Privatization, Power and Place: The Distributive Politics of Asylum Dispersal in England

We argue that governing parties can use privatization as a political discipline mechanism to reward core constituents and swing voters by diverting unwanted social and economic costs to other places. We test this argument by analysing the dispersal of asylum seekers across English local authorities before and after the Conservative Party-led privatisation of the dispersal system in 2011. Our findings suggest that asylum dispersals to Labour Party “ core ” constituencies increased following privatization, but that dispersals to “swing” districts were unchanged. Dispersals to places with high institutional capacity decreased, despite contractors being expected to settle asylum seekers in such areas.


Introduction
Scholars have long recognised that electoral politics can play a decisive role in determining the geographical allocation and distribution of public goods (e.g Lasswell, 1936;Olson, 1965;Johnston, 1977).Defined as "projects, programs, and grants that concentrate the benefits in geographically specific constituencies" (Weingast, Shepsle and Johnsen, 1981, 645), "distributive political decisions" are usually intended to boost governing politicians' prospects of re-election by allocating goods to their "core constituents" and/or "swing voters" (Golden and Min, 2013).The dynamics of such "pork-barrel" politics occur across multiple levels of territorial governance (see Dellmuth, Schraff, and Stoffel, 2017), and are found within multiple policy areas (Golden and Min, 2013).Despite growing interest in the salience of distributive political decisions (Livert and Gainza, 2018;Luca and Rodriguez-Pose, forthcoming;Palmer-Rubin, 2016;Rodriguez-Pose, Psycharis and Tselios, 2016a;2016b), surprisingly little research systematically investigates the distributive dimension of the geographical allocation of social and economic costs.Moreover, to date only limited attention has been paid to the variety of institutional mechanisms through which distributive goals can be achieved.Drawing on the core and swing voter models of distributive politics, our main contribution in this paper is to investigate the role that privatisation can play as a political discipline mechanism in ensuring that partisan opponents to the governing political party are burdened with unwanted social and economic costs, and that areas prone to close election results are not.
According to the core voter model of distributive politics, governing parties reward constituencies that are loyal to them by approving more grants, higher expenditures, and lower taxation in the places in which their support is strongest (Cox, 2009).By contrast, the swing voter model of distributive politics indicates that governing parties may seek to induce moderate and nonpartisan voters to support them by distributing benefits to marginal constituencies (Lindbeck and Weibull, 1987), and in doing so take the loyalty of core constituencies for granted (Dixit and Londregan, 1996).Both models assume that national policy-makers have the capacity to make meaningful decisions about the geographical distribution of benefits.Although the swing voter model is sometimes thought to be especially applicable to majoritarian political systems, such as the United Kingdom (Ward and John, 1999), support for the core and swing voter hypotheses has been found in majoritarian systems, such as the United States (Dixit and Londregan, 1996), and in systems that utilise proportional representation, such as Italy (Golden and Picci, 2008) and Turkey (Luca, 2018).
For both models of distributive politics, the propensity for ruling parties to target benefits towards a core or swing constituency "could just as well be taken as [signifying] negative effects on opposition constituencies" (Asher and Novosad, 2017).Building on this idea, we propose an extension to the core and swing voter models to suggest that a governing party may also reward "loyalists" and "swing voters" by diverting unwanted social and economic costs away from them and instead imposing them on places populated by "partisan opponents".We further propose that centralized privatisation processes can serve as a political discipline mechanism that facilitates distributive politics of this kind (Bertelli and John, 2010).Firstly, by extending the formal rule-making authority of the governing party to monitor and manage the effects of incipient marketization.Secondly, by establishing armslength arrangements for regulating local policy choices that appear to de-politicise policy decisions (Foster et al., 2014).
To investigate whether governing parties might use privatization as a political discipline mechanism, we examine the dispersal of asylum seekers across English local authorities during the period 2004-17.The 1999 Asylum and Immigration Act introduced the forced dispersal of asylum seekers across the UK to reduce the numbers of migrants claiming welfare benefits and living and working in London and the South East of England.To effectively manage dispersal, local authorities collaborated together in regional consortia to contract with the Home Office to provide asylum seeker accommodation.In 2011, the Conservative-led coalition government privatised the dispersal system by awarding Commercial and Operational Managers Procuring Asylum Support Services (COMPASS) contracts for the provision of asylum accommodation to three companies: Clearel, G4S and Serco.Under this scheme, local authorities decide whether to opt-in to the services offered by the company operating in their region.The switch from a locally-managed partnership approach to one managed through centrally-appointed contractors enables us to examine whether privatization can facilitate the diversion of unwanted social and economic costs to core partisan opponent constituencies and away from swing voter "battleground" areas.
To analyse the potential for privatisation to serve as a mechanism facilitating distributive political decisions, we employ count regression techniques.Specifically, we use Zero Inflated Negative Binomial models, which account for over-dispersion and an excess of zeroes in our dataset (i.e.local authorities receiving no asylum seekers at all) (Cameron and Trivedi, 1998).By focusing on the sheer numbers of asylum seekers dispersed annually to local authorities across England, we can identify geographical patterns of dispersal before and after the privatization of the dispersal system in 2011.In particular, because local government in England is largely a two-party political system, we are able to test the core and swing voter models of distributive politics.We do so by estimating levels of asylum dispersal to: i) local authorities firmly in the control of the main opposition party (the Labour Party) versus dispersal to authorities controlled by the governing parties responsible for the privatised dispersal system (the Conservative Party and Liberal Democrats); and, ii) local authorities that are "swing districts", with a narrow margin of victory for the ruling party versus dispersal to all other authorities.Furthermore, we include several key variables likely to influence dispersal in our regression models: house prices; housing availability; institutional capacity; foreign-born population; refugee charities; population size; urban location; and region.
Our analysis suggests that the dispersal of asylum seekers to Labour-controlled authorities has grown sharply under the privatized system.By contrast, dispersal of asylum seekers to places with strong institutional capacity and better housing availability decreased following privatization, even though private contractors are formally required to prioritize such areas and can cut their costs by doing so.Further analysis indicates that dispersal to core Labour constituencies increased following privatization, but that swing voter districts accommodate no more or less asylum seekers than other areas.These findings provide support for the idea that centralized privatization can serve as a political discipline mechanism for governing parties' engaging in distributive politics.

