Migration and redistributive spending:evidence from local authorities in England

The inflow of migrants can impact public spending through its effect on local preferences for redistribution and through changes in demand for local services brought about by migrants’ distinct characteristics. In this paper, we analyse the effects of the migration wave from Central and Eastern European countries following their EU accession in 2004 on local level redistribution in England, specifically disentangling these two channels. We apply a differencein-differences estimation strategy and find that migrants did not have an effect on local authorities’ total service provision per capita. Once we zoom in on the different expenditure items, we find that local authorities experiencing relatively larger migration inflows saw their spending on means-tested social care services decrease in relative terms, while spending on education services increased. Analysing changes in local Council compositions and internal migration flows in response to the arrival of outsiders, we find no evidence that spending shifts are driven by a change in the local willingness to redistribute income. Rather, our results suggest that, due to migrants’ young age at the time of arrival, migration following the 2004 EU enlargement alleviated some of the pressure social care spending in England faces.

tion wave from Central and Eastern European countries following their EU accession in 2004 on local level redistribution in England, specifically disentangling these two channels. We apply a difference-in-differences estimation strategy and find that migrants did not have an effect on local authorities' total service provision per capita. Once we zoom in on the different expenditure items, we find that local authorities experiencing relatively larger migration inflows saw their spending on means-tested social care services decrease in relative terms, while spending on education services increased. Analysing changes in local Council compositions and internal migration flows in response to the arrival of outsiders, we find no evidence that spending shifts are driven by a change in the local willingness to redistribute income. Rather, our results suggest that, due to migrants' young age at the time of arrival, migration following the 2004 EU enlargement alleviated some of the pressure social care spending in England faces.

Introduction
Concerns about redistributing income to what are considered outsiders has featured as a salient issue in the run up to the 2016 Brexit referendum that ultimately saw the UK leaving the European Union (EU) by popular vote. Both UK transfers to the EU and the pressure EU immigrants allegedly put on public service provision in the UK were platforms the "Vote Leave" campaign heavily relied on to mobilise its supporters [Gherghina and O'Malley, 2019, Goodwin and Milazzo, 2017, Becker et al., 2017.
In this paper, we investigate if the link between the inflow of EU "outsiders" and a local loss of appetite for redistributing income was visible in the UK before the Brexit referendum took place. Specifically, we focus on the time period after the 2004 and 2007 EU enlargement when the historically anchored EU scepticism in the UK took a turn against migration following the granting of free movement rights to citizens from Central and Eastern European countries and the large inflow that followed into UK territory [Gherghina and O'Malley, 2019].
We analyse the effect of the unexpected and spatially heterogeneous migration wave from the 12 post-2004 Accession Countries (AC-12) on English local authority level public spending and revenue to answer the question if the local presence of migrants is indeed associated with less redistributive spending patterns. We combine detailed local authority revenue and expenditure data from the Chartered Institute of Public Finance and Accountancy (CIPFA) with annual data on local authority level migrant stocks calculated from the UK Labour Force Survey/Annual Population Survey (LFS/APS) obtained under a special licence agreement over the 2000 to 2015 observation period. Due to the very low number of AC-12 migrants present in England in the pre-enlargement period, the estimated coefficients we obtain from our two-way fixed effects panel regressions correspond to a difference-in-differences research design, allowing us to recover treatment-on-the-treated effects of AC-12 migration on local authority revenue and spending.
Taken at face value, we find ambiguous results with regards to the hypothesis of a closed in-group that cuts down on redistributing income when faced with outsiders. On the one hand, AC-12 migration did not affect local authorities' per capita spending. However, once we zoom in on local revenue and spending patterns, we find that the presence of AC-12 migration is associated with a decrease in locally generated revenue and the unchanged per capita spending was heavily supported by an increase in funding local authorities received from the central government. AC-12 migration is further associated with both a decrease in means-tested social care spending per capita and an increase in education expenditure per capita, an expenditure item where inter-group transfers are likely to be relatively more salient compared to other non-means-tested services [Tabellini, 2020a, Speciale, 2012.
To further explain these results, we then disentangle local preferences for less redistribution from mechanical changes in demand for local services brought about by the distinct characteristics of AC-12 migrants. Our results show that the strong association of AC-12 migration with a decline in social care and a rise in education expenditure per capita is driven by changes this migrant group caused to local demographics. In fact, social care expenditure per population aged 65 and above, the main recipient population of social care services, increased strongly in response to AC-12 inflows. Similarly, when normalising education expenditure per the rising local number of pupils in areas more strongly affected by AC-12 migration, expenditure remained vastly unaffected. Thus, the effects AC-12 migrants had on local authority expenditure patterns were in large parts due to the shifts to local demographics these migrants caused and the corresponding institutional responses that were triggered by the resulting changes in local service demand. The relative shifts from social care towards education expenditure further explain the observed dynamics on the revenue side: In England, local authority education expenditure is almost entirely funded through central government grants, while a larger share of social care spending comes from locally generated revenue [Phillips, 2018]. It thus appears that migration decreased the pressure local authorities faced on social care spending over our observation period, allowing local authorities to take better care of their vulnerable older population while decreasing the need for raising revenue locally. On the other hand, we do not find evidence that would support the hypothesis of a shift in local preferences towards less redistribution in response to migration inflows, which we test by analysing local voting patterns and native out-migration ("voting with their feet") following the 2004 EU enlargement and the subsequent wave of AC-12 migration. Our results show that a larger presence of AC-12 migrants is associated with a rise in local Council seats held by the more redistributive Labour party (rather than the Conservatives) and a decline in net-migration outflows. We note that the latter result, coined "fiscal externality" by Tabellini [2020b], may also provide a partial explanation for the changes in local demographics associated with AC-12 migration.
England is a suitable test bed for the local level link between migration and public spending for a number of reasons. First, the country experienced large waves of migration in recent decades, including the both unprecedented and unexpected wave of AC-12 migrants following the 2004 enlargement of the European Union [Becker et al., 2016]. These successive waves have led scholars to test the impact of the intensity of migration flows to the UK on numerous outcomes including crime rates [Bell et al., 2013, Jaitman and, house prices [Sá, 2015], hospital waiting times [Giuntella et al., 2018], public budgets and the wages and employment of natives [Card et al., 2012, Manacorda et al., 2012, Dustmann et al., 2013, Becker et al., 2018. In this study, we exploit the large and spatially heterogeneous shock to local migration stocks that stemmed from EU enlargement in 2004 and led to more than one million people migrating from Central and Eastern Europe to the UK. The fact this stemmed from the granting of free movement rights to new EU citizens explains the size and suddenness of the migration wave and distinguishes it from other migration waves because of the rights framework that enabled movement from the AC-12 to England post-2004. Second, the discretion in raising revenue and spending decisions at the hands of local 5 authorities in the UK makes it an appropriate case study. Funded through a mix of central government grants and locally raised revenue, England, and the UK more generally, is one of the European countries where local governments have discretion over spending decisions that encompass several public expenditure items. UK local authorities are responsible for policies concerning education, social care services, highways, roads and transport, housing, cultural services, environmental services, planning and development and protective services [Gavazza et al., 2019, Sandford, 2018, Phillips, 2018. Local authorities are required to balance their budget but can increase or decrease their total spending through steering the local council tax, a property tax. They can further shift their spending between more or less redistributive spending items, with spending on means-tested social care services in particular reflecting the redistribution of income towards the relatively less wealthy. Third, due to the limited scope the UK central government had under EU law to target EU migrants directly over our 2000 to 2015 observation period, the central government could not use restrictive migration policies in the years prior to the Brexit referendum, such that cutting public spending was the only possible response to a decrease in appetite for redistributing resources. Finally, the wealth of information available allows to disentangle local preferences and fiscal externalities from a more mechanical migrant demand channel, a mechanism frequently neglected in the literature when studying the effect of migration on preferences for redistribution.

