Media Independence through Routine Press-State Relations: Immigration and Government Statistics in the British Press

When and how does press coverage maintain independence from governments’ preferred language? Leading scholarship argues that elites shape media content, especially in foreign affairs settings where journalists rely on official sources. But do media push back in domestic policy contexts? Focusing on immigration in Britain, we find press coverage exhibits signs of autonomy that rely on the state’s administrative branches. Our evidence comes from automated linguistic analysis of 190,000 items of migration coverage in nineteen national British newspapers from 2006 to 2015, and press releases published by the U.K. Home Office between 2010 and 2015. We show that the press increasingly portrayed immigration in terms of its scale. Then, by comparing the dynamics of a key government policy—lowering “net migration”—in press and Home Office rhetoric, we illustrate the limits of the government to insert its desired language into the press. Finally, we argue routine press interactions with the nonpolitical Office for National Statistics enabled coverage that diverged from politicians’ preferred lines. Our study contributes to press-state theory by providing evidence of media semiautonomy in a domestic policy arena, and highlighting the often-overlooked role of routine, bureaucratic procedure in supporting that autonomy.

Mass media play critical roles in many normative theories of representative democracy. Journalists serve as watchdogs against corruption by acting as conduits for information about politicians and policies which allows citizens to hold elected officials accountable for their actions (Curran 2005). In this model of basic democratic responsiveness, governments act, reporters cover those acts, and the public responds. More broadly, journalists might aim to provide the citizenry with "a more thorough, complete, objective, and reliable portrait of relevant public life" (Schudson 2015: 256).
Leading theories of press-state relations present a contrasting, and troubling, picture. Elites dominate media coverage. Investigative journalism is rare. Most information is provided in terms and frames chosen by elites. Journalists rely on official government sources, either out of choice or necessity (Cook 1998). In this manner, media coverage becomes "indexed" to power (Bennett 1990), or takes cues from governing elites in a process of "cascading activation" (Entman 2004). Of course, elite domination of media frames is not complete. There are examples of media "semiautonomy" (Bennett et al. 2007), when cascades move upstream against the prevailing current. So, rather than asking about press independence in absolute terms, the question for theories of press-state relations becomes how and when do departures from elite domination occur (Lawrence 2012)?
Our analysis of British press coverage of immigration shows evidence of media independence. We argue this independence is rooted in an underappreciated but theoretically important information source: the administrative arm of the state. Prior accounts have found media autonomy arising from exceptional events or unusually effective oppositional strategies (Bennett et al. 2007;Lawrence 2012). In contrast, we demonstrate that media independence can be grounded in the routine coproduction of news (Cook 1998) by journalists with government bureaucrats. Specifically, we point to official data produced and disseminated by civil servants-who are guided by professional norms and incentives to maintain bureaucratic independence-as sources for media coverage. We find elements of news coverage that remained independent from political leaders, offering the public a tool to hold government accountable for failing to meet a prominent campaign promise.
We address two empirical questions. First, how did the British press talk about immigration between 2006 and 2015, and did key trends follow or depart from the preferred language of powerful governmental actors? Analyzing a comprehensive corpus of British newspaper coverage of immigration, we find sharp growth in discussions of immigration in terms of its scale. Comparing these findings to government communications, we find divergence in language between press and policy makers, and signs of media holding government accountable for a major campaign promise (a numerical target for "net migration"). Specifically, the language of scale increased in the press just when government communications shifted to downplay immigration numbers.
Second, how did this instance of media semi-independence develop? We argue that the administrative state, exemplified by the United Kingdom's Office for National Statistics (hereafter ONS), played an important role in sustaining press independence through routine rather than exceptional actions. Analyzing the relationship between press behavior and regular public releases of migration data, we illustrate how nonpolitical administrative actors provided alternative touchstones for journalists to frame immigration differently from top political actors.
