‘ Left behind ’ people, or places? The role of local economies in perceived community representation

Modern economics attributes importance to spatial inequality: yet in studying discontent with politics, existing research has mostly neglected local contexts and attitudes people hold about them. I use British Election Study data to investigate the factors leading people to believe their community is ignored by the political process. Firstly, real economic contexts play a role, since residents of low-income communities tend to take more negative views about how well their community is represented. Secondly, negative perceptions of the local economy are associated with more negative views of community representation, whereas equivalent ‘ egotropic ’ measures of people's personal economic situation have no such e ﬀ ect. Thirdly, I observe a ‘ grievance ’ e ﬀ ect wherein people are particularly negative about community representation when they believe that the national economy is more successful than that of one's local community.


Introduction
Since the Brexit vote in the United Kingdom, the notion of political discontent reflecting geographic divisions has found a degree of favour within the discipline of political science. Coyle and Ford (2017, p. 67), in their account of Brexit, argue that political 'alienation' has become entrenched in 'left behind' areas since the 1980s, due to the fraying of the 'economic and social fabric' and the failure of successive governments to reverse this decline. Specifically, they argue for 'the devolved nations and England outside of the southeast' as being the locus of discontentidentifying regional divisions as the wellspring of political anger. However, while this paper argues that this perspective has real value, the geographic account as it exists at present requires further explanation, examination and refinement.
The sole existing test of this thesis has been conducted by Jennings et al (2016). They looked at two types of area, 'cosmopolitan' and 'backwater', which they defined as having different levels of 'access' to high-skilled jobs and 'connectedness' to the global economy. Looking at fifty Parliamentary constituencies for each, they found that 'backwaters' were 'more inward-looking, illiberal, negative about the EU and immigration, nostalgic and more English in their identity ' (p. 372). However, such differences did not extend to 'expressions of anti-political sentiment': from distrust in MPs and politicians, to dissatisfaction with UK democracy, 'both sets of populations look remarkably similar to the average voter ' (p. 376). This leads them to reject the idea that local economies currently affect political discontent, although they do not shut the door entirely, emphasising instead that poor areas are nonetheless fertile for the 'growth and spread of political disaffection' (p.380).
This paper re-examines the fundamental question posed by Jennings and Stoker. It does so in three ways. Firstly, there is reason to believe that a more discriminating analysis, based on much smaller geographic areas, may be required to properly identify the effects of economic context. A large geographic unit such as the constituency is not the 'context' that most people see in their day-to-day lives. Research on economic inequalities in the United Kingdom (e.g. Dorling and Pritchard, 2010) has emphasised the spatially concentrated nature of disadvantage, such that the economic context of one's 'community' may be very different to that of one's constituency or region. Equally, nor is it likely to be the 'context' they consider meaningful to themselves. Wong et al. (2018) show that the self-defined community tends to denote people's neighbourhoods, and the people and places they see most frequently: consequently, its physical size is often quite small. Secondly, the kind of attitudes which Jennings and Stoker expect to vary spatially may in fact be ill-suited to picking up real variations in underlying sentiment: attitudes that are more specifically about how the local community is represented are more likely candidates. Finally, the existing analysis is subject to the typical problems of ecological inference: properly analysing the contextual effect of local economies is of real importance.
Drawing on an innovative method of gathering public perceptions related to the local community embedded in a large UK-based public opinion survey, and integrating this with official small area economic data, this article provides empirical evidence of the relevance of the local, and proposes how these relationships should be conceptually understood. It innovates in using the local community as an analytical frame to study discontent and its causes: thus, taking up the challenge put to the discipline by Mutz and Mondak (1997, p. 302) that we should study how people relate to the 'countless subnational collectives' between the individual and national. It does so both with reference to the attitude put under the microscope, 'perceived community representation', and in terms of the causal theories proposed. This article finds evidence for three distinct mechanisms driving perceived community representation, demonstrating not just that context (real or perceived) matters but how it matters.
Specifically, these are as follows. Firstly, real economic contexts may also hold a degree of importance, since living in a low-income community is associated with more negative views of 'community representation'. Secondly, I find evidence for a basic sociotropic dynamic at community-level: negative perceptions of the local economy are associated with more negative views of 'community representation'. Finally, I find negative views of 'community representation' are associated with 'grievance' that one's community is perceived to be facing worse economic conditions compared to the country at large.

