005    76.1    21.7      1.2       0.2      0.0
    2006    71.0    26.1      1.9       0.3      0.0
    2007    67.2    29.1      2.7       0.5      0.0
    2008    63.0    32.2      3.5       0.7      0.0
    2009    53.4    39.6      5.3       1.2      0.1
    2010    49.7    42.0      6.3       1.7      0.1
    2011    48.0    42.4      7.0       2.1      0.1

The revisions to the previous UMBR estimates are greatest in 2011, when
the population and household data was furthest projected since the last
Census in 20001. In 2011, just under half of all areas had revisions of
less than 1% (0.01) in their UMBR score. However, even in that year,
less than 10% of all areas had revisions of greater than 5% (0.05).

Regional pattern of UMBR revisions {#sec-4-2}
----------------------------------

The chart in Figure 1 shows the pattern of revisions by region
(separating Inner and Outer London) by year. It shows the size of the
revision at the 5th, 25th, 50th (median), 75th and 95th percentiles.

![UMBR-14 revisions by year and region, values at selected
percentiles](regional-revisions-fan.pdf)

In all regions, the median revision lies close to zero; the largest
median revision is in Outer London in 2010 (+0.56%). At regional level
the revisions do not point to substantial bias in the earlier estimates.
The largest revisions are seen in Inner London, where several features
of the population make inter-censal population estimates more difficult.
The relatively large revisions in Scotland are likely to be attributable
to the smaller mean population of the areal units (Datazones) there;
equivalent absolute revisions in the population estimates result in
larger revisions to the UMBR rate.

Typological pattern of UMBR revisions {#sec-4-3}
-------------------------------------

The chart in Figure 2 uses the 2001 Census Area Classification[^4] to
analyse the pattern of revisions by area type.

![UMBR-14 revisions by year and neighbourhood type, values at selected
percentiles](nhoodtype-revisions-fan.pdf)

This again shows the greater overall scale of revisions to inner-city
areas, notably the “Multicultural City Life” type. More importantly,
there are significant downward revisions at the median for the
“Disadvantaged Urban Communities” type. This implies that with the
advantage of revised population estimates, UMBR-14 estimates for this
type of area are generally somewhat lower than those in UMBR-12. The
median revision in 2011 is -1.5%.

Changes to statistical geographies {#sec-5}
==================================

The results of the 2011 Census are of unparalleled importance in
improving the accuracy of the estimates in a dataset such as UMBR.
However, the publication of Census 2011 results has entailed changes to
the geographic boundaries to which small-area aggregate data are
published, according to the rules and parameters for these small-areas
set by ONS[^5].

In England and Wales, Output Areas as well as Super Output Areas have
been redesigned, where necessary, on the basis of the 2011 Census
results[^6]. In Scotland new Output Areas have been created where
necessary for the 2011 Census results, but changes to the standard
Scottish small-area geography used in UMBR, the Datazone, have not yet
been finalised[^7].

Other government agencies, such as DWP, have not yet adopted the new
Census 2011 geographies, and continue to publish administrative data to
the 2001 boundaries. UMBR-14 must therefore combine data published to
different sets of boundaries which are largely, but not wholly,
compatible. The approach to combining statistical geographies follows
these principles:

-   to use standard geographies (LSOAs) in their current (2011) version
    for output;

-   to enable typical analyses to proceed with little or no alteration;

-   to provide a data set that includes all values from the source data,
    consistent with figures for higher-level areas;

-   to give values for as many geographies as possible, where the
    available data directly provide figures or enable reasonable
    estimation;

-   to mark clearly where figures are unavailable or subject to
    estimation.

More detailed information on how ONS boundary changes are handled in
UMBR is provided in the .

Effects of spatial unit boundary changes {#sec-5-1}
----------------------------------------

[sec:geog-changes] Each case in UMBR-14 has a field `Geog.Indic`, which
indicates whether and how that area has been affected by boundary
changes. The codes and possible types of changes are as follows:

A
:   areas composed only of one or more whole source data geographies, no
    special techniques used. Note that such areas are still affected by
    rounding, best-fit and disclosure-control techniques used in the
    source data.

B
:   areas affected by irregular boundary changes by ONS, and/or composed
    of smaller geographic units where some values have been apportioned.