Contribution to the Literature
Our results contribute to the literature on the impact of migration on redistribution in destination areas. Pioneered by Freeman [1986], Alesina et al. [2001], Alesina et al. [2004] and Easterly and Levine [1997], an important stream of literature argues that redistributive policies are supported more strongly by homogeneous groups. These findings are driven by in-group biases which translate into greater immigrant diversity lowering preferences for redistribution. Following this work, several scholars have analysed the relationship between 6 immigration and preferences for redistribution in the context of migration to the US and European countries with results pointing towards a negative association of the two [Senik et al., 2009, Hopkins, 2010, Speciale, 2012, Halla et al., 2017, Dinas et al., 2019, Steinmayr, 2020. For example, Senik et al. [2009] finds some evidence of a negative association studying Europe as whole. Speciale [2012] leverages the variation in inflows of migrants to EU-15 countries stemming from the 1990s Balkan wars to study the impact of migration on education expenditure and finds a small and negative association between perceived migration and support for the welfare state. Both authors document considerable heterogeneity across countries and stress the importance of sub-national studies to understand the mechanisms at play and allow for a causal investigation. Dahlberg et al. [2012] exploit a refugee dispersal mechanism in Sweden and find a significant negative effect of immigration on the local support for redistribution. In Denmark, Harmon [2018] uses an instrumental variable strategy based on historical housing stock data and finds that greater migration inflows leading to increases in local ethnic diversity shift election outcomes from traditional "big government" left-wing parties towards anti-immigrant nationalist parties. Both studies suggest immigration may lower the level of redistribution or public spending but identify further examination of the effect of immigration and ethnic diversity on more direct measures of redistributive spending as an important topic for future work. Similarly, in more recent work studying the effect of extending the voting franchise to non-natives, Ferwerda [2021] stresses that while evidence points towards European citizens preferring less redistribution with greater migration, the evidence that this leads to a reduction on public good provision is less well understood. Our work contributes directly to filling this gap by measuring the impact of a large migration shock to England on local level redistribution.
Closest to this present work are studies by Tabellini [2020b], Tabellini [2020a], Gerdes [2011] and Jofre-Monseny et al. [2016]. Tabellini [2020b] studies the first "Great Migration" when 6 million black Americans migrated from the South to the North of the US. The author specifically focuses on its impact on local public finances due to changes in racial heterogeneity in Northern US cities between 1915 and 1930. After collecting data on local finances for the years 1910, 1919, and 1930 and deploying a version of a shift-share instrument based on historical settlements similar to Card [2001] and Boustan [2010] to predict black migration, the author finds that larger inflows had negative impacts on both public spending and tax revenues. The author then investigates whether this result is driven by a change in local preferences or by second-order effects black migrants had on out-migration of white Americans. While Tabellini [2020b] acknowledges that these mechanisms are not necessarily mutually exclusive, the author argues that the study's results are rather driven by a negative fiscal externality due to white flight, corroborated by the fact that cities did not change their allocation of spending. In a second related study, Tabellini [2020a] jointly investigates the political and economic impact of European immigration to the US between 1910 and 1930, using a similar shift-share instrumental variable strategy. The author finds that reductions in redistribution stemming from greater migration inflows were more likely driven by natives' preferences and cultural distance. Our study distinguishes itself from these contributions by investigating the role of migration at a time where levels of discretion of municipal councils in Europe differ from the early 20th century US. In addition to the mechanism linking migration to local expenditure and revenue offered by Tabellini [2020a] and Tabellini [2020b], we are able to study a third potential mechanism, namely the change stemming from mechanical demand for local services induced by distinct demographic characteristics of migrants.
Exploiting a refugee placement policy in Denmark, Gerdes [2011] examines the effect of immigration on municipal redistributive spending and does not find any evidence of a change in public sector spending. However, as highlighted in Harmon [2018], the author's empirical strategy might suffer from the endogenous relocation choices of immigrants not covered by 8 the policy as well as the political discretion in assigning migrants to different municipalities.
We tackle the empirical issue of endogenous location choices of AC-12 migrants in England by presenting parallel trends comparing heavily and less heavily affected local authority areas in our main outcomes. We further show that all our results are robust to (i) matching local authorities on a wide range of 2001 Census characteristics and (ii) deploying a shift-share instrumental variable strategy based on historical settlement of AC-12 migrants in the tradition of Card [2001].
Jofre-Monseny et al. [2016] focus on the link between municipality-level variation in extra-EU immigrant density and local social spending in Spain. Instrumenting migration flows between 1998 and 2006 using the distribution of rental housing in 1991, the authors find that per capita social spending increases less in municipalities that experience the largest increases in immigrant density. While the authors report strong first-stage results in their instrumental variable regressions, one main concern with this particular identification strategy relates to the exclusion restriction. Municipalities with a relatively large supply of rental housing six years before the migration inflows are likely to also be poorer and larger than other cities and therefore may lie on differential trends that the authors do not account for.
Municipalities' public finances might be affected in a way that social services spending may slow down six years later for reasons other than migration. A second concern relates to the authors' interpretation of their results. While Jofre-Monseny et al. [2016] do not study the impact of elections on native outflows and electoral outcomes, they interpret their findings as evidence for a materialisation of a shift in preferences for redistribution among the native population. We argue that this interpretation, although in line with predictions of in-groupout-group theories, is only one possibility. A decrease in redistributive public spending could also reflect a mechanical relationship introduced by migrants' socio-economic characteristics and/or their differing propensity to take up social services if inflows are sufficiently large and migrant characteristics are distinctly different from the local population.

9
The remainder of this paper is structured as follows. In the following section 2, we introduce the institutional setting of local authority spending and describe the nature of AC-12 migration flows. Section 3 discusses our data sources. Section 4 lays out our main identification strategy. Section 5 presents and discusses our results. Section 6 discusses alternative identification strategies and robustness. Section 7 discusses the main mechanisms of our results and section 8 concludes.

Institutional setting
In this section, we first provide an overview of the level of discretion local authorities hold over spending and revenue in subsection 2.1. We then describe the nature of AC-12 migration to England in more detail in subsection 2.2.