Capturing the local in representation perceptions: introducing 'community representation'
In order to approximate the theoretical construct at hand, I use a dependent variable quite different from those used in the literature relating to political discontent. The stalwarts of the discipline (external efficacy, institutional trust, democratic satisfaction and so on) are poorly matched to the task. A central question in this area is distinguishing how people view different political 'subjects': government versus regime; parliament versus parties (Easton, 1975;Klingemann, 1999;Dalton, 2004;Esaiasson et al., 2015). However, scholars have shown more limited interest in how people assess the 'relational element between the entity that represents and the entity that is represented' (Castiglione, 2006, p.845). In order to make sense of public attitudes to their representation, research must address how said attitudes depend on what is presented as the 'objects': i.e., the represented group. However, this is a component of question design that has been given only perfunctory attention.
The issue is not confined to them, but for the sake of illustration, let us consider the classic 'external efficacy' items from the American National Election Study: 'People like me don't have any say about what the government does', and 'I don't think public officials care much what people like me think'. 1 There are two main ambiguities here. Firstly, do people bring 'egotropic' considerationsthose related to how an individual perceives their own situation -to the table, as Campbell et al. (1954) seemed to believe? Or, alternatively, are their responses based in perceptions of wider societal or 'sociotropic' conditions, as Ulbig (2008) suggests? Secondly, if their responses are sociotropic, the ambiguities in what is meant by 'people like me' suggests that people can bring any one of their heterogenous identities to bear, which they in some way sample from or weigh-up (Zaller and Feldman, 1992) in a way that is unknowable from the researcher's point of view. The role that local factors will play is unclear and liable to inconsistency. Furthermore, the survey context may activate some of these identities while leaving others dormant: asking dozens of questions about party politics could make someone who, day-to-day, thinks of themselves as a Mancunian or Londoner first-and-foremost, act as a down-the-line partisan in their responses.
This study is the first to position 'local community' as the 'objects' of representation. Specifically, the survey question used asks 'How much do you think the interests and views of people in your local community are listened to when important decisions affecting them are made by national government?'. This question, while innovative in some respects, mirrors the concept of 'perceived responsiveness' in Esaiasson et al. (2015), developed from theoretical literature on representation. It incorporates the insights of Hibbing and Theiss-Morse (2001), who states that the way policies are produced matters as much (if not more) to people than the policies themselvesgiving reason to believe process questions such as this generate responses with real affective content. It also specifies that the decisions taken have real stakes for the local area. Thus, negative responses are more likely to communicate something about a failure to meet normative expectations. 2 Although the dependent variable is perhaps more specific, for the purposes of this article I will refer to this as 'perceived community representation'.
Specifying 'national' government is useful because it is should be easily understood that people's views of their local governmentwhich, in the British context, are usually more positive than those of the central government (Hansard Society, 2019)are not sought here. I nonetheless acknowledge certain ambiguities in the chosen dependent variable. It may be that the concept of perceived 'community representation', as expressed through this variable, captures attitudes to both elected government officials and unelected officials such as civil servants. Scholars have noted that senior civil servants are increasingly part of the 'public face' of government in the UK and are sometimes drawn into effective advocacy for it (e.g. Grube, 2015). It is not undesirable if attitudes to non-elected officials are captured, however, as some (e.g. Jones and Stewart, 2012) have argued that Britain's central civil servants are to some extent responsible for the degree of responsiveness to local need (and why it is often lacking, as they are accused of operating a system of 'command and control' over local government). A more genuine challenge in using this variable is that it is not entirely clear whether it refers to 'specific' or 'diffuse' evaluations: a negative response about representation by 'national government' could conceivably capture attitudes both towards the incumbent, and a general sense of a remote, uncaring state 'regime'. Because I am more interested in understanding the roots of the latter than the former, I control for government approval, which implies that the remaining variance to be explained refers to a more 'diffuse' attitude.
In short, perceived 'community representation' builds on existing literature while incorporating an original focus on community. It refers to perceived responsiveness in the policy process to local people by central government actors, potentially non-elected as well as elected.
Where there are undesirable ambiguities in the question wording, these are addressed where possible by decisions around modelling and controls.

Bringing geography in
The United Kingdom is an appropriate test-case for the effects of spatial inequalities. The UK is among the top ten OECD countries for inter-region gaps in disposable income (McCann, 2016). Long-term processes, related to the legacy of deindustrialisation and the increased importance of property wealth, are compounded by shorter-term effects brought on by the Great Recession of the late 2000s and early 2010s. For instance, Forth finds that in the period between 2007 and 2011, the decline experienced by the economy of South Yorkshire (as measured by GDP per capita) was almost as large as that of Greece; and it has since experienced very little of the return to growth found in areas such 1 In the British Election Studies, these questions are adapted by changing the subject, but not the object: for instance, successive iterations have asked respondents to indicate their agreement with the statement that 'politicians don't care about people like me'.
as London (Forth, 2017). Wealth inequalities between areas are even more stark: as of 2013, in Kensington and Chelsea, the unitary authority in the UK with the highest property prices, these were nearly twenty times those seen in Blaenau and Gwent (Savage, 2015). It is recognised that these divergent economic outcomes have broader social impacts: Buchan et al. (2017) find compelling evidence of the North-South divide in a large and widening gap in early mortality since 1995. Surveys suggest public awareness of the regional dimension of UK inequalities: in a 2014 survey, nearly half of respondents agreed that 'Britain's economy is more regionally divided than it was 30 years ago' (Johnson, 2015).
Yet the focus on inequality as a regional phenomenon may obscure how spatially concentrated disadvantage is in the United Kingdom. While it is generally the case that the North is home to most of the UK's poorest areas, the most deprived place in England is the Essex seaside village of Jaywick (Gill, 2016): parts of Thanet in Kent and Great Yarmouth in Norfolk join it in the top twenty. It is not merely that a large or well-populated area can have its outliers, but also that, in many such areas, there is no 'normal': as Dorling and Pritchard (2010) put it, 'poverty and wealth are fractal in their geographies ' (p. 90). At the level of Lower Super Output Areas (LSOAs), a small statistical area comprising around 1-3000 residents, areas in Glasgow, Liverpool and Manchester at the 5th decile of deprivation are on average bordered by at least one LSOA where 35% of people are income deprived and one LSOA where just 10% of people are income deprived. In the city of Glasgow, even areas with the least deprivation are on average bordered by at least one area where 30% of people are income deprived (Livingston et al., 2013). The extent of concentration means that, to explore the effects of people's lived reality, economic context should be considered in a more localised fashion than has so far been utilised in research into discontent.
However, it remains to be demonstrated what the relevance of local conditions really is. Why is it that people should care about community conditions, rather than merely their personal situation? Do community conditions matter only in an absolute sense, or is the perception that one's area is being 'left behind' (by the country at large) also important? In the following sections, these questions shall be addressed by rethinking the existing literature on discontent that addresses economic factors. Since I have no strong theoretical priors about which 'discontent' attitudes perceived community representation is most related to, I draw from the literature liberally: while I recognise that there are important differences between (for instance) the literatures on trust, democratic satisfaction and external efficacy, these may all give useful guidance as to how local factors may function.