C and Cx
:   areas with major population increases or decreases, and thus
    multiple new areas in source data. UMBR data is reported jointly for
    a composite of several contiguous LSOAs within one MSOA, all of
    which have the same rate. Count data (claimants and households) for
    areas marked **Cx** is missing (“NA”) and held for all the areas
    under a corresponding **C** area, noted by the `Geog.XRef` field.

Areas marked **B**, **C** and **Cx** are likely to suffer additional
estimation error and in some cases unreliable estimates. The proportion
of areas affected is shown in Table 2:

[ht]

                   A     B     C    Cx
  ---------- ------- ----- ----- -----
     England    97.6   0.6   0.7   1.1
    Scotland   100.0   0.0   0.0   0.0
       Wales    98.1   0.7   0.5   0.8

Note that in Scotland, five Datazone codes do not appear in UMBR-14. As
a result of large decreases in population between 2001 and 2011, these
Datazones now have no 2011 Census data attached to them by the best-fit
method used to produce the Scottish Census results.

Using the geography changes in analyses {#sec-5-2}
---------------------------------------

Changes to geography are intended to be for most uses of UMBR
transparent to the user, so that no special handling of cases is needed.
For higher-level analysis, one can simply aggregate and sum `All.MTB`
and `Hholds` to higher geographies such as MSOAs, Wards, Local
Authorities or Parliamentary Constituencies, omitting missing values.
The UMBR rate for the higher geography is then simply the summed
numerator (`All.MTB`) divided by the summed denominator (`Hholds`).

Analyses of rates over time, and codings of other datasets using UMBR
should be in most cases robust to the effects of the changes. Where the
results of interest are at the boundaries of significance, it may be
useful to repeat the analysis omitting areas affected by boundary
changes.

Technical Information {#sec-6}
=====================

[sec:techinf]

The Numerator: Means-Tested Benefit Claimants {#sec-6-1}
---------------------------------------------

The numerator of UMBR is the sum of claimants of the following benefits:

Job-Seeker’s Allowance (JSA)
:   both Income-based and Contribution-based claimants are included

Income Support (IS)
:   welfare changes mean that several groups of Income Support claimants
    such as older people and some lone parents were transferred to other
    UMBR benefits during the period

Pension Credit (PC-GC)
:   2003 onwards; only claimants who claim the “Guarantee Element” are
    included

Employment Support Allowance (ESA)
:   2008 onwards; both Income-based and Contribution-based claimants are
    included

Universal Credit (UC)
:   2013 onwards, pilot areas only

Quarterly data for JSA, IS, PC-GC and ESA are extracted from NOMIS[^8]
and averaged over four calendar-year quarters. Data for newly introduced
benefits (PC, ESA, UC) are still averaged over a full year, even where
in the year of introduction data for only some quarters are available.

The heavily delayed introduction of Universal Credit means that over
2013, when the first pilots were started, there was a monthly average of
only around 1,200 claimants across Great Britain, heavily concentrated
in a small number of pilot authorities (Wigan, Oldham, Tameside and
Warrington). Local authority and LSOA claimant data for Universal Credit
has not yet been made publicly available. UMBR-14 therefore takes the
four Job Centres with more than 100 UC claimants as at the end of 2013
and allocates UC claimants at these to the corresponding local
authority. Since in the pilots, only single people without work were
handled under UC, local authority counts of UC claimants are distributed
across that local authority’s LSOAs in proportion to the LSOA’s 2013
share of JSA claimants.

In general, the stability of a proxy measure like UMBR and its
comparability over time depend in part on the stability of the
administrative system from which the data are drawn. Clearly, while
there has been some continuity in the British welfare system in the
2000s, there has also been major changes in benefits for low-income
people of retirement age, lone parents and those unable to work through
sickness or disability. Further wide-ranging changes have been
introduced since 2010, which have only been partly thus far implemented.

### Estimation of OA-level benefits data {#sec-6-1-1}

DWP’s counts of benefit claimants, which provide the numerator for UMBR,
are not generally available for Output Areas or 2011 LSOAs, only 2001
LSOAs. This means that for areas whose boundaries have changed,
published 2001 LSOA figures must be assigned to 2011 geographies.

To do this, the means-tested benefit claimant counts in the source 2001
LSOA data is first distributed among the OAs that constitute that LSOA
(see below), before being re-summed to 2011 boundaries.