Local authorities in England
In total, there are 353 local authorities in England. England's local government structure is not homogeneous across the country. Local governments either function under a two-tier or a single-tier regime. The two-tier local authorities consist of an upper-tier, the county councils, and a lower-tier, the district councils. Single-tier authorities encompass 55 unitary authority councils, 36 metropolitan boroughs, 32 London boroughs, the Common Council of the City of London and the Council of the Isles of Scilly for a total of 125. Of these, we exclude nine new unitary authorities from our analyses that were only formed out of tow-tier authorities between 2007 and 2009. Two-tier authorities consist of 27 county councils which in turn are divided into 201 district councils.
While single-tier authorities bear the responsibility for all service spending decisions, county councils and district councils divide responsibilities between themselves in two-tier authorities [Sandford, 2018]. To make local authorities comparable across England, we aggregate all lower-tier authority spending up into the upper tiers. This is unproblematic for two reasons. First, the areas where spending decisions are not clearly distinguishable only concern spending on cultural goods such as museums, transport planning, economic development and tourism. Second, spending decisions on the largest expenditure items such as education and social care -the focus of our study -are made on the upper-tier county council level. We provide more information on how areas of responsibility are divided between the two tiers in B of the appendix.
On the revenue side, UK local authorities obtain their funding via a mix of specific, ring-fenced grant funding, general grant funding, the collection of business rate revenue and income from a local council tax. While there is currently a trend towards more devolution in revenue, particularly regarding the retention of local business rates, the discretion of local authorities was limited to steering the local council tax rate over our main observation period from 2000 to 2015 [Phillips, 2018]. The council tax is a property tax collected by local authorities and its amount is based on the value of property. Each property is categorised into one of eight bands (A to H) and the tax is then due annually as a fixed fraction of local authority defined baseline tax band D. In 2001, the average band D council tax rate stood at GBP 898, rising to GBP 1,459 in 2015. Total tax income collected by local authorities is then simply the multiple of the band D tax rate and the tax base, i.e. the number of band D equivalent dwellings. Over our observation period, council tax income covers about 25% of annual total service expenditure. We summarise the other main sources of income from central government grant and centrally redistributed business rates in a 'central government transfers' measure.
The discretion of the elected Councillors over local authorities' spending is not limited to revenue collected from council tax, as only a small share of funding from central government grants is explicitly ring-fenced [Phillips, 2018]. England, and the UK more generally, is one of the European countries where local governments have discretion over spending decisions that encompass several public expenditure items. UK local authorities are responsible for policies concerning education, social care services, highways, roads and transport, housing, cultural services, environmental services, planning and development and protective services such as fire and rescue [Sandford, 2018]. In this context, it is important to note that, unless local authorities can temporarily draw on previously accumulated reserves, they are required to balance their budgets and are unable to borrow on financial markets.
In 2000, British local public spending represented GBP 113 million, a value that increased to GBP 198 million in 2015 in current prices and represents approximately 25% of total government spending. 1 English local authorities spend by far the largest share of their total service expenditure on education. In 2001, education made for 53% of all service expenditure, a value that has decreased slightly over time. The second largest share of total expenditure is spent on social care services (23.5% in 2001) which has increased over time.
All other expenditure items combined make for less than one fourth of total service expenditure throughout our observation period.
In our analysis, we are particularly interested in means-tested services for which local authorities have discretion during our sample period. Social care services in particular fulfil these criteria while education partially fulfils them.
Social care services in the UK consist of adult social care and child social care. Adult social care entails a range of support services available to the physically or mentally impaired as well as other items where the level of uptake is most highly correlated with advanced age. It also assists disadvantaged groups such as asylum seekers or substance abusers. Adult social care accounts for the bulk of social care service expenditure in England over our main observation period (approximately 70% on average). Social care service minimum eligibility criteria are set by the central government but the amount spent on social care is at local authorities' discretion [Simpson, 2017]. Phillips [2018] notes that Councils' discretion extended to determining what kind of services were offered, needs' assessments and eligibility criteria over the 2000 to 2015 observation period. The latter included different thresholds for what Councils considered the risk for an individual in absence of treatment. A detailed overview of local authority social care services and its means-testing criteria is provided in Table 16 of appendix B.
While there has been a trend towards centralisation, the sample period we study still left room for local authority discretion when it comes to education expenditure. Education expenditure in the U.K. has traditionally been locally-managed although school funding was ring-fenced as of 2006 via the Dedicated Schools Grant (DSG), which de facto represents a minimum threshold below which school spending cannot fall for most of the sample period we study. However, local authorities could still exert upward discretion and use their own revenues to additionally fund education. Recent moves by the Conservative government to remove local governments from the education expenditure formula and only have a national funding formula, whereby schools with similar characteristics receive equal funding, are yet to come into place. As highlighted by Phillips [2018], education expenditures was still partially locally managed in our sample period.
In summary, for the purposes of this paper, two observations are therefore important: First, local authorities do have a significant amount of discretion over both the revenue they collect and the allocation of their funds across different spending areas but need to balance their budgets. Second, 1100 statutory spending requirements limit local authorities in their spending decisions to some extent and do not always allow for a clear distinction between mandatory and discretionary spending [Gray and Barford, 2018]. This means that the larger the level of disaggregation of spending items, the more detailed knowledge of statutory spend-13 ing requirements is necessary. We conduct our analysis on expenditure items aggregated at a relatively high level within the different spending areas to minimize this risk.
A local fiscal response to migration in England that reflects a change in redistribution could thus become visible through two main channels. First, migration could affect total spending per capita and revenue. Second, less redistribution could also become visible through a shift between expenditure items more strongly associated with redistribution and those less associated with redistribution. In the analysis below, we focus on social care spending per capita as the main redistributive item due to its free availability only to low-income individuals. It is less clear to which extent education spending falls under redistribution.
Some authors have argued that inter-group transfers are more salient in education expenditure than in other non-means-tested services [Tabellini, 2020a, Speciale, 2012. In our analyses, we therefore leave education as a separate expenditure item. We then aggregate all expenditure that falls outside of social care and education into a third category. measures regulating access to their labour markets for nationals of A-8 countries for up to seven years" [Kennedy, 2011, p.5].

The AC-12 migration shock
All EU Member States but Sweden, Ireland and the UK applied these regulations on A-8 migrants [Anderson et al., 2006]. Becker et al. [2016, p.11] explain that the decision by Tony Blair's Labour government not to limit labour market access to A-8 migrants was driven by "a thriving economy and a misunderstanding of the consequences of decisions by other big EU countries to keep their borders closed to Eastern European workers for a transition period". their border immediately and estimated that "only around 5,000-13,000 Eastern Europeans would arrive to the UK per year" [Dustmann et al., 2003] (as cited in Becker et al. 2016, p.11). However, between 2004 and 2007, more than 500,000 migrants arrived to the UK from Central-and Eastern Europe, vastly exceeding the initial projections [D' Auria et al., 2008]. 3 Thus, overnight on 1 May 2004, workers from the A-8 accession countries, Malta and Cyprus obtained full rights to live and work in the UK, including the right to stay permanently and the right to be joined by dependants [Anderson et al., 2006]. Workers from A-8 countries (but not Malta and Cyprus) were obliged to register on the so-called Workers Registration Scheme (WRT). Registration on the WRT gave A-8 workers access to in-work benefits but they only had access to out-of-work benefits after they had been in registered employment for 12 months [Kennedy, 2011]. For the purposes of this study, it is important to note that local authority services, including social care and education, were accessible to AC-12 workers as soon as they were registered on the WRT.
2 One of the other motives cited for the Blair government agreeing to immediate free movement rights for A-8 citizens was that the UK saw a wider Europe as a way to provide the UK with new allies within the EU against the traditionally more pro-integration "old-Europe" States (see e.g. Bulmer [2008]) 3 It has been argued that the post-Brexit registration scheme figures suggest that the number of Central/East European migrants might have been under-estimated (see e.g. migrationobservatory.ox.ac.uk) Table 1 summarises the AC-12 migrants' socio-economic characteristics relative to British natives, EU-15 migrants and the rest of the world based on the 2011 UK Population Census.