The Politics of Privatization
Much of the existing scholarship on the motivations behind privatization has focused on the technical criteria governments utilise to explain the contracting out decision, especially arguments about the potential for achieving efficiency gains through competitive tendering (e.g.Alonso, Clifton and Diaz-Fuentes, 2017;Domberger and Jensen, 1997;Pinch and Paterson, 2000).In general, contracting out is assumed to result in such gains because private contractors have an incentive to cut costs in order to improve their profit margin (Savas, 1987;Shleifer and Vishny, 1994).Nevertheless, despite the on-going salience of the economic criteria that shape privatisation programmes, government contracting also entails a variety of political considerations that go beyond a simple preference for cost-savings (Brown, Potoski and Van Slyke, 2006;Wallin, 1997).
Besides ideological influences on the contracting-out decision (see Bel and Fageda, 2017), ruling political parties may choose to privatize public services to achieve strategic political aims, such as strengthening their support among core voters or appealing to the preferences of swing voters (Biais and Perrotti, 2002).For example, in the UK, one objective of the Conservative government's 1980 right to buy legislation was to convert Labour-voting council tenants into property-owning Conservative voters (Williams, Sewel and Twine, 1987).More generally, privatization schemes can be regarded as a means for political parties to achieve distributive policy outcomes that are consistent with elite interests (McAllister and Studlar, 1989).Hence, it is conceivable that privatization can be used by governing parties (and political elites) as a means to punish and discipline as well as reward certain constituencies.
In theory, privatization and the contracting out of public services rely upon the assumption that the market for public goods and services can allocate the costs and benefits associated with public policies more efficiently than the state (Savas, 1987).In practice, public service markets are imperfect and public contracts are incomplete (Brown et al., 2006), which, in turn, incentivises various forms of opportunistic behaviour by contractors (Koning and Heinrich, 2013).In particular, contractors may seek to "park" difficult tasks and opt to fulfil contracts in the most technically (and politically) straightforward and cost-efficient way (Bruttel, 2008;Finn, 2008).From this perspective, central control of privatization processes can be used to incentivise the distribution of unwanted social costs and burdens away from constituencies that support (or could support) governing parties and towards the places in which partisan opponents reside.For policy fields in which there is an electoral connection between policies and voting behaviour, privatisation might thereby serve as a political discipline mechanism signalling that loyalists to the government are rewarded and adversaries punished: an approach that could also see disciplinary efforts divert costs away from undecided swing voters.
According to Bertelli and John (2010), a political discipline mechanism is "a series of bilateral relationships operating though rules, incentives, and anticipated reactions that permit distributive politics to enter a perceived area of technocratic decision making" (p.546).The central privatisation of policies that are ultimately implemented at the local level therefore seems especially likely to make a political discipline mechanism possible.In particular, it ensures that the superordinate (national) government can exert more influence over outcomes with potential electoral consequences that were previously the preserve of subordinate (local) governments.In addition, centralized privatization processes enable governing parties to establish greater control over local policy choices by establish arms-length arrangements that regulate and constrain those choices in ways that appear to be de-politicised (Darling, 2016a).
Both these features of a political discipline mechanism seem to be applicable to the case of asylum privatization in the UK.