Resource effects and their application to communities
In the trust literature, scholars frequently invoke 'rational choice' explanations (Cook and Santana, 2018), wherein trust is based on selfinterest. Such a view is echoed by Newton et al. (2018, p. 48) who state simply that the affluentas society's 'winners'should be less likely to be dissatisfied with 'social arrangements that have served them well', and thus more likely to trust the system. For the most part, existing studies suggest that this assumption holds. 3 In the United States, some studies have detected income effects (Soss, 1999;Jackson, 1995), which connected higher levels of household income to higher levels of 'external political efficacy' (i.e. more positive views of whether politics is responsive to people like oneself). Similarly, studies with large crossnational data sources have demonstrated some association of higher income with increased trust (Van Der Meer and Dekker, 2011) and higher satisfaction with democracy (Castillo, 2006).
It is notable that income appears to have the same effects across these different forms of political satisfaction/discontent. This is not the case for another key 'resource', education, where higher education tends to increase 'external efficacy', but has been found to decrease satisfaction with democracy, in the sense of falling short of citizen expectations (Norris, 2011). As such, there may be something particular about the experience of living on a low income that catalyses a somewhat generalised discontent with politics.
This may be derived from various sources. In part, it may stem from inequalities of representation. Though not as comprehensively tested in the United Kingdom, Gilens and Page (2014) have demonstrated for the United States that 'economic elites' and business interests have substantial impacts on government policy, while 'average citizens' and mass interest groups do not. Peters and Ensink (2014) show for Europe that government welfare state effort is more responsive to the preferences of better-off groups than to the poor, especially where their preferences diverge most. Secondly, economic change, whereby those on low incomes have experienced stagnating wages (Lansley, 2011), may cause people to perceive that politicians fail to deliver for them, regardless of policy congruence. There may be an element of social distance from professional politicians: MPsand cabinet ministers even more soare high earners by definition and so may be seen by people on low incomes as less capable of understanding their needs. Allen and Sarmiento-Mirwaldt (2015) find that those with incomes below £40,000 were significantly more likely to feel a 'discommunion of interests' between themselves and MPs: that is, to feel that MPs were less affected than them by economic conditions and less reliant on public services. Finally, a phenomenon like political trust is partly 'anticipatory': trust is highest when one can expect the best in the future (Scheidegger and Staerklé, 2011). Yet many people on low incomes are never far from circumstances of real poverty and tend to face greater challenges of employment insecurity: they may therefore have less confidence that any success they experience will be sustained.
Although we lack evidence on the effects of low income as a contextual factor, it is possible that it matters at a contextual level for much the same reasons: under-representation of the poor in policy, an apparent failure of politics to deliver for the community, a social distance from its people, and an inability to expect a good future for the community. Indeed, it may be easier for the public to diagnose systemic failings from observing patterns of experience in their community (repeated job losses, business closures, homelessness, evictions and so on) than it is for them to connect their own, limited experience to political failure, as personal experience is often 'compartmentalized' (Sears, 1993, p. 144) and 'depoliticised' (Mutz, 1992).
While people will inarguably have imperfect knowledge of their community's 'resources', the spatial proximity of poorer and richer areas in the UK provides people with some means of gauging conditions: most people are likely to have some yardstick for what a community that is doing well or doing badly would look like. As such, it is likely that people have at least a broad understanding of economic conditions in their area. The above leads to my first hypothesis.

H1. A higher average local income is associated with positive views of community representation.
It is essential to acknowledge at this point that the take-home pay of locals may not be the specific cause of an income effect. For the purposes of this paper, any effect of income should properly be considered as a proxy for the effect of occupying a broadly higher or lower position in a hierarchy of local economies: that is to say, being advantaged in the distribution of economic resources. For this reason, I refer to this explanation as the community resource model of perceived community representation.

The role of 'sociotropic' perceptions and why the local should matter
While the processes I investigate here should not be devoid of a material foundation, how individuals perceive their world is crucial. Perhaps surprisingly, the evidence for the importance of economic perceptions to relevant evaluations is somewhat more circumspect than that for concrete individual 'resources'. At the macro-level, Alesina and Wacziarg (2000) provide a compelling explanation for the long-term decline of trust across most nations in the developed world, arguing that the end of the post-war years of growth and stability has led to long-term loss of confidence in the government's ability to provide good economic conditions, although Dalton (2016) disputes this. In crosssectional, individual-level studies, findings are inconsistent as to whether poor national economic perceptions are responsible for a lack of trust (Hetherington and Rudolph, 2015): for instance, Citrin and Green (1986) found that their influence in the USA varied dramatically across a four-year period in the early 1980s.
While it remains an open question as to how far economic perceptions affect political discontent, there is a near-consensus that, if they do, rather than 'egotropic' evaluationsthose related to how an individual perceives their own situation, it is 'sociotropic' evaluationsthose that pertain to national economic conditionsthat matter. The public, in the view of Anderson and Mendes (2005, p. 9), 'blame the government less for their personal economic difficulties than for failing to produce good economic outcomes for the country as a whole'only the former, they find, influences trust in government.
What is left mostly unaddressed by the literature is whether perceptions of the local economy also hold any importance. However, there is a firm theoretical foundation for this expectation. The basis for the importance of national-level evaluations to political attitudes is believed to be two-fold: first, it may be based on 'pro-social' concern for others; second, it may be based on an expectation that national conditions indicate the economic risks and benefits that one can personally expect in future (Kinder and Kiewiet, 1981, p. 132). Both, however, indicate that perceptions of local circumstances should be of real importance. The pro-social factor should be present: if anything, in-group sentiment is likely to be stronger in the real community (populated by people we know and care about) than in the 'imagined community' of the nation-state. Secondly, since people usually work close to their home, people are likely to give some weight to local economic conditions when they consider their personal prospects. 4 Beyond the theoretical, this argument is supported to some degree by existing empirical research. Firstly, in the context of the United States, Rogers (2014) demonstrated that 'communotropic' economic perceptions affected Congressional approval, controlling for both personal and national equivalents. Since Congress is a key representative body, it is plausible that this reflects local circumstances shaping perceptions of the underlying quality of representation. Secondly, scholars in the field of electoral geography have shown that localised considerationssuch as views of whether one's local area has been getting more or less prosperous than other parts of Britaininfluence people's political judgements in a UK context, namely their voters at General Elections (Pattie and Johnston 1995, Tunstall et al., 2000, Pattie and Johnston, 2006. Thus, localised conditions have a proven importance in UK politics, and have been shown (albeit in a different national context) to influence considerations connected to representation.
From this follows a basic proposition: evaluations of the representation of one's community should be influenced by so-called 'communotropic' perceptions of the local economy (Rogers, 2014). This general expectation takes the form of two specific hypotheses.
H2a. Negative views of recent change in the local economy are associated with negative views of community representation.
H2b. The perception of a higher rate of unemployment in one's community is associated with negative views of community representation.
This explanation shall henceforth be referred to as the communotropic model of perceived community representation.