Characteristics of the AC-12 migrants
The table shows age and employment characteristics that could potentially have had an impact on the local population's attitude towards AC-12 migrants. Both employment and young age may thereby reflect a lower likelihood of welfare dependency [Gherghina and O'Malley, 2019, Goodwin and Milazzo, 2017, Becker et al., 2017. As noted above, these characteristics may also reflect the group's need for local authority services. The migration inflow to the UK stemming from the EU enlargement was indeed both sizeable and characterised by the distinctive features of AC-12 migrants in terms of age, education and employment.
Most of the AC-12 migrants living in the UK in 2011 were in the 25 to 39 years' old categories. This contrasts sharply with the age structure of AC-12 migrants we observe in the 2001 UK Population Census where the distribution was much flatter. Other migrants from EU-15 countries and the rest of the world are also younger than the UK born on average but their age-distribution is less skewed to the left than that of AC-12. Cross-checking these numbers with those from the 2001 Census shows little to no movement. Taken together, this suggests that the inflow of migrants from AC-12 countries was distinctively younger than AC-12 migrants living in the UK pre-enlargement and that this change in the pattern of their age-structure was distinct to this group of migrants.
In terms of qualifications, the bulk of AC-12 migrants living in England in 2011 were categorised in the "apprenticeships and other" qualifications section that does not directly translate into the UK system of qualifications but is indicative of relatively low and medium skills. It is further worth noting that the 24% share in the highest skill category did not translate into a similar share of employment in high-skill professions [Drinkwater et al., 2009]. Most of the residents born in AC-12 countries as per the 2011 census were working in routine or semi-routine occupations, again contrasting with the 2001 situation for AC-12 migrants residing in England in 2001 and other groups in both 2001 and 2011. Thus, at least over the years following the 2004 accession, the AC-12 migration flow was in its majority a labour migration flow into low and medium skills' professions. 80% of AC-12 migrants were economically active in 2011, a share significantly above that of UK born and other migrant groups.  Finally, it should be noted that AC-12 migration inflows into the UK were also geographically and compositionally different from migrants of AC-12 countries that had settled in the UK prior to the 2004 enlargement shock [Becker et al., 2018]. Before the AC-12 countries joined the EU, the stock of individuals who were born in any of the ten Central-and Eastern European accession countries was around 193,180. Unlike AC-12 migrants arriving in the UK after 2004, these migrants were mostly concentrated in the London area [Becker et al., 2016]. Approximately 30% of this group had arrived before 1981 and consisted mostly of people born in Poland, who made up 42% of the stock of Eastern Europeans having arrived prior to 2004 [Becker et al., 2016]. The number of these Polish-born migrants increased by a factor of seven, and the number of Eastern Europeans in the UK by a factor of five, such that the number of AC-12 migrants living in England represented approximately 2% of the English population in 2011, reaching 1,085,351 inhabitants.

Sampling frame and data sources
In this section, we first describe both the data on local authority expenditure and revenue and the migration data we use for our subsequent analyses, in subsections 3.1 and 3.2 respectively.
We then describe additional data sources we draw on to obtain control variables and for the analyses of mechanisms that may explain the obtained results in section 3.3.

Local authority expenditure and revenue
Detailed panel data on local authority public finances is available from the Chartered Institute of Public Finance and Accountancy (CIPFA) website which gathers local authority budgeted expenditure and revenue data. We use this data set to identify exactly how the 116 singletier authorities and 27 authorities operating under a two-tier regime allocated their funds between all different spending areas annually. We obtain these data for the period from 2000 to 2015 such that our sample consists of a total of 143 local authorities observed over a 16-year period.

Migration data
We draw our yearly data on population for each local authority from a special license of the

Additional data sources
Our main specification measuring the effect of AC-12 migration controls for a number of local authority level time varying characteristics. First, we control for local area population obtained from CIPFA to account for potential scale effects in service delivery, especially regarding education. Since the effect of AC-12 on local redistribution may also be conditional of the existing composition and heterogeneity of the population, we further control for the 20 share of EU-15 and non-EU migrants obtained from the LFS/APS [Alesina et al., 2019, Tabellini, 2020b. Finally, we also control for local unemployment rates to account for the fact that local economic conditions are an important pull factor for labour migrants such as those originating from AC-12 countries.
We then construct additional dependent variables to understand the mechanisms driving our main results. These variables include information on the number of pupils per local authority, and the age structure of the population (both obtained from CIPFA). To test for a change in political preferences, we use yearly data on the political composition of local We further substantiate the strong differences between the Conservative and Labour party on redistribution in Figure 1.
5 The data can be accessed on http://www.electionscentre.co.uk. 6 See https://www.gov.uk/understand-how-your-council-works/local-councillors-and-elections for details. are calculated from the question "Should the government redistribute income?" to which respondents could respond on a scale from 0 ("make much greater effort") to 10 ("be much less concerned"). Respondents answering 0,1,2 or 3 are categorised as "favours more redistribution". Respondents answering 7,8,9 or 10 are categorised as "favours less redistribution". N=2840.

22
The figure based on data from the March wave of the British Election Study 2015 shows that more than 50% of Labour voters favour more redistribution of income, whereas that share stood at less than 20% among Conservative voters, who tend to favour less redistribution.
Voters of Liberal Democrats, SNP and UKIP fall in between.
Unfortunately, the Council composition dataset aggregates information on the Council seat share of the UK Independence Party (UKIP) into an "other" category. Founded in 1991 and until the 2016 Brexit referendum, UKIP was essentially a single-issue party campaigning for an exit from the EU and made strong gains in European elections at the expense of the Conservative Party over our sample period [Ziebarth, 2020, Becker et al., 2016, Fetzer, 2019.
However, it is is not clear that more votes for UKIP are a signal of preferences towards less redistribution, as highlighted by Figure 1 . Nevertheless, greater vote shares for UKIP in the face of AC-12 migration still suggest greater distaste for migration, which could potentially result in changes in local Councillors' spending decisions. Therefore, we additionally match our database with information on all local election results from Fetzer [2019], which contains election results of almost all parties including UKIP. Finally, we also use information from the second quarter of each LFS wave as well as NHS data to build local authority level measures of internal inflows and outflows.
We note that not all data we use for our analyses of mechanisms is available for the entire 2000 to 2015 observation period. We summarise data availability, sources and the measures we construct in Table 11.

Empirical strategy
In this section, we first lay out our main empirical strategy in subsection 4.1. We then turn to the main empirical issue in our setting at hand, the potentially endogenous location choice of migrants within England in subsection 4.2.

