Geographical distribution of disability living allowance and attendance allowance and income poverty

Abstract

This paper concerns the relationship between the spatial distribution of older beneficiaries of Disability Living Allowance and Attendance Allowance and that of Income Deprivation in England, UK. Using data for 32,482 Local Super Output Areas in England, we used spatial regression methods to investigate whether these allowances would be benefiting more deprived geographical units.

Keywords:

• benefits
• income deprivation
• elderly
• lower super output areas
• spatial econometrics

1. Introduction

In the United Kingdom (UK), there are two social security benefits for people who are severely disable and are assessed as having personal care, including supervision, needs or mobility needs – the Disability Living Allowance (DLA) and the Attendance Allowance (AA). The benefits provide a financial contribution towards the extra costs these people experience as a result of their disability.

AA can be claimed from the age of 65 and DLA can be claimed up to that age, although once awarded, beneficiaries can remain receiving this benefit beyond the age of 65 (around 38% of recipients of DLA are aged 65 or over). Both are tax-free, non-contributory, non-means tested benefits. The higher rate for each benefit stands at £77.45 a week, which can buy approximately 13 hours of home care services.1

Given that all beneficiaries of AA are aged 65 or over and that, as mentioned above, so are 38% of recipients of DLA, in this paper we concentrate on this age group.

Research has noted that disability and living in poverty are closely related (Brewer, Muriel, Phillips, & Sibieta, 2008; Palmer, Macinnes, & Kenway, 2006; Parckar, 2008). Therefore, in this paper we investigate whether there is an association between the geographical distribution of DLA and AA and that of income deprivation among older people. If such an association exists, even though the benefits are not means-tested they would be to some extent alleviating the material and social conditions of some of the older people in the most deprived areas of the country. Or, to look at this issue from a different perspective, if this were the case, reducing or eliminating these benefits would have a more deleterious impact the poorer the neighbourhood.

The paper is structured as follows. Section 2 describes the data. Section 3 presents the methods and Section 4 the results. We include some final remarks in Section 5 whilst in Section 6 we describe the maps.

2. Data on benefits and income deprivation

The UK has four constituent countries: England, Wales, Scotland and Northern Ireland. The most geographically disaggregated level at which data on AA and DLA are available is the Lower Super Output Area (LSOA) (Office for National Statistics, 2010). In total, there are 32,482 LSOAs in England and 1896 in Wales, with a mean estimated population by mid 2010 of 1600 people in each country.2 In Scotland, small area data are produced by units referred to as Data Zones; however, their mean estimated population by mid 2010 was 803 people.3 Northern Ireland only produces data for Super Output Areas (SOAs), not LSOAs, with a mean estimated population by mid 2010 of 2022.4 Hence data by LSOAs in England and Wales are not truly comparable to small area data for Scotland and Northern Ireland.

Data on claimants by LSOA and age group for each benefit as of November 2010 was published by the Department for Work and Pensions (Department for Work and Pensions, 2010). As already mentioned, we used data for all recipients of AA and those beneficiaries of DLA aged 65 or over. To obtain the proportion of claimants for each LSOA as total resident population aged 65 or over, we use the population estimates for mid 2010 available from the Office for National Statistics (Office for National Statistics, 2011).

With regards to income deprivation among older people, we use the Income Deprivation Affecting Older People Index (IDAOPI) for 2010 produced by the Social Disadvantage Research Centre (SDRC) at the University of Oxford (Department for Communities and Local Government, 2012). The IDAOPI represents the proportion of people aged 60 and over living in income deprived households. Income deprived households are those households with adults and children in receipt of benefits for people in low income such as the Income Support, the Income-Based Jobseeker's Allowance, the guarantee element of the Pension Credit or the Child Tax Credit, and also those with asylum seekers in receipt of subsistence support, accommodation support, or both. Importantly for the purposes of this paper, the definition of the IDAOPI does not include the receipt of AA or DLA. The SDRC also produces income (and other) deprivation indices for Wales at LSOA level, but not the IDAOPI.

Given these data considerations, and also that even though social security is defined for the UK, urban regeneration, local economic growth, and other anti-poverty policies have been devolved to the administrations in Wales, Scotland and Northern Ireland and therefore vary across the UK, we restrict the analysis to England.

3. Methods

We followed the principles of reproducible research (Koenker & Zeileis, 2009) which is based on the assertion that effective communication of research depends upon an integrated approach comprising of software, data, empirical analysis and documentation. With this aim, and that of facilitating the replication of the results presented in this paper, we used the open source software R (R Development Core Team, 2007).

Due to the geographical nature of the data, a statistical analysis which did not account for spatial effects would be incurring into model misspecification: as each observation corresponds to a location (in our case, a LSOA), instead of assuming that the data generating process corresponds to statistically independent observations (as in non-spatial regression models), we need to control for the possibility that each observation may depend on the values of its neighbouring LSOAs (i.e. spatial dependence). This is the main assumption behind spatial econometric methods, the approach we followed. The most popular among these methods is the Moran's I test on the residuals of a regression on the variables (here, a regression of the proportion of beneficiaries on the proportion of people living in income deprivation). This test would detect any spatial autocorrelations and would therefore render coefficients accounting for any spatial effects that otherwise would bias the results.

Therefore, spatial autocorrelation models correct the prediction of a dependent variable Y as a function of a number of independent variables Y, which would result from regression analysis:

The correction involves the addition of a weighted average of the values on neighbouring observations (Pace & Barry, 1997).

where α is the autoregressive parameter, D is the (n × n) weighting matrix with 0s on the diagonal, and e is an (n × 1) vector of error terms.