Distributive justice and a theory of 'grievance'
While the communotropic dimension is likely to be a relevant one in and of itself, it may be that its importance also derives from how the community is seen in relation to the norm. This expectation is derived from core principles of the 'distributive justice' literature, particularly the class of distributive justice theories focused on 'fraternal' or groupbased 'relative deprivation', or RD (Runciman, 1966). Under principles of distributive justice, when a group is not conferred the benefits that accrue to 'salient others' (Tyler and Lind, 1992, p. 122), members of that group tend to experience a sense of grievance, which informs their confidence in the system that is perceived as responsible for distributing resources (in this instance, government or the political system as a whole). As Hooghe et al. (2017) have lamented, the distributive justice perspective is rarely applied in research on trust and, similarly, the literature linking distributive justice to other discontent outcomes is rather barren. However, there is a substantial body of work connecting relative deprivation to engagement in protest (Van Zomeren et al., 2008) and non-co-operation with authority (Verboon and Goslinga, 2009), from which feelings of discontent with authority can most likely be inferred. Indeed, recent work by Zhang and Zhou (2018) has made the direct connection: they show using experiments that trust in government mediates the effect of group-based distributive justice on noncooperation with authorities. It should be acknowledged that RD, as operationalised in these studies, does not specifically refer to intergroup comparisons, such as the 'local-national' dynamic investigated here. Focusing specifically on intergroup comparisons, Smith et al. (2012, p. 216) find that across nine studies, RD was 'solidly associated with negative attitudes towards the larger system', such as support for political authorities.
Applying this basic principle to the local/national dynamic, it follows that people may be especially likely to experience discontent with the representation of their community when they perceive the economic situation in their local area as worse than that which exists nationally. This does make an assumption that 'national conditions' are indeed a relevant 'reference group' for social comparison. However, this is also a common issue in the RD literature in general: as Budria and Ferrer-I-Carbonell (2018, p. 10) have noted, 'large-scale surveys do not contain direct questions about the composition of the reference groups'. Instead, the relevant reference group is often inferred from results (Delhey and Kohler, 2006).
One of the necessary pre-conditions for a 'grievance' effect is that people actually attribute the responsibility for economic conditions to 'politics', in some respect. It is generally acknowledged that attributions are 'important determinants of judgements, decisions and behaviour' (Ige, 2014(Ige, , p. 2180. It might be argued that the attribution of responsibility for the local economy is not as clear as for the national economy, and so they are less likely to blame the authorities. It might also be argued that, even if both the quality of the local and national economy were attributed to politics, people might not hold government responsible for unequal prosperity. However, Johnston and Pattie (2002), using the 1997 BES, show that most people (56%) do indeed hold 'government policies' responsible for how economic development at the local level had compared to 'other parts of Britain' over the previous five years. The existence of attribution for local, national and relative conditions suggests that the 'grievance effect' is a plausible result of relative deprivation perceptions.
Attribution is also relevant, in that it points towards a specific formulation of this general idea. Theoretically, this is based in 'group justification' theory, wherein people experience a psychological need to 'develop and maintain favourable images of one's own group', in this instance members of one's geographic community (Jost and Banaji, 1994). People who believe they live in a poorly-performing area, and believe that the national situation is significantly better, might seek external causes for the relative deprivation of their in-group, which do not reflect on the qualities of local inhabitants. They may, therefore, be more likely to understand the area's underperformance as an unjust result of political failure. However, those who believe they live in a high-performing area, but that the national economy is performing poorly, may be less keen to attribute their relative 'gratification' to political choices that have benefited their area and its inhabitants. In other words, people tend to blame political processes rather than credit them, in line with the general perspective of 'grievance asymmetry theory' in the economic voting literature. 5 Indeed, Johnston and Pattie (2002) found that those who perceived the local economy to have got worse (compared to others) were substantially more likely to blame the government for what had happened in their area: as such, it seems that the asymmetries are likely to translate to 'communotropic' perceptions.
Thus, the following expectation presents itself: the more negative the view of local economic performance, the stronger the association between positive views of the national economy and negative views of community representation. This I shall term the grievance model of perceived community representation. Specifically, it is represented by the following hypotheses.
H3a. The more positive the view of the national economy, the stronger the association between negative views of the local economy and negative views of community representation.
H3b. The lower the estimate of national unemployment, the stronger the association between higher estimates of local unemployment and negative views of community representation.