Main specification
The AC-12 migration wave to England, and the U.K. more generally, was a grand-scale natural experiment due to its large and unexpected nature. To capture its magnitude by local authority, we closely follow Becker et al. [2018] and build the following shock measure: 7 where AC12migrants In equation 2, P ost 2004 is a dummy equal to 1 after 2004 such that its interaction with Shock c,t identifies a treatment on the treated effect of AC-12 migration on the outcome measures. X c,t represents our vector of controls which includes the local share of EU-15 migrants, the share of non-EU migrants, the local authority population and the local unemployment rate. We further include year fixed effects β t and local authority dummies α c to account for unobservable year-specific variation common to all local authorities and time-constant local authority level variation respectively. Due to the spatial nature of our data, we cluster the error term c,t at the local authority district level in all analyses to correct standard errors for within local authority correlations [Bertrand et al., 2004]. Our estimation is therefore akin to employing a two-way fixed effects panel model that combines features of a continuous Difference-in-Differences and event study as we set the pre-accession AC-12 stocks to zero.
We do this due to the to the strong differences between AC-12 pre-and post-2004 and expect a gradual divergence of treated and untreated local authorities in the post treatment period, as not all migrants moved to the UK at the start of our treatment period.

Endogenous location choices
The main concern for causal interpretation of γ is the endogenous location choices of AC-12 migrants. Despite the exogenous nature of the shock at hand, immigrants were not randomly allocated across local authorities and sorting into local authorities might be endogenous to migrants' underlying preferences for redistribution or other unobserved characteristics that determine both migration and trends in local authority spending. We first address this concern by presenting evidence in favour of the common trends assumption by showing that all measures of interest did not move systematically differently between affected and unaffected local authorities prior to 2004 in 5 and 6.
Unfortunately, the fact the treatment increases in all groups means we cannot use the al- We show that our results are practically unchanged when matching local authorities affected by AC-12 migration inflows to unaffected local authorities using propensity score matching. To do so, we follow an approach similar to Becker et al. [2018] and first apply a corrected Akaike Information Criterion (AICc) algorithm to a large set of local authority characteristics we gather from the 2001 Census to find the best predictors for AC-12 migration inflows and then match our sample based on these variables (with replacement).
While this approach reduces our sample size, it mitigates remaining concerns that destination choices were not picked at random and should be treated as complementary to our main difference-in-differences results.
Finally, we proceed to showing the results when using a shift-share "Bartik" instrumental variable approach based on historical settlement [Bartik, 1991]. The problem of endogenous sorting of migrants has indeed often been tackled by using such an approach pioneered by Card [2001]. It is based on the premise that immigrant networks are an important determinant of locational choices and allows to identify local average treatment effects of current migration inflows induced by these historical settlement patterns. In the context of the UK, this instrument has been used by Bell et al. [2013], Sá [2015] and Giuntella et al. [2018] in their studies on the impact of migration to the UK and its effect on crime, house prices and hospital waiting times respectively. The validity of such an instrument relies on the assumption that the past settlement of immigrants is uncorrelated with changes in economic outcomes between local authorities prior to 2003. Thus, it assumes that immigrant settle-

Results
In this section, we turn to the results and split our analyses into two parts. Subsection 5.1 provides evidence in support of the identifying common trend assumption. Subsection 5.2 then shows the main results.

Pre-and post trends
In this subsection, we analyse the main outcomes of interest before and after the opening of UK borders to AC-12 migrants in 2004. Figure 3 shows that the share of AC-12 migrants 28 indeed gradually picked up after 2004, starting from a level close to zero before the accession of the AC-12 countries.  To test for pre-and post-trends in our main outcome variables of interest, we define local authority districts that were in the top 25% (Q4) of the 2015 migration shock distribution as treated ("affected") . We further define all local authority districts that are within the bottom 25% (Q1) of the 2015 migration shock distribution as untreated ("unaffected"). Formally, we The categorisation is motivated by the distribution of the AC-12 shock measure in our   In sum, three main findings emerge from this first descriptive analysis. First, total expen-diture per capita remained unchanged following AC-12 migrant inflows. Second, the funding mix slightly shifted away from locally raised budget towards central government transfers.
And third, AC-12 migrants shifted the expenditure mix away from social care expenditure towards education expenditure.

Main results
In this subsection, we turn to the regression results based on the empirical specification introduced in section 4.    (1) and (2) show that the association between the AC-12 shock measure and total service expenditure per capita is positive, albeit not significant at any conventional level. Similar to the descriptive analysis in subsection 5.1, local authority revenue sources were indeed impacted by AC-12 migration: Column (3) (1), (2) and (3)   Standard errors clustered at the local authority level in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Notes: Outcomes expressed in per capita terms (pc). The AC-12 shock is defined as the difference in AC-12 population shares in a given local authority-year and its 2001 baseline share as defined in equation 1 of section 4.1. All regressions include local authority fixed effects. The full set of controls refers to the share of EU-15 migrants, the share of non-EU migrants, the local authority unemployment rate and the total local population. LFS/APS refers to the UK Labour Force Survey and its boost samples in the Annual Population Survey. The observation period is 2000 to 2015.  1, 14) is the result of an increase in additional spending on education associated with this particular migrant group.
Neither EU-15 migrants nor non-EU migrants appear to be associated with changes in social care expenditure (column 4, Table 14).
In sum, three main results emerge. First, the inflow of AC-12 migration does not show a statistically significant association with total service expenditure per capita. Second, AC-12 migration did impact on local authority revenue sources: They are associated with an increase in funding received from the central government but decreased locally generated income. Finally, AC-12 migration decreased social care spending per capita and increased education expenditure per capita, while other expenditure items remained largely unchanged.
We note that these findings do not allow for making definite statements about the mechanisms at hand: A reduction in the mostly discretionary means-tested social care services and the relative reduction of locally generated revenue in response to AC-12 inflows could indicate both a shift in local preferences towards less redistribution but could also capture changes in local service demand brought about by distinct socio-economic characteristics of newarrivals, with repercussions for necessary funding. We analyse these mechanisms in more detail in section 7, after testing the robustness of our results in section 6.

Robustness tests
In this section, we test the robustness of our main results along a range of dimensions. Sub- WRS stipulated that AC-8 migrants -that is, migrants who originate from the 2004 Central and Eastern European EU joiners -could claim out-of-work benefits and tax credits on the same grounds as other EEA nationals only after being registered to the WRS and in contin-uous employment for 12 months. Giua [2020] shows that the expiration of the WRS had a positive impact on the probability of claiming out-of-work benefits by these migrants. Access to some of the local authority services such as social care services only required registering on the scheme, mitigating the concern of a pick-up in demand after the expiration of WRS [Kofman et al., 2009]. However, the pick-up in claims of out-of-work benefits by AC-8 migrants post 2011 may have had a second-order effect on demand for means-tested local authority services due to its direct effect on income. A general link between local unemployment rates and demand for means-tested social care services is indeed visible in our regressions (column 4, Table 14, appendix A). To account for these two important regime changes, we therefore test our results for robustness when excluding post-2010 years. Table 6 shows the results estimated in our preferred specification when excluding all years post 2010.  On the other hand, the more gradual increase in the demand for education services in local authorities more heavily affected by the migration shock is in line with the observation that many AC-12 migrants were of child-bearing age when they arrived to the UK, or required additional spending due on language schooling. We analyse these hypotheses in more detail in section 7.