Asylum Dispersal in England
To establish a more equitable distribution of the perceived "burden" of accommodating and supporting refugees across host communities, national governments across Europe have implemented a range of initiatives intended to divert asylum seekers away from large cities (Robinson and Andersson, 2003).While the European Union has endeavoured to create a common approach to the distribution and dispersal of asylum seekers (Moraga and Rapoport, 2015), the systems of dispersal in operation vary considerably across the continent.Some countries, such as France and Sweden, are more committed to giving refugees the opportunity to select whichever community they prefer, whereas, in others, the central or federal government seeks to play a stronger role in determining the placement of asylum seekers within particular areas (e.g.Denmark and Germany) (Hernes, 2017;Thieleman et al., 2010).
In practice, most countries operate a mixed approach to dispersal through which local administrative units work together with central agencies to host asylum seekers (Asylum Information Database, 2019) that, in some cases, is delivered through partnerships between municipalities and civil society organizations, as in Italy's progetti territoriali (Fratesi et al., 2019).
In the UK, the 1999 Asylum and Immigration Act gave the Home Office the responsibility to enact the forced dispersal of asylum seekers across the UK.This policy was intended to reduce the numbers of migrants residing in London and the South East of England, and to formally recognise asylum seekers' entitlement to means-tested support for accommodation and/or subsistence co-ordinated by their host local authority (Hynes, 2011).
To effectively manage the dispersal of asylum seekers in the wake of the 1999 Act, local authorities joined regional consortia to contract with the Home Office for the provision of refugee accommodation.Nevertheless, concerns about the costs of the dispersal system, led the Labour national government of the time to review the potential for introducing competitive tendering for accommodation services.This culminated in the launch of the Commercial and Operational Managers Procuring Asylum Support Services (COMPASS) project by the UK Border Agency in July 2009 (National Audit Office, 2014).
Following the Labour Party's exit from office in 2010, the Conservative-Liberal Democrat coalition government brought the COMPASS project to completion, awarding the first central contracts for the provision of asylum accommodation services in 2011.This privatised asylum dispersal regime is unique in Europe, and is rare in countries elsewhere in the world.For example, contracting out of asylum dispersal is a core feature of the Australian Department of Home Affairs' Humanitarian Settlement Program, but dispersal services under that scheme are undertaken by non-profit rather than profit-making service providers.Three private sector companies specialising in outsourcing solutions for government successfully bid for COMPASS contracts to co-ordinate refugee accommodation across large tracts of the UK: Clearel (London, South East England, South West England, Wales); G4S (East of England, the Midlands, North East England, Yorkshire and Humberside); and Serco (North West England, Northern Ireland, Scotland).
Under the COMPASS scheme, the companies act as a prime contractor, with local authorities and voluntary organizations, in principle, deciding whether or not they wish to opt-in to the property management services offered by the company operating in their region.
However, in practice, local authorities already accepting asylum seekers have rarely opted out due to Home Office pressure for them to continue, and few authorities have chosen to opt-in, as they have not been encouraged to do so by the Conservative-Liberal Democrat and Conservative governments (see Darling, 2016).In fact, the Home Office has ignored the advice of the Home Affairs Committee that it mandate the involvement of all local authorities in dispersal, even though the Home Secretary retains the right to compel the co-operation of authorities (Home Affairs Committee, 2018).Moreover, the Home Office has repeatedly rebuffed recommendations that it involve local authorities more closely in managing and monitoring the provision of asylum accommodation (Independent Chief Inspector of Borders and Immigration, 2018).All of which contributes to the impression that the Home Office ultimately guides patterns of dispersal in consultation with the contractors.As a result, the privatised system can be seen as a process of 'governmental depoliticisation' that can advance the "state's agenda through the buying in (or buying off) of other organizations" (Foster et al., 2014, p.239).
The first tranche of COMPASS contracts was awarded on a five-year basis, with a possible two-year extension based on satisfactory performance (National Audit Office, 2014).