Our analysis contributes to understanding press-state relations in several ways. Empirically, we advance efforts to understand how and when mass media can retain independence. We show that politically neutral civil servants can play a distinct role in press-state relations, complementing prior studies that cast politicians and politically appointed bureaucrats as primary contributors to the coproduction of news (Ciboh 2017;Cook 1998;Davis 2007;Figenschou et al. 2017;Rice and Somerville 2017). Methodologically, we demonstrate the value of using automated corpus linguistic methods to document changes in language use, especially when coupled with analysis of concurrent political changes. Finally, we conclude in a normative vein, linking our findings to broader arguments about how media should engage with politics and society. We point out how journalists, as they coproduce news with a variety of political sources, can benefit from looking toward institutions that generate independent, expert knowledge (Schudson 2010(Schudson , 2015.

Theorizing Press-State Relations: Exceptions to Indexing
Current understandings of press-state relations build on Bennett's indexing hypothesis and Entman's related model of cascading activation. According to Bennett (1990), news stories tend to revert to ("index") narratives provided by official sources, especially in foreign policy topics where journalists might be more likely to rely on higher-ranking officials. Entman's (2004) cascading activation model also argues that frames normally spread downward from top government officials to journalists and news organizations.
A limited set of studies showing evidence of semiautonomy give prominence to the role of exceptional circumstances. Bennett et al. (2007) suggest that three factors reduce reliance on indexing: "unexpected and dramatic news events," "skillful oppositional communications strategies," and "investigative journalism" operating outside of "routine daily reporting" (see also Lawrence 2012). Each of these factors depends on a nonroutine occurrence, whether it is an event that catches government officials off-guard and disrupts their "spin," an unusually "creative" oppositional tactic (Bartholomé et al. 2015), or an investigative effort that goes beyond everyday journalism. Glazier and Boydstun (2012) add time as another factor that can reduce elite dominance of media frames, showing that U.S. Presidents' ability to frame a foreign policy crisis decays after the onset of crises. Otherwise, past research tends to associate media autonomy with exceptional cases. This fits with the logic of indexing theory, which developed from accounts of journalistic routines themselves.

From the Rare to the Routine: Government Agencies in Press-State Relations
Our account both adds to and differs from these efforts by emphasizing relatively lowranking but professionalized civil servants as information sources who bolster journalists' abilities to independently frame issues. We show that the British press devoted increasing attention to official statistics tracking "net migration"-the annual difference between immigration and emigration-after it disappeared from politicians' own communications. This press coverage, running counter to governing elites' interests and revealed preferences, was consistently anchored in the information that government agencies produced and communicated. Specifically, coverage was tied to regular, routine reports by the public statistics authority rather than unusual events, investigative journalism, or strong oppositional strategies.
Accounts of indexing and cascading activation do not regard civil servants as serious barriers to political elites' dominance of media framing. Bennett et al. (2007: 28) view such "lower-ranking officials" as unlikely to have the political and organizational power to influence media coverage. Meanwhile, Entman (2004: 11) portrays the entire U.S. executive branch in political terms, as part of the presidential administration. Undoubtedly, these contentions are often accurate, especially in foreign policy issues (see also Robinson et al. 2010).
Institutional models of press-state relations do pay more attention to bureaucrats' role. Cook (1998: 141-49) observes how journalists use executive branch bureaucrats as sources in the coproduction process. However, his empirical and theoretical focus remains on political appointees, who are much like elected officials in that making news is intertwined with making policy. Likewise, in the United Kingdom, Davis's (2010) work shows the close relationships between journalists and Westminster (the home of Parliament) rather than civil servants, despite the key policy role the latter group played. For Davis, these relationships became too cozy, creating an insular London-centered news discourse suffused with political spin rather than actual policy content. In general, research on state-press relations has foregrounded the political rather than the administrative state. As Liu and her colleagues (2012: 227) note, "[s] cholars examining government communication generally have focused on the elected side of government and its interaction with the media, largely ignoring the public administration, bureaucratic side of government." Indeed, recent empirical studies in several settings tend to stress the influence of political appointees, at the direct expense of civil servants. In Northern Ireland, political Special Advisors intervene in communications between media and civil service Government Information Officers (Rice and Somerville 2017). Meanwhile, in Nigeria, political aides are deeply involved in managing the coproduction process, at times demanding positive coverage while offering information subsidies or even direct payments to journalists in return (Ciboh 2017). Sometimes, the relative newness and weakness of democratic institutions and norms allow for very direct political intervention in news production. However, in Norway, similar outcomes arise simply from journalists' choices of sources: reporters seek out political ministers more than career civil servants, even for questions that call for policy-specific expertise of the type that civil servants are likely to provide (Figenschou et al. 2017).