Matching
The pre-trends we present in Figures 5 and 6 show that AC-12 migrants did not systematically sort into local authorities based on trends in local spending or revenue regimes.
However, a concern remains that destination choices were not picked at random: If AC-12 migrants moved into local authorities of specific underlying unobserved characteristics, these characteristics could have medium-to long-term consequences for local spending and revenue patterns that are then picked up by our estimates. For example, since AC-12 migrants were primarily labour migrants of a distinct skill profile, their destination choices were likely based on labour demand from specific industries; if the presence of these industries was linked to future local economic development in local authority areas, the association of AC-12 migrant shares and local area spending may be spurious.
In this subsection, we address this concern by matching affected and unaffected local authorities based on a wide range of local authority characteristics we draw from the 2001 UK Census. The local area characteristics we use for the matching procedure range from local authority expenditure and revenue measures to local household wealth indicators, local industry composition and demographics. A full list of variables is provided in appendix D.
The simple matching-with-replacement estimation proceeds in three steps. Similar to our pre and post-trend analysis, we first divide local authorities into affected (=treated) and unaffected (=untreated) based on the increase in AC-12 migrants they experienced between 2004 and 2015. To increase the pool of potential matches, we define the top 50% receiving areas as affected and the bottom 50% of receiving areas as unaffected for the sake of the matching exercise. In a second step, we then use a simple stepwise Akaike's corrected information criterion (aicc) to select variables that best predict whether or not a local authority was treated, balancing an increase in the goodness-of-fit against the additional information required to achieve it [Cavanaugh and Neath, 2019]. The variables selected by the algorithm are highlighted in the full list shown in appendix D. In summary, AC-12 migrants were most likely to migrate into local authority areas with relatively larger manufacturing, hotel, health, fishing, financial and domestic work industry shares. Further variables that predict AC-12's propensity to migrate into specific areas include household shares deprived in one dimension and the share of social housing provided by the local Council. In the final step, we then use propensity score matching to match treated and untreated local authorities based on the selected variables. Table 7 shows the results of the matching regressions.  The drop in local service expenditure outside education and social care in response is more precisely estimated in the matched sample (p<0.05). Overall, the results suggest that spatial sorting among AC-12 migrants based on observable local authority characteristics is not a large concern in the post-accession English setting. To further robustify this finding, we next turn to an an instrumental variable approach in subsection 6.3.

Instrumental variable approach
As discussed in subsection 4.2, a common way of dealing with endogenous destination choices is to instrument contemporaneous migration inflows by historical settlement patterns. This instrumental variable was pioneered by Card [2001] and is based on the premise that immigrant networks are an important determinant of locational choices [Altonji and Card, 2018].
In the context of the UK, such a "shift-share" instrument has been used by Bell et al. [2013], Sá [2015] and Giuntella et al. [2018] in their studies on the impact of migration to the UK and its effect on crime, house prices and NHS waiting times respectively. In our setting, the identifying assumption of such an instrument is that the historical settlement of AC-12 immigrants is correlated with post-2004 changes in local authority spending and revenue only through their effect on post-2004 inflows of AC-12 migrants. In this subsection, we use 1991 UK Census data to construct the instrument similar to Giuntella et al. [2018]. Specifically, we define λ c,1991 as the share of AC-12-borns living in local authority c in 1991 and calculate: 43 Thus, for each local authority c and year t, we multiply the 1991 share of AC-12 migrants residing in that local authority by the aggregate national level stock of AC-12 migrants in year t to project the new stock of AC-12 migrants. We then divide this number by the local area population P op c,t to derive the projected shares.
We do not use the IV approach as our preferred specification for a number of reasons. With these caveats in mind, Table 8 displays the results of the IV-regressions.  Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01 Notes: Outcomes expressed in per capita terms (pc). The AC-12 shock is defined as the difference in AC-12 population shares in a given local authority-year and its 2001 baseline share as defined in equation 1 of section 4.1. DiD -IV refers to a difference-in-differences approach where the AC-12 shock variable is instrumented by the historical distribution of AC-12 migrants across England as defined in 4. All regressions include local authority fixed effects. The full set of controls refers to the share of EU-15 migrants, the share of non-EU migrants, the local authority unemployment rate and the total local population.LFS/APS refers to the UK Labour Force Survey and its boost samples in the Annual Population Survey. The observation period is 2000 to 2015. The F-test of the first-stage regression is the Kleibergen-Paap rk Wald F statistic. It is reported only for the post-shock period.
The LATE estimates obtained from the instrumental variable regressions confirm the main results. When interpreted causally, the results show that the subset of AC-12 migrants pulled in by historical networks had sizeable effects on both local authority revenue sources and the spending mix. All estimated coefficients are between five and eight times larger than those obtained in our preferred specifications of Tables 4 and 5. We note that the difference in the magnitude of the obtained coefficients allows for two possible interpretations: First, the treatment-on-the-treated effects obtained from the difference-in-differences specification underestimate the effects AC-12 migration had on local authority budgets and service provision. A second interpretation is that the estimated LATE is identified based on a subset of migrants that differs significantly from the average group characteristics of AC-12. For the reasons outlined above, we believe that the second is more likely to hold true. We conclude that the estimates obtained in Tables 4 and 5 are likely to be closer to the true effect AC-12 had on local authority service provision. 9

Census data
Despite being relied on heavily in the UK-migration literature as a source of tracking local authority level migration stocks over time in the UK [Bell et al., 2013, Sá, 2015, Giuntella et al., 2018, the LFS/APS data we use as our primary data source has some disadvantages regarding its ability to accurately capture migrants of specific countries of origin on a granular geographical level [Cangiano, 2010]. The absolute number of AC-12 migrants or other migrant groups sampled in each local authority-year is sometimes too low to rely on in our empirical analyses, leading us to discard observation points when these counts fall below 10. In this section, we therefore confirm the main results using 2001 and 2011 Census data instead of the more volatile LFS/APS. Both censuses provide data at the local authority level of residents by country or region of birth while the 2011 Census provides information on the time of arrival 9 In theory, the relatively weak first stage F-test (F=5.93) may both lead to a bias of the estimated coefficient towards OLS -in which case the true coefficients would be even larger than those obtained -and an underestimation of the size of the standard errors around these point estimates [Murray, 2006]. of 2011 residents in two-or three-year brackets. As we do not have information within these year brackets, we calculate the stock of migrants as the last year of each bracket: For example, a migrant reporting to have entered a local authority district five to seven years ago in 2011 enters our stock calculation for the year 2006. This way, we obtain stock data for AC-12 migrants in every local authority district for the years 2000, 2003, 2006, 2009 and 2011. A disadvantage compared to our main AC-12 shock measure is that all results are estimated exploiting year-on-year variation within local authorities only when stock data is available.
The way the stock of residents is reported in the Censuses further means that it does not count migrants who arrived in England before 2011 and then left England before 2011 or who died before 2011. The stock of migrants reported for every year is therefore a lower bound estimate of the true migration inflow. The alternative shock measure is summarised in Table   9.