Each OA’s share of its LSOA parent’s claimants is calculated from two
indicators:

1.  The 2001 Census count (from table CAS030) of all people aged 16-74
    in households who are either:

    -   Unemployed;

    -   Permanently sick or disabled; or

    -   Looking After Family/Children *AND* not living in a couple

2.  The rounded count of Out-of-Work benefit claimants in that OA, in
    the data published by DWP at that level for 2013 only (four-quarter
    average).

The actual indicator for an OA in a given year is a time-weighted
average of these two numbers. They are evenly weighted in 2001, and the
weight of the Census 2001 data is then steadily reduced to zero in 2013.
The indicators and weighting were selected by manually maximising the
Pearson correlation between them and the LSOA count of all means-tested
benefit claimants from 2001 to 2013. The final indicator has a
correlation of ≥0.94 in all years.

Household Estimates {#sec-6-2}
-------------------

The details of the procedure for estimating numbers of households in
each LSOA and Datazone in each year is as set out previously in a SPCC
Research Note[^9]. In overview it entails:

1.  Starting with the annual small-area adult population estimates
    produced by ONS (England and Wales) and GROS (Scotland), banded into
    twelve (ten in Scotland) groups by age band and sex.

2.  From this, deriving the population living in households, by
    deducting the number in each age/sex group living in communal
    establishment in the 2001 and 2011 Censuses

3.  For each age/sex group in each area, calculating the proportion of
    that group who are household representative persons (the “household
    representative rate”) in the 2001 and 2011 Censuses

4.  Multiplying the household population in each age/sex group by the
    household representative rate for that group to give a number of
    households, then summing across all age/sex groups in each area

5.  Constraining the totals for all areas in each local authority to the
    published estimate or projection of households in that area.

Small-area population estimates are not yet available for 2013. The base
household counts in each area are thus instead calculated by applying
the four-year (2008 to 2012) annualised change rate for that area to the
2012 total before constraining to local authority projections as normal.

For all three countries, the latest national household projections based
on 2011 Census results are used. In Scotland, these indicate an
implausibly large shift from recent trends in some areas, and so they
are averaged with the results of a straight-line projection. In England,
the latest local authority household figures do not include values for
2002 to 2010. These are interpolated per local authority, with the rate
of change between 2001 and 2011 proportional to the all-England change
in population.

Geographic matching of main datasets {#sec-6-3}
------------------------------------

The source datasets are used with the following geographic
transformations or re-codings:

2001 Census data (households, residential establishments)
:   summed from 2001 OAs to included 2011 LSOAs by one-to-one matching

2011 Census data (households, residential establishments)
:   direct 2011 values used, summed to included LSOAs

DWP Benefits
:   apportioned from 2001 LSOAs to 2001 OAs, then re-summed to 2011
    boundaries

Inter-censal small-area population estimates
:   Now published by ONS to 2011 boundaries in England and Wales

All Scottish data
:   consistent 2001 Datazone estimates used, five deleted Datazones are
    omitted

For these, the approach taken to converting data from 2001 to 2011 LSOA
geographies is to work with data at the smaller 2001 OA level, and then
match these more precise boundaries to the larger 2011 LSOAs. Even where
boundaries have been redrawn due to housing and population change, the
large majority (≈90%) of OAs affected are unambiguously assigned to a
single 2011 LSOA. [^10]

A small number of 2001 OAs have been split into multiple 2011 LSOAs.
This normally occurs where large quantities of housing have been
completed and occupied on sites that had in 2001 few or no occupied
dwellings. For these areas there is insufficient available to attribute
data to one of the resultant LSOA.

### Allocation of Output Areas to LSOAs {#sec-6-3-1}

The 2001 Output Areas are matched to 2011 Output Areas (which may be
one-to-one, one-to-many or many-to-one). These are then matched onto
2011 LSOAs (which is always one-to-one). The matching is done as
follows:

-   2001 Output Areas belonging to only one 2011 LSOA are attributed to
    it. These correspond to those labelled **A** or **B** in the UMBR
    dataset

-   2001 Output Areas belonging to two or more 2011 LSOAs within the
    same MSOA: the LSOAs are combined and reported together. These
    corerspond those labelled **C** or **Cx** in the UMBR dataset

-   2001 Output Areas belonging to two or more 2011 LSOAs in two or more
    MSOAs: assigned to an LSOA in the MSOA where the majority of the
    corresponding 2011 Output Areas lie, combining 2011 LSOAs if
    necessary. In the case of ties (4), random assignment is used. These
    are labelled **B**, **C** or **Cx** in the UMBR dataset.