Our argument complements, rather than contradicts, these prevailing depictions of press-state relations. As shown above, much evidence supports the notions that media coverage is both frequently indexed to power and created in processes of coproduction involving elected officials and political appointees. However, we suggest these processes and outcomes can be disrupted not only in exceptional cases but also through journalists' use of routinely available information from government sources that come from administrative rather than political branches of the state. This routine availability of independent information might also help explain why (some) domestic issues are less susceptible to indexing than the paradigmatic cases of foreign policy crises (Bennett et al. 2007;Entman 2004;Robinson et al. 2010).
Of course, that governments produce and publish data is not surprising: "Much of 'what government does' consists of obtaining information" (Hood and Margetts 2007: 21). While such information might be suppressed or altered for political ends in autocratic or developing countries (Sadiq 2005), bureaucrats in consolidated democracies usually share commitments to professional norms and their organization's missionand maintain considerable autonomy to pursue that mission (Hood and Lodge 2004;Wilson 1989). Substantive competence and procedural independence are critical to sustaining these organizations' reputations (Carpenter and Krause 2012), becoming powerful career incentives for individuals working within them (Mulgan 2007). In government agencies charged with producing credible data, "independence and professional integrity" (Majone 1997) are especially critical. The ONS-a pivotal player in our story-enshrines these values in its mission statement, as later detailed.
Given the existence of independent and credible information sources, journalists can depart from rather than replicate the preferred language of power-holding elites by using civil servants and the data they generate. By making this claim, we follow Schudson's (2010Schudson's ( , 2015Schudson's ( , 2016) recent arguments about the democratic value of "political observatories" (Lippmann 1920), or sources of expertise in socially and politically important matters that might be housed within universities, government, or civil society.
Two qualifications are important to clarify our argument. First, civil servants and government data do not always support alternatives to governing elites' preferred frames. Indeed, data may actually enhance the government's narrative. When these factors do not diverge, journalists' use of civil service sources will not alter the outcomes predicted by indexing or described by accounts of coproduction with policy makers. Second, we do not claim that government data reflect truth or reality in unproblematic ways. Political observatories-within or outside of the state-may produce expert knowledge, but experts can be mistaken and procedures for generating data can be flawed or incomplete. In our case, headline ONS statistics on net migration have been criticized for having large margins of error, potential built-in biases, and limited usefulness for policy makers concerned with the effects of different types of migrants (Allen et al. 2018;Cangiano 2016). The point is not that civil servants produce perfect knowledge but rather that they produce information using procedures independent of partisan politics.
That said, our main contribution is to show how routine press-state interactions can sustain a degree of media independence. Data are regularly gathered and disseminated by lower-ranking state bureaucrats, who deliberately cultivate reputations for independence and competence rather than pursuing political ends. This type of information became a critical source for journalists covering immigration in Britain over the last decade, to the detriment of government-preferred frames. While previous studies might suggest forms of news coproduction with partisan politicians are more typical, our findings point toward an important alternative rooted in different sources of information.

The Case of Immigration in Britain's Media and Political Systems
Before proceeding to our study, we will briefly introduce the key elements of our empirical case. Immigration was a highly salient issue in British politics throughout the period covered by our newspaper corpus (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), Since the early 2000s, the British public has consistently ranked immigration among the three most important issues facing the country, with majorities heavily favoring increased restrictions (Duffy and Frere-Smith 2014). In the 2010 General Election campaign, then-Conservative Party leader David Cameron pledged to reduce annual net migration from "the hundreds of thousands to the tens of thousands" (Conservative Party 2010). This positioning was arguably an important factor in the Conservatives' (plurality) victory in 2010 (Evans and Chzhen 2013). Upon taking power, the Conservative-led coalition government designed and implemented policies aimed at reducing immigration to Britain via three major routes: labor, student, and family migration (Cangiano 2016).