Variable
Mean Std. Dev. AC-12 migration shock (alternative) 0.013 0.017 N 2028 Table 9: Summary statistics Table 10 then turns to the results using the alternative AC-12 migration shock measure based on Census data.  48 The results confirm the main results shown in Tables 4 and 5. The estimated coefficients on total service expenditure per capita (column 1), central government transfers per capita (2), social care spending per capita (4) and education per capita (5) are by a magnitude of 1.5 to three times larger than those estimated in the baseline specification. All coefficients are slightly more precisely estimated. Due to the shortcomings of the Census data discussed above, we do not overinterpret these differences in effect size; however, the results show that imprecisely measured AC-12 migration stocks are unlikely to be the drivers of the association between AC-12 migration and local authority spending and revenue patterns.

Mechanisms
The results found in this paper document how a large migration shock translates into the provision of different public goods. In aggregate terms, our results suggest that the AC-12 migration shock only marginally affected the overall supply of public services in both pre and post-austerity years. On the revenue side, our results show a relative increase in central government transfers and a relative decline in locally generated revenue in local authorities affected by AC-12 migration inflows. We also observe important shifts in the types of public goods local authorities provide following the migration shock brought about by AC-12 migrants. Local authorities affected relatively more by migration inflows spent relatively less on means-tested social care services and significantly more on education services.
In this section, we investigate two important channels that might explain our observed results. On the one hand, the change in local authorities' populations stemming from AC-12 migration could lead to changes in redistribution via a change in preferences towards less redistribution or native flight lowering the tax base and local authority revenues. On the other hand, these changes could also reflect a more mechanical response. Migrants' have different socio-economic characteristics and different propensities to uptake services, which could lead to changes in redistribution without reflecting any migration-specific altruistic component.
In this case, changes in redistribution from the rich to the poor do not discriminate between native-born versus foreign-born poor and simply reflect changes in the socio-economics characteristics of the population. Conditional on an institutional response by local authority councils, changes in the allocation of spending would then follow a mechanical response. We discuss both these channels separately in the following subsections. While we acknowledge that both channels are non-exhaustive and could be at play simultaneously, we first discuss each of these channels and the hypotheses one can derive from them before arguing, through a number of tests, that the demographic channel better explains the observed changes in redistribution following AC-12 migration. For an overview of all variables we analyse as mechanisms and their source, data availability and construction, see Table 11 .

Migration and the altruism towards co-nationals: the preferences channel
The shift in expenditure from social care services towards education and the decline in the locally raised budget we identify could be interpreted as local authorities responding to a change in local preferences towards less redistribution following the migration shock. The link between migrant inflows and local preferences for redistribution could be a reflection of people from different groups disagreeing on the optimal amount and composition of local spending, or the dominant native group being less willing to redistribute towards non-co-nationals.
While one has to be careful with using vote shares as an indication for an underlying preference for redistributing wealth or income in the population, we showed in Figure 1 that in England, it is clearly the Conservative party that has support from voters less in favour of redistribution. We can thus test whether such a "demand" for less redistribution effect could explain our results by measuring the impact of AC-12 migration on the composition of local Councils and test whether the Conservative less pro-redistributive platform increased its seat share following a stronger migration wave. In this context, it is also worth noting that all EU (and Commonwealth) migrants to the UK have voting rights in local elections.
Thus, an increase in the seats' share of Conservative Councillors, who represent the interest of voters with preferences for less redistribution, could also reflect a change in the aggregates populations' preferences if migrants exerted their voting rights.
A second more indirect mechanism through which distaste for outsiders could lead to less redistribution is a mechanism often referred to as "native flight" (see e.g. Cascio and Lewis [2012]) or "voting by feet". In addition to a direct reduction in demand for redistribution translating into changes in local spending, distaste for living near foreigners can also affect redistribution via natives moving into different local authorities. Such changes can indeed lead to a lowering of the tax base as the number of taxable properties decline, leading to a deterioration of local authorities' budgets and eventually a decline in local authority government spending. To analyse such second-order internal migration in response to international inflows of migrants, we therefore report the impact of AC-12 migration on the number of taxable dwellings per capita (the tax base) as well as on the Band-D council tax rate, a lump-sum tax due to be paid annually by the dwelling owner. On average, this Band-D council tax stood at GBP 1278 over our observation period. To test the native flight hypothesis, we create two measures of internal migration. First, we draw on National Health Service (NHS) registration data. Getting basic access to free public health care in the UK requires registration with a local general practitioner. If individuals change their residence within the UK, they are required to provide their new general practitioner with any previous registration. The data thus allows to calculate both registrations and de-registrations for any given local authority and year. We then subtract internal migration inflows from internal migration outflows to construct a measure of net outflows of internal migrants at the local authority year level. Since the measure does not allow us to distinguish internal migrants by country of birth, we construct a second similar measure for UK native borns only, following Giuntella et al. [2018]. Between 2000 and 2013, the second quarter survey of the UK-LFS 51 contained a variable that asked respondents for their local authority of residence in the year prior to the survey, as well as their current local authority of residence. We use this information and calculate a "net internal native out-migration" variable for UK born individuals for all local authority years. Unlike the measure based on administrative NHS registration data, which is comprehensive, we normalise the UK-LFS data by the total number of respondents in each local authority-year to account for fluctuations in the sample size of the survey over time.  [Phillips, 2018]. To investigate this institutional response, we begin by analysing changes in local demographics associated with AC-12 inflows by constructing a measure for the population share of pupils and a measure for the share of the population aged 65 and above to proxy demand for education services and social care services brought about by demographics respectively. In a second step, we then normalise changes in spending by these largest relevant consumer groups for both total spending on education and social care instead of using the previous per capita measure to test whether the observed changes were mechanical.
A second potential explanation for the uncovered association between AC-12 migration inflows and changes in local authority spending and revenue is a second-order effects brought about by an internal migration response to the new-arrivals: If, for example, the availability of relatively cheap labour creates local economic opportunities, this could draw in additional migrants from within the country. This, in essence, is the opposite of the "native flight" mechanism and can therefore be tested using the same internal migration measures as outcome variables as explained in subsection 7.3. 10 7.3 The relative importance of the demographic channels: the easing off of pressure on social care stemming from AC-12 migration Table 12 reports the results testing the hypotheses we have brought forward to discuss the relative relevance of the two main channels discussed above.
10 An additional mechanism we do not consider due to the lack of reliable data is migrants' propensity to demand local services, even once demographic differences are accounted for. For example, the "healthy migrant effect" could decrease the demand for social care services [Abraido-Lanza et al., 1999]. While data on social care referrals in England are available, these cannot systematically be divided into self-referrals and referrals by doctors. This is problematic since doctors are likely to refer their patients to social care services based on their understanding of availability of these services. A further issue with data on referrals is that they do not indicate the type (and thus, cost) of services requested.   Table 11. The number of observation varies due differences in data availability shown in detail in the same table. The AC-12 shock is defined as the difference in AC-12 population shares in a given local authority-year and its 2001 baseline share as defined in equation 1 of section 4.1. All regressions include local authority fixed effects. The full set of controls refers to the share of EU-15 migrants, the share of non-EU migrants, the local authority unemployment rate and the total local population. LFS/APS refers to the UK Labour Force Survey and its boost samples in the Annual Population Survey.
We first turn to columns (1) to (4) to analyse the local demographic channel. AC-12 migration is strongly associated with an increase in the local population share of pupils (p<0.01; column 1) and is similarly associated with a decline in the local population aged 65 and above (p<0.01; column 2). The estimated coefficients indicate that a one percentage point increase in the share of local AC-12 migrants relative to the 2001 AC-12 population share changes the share of pupils in the population and the share of individuals aged above 65 by 0.15 percentage points and -0.26 percentage points respectively. Columns (3) and (4) then show the spending on education and social care relative to their main consumption groups. The coefficient estimated on education expenditure by pupil (3) shows that the large increase in education expenditure per capita we documented in the previous analyses is likely due to the change in underlying demographics associated with AC-12 inflows. The estimated coefficient is still positive, but no longer differs from zero at any conventional statistical level. In column 4, we show that the decline in social care service spending per capita associated with AC-12 inflows disappears when spending is calculated as a share of the main recipient group. In fact, increases in the local AC-12 migration share are associated with a large inrease in social care spending by population aged 65 and above (p<0.01). The estimated coefficient shows that a one percentage point increase in the AC-12 migrant shock measure leads to a GBP 56 increase in social care spending by the population aged 65 and above.
We next turn to the local preferences for redistribution channel in column (5) and (6) of Finally, columns (7) and (8) show the association of AC-12 migration with net total UK-internal migrant outflows and the net internal outflow of UK-borns respectively. Unlike previous research by Sá [2015] and Giuntella et al. [2018], who analyse migrants as a homogenous group, we find no evidence for a "native flight" following the inflow of AC-12 migrants into local authority areas. Instead, our estimates in fact suggest a decline in the net outflow of both total and UK native-born internal migrants in response to AC-12 migration.
Taken together, we interpret these results as strongly suggestive of a greater role played by the demographic channel. While the results on native flight and the increase in the more pro-redistributive platform go against the preferences channel, the fact AC-12 migrants were on average younger and more likely to be in employment suggests the demographic channel played a significant role in explaining the reduction in social care spending per capita. Furthermore, education spending per pupil remains stable and can explain the rise in central government transfers per capita for this partially ring-fenced expenditure item. While we cannot fully test this argument, it is also worth noting that this increase in pupils per capita should not be fully attributed to a rise in pupils from AC-12 countries only. As shown in Table 1, AC-12 migrants were not particularly more likely to have more children than other groups, such that the relative increase in pupils may also reflect the second order effect of a reduction in net outflows of potentially young natives with children.  The results shown in Table 13 suggest that, while it was indeed the case that the housing stock did not keep up with the increasing number of migrants (column 2), AC-12 migration was also associated with a relative decrease in the local council tax (column 1, p<0.1).
In light of the results on local voting patterns and internal migration responses to AC-12 inflows shown in columns (5) to (8) of Table 12, we do not interpret these effects as reflecting the importance of the preferences channel. The provision of local housing units naturally lags changes in the local population size, explaining the decline in the tax base per capita as affected local authorities' populations increased drastically. This decline in the number of taxable dwellings relative to the local population did likely not require a compensation by increasing the local council tax. This interpretation is corroborated by the fact that the reduction in social care spending per capita was not conducted at the expense, but rather at the benefit of the most vulnerable populations (column 4, Table 12). Indeed, the most plausible interpretation of our findings is that the inflow of a dynamic and young population eased pressure on local authorities who could now spend relatively more on the social care of those aged 65 and above.
In summary, our analysis of mechanisms that explain the shifts in local authority expenditure and revenue associated with AC-12 migration inflows leads us to three main conclusions: First, the changes AC-12 migrants caused to local authority expenditure and revenue patterns were in large parts due to the shifts these migrants caused to local demographics and the corresponding institutional responses that were triggered by the resulting changes in local service demand. Second, AC-12 inflows were significantly associated with a relative increase in internal migration. The increasing share of pupils associated with AC-12 migration was likely linked to internal migrant inflows into the same local authorities leading to increased education spending per capita. Finally, the relative drop in demand for local social care services associated with AC-12 migrants did not just result in a large inrease in social care spending by the population aged 65 and above, but also allowed affected local authorities to keep their council tax rates relatively low. Overall, our results suggest that the demographic channel was most relevant to explain the impact of AC-12 migration. Thus, they further strengthen previous research findings of an overall net positive fiscal contribution migrants from 2004 EU accession countries made to the UK government budget [Dustmann and Frattini, 2014].