Data and dependent variable
I use a merged dataset, combining survey data on respondents in England and Wales from Wave Three of the British Election Study Internet Panel (Fieldhouse et al., 2016: fieldwork September to October 2014 with official statistics at the Middle Super Output Area level. Wave Three applied an innovative survey method to elicit specific respondent considerations about their 'community'. Respondents are asked to draw on a digital map the area which they consider defines their 'community', with the map initially centred on their household. Respondents were subsequently asked 'How much do you think the interests and views of people in your local community are listened to when important decisions affecting them are made by national government?', rating this on a scale of 1-4 from 'not at all' to 'a great deal', along with a wider battery of questions. Although the measure is ordinal, ordinal models violated the proportional odds assumption, according to common tests (e.g. Brant, 1990), potentially creating 'a misleading impression of how the outcome and explanatory variables are related' (Williams, 2016). Although certain statistical workarounds exist (such as generalised ordered logistic regression), these are cumbersome to present and often poorly understood (Williams). As such, I chose to dichotomise the DV. However, due to the distribution of responses, standard 'median split' practice for dichotomisation did not offer a single logical cut-point. The DV could be cut between 'none at all' and 'not very much', or between 'not very much' and 'somewhat'. I chose the former. Feeling 'not at all' listened to is a clearer signal that a democratic expectation is being violated than 'not very much' and should be more clearly associated with discontent. This is supported by the fact that 'not very much' respondents look more different to 'not at all' respondents than they do to 'somewhat' and 'a great deal' respondents, in terms of their higher levels of more general 'discontent' as measured through efficacy, trust and democratic satisfaction. 6 From the standpoint of model fit, dichotomising between 'not at all' and 'not very much' also performed better than the alternative.
The choice of dependent variable is primarily justified on theoretical grounds. However, in Appendix 1, I undertake an empirical validation of the dependent variable, using EFA to test the principle that it measures a distinctive attitude to those yielded by the existing repertoire of questions (satisfaction with democracy, external efficacy, government satisfaction, perceptions of politicians and so on). The analysis did not yield a satisfying factor solution, but the analysis of Squared Multiple Correlations between the items displayed very high specific variance (Child, 2006) for the dependent variable. As Samuels states, this is often indicative of an additional factor that requires further items to detect in EFA. Considering this, it is unlikely that the variable taps a more generalised discontent, and more likely that it measures something genuinely new and distinctive.

Independent variables
The independent variables are as follows. Firstly, I model the effects of retrospective economic perceptions, for both national and community-related perceptions. The former is measured by asking 'How do you think the general economic situation in this country has changed over the last 12 months?'. For the latter, respondents are asked 'Thinking about your local community, how do you think the general economic situation has changed over the last 12 months?'. These scales are reversed such that higher numbers denote increasing negativity about each kind of economic perception. For simplicity of interpretation, these are treated as continuous in the models. Perceived unemployment rates in one's community/the country as a whole are measured by a question asking respondents to place a marker for each on the same slider from zero to 100 per cent, encouraging respondents to gauge one against the other.
In addition, objective context is measured at the Middle Super Output Area level for England and Wales: small census areas of between 5,000 and 15,000 people. The BES offers a wide variety of potential contextual 'containers', linking respondents to large Government Office Regions, local authorities, Parliamentary constituencies, and, at the smallest level, MSOAs. My decision to use MSOAs is informed by two factors. Firstly, it most accurately captures the scale of objective spatial inequalities in the United Kingdom, which are most evident the less aggregation is undertaken. Secondly, small areas are necessary because they most closely approximate the kinds of areas which people consider to be their 'community', according to prior research. Wong et al. (2018) find that, using the mapping method replicated in the BES, Canadians' community maps had a median area of twelve square kilometres; MSOAs in England and Wales have a mean of twenty-one squared kilometres in area. By comparison, the mean constituency in England and Wales encompasses 263 square kilometres, which indicates the utility of far more granular data.
The specific variable at MSOA-level is the median net household income, based on model-based estimates for 2014 (Office for National Statistics/ONS, 2016), which include all major income sources net of various outgoings such as taxes, rates and housing costs. 5 See Mueller (1970) for the original statement of the theory. 6 For example, the mean of satisfaction with UK democracy (1-4 scale) by levels of perceived community representation is 2.05 for 'not at all' respondents, 2.48 for 'not very much' respondents and 2.68 for 'somewhat' respondents.

Control variables
In all models, individual-level socio-economic controls are included, in order to assess the presence and magnitude of their effects relative to those deriving from local factors. The principle decision taken in this regard was to control for household income: based on the ordinal variable of self-reported gross annual income by income category in the British Election Study. However, as Pickett and Pearl (2001, pp. 119-20) have noted, 'in studies with only a single individual-level SES [socio-economic status] variable, the neighbourhood-level SES variables may be capturing unmeasured individual-level variation'. As such, I also include a four-category variable for respondent qualifications, another important component of SES, and a three-category measure of working status (in work or study; working-age out of work; retired). Gender, age, and marital status are included as further demographic controls. Additionally, all models control for government approval (on a 1-4 scale), and for party identification, since the 'winner-loser gap' in levels of discontent is a 'consistent and persistent' feature of public opinion in democratic systems (Anderson et al., 2005, p. 60).
At the contextual level, I include controls measuring urban-rural status of the community, which has been a particular concern in more recent research. Research suggests that places outside of major cities often perceive 'relative social, political and economic deprivation' (Jennings, 2017). Specifically, I use a measure of the population density of the MSOA. While official measures of urban-rural status exist, the number of categories is too smallfailing to capture, in particular, the differences between cities and small towns. Although a continuous measure of population density has its weaknessessuch as failing to identify a potential non-linear relationship whereby towns are highest in discontent (Jennings, 2017) it ought to suffice as a control. Indeed, it is correlatedalbeit weaklyto the measure of local incomes, such that the denser a place is, the worse-off it is. Thus, the inclusion of this control should aid in the measurement of the contextual effect of local economy.