Aside from its political salience, immigration in the British context has broader relevance for press-state relations, potentially speaking to other issues and media systems. Notably, immigration does not stand out as a particularly easy case for finding media independence. We acknowledge that domestic policy issues are less favorable to indexing than foreign policy issues (Lawrence 2012), and that salient, contested issues such as immigration may operate differently from less visible or contested ones. However, research on British migration policy making paints a picture of elite dominance with minimal autonomous media influence: newspapers serve as venues for claims-making by other actors rather than as actors in their own right (Statham and Geddes 2006: 256). Others may ascribe larger roles in the policy process to thinktanks, NGOs, and broader epistemic communities, but they still discount any direct role for the media (Balch 2009; Somerville and Goodman 2010). Second, at the country level, Britain's news media system shares a number of features that are widely shared (but not universal) in other consolidated democracies. Television and radio are led by public broadcasting as in much of Europe, while the newspaper sector is more market-oriented, making Britain difficult to classify as a media system (Davis 2014). British newspapers represent various ideological viewpoints, although they are not as explicitly tied to political parties as in Hallin and Mancini's (2004) Southern European polarized pluralist model. Despite this ideological bent, norms of objective, fact-based reporting still guide journalists' behavior (Davis 2007). Therefore, the indexing hypothesis and other U.S.-based conceptions of press-state relations are potentially applicable to the British case.
Third, politically, Britain is a consolidated democracy with a heavily majoritarian electoral system and, crucially for our analysis, a history of an independent and competent civil service (Hood and Lodge 2004). The values of the traditional "public service ethos," including "integrity, honesty, and impartiality" remain central to the self-definition of Britain's civil service, despite changes under Thatcher and New Labour that introduced other values such as efficiency (Horton 2006). These aspects of government are far from unique, but neither are they universal. Therefore, our analysis is most applicable to consolidated democracies with a strong, independent civil service and a liberal, market-based media sector featuring well-protected freedom of the press. Within the British state, two institutional actors play crucial roles in immigration politics and in our analysis. The first is the U.K. Home Office, a large government department headed by the Home Secretary who is a political appointee. This was the post held by Theresa May from 2010 until her rise to Prime Minister in 2016. The Home Office is central to formulating and implementing British immigration policy (Statham and Geddes 2006). The second key institution is the ONS, an independent agency that produces and publishes official statistics on Britain's people and economy. Unlike the Home Office, the ONS mission statement stresses its independence from political ministers and outlines aims of meeting goals for quality and professional standards (ONS 2013). Both the ONS and Home Office engage with the public through the media, using press releases and other forms of communication to reach their intended audiences. However, their different goals lead to outputs with different content, including varying language surrounding immigration.

Data and Method
Our study investigates a series of questions about British media coverage of immigration. We begin with descriptive questions: what is the nature of this coverage in British newspapers, and how has it changed over time? Then, we assess whether these trends conform with or diverge from preferred government discourse as seen in Home Office publications. Finally, to the extent that our data allow, we investigate how and when divergences occur.

Data Sources
We primarily rely on a large corpus of British press coverage from 2006 to 2015. We examine newspapers rather than broadcast or Internet media sources in part because of the better fit with our text-based method, and the prohibitive costs of collecting a comprehensive corpus of television or radio transcripts. This choice brings limitations. For example, we cannot consider the role of images or videos in media depictions of immigration. However, newspapers still play a leading role in the British media landscape, despite declining circulation of their physical editions. Studies of intermedia agenda setting acknowledge reciprocal influence between traditional and online media but find that online and social media are often dependent on original content generated by newspapers and television news (see Weaver and Choi 2017, for a review).
The corpus, collected from Nexus U.K. and Factiva archiving services, covers all nineteen national U.K. daily and Sunday publications that were continually published during 2006 to 2015. 1 A full list appears in Table 1. It contains all articles retrieved using the following terms appearing anywhere in the body or headline of the texts, where the "!" symbol was a wildcard that included variations: (refugee! OR asylum! OR deport! OR immigr! OR emigr! OR migrant! OR illegal alien! OR illegal entry OR leave to remain) NOT (deportivo OR deportment). Throughout this article, italicized terms refer to the words themselves rather than the actual concept or object.