Contractors were formally required to consider a range of community factors alongside cost when proposing properties to be used for dispersal accommodation, and to negotiate with local organizations to establish a well-functioning property supply chain.In doing this, contractors' efforts have been measured against a series of Key Performance Indicators, especially the speed with which asylum seekers were allocated accommodation, and the quality of the properties to which they were dispersed (National Audit Office, 2014).
According to theories of contracting, performance-based monitoring would be likely to encourage contractors to "park" asylum seekers in those areas in which local institutional capacity is strongest and housing costs lowest to increase efficiency (Koning and Heinrich, 2013).However, because asylum dispersal is of such high political salience (e.g.Innes, 2010;Robinson and Andersson, 2003), and because the Home Office has so much influence over authorities' involvement in dispersal, distributive political considerations seem highly likely to be brought into play.In particular, the governing party may be reluctant to oblige "loyalist" local authorities to accept (more) asylum seekers, and there are only a few examples of Conservative-led authorities being required to even consider doing so (see https://www.bbc.co.uk/news/uk-england-york-north-yorkshire-44687424).Similarly, the national government has good reason to avoid compelling dispersal to swing voter areas.
Privatisation has thereby offered a mechanism through which decisions with potentially important distributive outcomes can appear to be depoliticised, but are actually just shifted to a new "venue" through which elite interests still control outcomes (Menz, 2010).
Because the national governing party is responsible for the contract specification, monitoring and renewal of asylum dispersal contracts in the UK, prime contractors have a strong incentive to place asylum seekers in areas that maximize the political advantage of the governing party.For instance, the renewal of Serco's contract for asylum dispersal services in 2019 was the largest government contract ever won by the company (https://www.serco.com/media-and-news/2019/serco-awarded-uk-asylum-support-services-contracts-with-an-estimated-value-of-19bn-sercos-largest-ever-contract-award).It was a contract that was also open to tender for a mere twenty-nine days (https://ted.europa.eu/udl?uri=TED:NOTICE:461664-2017:TEXT:EN:HTML ).
While there may be efficiency-based reasons for contractors to target placing asylum seekers in Labour-controlled areas, the Home Office has done little to require Conservative local authorities to participate in the dispersal scheme.At the same time, despite a past history of welcoming asylum seekers, Labour local authorities are not necessarily easy partners with whom contractors can work, with many threatening to withdraw from the dispersal scheme due to the lack of support from the Home Office and concerns about the quality of the service provided by contractors (Brady, 2018).Systematic analysis of the geographical distribution of asylum seekers before and after the introduction of the COMPASS contracts can therefore shed valuable light on whether dispersals have been determined by conventional economic incentives to "park" difficult clients in low-cost locations or if, indeed, privatization is serving as a political discipline mechanism through which distributive policy outcomes are being achieved.Hence, to build knowledge around the distributive politics of asylum dispersal, we test both the core and swing voter hypotheses.
Westminster-style cabinet system of political management, which is typically composed of senior members of the ruling political party.They receive most of their income from UK central government, and are responsible for co-ordinating the provision of social housing and a range of housing-related welfare benefits that form the bedrock for the local state-led support that can be made available to dispersed asylum seekers.The timeframe of our study covers most of the years in which the UK asylum dispersal system has been in place.Due to data availability constraints we are, unfortunately, unable to analyse asylum dispersals between 2000 and 2004, when the system was first established.The dispersal decisions made during this period, will have played a large part influencing the legacy of openness to asylum seekers in some local areas rather than others.It should also be noted that in April 2009 the two-tier element of the English local government system in rural areas experienced a relatively significant process of restructuring, with nine new unitary authorities created from the consolidation of a number of lower tier district councils. 1Pre-2009 data for all nine new unitary authorities is therefore based on aggregates 1 In five areas (Cornwall, Durham, Northumberland, Shropshire and Wiltshire) the districts were merged with the former county council which became the new unitary authority.The four remaining areas were formed as follows: in Cheshire, Congleton, Crewe and Nantwich, and Macclesfield merged to become Cheshire East, while Cheshire West was formed from Chester, Ellesmere Port and Neston, and Vale Royal; in Bedfordshire; of data from the district councils existing before the 2009 reorganization (see also Alonso and Andrews, 2018).