Testing against 'cultural' factors
Drawing on the literature on radical right voting (e.g. Norris and Inglehart, 2016), I take a particular focus on testing against 'cultural' factors, which constitute a plausible alternative set of influences on perceived community representation. In practice, the cultural factors identified by these studies usually relate to ethnic diversity, foreignborn populations and immigration (Amengay and Stockemer, 2018). It is possible that, as in the radical right literature, the effect of the (real or perceived) economic context is overestimated if measurements of 'cultural' context are omitted.
The set-up of testing economic factors against cultural ones is not intended to imply a total dichotomy between them: I acknowledge that perceived economic and 'cultural' context may be related, potentially reciprocally (Lamerís et al., 2018a and2018b). Nonetheless, it is unclear how strong their relationship is and how applicable the underlying theory is to the UK, while empirical tests suggest that in my data there are only modest correlations between the two. These issues are explored and tested in Appendix 2. As such, I maintain that the approach to use 'cultural' controls is preferable, regardless of the persistence of mild endogeneity concerns, given the alternate risk of omitted variable bias.
For the models which deal with the effects of perceptions of the local economy (i. e., the communotropic and grievance models), I use perceived cultural contexts as the relevant controls. Specifically, these are the respondent's estimate of the proportion of the local population which is White, the estimated proportion of those born outside the UK, and the respondent's perception of whether their community has become more diverse (on a 1-5 scale). The last of these is of particular utility, because the nub of the cultural explanation does not hinge on numbers per se, but on a 'cultural backlash' against the perceived decline in the 'social centrality' of white natives (Gest et al., 2017(Gest et al., , p. 1698. Should economic perceptions still have explanatory power, we would be in possession of genuine evidence that culture does not trump economics in explaining why people feel their communities are poorly represented. For the models that test the effects of real economic contexts, I include alternative 'real' cultural contexts. These are selected in order to parallel the perceived cultural contexts mentioned above. I use the percentage of the population of the MSOA in 2011 (the most recent census year) which is non-White and non-UK born. To estimate changes in diversity, I calculate the percentage point growth in the proportion of non-Whites between 2001 and 2011 in the MSOA. Due to the very large correlations between all three of these variables, separate models are specified.

Modelling contextual effects
The models specified below are at a single-level, including those attempting to assess contextual effects. This choice is, first and foremost, a consequence of the sparse data structure encountered, which identifies 6,023 respondents in 3,526 MSOAs (for an average of 1.7 respondents per MSOA). The literature suggests that the utility of multilevel (ML) models may be compromised at such degrees of sparseness: for instance, Clarke and Wheaton (2007) recommends a minimum of five observations per group for ML modelling. Furthermore, it can be assumed that no local-level clustering (which would necessitate ML) occurs through the data collection method, as might occur with a multistage sampling method: for instance, where individuals are selected within neighbourhoods.
Utilising single-level models does result in one important limitation, in that it is not possible to estimate and compare the between-places and between-individuals variance, nor is it possible to assess the degree to which between-places variance is explained by economic context. As such, this article makes no claims as to the extent of the importance of context, nor whether economic context is the dominant factor in local context.

The effect of real conditions
In the first instance, it is crucial to understand how far attitudes vary predictably according to real economic resources in the communitywhich should serve as the most direct test of whether spatial inequality is a genuinely important phenomenon determining how people feel about representation. In order to test the validity of the 'community resource' model, I specify Models 1a-c. (Table 1).
Models 1a-c deliberately omit the local economic perceptions (of change and unemployment), which are expected to be partly derived from the 'real' local conditions as captured through the variable for local average incomes. It should be apparent that the results support the community resource model. Models 1a-c show that we can reject the null in the case of H1, which proposed that a higher average local income is associated with positive views of community representation. Average weekly income after housing costs had a highly significant effect, and in the expected direction: higher local incomes are associated with a lower probability of believing one's community is 'not at all' listened to (see Fig. 1). Notably, this effect of income is not found at an individual-level: people who estimate their household income in the higher brackets do not tend to believe their communities are better represented.

Communotropic versus conventional
The second line of enquiry pursued here concerns the basic relevance of 'communotropic' perceptions. These shall be contrasted with the 'conventional' model, which expresses the contribution of the range of predictors discussed in the established literature. Specifically, it encapsulates the effects of individual-level resources (household income, level of education, working status), personal experience (retrospective household finances), and national-level economic perceptions (retrospective general economy, and perceived national unemployment rate).
It was argued that negative views of the local economy should be associated with higher discontent with the representation of one's community by central government. This general theory was expressed by two specific hypotheses. Firstly, H2a stated that negative views of recent change in the local economy are associated with negative views of community representation. Model 3 (Table 2) indicates that it is indeed possible to reject the null hypothesis. The effect of more negative opinions about the local economy's performance in the past year is positively-signed, and highly significant: thus, negative views of the local economy are associated with negative views of community representation. Furthermore, Model 3 indicates that this relationship is not item-specific. H2b stated that the perception of a higher rate of unemployment in one's community is associated with negative views of community representation, and in this instance we can also reject the null hypothesis. From Model 3, we can observe a positive relationship between the respondent's estimated level of unemployment in one's community and negative views of community representation. Hence, Model 3 provides considerable support for the communotropic model.
To illuminate the contribution of communotropic factors, three further findings can be highlighted. Firstly, the effect size of local retrospective perceptions is substantial. Moving from the most positive to the most negative perception increases the predicted probability of expressing high discontent from 0.10 to 0.39. Secondly, the communotropic model appears to have substantially better fit with the data compared to the 'conventional' model, according to all three fit statistics tested (AIC, BIC, and R 2 ). Thirdly, whereas in the conventional model (2) perceptions of the change in the national economy had a significant effect, this does not occur when the variables measuring communotropic factors are included in Model 3. It follows from the above, as well as from the support found for H2a and H2b, that the communotropic model is of significant value.

The role of grievance
The third theoretical proposition was the 'grievance model', which posited that people would make more negative judgements of community representation if they saw their community as being 'left behind' compared to the nation as a whole. In Models 4 and 5, I test the grievance model over two distinct variants of economic perceptions. Firstly, I include a variable for level of economic negativity: one relating to negativity about one's local community, and one relating to the national situation. H3a states that the more positive the view of the national economy, the stronger the association between negative views of the local economy and negative views of community representation. Support for the grievance model, in this case, would arise if the interaction term between these variables were negative, meaning the lower the level of negativity about the national economy, the larger is the effect of negativity about the local economy in increasing the propensity for negative views (and vice versa). From Model 4, we observe that the expected relationship is not found in the case of local and national economic evaluations. As such, we fail to reject the null hypothesis for H3a.
Secondly, in Model 5, I model the interaction of national and local unemployment. Similar to the above, H3b states that the lower the estimate of national unemployment, the stronger the association between higher estimates of local unemployment and negative views of community representation. Again, support for the grievance model would arise if the interaction term were negative, indicating that the lower the estimate of national unemployment, the larger is the effect of Table 1 The effects of economic context on the probability of high discontent with community representation.