Conclusion
In this study, we investigated the effect of the large and unexpected wave of Central and Eastern European migrants starting in 2004 on local authority redistributive spending in England. Our results suggest that AC-12 migrants indeed impacted on local authority revenue sources and the local spending mix. We do not find evidence in favour of the hypothesis that these large migration inflows impacted on local preferences for redistributing income.
We neither observe voting patterns in local elections that would indicate such a shift in preferences, nor did local residents "vote with their feet" in response to migration inflows. While our results clearly favour shifts in demographics as an explanation for the observed changes in local authority revenue and spending, a limitation of our study is the lack of survey data that could capture preferences for redistribution more precisely.
We interpret our results as a word of caution when relating them to the existing literature. A decrease in public spending can mean a lack of demand from newcomers due to their distinct socio-economic characteristics, rather than the outcome of an increasing local insider-outsider dynamic. Thus, our findings rather lend further support to Dustmann and Frattini [2014] who show that AC-12 migrants were positive net contributors to the UK public purse.
It is worth reflecting on our results in light of the Brexit vote, where anti-immigrant sentiment has played an important role [Meleady et al., 2017, Dennison andGeddes, 2018].
A possible interpretation of our results is that the national-level distaste for immigrants expressed in the Brexit vote was not driven by local level exposure to foreigners. This explanation finds strong support in a recent study by Becker et al. [2017], who show that local education, income and unemployment levels are strongly correlated with the local Vote Leave share, whereas exposure to EU migrants has little explanatory power. Such an interpretation would further corroborate the necessity to conduct studies relating the presence of migrants to preferences for redistribution on the subnational rather than the national level if the aim is to study direct exposure: On the national level, changes in preferences for redistribution may capture an increased fear of foreigners in areas not necessarily exposed to migration.
Finally, the interpretation of our findings with regards to the sustainability of social care services in England requires a careful reflection. On the one hand, we show that the distinct demographics of AC-12 migrants eased the pressure these means-tested services face in England in the short-term. On the other hand, those migrants who arrived as part of the post-accession waves are increasingly getting older and will likely demand social care services in larger numbers in the future. Using migration as a tool to permanently ease the pressure on local service provision would thus require a continuous inflow of migrants.
However, migration inflows from EU countries to the UK have slowed down in response to the end of free movement for EU citizens and recent political developments in the UK, with net migration flows from the 2004 Central and Eastern European countries turning negative in 2018 for the first time [Sumption and Vargas-Silva, 2021]. These developments are likely to have repercussions for local authority revenue and spending in the near future and may call for new reforms to regulate the flow of migrants in the absence of free movement of EU citizens.

Appendices
A Main results with controls -full table   Table 14 shows the coefficients of the AC-12 shock as well as the coefficients of all our control variables on our main outcomes of interest.         E UKIP results Table 18 shows the impact of our AC-12 shock measure on the electoral results of UKIP in local elections that took place between 2000 and 2015. (1)