This query was developed and validated as part of the "Refugees, Asylum Seekers, and Immigrants" (RASIM) project reported in Gabrielatos (2007). It intentionally avoids including the term migration because this could retrieve articles only referring to movement of nonhuman types, such as bird or data migration. Therefore, we can be confident that, when we call up and analyze instances of migration from the data set, these cases are actually referring to human mobility. The corpus still includes many uses of migration in articles that also include at least one term from the query. There is some risk of omitting articles that mention human migration while avoiding all other query terms. This dilemma represents the trade-off between precision and recall in information retrieval. We follow Gabrielatos (2007) in siding with precision by excluding items about nonhuman migration that might distort our findings. Articles that mention human migration but no other search terms are likely to be referring to migration only tangentially.
In total, the corpus contains 190,657 articles, comprising 157.4 million words, with exact and near-duplicate articles removed. 2 Most full-text articles were available using NexisUK. However, for two publications (The Express and Sunday Express) between 2006 and 2008, full-text articles were only available using the Factiva service. For a small number of months (0.8 percent of cases), some publications' archives had incomplete data, so they were dropped from the corpus. Comparing results of other publications between the two services did not reveal any obvious systematic biases or gaps in archiving for the ten-year period under scrutiny.
To identify patterns in government-preferred language, we also use a corpus of press releases and news items relating to migration that were produced by the Home Office. Published on the Home Office website, these documents contain government policy announcements, positions, and reactions to external events. For this study, we assume these documents accurately represent the priorities and preferences of the Home Office since they were drafted and approved by staff for public release and web publication. Automatically scraping all available documents from 2010 to 2015 that were classified by the Home Office as relating to borders and immigration captured 348 items.

Method: Corpus Linguistics
Our methodological approach comes from computer-assisted corpus linguistics. Corpus linguistics is a way of studying language by analyzing collections of texts ("corpora") to look for usage patterns, among other aspects, that may not be apparent when researchers read a small set of items (Baker 2006). Often, specialist software aids this process by organizing, processing, and quantifying the text data. Alongside this computerized assistance, researchers can view and interpret the quantitative patterns in context. Our analysis proceeds in both "bottom-up" and "top-down" directions (Tognini-Bonelli 2001). The former relies on identifying quantitative patterns in the corpora without predetermined goals. The latter approaches corpora with specific terms or questions in mind. In both approaches, we use three kinds of analysis. The first is frequency analysis, where specified terms and the articles in which they appear are totaled up to show how salient they are in a given corpus or subcorpus (which is a researchdefined subset of the larger body of text). The second is collocational analysis, a method that can use grammatical rules as well as statistical tests to determine how strongly one word is linked with a target word, as opposed to them appearing together by random chance. Conventionally, collocation is defined as "a co-occurrence relationship between two words" (McEnery and Hardie 2011). The third is concordance analysis, which produces "a display of every instance of a specified word or other search term in a corpus, together with a given amount of preceding and following context for each result" (McEnery and Hardie 2011: 241). Concordances are helpful in sorting data and disambiguating usages of words with multiple meanings, as well as qualitatively seeing how and when chosen phrases or collocations operate (Baker 2006).
The data set was organized, stored, and analyzed using the web-based software Sketch Engine (Kilgarriff et al. 2014). Our analysis relies on part-of-speech (POS) tagging, which attaches grammatical information for each word, enabling researchers to look for patterns in usage. We use this feature to identify all modifiers of migration and immigration in the corpus, for example. The specific set of tags used by Sketch Engine come from Marcus et al. (1993).

Changes in Press Coverage of Immigration, 2006-2015
To identify any divergences in language between press and government sources, we first need to establish how the British press characterized immigration between 2006 and 2015. By looking at modifiers of the terms immigration and migration, we can identify aspects of immigration that the press tended to feature. Modifiers are words The Sunday Times that describe, characterize, or intensify another word, usually an object. 3 In the corpus, the nouns immigration or migration appear 136,311 times. In about one fifth of those cases (22.6 percent), another word modifies the noun. Figure 1 shows the top ten modifiers of these terms ranked by their share of all modifiers. The numerator in this case is the number of times that the given word modifies either term. The denominator is the total number of modifiers for the two nouns. 4 The word net was the most common modifier in the corpus, constituting 14.5 percent of all modifiers of immigration or migration over the 2006-2015 period. This was closely followed by mass (14 percent of all modifiers) and illegal (9.8 percent).