Dependent variable
The dependent variable is an annual count of the number of asylum seekers in dispersed accommodation by local authority.Data on the number of asylum seekers was obtained from the UK Home Office.Data sources and descriptive statistics for all the variables included in our analysis are reported in Table 1.East, West Midlands and Yorkshire, such as Liverpool, Manchester, Newcastle, Birmingham and Leeds.Under COMPASS it seems that, on average, there is some continuity in the geography of dispersal, though local areas in the North West of England, in particular, seem to have taken an even higher number of asylum seekers under COMPASS.To illustrate these regional differences, the dispersal trends are depicted by region in Fig. A1 (in the Appendix).This plot suggests that there are indeed substantial differences between regions, both before and under COMPASS.While the number of asylum seekers remined stable (and low)

Independent variables
For our first model, we include a dichotomous variable, labelled Labour Control, which takes a value of 1 if the Labour party controls the local authority and 0 otherwise (model 1).Our basic assumption is that local authorities ruled by the Labour Party will take more asylum seekers than those with non-Labour party rule in the period after the Conservatives took office alongside the Liberal Democrats.Fig. 2 indicates that this seems to be the case, and that Labour authorities were already accommodating more asylum seekers than other authorities prior to the introduction of the COMPASS contracts, perhaps reflecting a legacy of "municipal socialism" in some areas (Mynott, 2014).
To test the "core" and "swing" voter hypotheses, our second model includes two political dummy variables instead of one.In doing so, we first construct a dummy variable labelled Swing district which takes a value of 1 for districts having a share of Labour seats within a ± 6.2 percentage points margin to the 50% threshold.To select the margin above and below the 50% threshold to consider a district as a swing voter "battleground", we have computed the difference in seat shares between the first party (winner) and the second (main contender) for each local authority, and chose the value of the 10% lowest decile (i.e, the decile closer to the 50% threshold) as the baseline margin to define a district as a swing voter "battleground".The second political dummy variable, labelled as Labour core constituency, takes a value of 1 if the Labour Party controls the local authority and that local authority is not among the swing districts.

[FIGURE 2]
Besides the political context, previous research suggests that market logics, such as efficiency and cost-cutting, have shaped the asylum dispersal system in the UK since its implementation in 2000 (Darling, 2016a;2016b).This reliance on market logics and contractors' predilection for "parking" might have led to greater accommodation of asylum seekers in low-cost places (Darling, 2016b).We attempt to account for this possibility by including in both models 1 and 2 the deflated median housing price in each local authority at the year's end.
To account for the impact of local capacity on dispersal decisions we include in our models two different indicators: i) the availability of housing accommodation; and, ii) the administrative capacity available to provide support to asylum seekers.To measure the former, we gauge the number of vacant dwellings per one thousand inhabitants for each local authority and year.As a proxy for local authorities' capacity to support asylum seekers, we deploy an index of administrative capacity which captures the resources devoted by local authorities to central services (e.g., finance, internal audit) and management and support services (e.g., human resources, IT).This indicator gauges the volume of administrative reserves potentially available to better manage dispersal, and is constructed by normalizing the expenditure on central administration services in a 0,1 range for each year, with a higher score indicating higher capacity.Formally, , where cs represents expenditure on central administration per capita.2

Control variables
Our estimates are adjusted for a set of covariates other than political control, housing prices and local capacity that might explain asylum seekers' dispersal.First, we attempt to control for the possibility that some areas are intrinsically more open and welcoming to immigrants.
To do so, we include in our models two additional variables that might capture the relative degree of "welcome" for asylum seekers in an area.Firstly, to proxy for a history of openness to asylum seekers, we add a variable constructed from census data gauging the percentage of local residents whose country of origin is located in Africa, the Middle East or Asia, labelled "foreign-born population".Secondly, to measure the strength of the infrastructure of support made available by the third/voluntary sector, we include in our models the number of charities focused on serving asylum seekers operating in each area.To create this variable, we followed the approach of Mayblin and James (2019) and searched the Charity Commission's Register of Charities database for organizations whose charitable activities included "working for the prevention or relief of poverty" or "providing accommodation/housing" to "asylum seekers" and "refugees".
As an additional adjustment, we include in our models the population of each local authority.The Home Office established a general limit of one asylum seeker per 200 people, hence we anticipate a positive correlation between the number of asylum seekers and the population of each local authority.Next, we account for potential differences in asylum seekers' dispersal between urban and rural areas, by including in our models a dichotomous variable coded 1 for local authorities serving urban populations and 0 for those serving rural populations.To adjust for potential spatial effects from neighbouring local areas, we include in all models the number of asylum seekers in adjacent districts, computed using a rownormalized spatial contiguity matrix (common border between local areas).In addition, we adjust for potential regional disparities by including regional fixed effects in all models.
Finally, we include a time trend variable to adjust for temporal patterns in our data. 3 TABLE 1]

Methods
We fit our data by means of a Zero Inflated Negative Binomial (ZINB) regression model for two main reasons.First, preliminary analysis suggests that the variance of the dependent variable, i.e. the count of asylum seekers dispersed, is much larger than the mean (see Table 1), a condition known as over-dispersion.Second, there is clearly an 'excessive' number of zeroes in our data set (see Figure A2 in the Appendix).ZINB models extend the singleequation negative binomial (NB) model by introducing a two-stage process which can incorporate a potential excess of zeroes in the data-generating process (Cameron and Trivedi, 1998).
In our case, the first stage of the model includes a logit regression model predicting the probability of whether asylum seekers would be accommodated at all in a given local authority during the period under analysis.It seems plausible to expect that the potential factors explaining the presence of "always zeroes" in our data set, would be very similar to