Crucial to this is the construction of spatial weights – that is how neighbourhood is defined and how the effect of distance is modelled. With regards to the definition of a neighbour, there are many approaches, including graph measures (such as Delauney triangulation, sphere of influence, Gabriel graph neighbours or relative graph neighbours), distance thresholds, and k-nearest neighbours. We used the latter.

The weighting matrix D is row-standardised – i.e., it is scaled in such a way that each of its rows sums to 1. The elements of matrix D, that is, the weights, reflect the assumption that the closer the two units, the higher the influence they have on one another. It is further assumed that beyond a certain (Euclidean) distance, the influence becomes negligible – in other words, two spatial units become neighbours if and only if they are located at a certain distance d _i:m from each other. Hence, if w _ij represents the weight of, and d _ij the distance between, one unit i onto another unit j, we define:

By convention, we also assume that an observational unit is not a neighbour of itself (w _ii  = 0).

We initially set the number of neighbours (k) equal to one, which leads to asymmetric neighbours but ensures each unit has at least one neighbour, precluding the existence of unconnected observational units – so-called islands. As Anselin and Le Gallo (2006) indicate, there is little formal guidance as to how to choose the correct number of neighbours and it becomes an empirical matter. Consequently, we also estimated models with other configurations for the matrix, with k = 2, 3 and 4. These alternative definitions of neighbourhood result in increasingly more sparse weighting matrices as k increases.6 In our case, we fail to find any significant differences in the results using the four different weighting matrices. The following section, then, reports those stemming from k = 1.

4. Results

The map shows LSOAs which exhibit both high proportions of older beneficiaries of either benefit and of older people in income poverty. We defined a LSOA as having a ‘high’ proportion of each variable if it falls within the arbitrary cut-off point of the highest 10% concentration of the respective variable along its statistical distribution by LSOA. The map would suggest the presence of a strong association between older people receiving disability-related benefits and older people enduring income deprivation.

Nevertheless, we further estimated a spatial regression model and the results would also indicate that there is a significant spatial effect (rho = 0.201, p-value < 2.2e-16) and that once this is accounted for, there is still a strong, positive relationship between proportion of beneficiaries and proportion of older people in poverty: coefficient = 0.66, p-value < 2.2e-16.

A spatial generalised moments estimator model rendered similar results (coefficient = 0.748, p-value < 2.2e-16).7 The Breusch-Pagan test failed to detect any heteroscedasticity; however, to further check, we ran a robust variance-covariance matrix method regression (Zeileis, 2004) and we obtained statistically insignificant changes in the standard errors of the estimates.

5. Conclusions

There is a greater concentration of DLA and AA recipients aged 65 or over living in deprived areas than in more affluent areas. Nearly 30% of beneficiaries aged 65 and over live in the 20% poorest of areas and approaching two-thirds (62.4%) live in the poorest half of areas. These allowances are therefore benefiting communities where a higher proportion of older people are deprived. Of course, it would be a textbook example of ecological fallacy to infer from this that the allowances are benefiting more older people in lower income, but we can safely conclude that these benefits, although not means-tested, would be partially addressing the geographical inequalities in income of older people across England.

6. Maps

The source for the digital boundaries with which we created the maps is: Office for National Statistics (2004). Crown copyright material is reproduced with the permission of the Controller of HMSO. We used the full resolution Lower Layer SOA boundaries (Extent of Realm) in shapefile format.

LSOAs are built from groups of contiguous Output Areas (typically, between four to six OAs) and are constrained by the boundaries of Standard Table (ST) wards.

Output Areas are geographical units based on groups of postcodes and fit within the boundaries of electoral wards/divisions and parishes as at the end of 2002. There are 165,665 in England.

Standard Table wards are all electoral wards/divisions as at 31/12/2002, except for those with fewer than 1000 residents or 400 households which have been merged into neighbouring wards for confidentiality reasons.

Extent of the Realm is the statutory extent of the administrative areas on which super output areas, including LSOAs, are based. It usually extends only as far as the Mean Low Water mark, but in some instances it extends out to sea to include oshore islands.

As mentioned the spatial format of the file is shapefile. The other boundary file attributes are:

•	Spatial data type: Vector
•	Object type: G-Polygon
•	Object count: 32482
•	Map projection name: Transverse Mercator – British National Grid
•	Distance units: Metres
•	Datum: D-OSGB-1936
•	Spatial extent: Extent of the Realm (Full resolution files

Notes

The unit cost of home care for adults and older people was £17 per hour in 2010–2011. NHS (2012; Table 6.1).

Office for National Statistics (2011).

General Register Office for Scotland (2010, Table 1a).

Northern Ireland Statistics and Research Agency (2011).

In our specification, w _ij  = 1, which results in a binary spatial weighting matrix. More generally, we could have defined w _ij  = (d _ij )2.

The number of non-zero (that is, connected) components in the weighting matrices resulting from each definition of neighbourhood is, respectively: k ₁  = 8047; k ₂  = 162; k ₃  = 11and k ₄  = 4.

Under the k = 4 definition of neighbourhood, for example, we obtain fairly similar results: (rho = 0.282, p-value < 2.2e-16; poverty coefficient: 0.64, p-value < 2.2e-16).

Software

We used the open source statistical software R (R Development Core Team, 2007). The R package maptools (Lewin-Koh and Bivand, 2008) was used to read the spatial points in the shape file for LSOAs and to produce the maps; we used the R package RColorBrewer (Neuwirth, 2007) to colour them.

To obtain the Moran I coefficient we used the R package spdep (Bivand, 2008). We ran the Breusch-Pagan test to check for the presence of heteroscedasticity when the residuals from the original linear model are regressed on the right-hand-side variables with the R package lmtest. Finally, we used the R package sandwich (Lumley & Zeileis, 2010) for the robust variance-covariance matrix method regression.

To create the pdf files for the maps, we used the R package Cairo (Urbanek & Horner, 2009).