Prior research shows how press outlets in several countries, including the U.K., draw on language about illegality when describing immigrants and immigration (Blinder and Allen 2016;Burroughs 2015;Hogan and Haltinner 2015). But these studies either focus on people who migrate (immigrants) rather than the process itself (immigration), or elide differences between the two concepts. By focusing on the specific phenomenon of migration, Figure 1 reveals a distinct set of words related to the scale or pace of movement, including net, mass, uncontrolled, large-scale, high, and more. Grouping these words together, we find when the British press described immigration or migration in some way, 38 percent of the time it was in terms of scale or pace. 5 Examining concordances of these a Result for "European" includes the variations "EU," "intra-EU," and "Eastern/East European." kinds of words, including mass or high, revealed further support for grouping them together as ways of discussing immigration through frames of quantities. In all of the concordance examples that follow, we emphasized the target word and its relevant collocates by using boldface, a feature which was not present in the original quotes.
There were several reasons why Labour wanted mass immigration-and once started, nothing was going to stop them. (The Sun, December 12, 2012) We are concerned that high immigration could discourage employers from investing in adequate training for those who leave school at 16 and don't get jobs-a danger acknowledged by the Government. (The Times, April 4, 2008) The prevalence of scale modifiers associated with immigration or migration grew over this period, as This bottom-up analysis of the modifiers in the corpus associated with immigration or migration reveals how the scale of migration increasingly characterized British press content. But while these findings contribute a better picture of how the British press covered immigration over several years, they cannot by themselves address our larger questions about press autonomy in relation to powerful government officials. Therefore, we shift to top-down analysis of one of our key emergent results: the rise of the term net migration. We argue for two main points: first, the persistence of press semi-independence, and, second, the grounding of this semi-independence in routinized interactions with state institutions.

Divergence of "Net Migration" in the British Press and Home Office Rhetoric
A prominent result that emerged from the bottom-up analysis is the predominance of the term net as a way of describing migration, seen in Figure 1. Net migration became more visible in public debate once the Conservative Party pledged to bring this figure down to the "tens of thousands" in its 2010 General Election manifesto. This pledge drove policy action on immigration, including further restrictions, once the party was in coalition from 2010 to 2015. Figure 3 displays how often the term net modified immigration or migration at a monthly level in the press corpus. It shows that overall mentions of net migration remained relatively stable from 2006 to 2013: the rolling average during this period remains steady between six and twenty-seven mentions per month. But, from 2013 onward, the number of mentions substantially increases. The rolling six-monthly average peaks at 132 mentions in May 2015-a General Election month-while the single month with the most mentions was November 2014 with 242 mentions.
What might account for this rise? At first glance, our findings seem to confirm the power of elite government actors to shape media coverage. After all, the Conservative party came to power in 2010 having very visibly campaigned on a platform of reducing net migration. But examining the timing of changes in press usage of net migration among other scale and pace-related words rules out this explanation. Scale-related modifiers hovered around 35 percent of all modifiers of immigration and migration in the corpus between 2010 and 2012, as Figure 2 shows. This period includes the 2010 general election campaign, when Conservative leader Cameron made his net migration pledge, and the initial period of Conservative-led governance when policy makers were most actively pursuing that target. Net migration, like other modifiers related to scale, only rose in press mentions after 2012.
Furthermore, this change in media coverage occurred after a shift in government rhetoric away from the language of net migration. Figure 4 shows how frequently the Home Office mentioned net migration in its official news and press releases-specifically, those items that were classified as dealing with borders and immigration by the department itself-between 2010 and 2015. In this corpus, the phrase was more visible in 2010 and 2013, but declined in 2014 before nearly disappearing in 2015.