Baseline results
We begin our analysis by focusing on the whole dispersal period, that is, without distinguishing between observations before and after COMPASS.Table 2 reports ZINB estimates for models 1 and 2. Starting with the logit part of the model, it should be highlighted that the logit part of a ZINB model predicts the probability of membership in the always zero group (i.e., those local authorities who would not accommodate asylum seekers at all), which means that a negative coefficient should be interpreted as an increase in the probability of taking asylum seekers, and a positive coefficient should be interpreted as a decrease in the probability of accommodating asylum seekers.Our results suggest, first, that the political ideology of the ruling party influences the likelihood of a local authority refusing to participate in the dispersal system.In particular, the coefficient for Labour control is negative, suggesting that local authorities with left-wing party rule appear (on average) to be more likely to agree to take asylum seekers, when compared to those with any other form of political control.By contrast, the coefficient for housing prices is positive and, thus, negatively correlated with the likelihood of taking any asylum seekers.Consistent with our expectations, the number of vacant dwellings and administrative capacity increase the probability of participation in the dispersal system.Likewise, foreign-born population, the number of refugee charities, population and urban location are associated with acceptance of at least some asylum seekers.
The results of the NB part of the model predicting the number of accommodated asylum seekers are also reported in Table 2. Starting with the political variables, our results for model 1 suggest that, for the whole dispersal period for which data are available, Labour party control of a local authority is associated with a predicted increase in the number of accommodated asylum seekers by a factor of 1.28 (exp(0.25)),i.e. an increase of about 28%.
Turning our attention to Model 2, the estimates suggest that, for the whole period under analysis, the difference in the logs of the number of asylum seekers is expected to be 0.29 units higher (an increase of about 34%) for Labour controlled districts with a percentage of cabinet seats higher that 56.2%, i.e. those districts labelled Labour core constituency, while holding the other variables constant in the model.The coefficient for the variable Swing district is also positive, though the magnitude of the standard error suggests that our data are consistent with no effect over the whole dispersal period.
Regarding the second group of independent variables of interest, and consistent with our expectations, housing prices are negatively correlated with the count of asylum seekers for each local authority, and the number of vacant dwellings is positively correlated with the predicted number of asylum seekers.In particular, for a one unit change in housing prices (£1,000), the difference in the log of the number of asylum seekers is expected to decrease by 0.008, a decrease of about 1%, a finding in line with previous studies suggesting that asylum dispersal in the UK was focused in areas of low-cost housing (see, e.g.Phillips, 2006).By contrast, the coefficient for vacant dwellings is positive in all of the models (0.04), meaning that an one unit increase in the number of vacant dwellings per 1000 population is associated with an increase in the count of asylum seekers of about 4%.That said, the measure of administrative capacity does not seem to be related to the number of asylum seekers hosted in each local authority, since the magnitude of the standard error suggests that our data are consistent with no effect over the whole dispersal period.
Finally, conditional on our models and data, there seems to be a positive correlation between three of our adjustment covariates (population, the percentage of residents from the Middle-East, Asia, and Africa, and urban area), and the number of asylum seekers' hosted by a local authority.Our estimates also suggest that positive spatial correlation is present.

Determinants of asylum seekers' dispersal before and after COMPASS
We now turn our attention to the primary objective of this study, i.e. whether there might be differences in the patterns of asylum dispersal before and after COMPASS.To do so, we reestimate both regression models including a set of multiplicative terms interacting our independent variables of interest with a dichotomous variable which switches on for post 2011 observations, i.e., when the COMPASS contracts came into force.
Starting with model 1, our data are consistent with the existence of a substantial COMPASS effect as regards political control, the coefficient for the interaction term between the COMPASS dummy and the Labour Control dummy being positive (about 0.57), and the robust standard error suggests that these point estimates are statistically different from zero.
Indeed, the expected number of asylum seekers displaced to Labour controlled areas changes by a factor of 1.77 after COMPASS, i.e., an increase of 77%.Moving to model 2 (the "core" and "swing" voter hypotheses), our findings suggest that asylum seekers were dispersed mostly to Labour core constituency districts after COMPASS, with an increase of about 127%.By contrast, the interaction term between the COMPASS dummy and the Swing district dummy is not statistically significant, suggesting that, conditional on the data and model, there were no differences before and after COMPASS in the number of asylum seekers displaced to areas where the Labour Party won or lost by a narrow margin.The latter results, however, could be affected by the choice of the margin to classify local areas as swing districts (i.e., ±6.2 percentage points margin to the 50% threshold).As a sensitivity test, we therefore repeat the analysis using narrower and wider margins to select swing districts (±5% and ±10%).The estimates remain virtually identical across models using alternative margins (see Appendix A; table A5).
[TABLE 3] Regarding the potential influence of market logics and local capacity, the coefficients for the interaction terms reported in Table 3 suggest both, a degree of continuity between the mixed provision of the pre-COMPASS dispersal system and the COMPASS contracts regime, but also substantial differences; on the one hand, our data are consistent with the absence of a COMPASS effect as regards the influence of housing prices.By contrast, our data and models are consistent with the existence of a substantial COMPASS effect as regards vacant dwellings and administrative capacity.More specifically, the coefficients for both interaction terms are negative and statistically significant at the 10% level, which suggests that under COMPASS local capacity has become a less important factor explaining asylum seekers' dispersal.In non-linear ZINB models, interaction effects of continuous variables could be difficult to interpret though.Hence, to further facilitate our results interpretation, we provide in Appendix A, Fig. A3 a pictorial representation of the interactions between the COMPASS dummy and our continuous variables of interest, i.e.
housing prices, vacant dwellings and the administrative capacity of local authorities.These plots illustrate our findings that, while housing prices seem to be negatively correlated with the number of asylum seekers taken by each LA both before and after COMPASS, there are different patterns before and after COMPASS regarding the correlation between the number of vacant dwellings, administrative capacity and the number of asylum seekers.In particular, both slopes are clearly positive before COMPASS, but are nearly flat or negative under COMPASS.
In sum, our findings suggest that pre-COMPASS dispersal occurred, to a certain extent, in areas with higher administrative capacity and spare housing capacity.However, under COMPASS the political dimension of dispersal has become even more important than before, with dispersals to Labour Party-controlled authorities increasing under the privatized systemparticularly in those areas where the Labour Party is strongestwhile dispersals to places with high institutional capacity and housing availability have declined, even though contractors are expected to prioritize settling asylum seekers in such places.It is possible that the decreasing importance of administrative capacity is because local authorities no longer provide accommodation services under the privatised dispersal system, while the diminished role played by housing availability may reflect the growing salience of political considerations.