Variables
Model 1a Model 1b Model 1c Age ( Notes: Models include controls for party identification (not shown here due to number of categories and available space). *p < 0.05; **p < 0.01; ***p < 0.001. Fig. 1. The effect of average weekly earnings in one's Middle Super Output Area on the probability of high discontent with community representation. Notes: error bars represent 95% confidence intervals of the estimates from Models 1a-c (Table 1).
a high estimate of local unemployment in increasing the propensity for negative views (and vice versa). In this instance, the expected relationship does emerge. As such, we may reject the null for H3b, thereby offering support for the grievance model. These effects are most easily understood from a comparison of the slopes in Fig. 2. It has been argued by Ai and Norton (2003) that interaction effects in logit and probit models can be misspecified by conventional techniques. To rule out the possibility of a false positive, I calculate the 'cross-derivative' using inteff in STATA (Norton et al., 2004). This confirmed that the interaction was highly significant, in the correct direction, and equivalent in size. As proposed above, the most likely explanation for the interaction derives from the psychology of group attachment. People who see a much worse situation in their local economy than nationally will tend to want to displace the responsibility from the community itself and are likely to indict the government (or the political system) instead.
However, those who perceive their communities to be succeeding ahead of the nation are likely to use this in a process of psychological bolstering that gives credit to the in-group of local people, rather than to assess this as an example of extraordinary success by political authorities.
In spite of the support for H3b, there is a real problem of interpretation here, insofar as H3a was not confirmed. There are two viable explanations. First is that the sense that one's area is falling behind on a 'getting better'/'getting worse' question is not as damaging as falling behind on the unemployment question. In theory, people could see their area as failing to flourish in the short-term, but still possessing structural advantages over the rest of the country and thus retain the sense of being overall 'winners'. Indeed, I find that the ordering of 'better/worse' and 'higher/lower' unemployment rates were often at odds. Of those who rated the local economy as better than the national economy on the 'general' question, thirty percent deviated, and indicated that local unemployment was higher than national unemployment. Similarly, of those rating the local economy worse, some fifty percent thought there was lower unemployment locally. 7 It is unclear how much of this is due to respondents making a short-term/structural distinction, but it may be enough to blunt the impact of being 'left behind' on 'general' perceptions.
The alternative explanation is survey methodology. This is in two senses. First, the far more fine-grained 0-100 scale for unemployment perceptions may be more suited to the detection of interactions than the five-point scale. For instance, respondents who think the national economy is poor, but the local much worse, are left with little room to indicate the real gulf in their perceptions on the 'general' question but a lot more on the unemployment question. Again, the data supports this. Notes: Models include controls for party identification (not shown here due to number of categories). *p < 0.05; **p < 0.01; ***p < 0.001. Fig. 2. The effect of estimates of local unemployment on the probability of high discontent with community representation given perceived national unemployment. Note: Error bars represent 95% confidence intervals of the estimated probability of high discontent obtained from the results displayed in Model 5 (Table 2). 7 There is a slight tendency on the unemployment questions to think that things are better locally; on the economy 'in general', however, the reverse is true. Though unclear why, one speculative suggestion is that it may reflect media prominence given to the idea of a widespread use of unemployment and other benefits by the 'work-shy', which is perhaps sometimes disconfirmed at a local level while being maintained as a belief about the nation as a whole.
Nearly forty percent of respondents evaluated the local and national economies equally on the 'general' questions, but just eleven percent did so on the unemployment question. Of the forty percent, nearly nine in ten gave either higher or lower estimates for local unemployment than for national. 8 The unemployment question may also shake respondents out of survey-answering strategies which could suppress interaction effectsfor instance, 'straight-liners' and 'midpoint stickers' may have to give more consideration to how they answer the unemployment questions than the general ones. The second possible method effect is that the unemployment questions directly invited a comparison between local and national circumstances in a way that the 'general' questions did not. The local/ national unemployment questions were asked together, while the local/ national general economic perceptions were asked in different parts of the survey. Some respondents could therefore use the unemployment scales in a more expressive way than they could the 'general' scales (and as suggested, they could draw particularly dramatic comparisons).

Cultural factors, demographics and other controls
In the results discussed above, I find no evidence to suggest that 'culture trumps economics' in its influence on perceived community representation. Nonetheless, cultural factors are not irrelevant: although none of the variables measuring cultural context had effects, there was a significant effect (in all models) of perceiving one's community as being more White on feeling the community was not represented, and of perceiving a growth in community diversity had the same effect. How, precisely, a community's real or perceived cultural context affects discontent is an important question for future work. At the contextual level, I also observe an effect of population density. Specifically, this highlighted that people in less dense areas were more likely to be high in discontent, despite the relative affluence of rural areas.
In terms of the individual's demographics, two further points of interest emerge. The retired also appear distinct in their relative contentment with community representationalthough age is insignificant. This may be linked to greater involvement in community activities among retired than working people in the United Kingdom (Matthews et al., 2014) and the corresponding sense of influence that can come with local civic engagement (Johnson, 2014). Finally, gender stands out, with men more likely to perceive their communities as poorly represented. This has a potentially interesting implication. Gest (2016) suggests that political discontent is more acute among some men because they have struggled to adjust to both long and short-term changes in their local economy, often accompanied by the erosion of institutions (the union, the working men's club and so on) which they perceived afforded them status above their social class. This implies that gender and local economy may be interlinked in contemporary political discontent in the advanced economies. Given the role for local economies and for gender that emerges here, this is a plausible theory in the British context which, although out of scope of this paper, research might fruitfully explore.