This produces 1) a table linking every 2001 Output Area to a 2011 LSOA
and 2) a table for all 2011 LSOAs, with the LSOA code where its data is
enumerated (in most cases, itself). These tables are available on
request.

### Deleted Scottish Datazones {#sec-6-3-2}

The following five Scottish 2001 Datazones are omitted in the 2011
Census results, and thus do not appear in UMBR-2014.

-   `S01002296` (Edinburgh)

-   `S01003505`, `S01003031`, `S01003319`, `S01003548` (all in Glasgow).

The population estimate report[^11] reports three of these as having 0
population in 2011, with the two remaining having falls of \>50%. These
are thus areas which as a result of falling population no longer have
any Census Output Postcodes assigned to them by the best-fit method used
in Scotland to produce the Datazone Census figures.

Licence and Reuse {#sec-7}
=================

The accompanying data are adapted from data from the Office for National
Statistics licensed under the Open Government Licence v.3.0.

The package includes the R script files used to produce UMBR from the
source datasets. These are in the folder ’R’. They may freely under a
liberal licence (see below) to reproduce or adapt any part of UMBR.

The dataset was prepared using R 3.1.1, and requires the use of the
`data.table=[fn:7] package in particular to handle the relatively large datasets efficiently. The tables and figures in this *README* file can be recreated using the accompanying R or emacs-org files (=umbr-14-README.R`,
`umbr-14-README.org`).

Licence of files {#sec-7-1}
----------------

These files are released under the MIT licence, and are Copyright (C)
2015 Alex Fenton.

*Permission is hereby granted, free of charge, to any person obtaining a
copy of this software and associated documentation files (the
“Software”), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:*

*The above copyright notice and this permission notice shall be included
in all copies or substantial portions of the Software.*

*The software is provided “as is”, without warranty of any kind, express
or implied, including but not limited to the warranties of
merchantability, fitness for a particular purpose and noninfringement.
In no event shall the authors or copyright holders be liable for any
claim, damages or other liability, whether in an action of contract,
tort or otherwise, arising from, out of or in connection with the
software or the use or other dealings in the software.*

[^1]: The *Social Policy in a Cold Climate* programme was funded by the
    Joseph Rowntree Foundation, the Nuffield Foundation and the Trust
    for London.

[^2]: Available for download at <http://eprints.lse.ac.uk/46449/>

[^3]: The principles and empirical investigations underlying UMBR are
    set out in detail in Fenton (2013) *Small Area Measures of Income
    Poverty* CASE Paper 174 <http://eprints.lse.ac.uk/58053/>.

[^4]: The 2011 Census Area Classification has been recently released,
    but is currently only available for the Output Area Geography in
    England and Wales.

[^5]: The minimum and maximum household and population sizes for an LSOA
    are given at
    <http://www.ons.gov.uk/ons/guide-method/geography/beginner-s-guide/census/super-output-areas--soas-/index.html>

[^6]: Office for National Statistics (2012) *Changes to Output Areas and
    Super Output Areas in England and Wales, 2001 to 2011*
    <http://www.ons.gov.uk/ons/guide-method/geography/products/census/>

[^7]: Scottish Government (2011) *Datazone Consultation Response*
    <http://www.scotland.gov.uk/Topics/Statistics/sns/SNSRef/DZresponseintro>

[^8]: <http://www.nomisweb.co.uk>

[^9]: A Fenton (2012) *Post-censal household estimates for small areas*
    <http://sticerd.lse.ac.uk/dps/case/spcc/rn003.pdf>

[^10]: It is likely more accurate conversions could be made by analysis
    of the postcode records and postcode-level houshold and population
    data from the Census. This is the technique employed by Geoconvert,
    where, however, no conversions to 2011 geography are yet available.
    The complexity of such conversion is, however, beyond the scope of
    UMBR, and also leaves unresolved uncertainties about the
    inter-Censal years.

[^11]: National Records of Scotland (2013) *2011 Census Reconciliation
    Report – Small Area Population Estimates (SAPE) Scotland: Explaining
    the difference between the 2011 SAPE rolled-forward from the 2011
    Census and the 2011 SAPE rolled-forward from the 2001 Census*
    <http://www.gro-scotland.gov.uk/statistics/theme/population/estimates/special-area/sape/>.
    See esp table 3.5.