Discussion
Consistent with the arguments that we develop about the potential for privatization to facilitate distributive political decisions, we find that the centrally-controlled contracting out of the asylum dispersal system in England appears to have resulted in the diversion of social costs to partisan opponents of the governing political party.This finding applies to Labour core constituencies, but swing districts do not appear to be "benefiting" from the privatised dispersal system.Dispersals to core Labour-controlled local authorities have accelerated in the wake of privatization of the dispersal system: a pattern that seems unlikely to be simply the product of a moral commitment to receiving asylum seekers on the part of Labour authorities (see Mynott, 2014).At the same time, technical considerations have become less important determinants of dispersal decisions since the COMPASS contracts were introduced.These findings have important implications for theories of distributive politics and for public policy.
Our study advances distributive politics research by confirming the core voter perspective on the distribution of benefits to loyalist constituencies and extending it to the distribution of unwanted costs to places populated by partisan opponents.Using a measure of the dispersion of asylum seekers across England, a policy that is perceived negatively by voters (Robinson and Andersson, 2003), our findings suggest that distributive political decisions may be at work in shaping dispersal outcomes.We also contribute to distributive politics theories by clarifying the ways in which centralized privatization schemes can be deployed as political discipline mechanisms to achieve distributive policy outcomes.Prior research indicates that performance management regimes can serve as a disciplinary mechanism to distribute subnational political costs and benefits in ways that are advantageous to national governments (Bertelli and John, 2010;Bertelli et al., 2014).Our results imply that centralized control over public service contracting too can be deployed by governing parties as a device for realising distributive aims.In particular, by distributing unwanted social and economic costs to the core constituencies of partisan opponents.The weakening role of the institutional capacity supposed to guide the privatized dispersal system adds further weight to this interpretation of our findings, and highlights that future studies should always compare the relative salience of political and technical considerations.
In addition to the distributive politics literature, our findings can inform scholarship on privatization by highlighting that contractors may play a key role in translating the policy goals of governing parties.We are unable to pinpoint the precise mechanisms through which the UK government may have encouraged or allowed contractors providing dispersal services to target opponent constituencies on this occasion.The distributive dimension of dispersal is not formally acknowledged in the COMPASS contracts, but it may, of course, have informally influenced the award and renewal of the COMPASS contracts.Relationships between business and the Conservative-led Coalition Government were very close (Dommett, Hindmoor and Woods, 2017), with the outsourcing industry becoming an established base of elite power (Froud et al., 2017), albeit one often dependent upon political sponsorship (Greasley, forthcoming).In-depth research dealing with the relationships between the contractors providing asylum dispersal services and governing parties is therefore needed to fully understand the extent to which contractors respond to distributive signals from ruling politicians.
The results of our study highlight some of the challenges that distributive politics can pose for policy-makers at the national, regional and local levels.Scholars and practitioners have long draw attention to the issue of territorial justice in the distribution of public goods and services (e,g, Boyne and Powell, 1991;Kirby and Pinch, 1983;Smith, 1994).Our analysis illustrates that questions of territorial justice are also likely to apply to the distribution of the perceived burdens associated with certain public policies.In a centralized and majoritarian political system, such as that in the UK, it may be especially difficult to guard against the dysfunctional consequences of distributive politics.However, the potential for distributive decisions to distort public policy implementation could be mitigated by restricting the scope for manipulation of the funding mechanisms used to allocate resources to local authorities on the basis of need (John and Ward, 2001).Indeed, English councils have requested that the present funding regime be updated to acknowledge the additional resources needed to manage asylum dispersal (Home Affairs Committee, 2018).
The research also provides valuable lessons for the public organizations responsible for managing social policy implementation at the local level.While distributive policies may place additional and unforeseen "burdens" on local authorities, it is possible that those authorities can re-imagine the ways in which such burdens can be regarded and addressed.
For example, some UK cities have been effective in welcoming the presence of asylum seekers in ways that may be conducive to the growth of a "culture of refuge" (Darling, 2010).
Ultimately, though, the challenge of meeting new and unexpected human needs is likely to require additional human and material resources (Andrews et al., 2013).For English local authorities to increase their capacity to resist the potentially harmful effects of centralized distributive decisions they would therefore seem likely to benefit most from greater revenueraising powers and authority; including, through greater involvement in asylum dispersal decisions.
Overall, our study contributes to the growing literature on distributive politics by providing statistical evidence of the propensity of governing parties to utilise privatisation as a means to achieving distributive policy outcomes.We hope that our study provides a foundation for further theoretical development and empirical tests for other countries, policy areas and outcomes.
Single and lower-tier local authorities in England tend to be bigger than those in other countries and to vary considerably in size, mainly according to whether they serve urban or rural populations.Single-tier authorities correspond to the third level of the Nomenclature of Territorial Units for Statistics (NUTS) regions in the UK, with London and metropolitan boroughs being larger, on average, than unitary authorities (mean population of 274,639.7 compared with 212,362.1).District councils in rural England correspond to the European Union's Local Administrative Unit 1 statistical regions, and though large by European standards are small compared with the other authorities included in our study (mean population of 102,357.1).