Conclusion
I began by noting the way the narratives used to understand political discontent had begun to incorporate a spatial element in the political moment post-Brexit, but expressed concerns that this had not been accompanied by real scholarly understanding of this dimension. In this article, I have sought to identify the precise impact of community economiesboth from the standpoint of their real economic conditions and the perceptions people hold about themand the mechanism that links them to perceived community representation. Below, I summarise the major findings, exploring their potential implications, for how the discipline could explore this further, and in highlighting the general value of paying attention to geographic inequalities.
Firstly, people appear to be influenced by community 'resources'with low average incomes in one's area being associated with an elevated likelihood of discontent with community representation. Insofar as incomes are a good proxy for economic performance, this finding offers a counter-point to the finding of Jennings and Stoker that the 'two Englands' were no different in their degree of political discontent. While they posited that this might be an 'emerging trend' that they could not yet discern, the evidence presented here suggests instead that it had already manifested in some formalbeit, not necessarily in more generalised judgements about politics and politicians, but instead in more particular perceptions of community representation.
Secondly, replicating a common finding in the political trust literature (see e.g. Anderson and Mendes, 2005), personal circumstances and 'egotropic' economic perceptions are of minimal importance in perceived community representation. This can be seen in terms of the non-significant findings over two variables that capture different egotropic dimensions (household income, and personal economic perceptions). However, this study diverges from some previous literature, including Rogers, in that one's view of the national economy is not a significant factor. Despite its apparent utility in explaining political trust, the conventional 'sociotropic' perspective, which focuses on national perceptions, is not wholly satisfying in understanding why people perceive a poor quality of representation in their community.
Thirdly, this research has found that there is a substantial role for an explanation based on the 'grievance' associated with perceived relative deprivation between one's community and the country-at-large. This finding indicates that the common perception held by the public-atlarge, that different parts of the country are on different economic trajectories, has tangible effects on the sentiment that one's community is ignored.
Going forward, there may be opportunities to improve on the methodology utilised here to access 'grievance'. In particular, it will be important to interrogate the claim that being 'left behind' is typically processed as an injustice: work by Jost and Banaji (1994) on 'system justification' suggests that this will not always be the case and that often people will justify their disadvantage rather than let it drive them to anger. More broadly, it will be important to discover the affective content in 'grievance'. As Van Zomeren et al. (2008) argued, the 'cognitive component' in 'relative deprivation'the weighing up of one's own group's situation against another -is a far less powerful political motivator than the emotions of anger and resentment that may accompany such calculus.
Furthermore, there is a risk that working from people's nationallocal comparisons captures 'grievance' only in a limited sense. This study identifies those who feel their area is suffering compared to the country at large. However, an alternative source of grievance might be the sense that select areas of the country monopolise the fruits of economic growth. This way of seeing distributive politics may be potent politically, because it identifies specific beneficiaries in a way that might be associated with a degree of 'resentment'. Cramer (2016) shows how the resentment of big cities found in Wisconsin residents is profoundly connected to their views of representation: the relative economic failure of their own area is frequently explained by her interview subjects through a sense that successful places are subject to political favouritism.
It is not hard to see plausible parallels internationally, particularly in the United Kingdom. It may be that the high importance and ongoing ascendancy of London is especially important to the development and politicisation of grievance. Mandler (2016) has written that, along with struggling former industrial areas and coastal towns, ostensibly prosperous parts of the country perceive themselves as un-represented and deprived of power because of its monopolisation in London, 'the home of all of Britain's elites ' -financial, political, journalistic, professional. To acquire an improved picture of the nature and extent of grievance, as well as its ramifications for political discontent, future survey research might incorporate the perspective that people may see their area as struggling less in relation to the country and more in relation to specific centres of economic power.
The above analysis is, of course, merely a first step in understanding the effects of economic context on discontent. One principle limitation is that the average local income is only a static measure of the local economy. While this may capture something important, it is nonetheless somewhat abstracted from how people actively experience the economy: that is, as a dynamic system in which trajectories in prosperity and hardship may be as important as their levels, and may powerfully influence community psychologies. Given recent studies suggesting that localised decline is linked to cultural 'threat' (Colantone and Stanig, 2018;Carreras et al., 2019), which may be in turn linked to discontent, advances in this area may be especially important both in uncovering new facts and in connecting literatures that at present are generally presented as conflicting explanations.
Modelling the effects of change poses challenges of its own, of course: especially at small-area levels, where rates of growth are not estimated by official statistics. However, political science has recently begun to confront these obstacles. For instance, in order to study the economic drivers of Brexit, Jennings (2017) constructs an index of 'relative decline' at constituency level using a variety of indicators, including employment changes, business growth, population inflows and outflows, and changes in the proportions of degree-holders. Such a methodology may provide a useful precedent for how the relationship between economic trajectory and perceived representation might be approached.
It will also be valuable to explore how perceptions of 'community representation' can be better captured. This study used a single item, whereas for several reasons multi-item measures are generally preferred. Further, it would be useful to know to what extent people responsibilise and evaluate different political actors within central government for 'community representation', and in particular how they judge non-elected officials relative to government. In addition, the item used placed an emphasis on 'listening' -yet, as Esaiasson et al. (2015) argues, attitudes to representation also relate to how far governments 'adapt' policy decisions in line with public opinion, and to how well they 'explain' those decisions.
While the findings related to economic resources, economic perceptions, and the grievances of the 'left behind' are all of individual significanceand in their own way may contribute to academic and political conversationsthe wider perspective offered by this paper is that these findings reflect another dimension of the politics of inequality: the fact of real geographic divides with genuine consequences for residents' life experiences. While political science has not entirely eschewed this perspective, its application to research concerning political trust, democratic satisfaction, and similar political attitudes has thus far been limited. I contend that attentiveness to the geographic divides, and how they come to be understood by the general public, constitutes a perspective of real value to the study of questions of alienation, distrust and discontent that are central to our research, and that profoundly leave their mark on contemporary politics.

Declaration of interest
This research has been supported by the Economic and Social Research Council and the Hansard Society [grant number 1775110].