Fig. 1
Fig. 1 maps the geography of dispersal both before and under COMPASS.The spatial

Figure 1 .
Figure 1.Average number of asylum seekers accommodated by each local authority before and under COMPASS.

Figure 2 .
Figure 2. Number of asylum seekers by year and political control.

Figure A1 .
Figure A1.Number of asylum seekers by year and region.

Figure A2 .
Figure A2.Frequency distribution of asylum seekers, 2004-2017 in the East, South East and South West of England, other regions such as the North West, West Midlands, and Yorkshire and the Humber exhibit a substantial upward trend under COMPASS, i.e. after 2011.Bedford district council became a unitary authority, with Central Bedfordshire formed by merging Mid Bedfordshire and South Bedfordshire.
Poisson model to check the robustness of our results to different model specifications (reported in Appendix A; tables A3 and A4).
those factors explaining the actual number of asylum seekers that are accommodated by local authorities.Hence, we propose to include in the logit part of our ZINB models the main independent variables of interest, i.e., Labour control, housing prices, and local area capacity, plus the control covariates (percentage of foreign-born population, number of charities, population, and urban area).modeland the Poisson model (see tables 2, 3, A3, and A4).Although both the AIC and BIC point to the appropriateness of the ZINB model, we also estimate a single-equation NB model and a

Table 1
Descriptive statistics (2004Descriptive statistics ( -2017)   )UK´s Home Office, B; Rallings C & Thrasher M, various years, Local Elections in Britain: A Statistical Digest (LGC Elections Centre, University of Plymouth).C: Authors' calculations based on Rallings & Trasher data.D: Homes England.E: Authors´ calculations based on data provided by the Chartered Institute for Public Finance and Accounting.
F: UK Census data H: ODPM, Urban and Rural Area Definitions: A User Guide (ODPM, London).G: Charity Commission (2020) Register of Charities database.

Table 3
Determinants of asylum seekers' dispersal before and after COMPASS: ZINB estimates Regional fixed effects included in all models.Robust standard errors (SE) clustered at the local authority level.

Table A2
Determinants of asylum seekers' dispersal before and after COMPASS: ZINB Regional fixed effects and year dummies included in all models.Robust standard errors (SE) clustered at the local authority level.

Table A3 .
Determinants of asylum seekers' dispersal (2004-2017): NB and Poisson estimates.Regional fixed effects included in all models.Robust standard errors (SE) clustered at the local authority level.

Table A4
Determinants of asylum seekers' dispersal before and after COMPASS: NB and Poisson estimates.Regional fixed effects included in all models.Robust standard errors (SE) clustered at the local authority level.