Prior to 2013, the government consistently cast immigration as a problem of scale, opening opportunities to promote its pledge to tackle the issue on these terms. Initially, it had success in bringing net migration levels downward: according to quarterly reports published by the Office for National Statistics (2010Statistics ( -2015, provisional headline figures fell from 252,000 in May 2012 to 153,000 in May 2013. However, from this point onward, the figures rose dramatically to 336,000 by November 2015. Indeed, from mid-2013, it became increasingly clear that the government would fall far short of its self-imposed target, presenting a political problem for a government that had promised to reduce this figure by 2015 (Migration Observatory 2014).
Thus, the language of net migration and the scale frame, previously so politically useful to the Conservatives, began to work against them as the target became unattainable. Concordance evidence from the Home Office documents illustrates a qualitative shift in the way the government talked about net migration. At the beginning of the coalition government, the Home Office reiterated this campaign commitment, sometimes verbatim, even in the face of possible difficulties.
The Government remains fully committed to reducing net migration to the tens of thousands. (Home Office news item, December 20, 2010) Reducing net migration is a priority, Damian Green Immigration Minister said: "These statistics show why we must tighten our immigration system in order to reduce net migration to manageable levels." (Home Office news item, August 26, 2010) The 2013 peak in mentions corresponded with a drop in net migration-an achievement that was heavily reported by the Home Office, which eagerly claimed credit: "[O]ur immigration reforms are working, with net migration at its lowest level for a decade" (Home Office news item, July 22, 2013). An earlier news release also linked crime with migration by stating, "[p]olice and immigration reforms are workingcrime is down more than 10 percent and net migration has been cut by more than a third" (Home Office news item, June 26, 2013).
But by 2014, as the trend in aggregate net migration reversed, Home Office language shifted to focus on levels of migrants arriving from outside the European Union, a subcategory that did continue to fall. In total, 50 percent of the mentions of net migration in the Home Office documents during 2014 actually relate to non-EU or non-European Economic Area (EEA) migration, compared with only about 4 percent between 2010 and 2013.
The figures also showed net migration from outside the EU is down to levels not seen since the late 1990s, demonstrating that reforms to curb non-EU migration and tackle abuse in the work, family and student visa routes are having the intended impact. (Home Office news item, May 22, 2014) Our reforms have cut net non-EU migration to levels not seen since the 1990s and slashed overall net migration by a third since its peak under the last government. (Home Office news item, October 7, 2014) Finally, during 2015, net migration only appears twice in the Home Office corpusboth times in the same press release soon after the May General Election. The technocratic discussion and language around these mentions were a far cry from the bold statements in previous years: The Prime Minister has announced a new blueprint is being drawn up to reduce demand for migrant labour-a crucial part of the government's plan to cut net migration, and make sure Britain's immigration system delivers for working people . . . The MAC [Migration Advisory Committee] commission, which will examine how the Tier 2 (Skilled Work) visa system functions, follows the first meeting of the Prime Minister's newly-formed Immigration Taskforce, which has been tasked with reducing net migration and focuses on the domestic measures that the government can take to achieve this. (Home Office press release, June 10, 2015) When placed alongside official migration statistics, this concordance evidence suggests that the Home Office changed its communications strategy as quarterly figures diverged from the government's narrative of successfully reducing migration numbers. At first, it emphasized successes in reducing non-European migration, and then dropped the "net" language almost completely from news and press releases. And yet, as seen in Figure 3, the press continued using this language well into 2015, suggesting that newspapers were increasingly using language that the government was abandoning. The indexing hypothesis would not predict the dramatic rise in the use of net migration from 2013 to 2015. Instead, the press highlighted the government's failure to meet its own net migration target, making it possible for citizens to hold the government accountable for a campaign pledge. In this sense, British newspapers' muchcriticized coverage of immigration fulfilled one of the key functions ascribed to the media in normative democratic theory (Curran 2005).

Communicating Information: The Role of the ONS
We have shown how British newspaper coverage of migration changed over time, and argued that a pivotal change after 2013 worked in direct contrast to the government's preferred narrative. But how and why did this happen? In this section, we return to the overlooked role of the administrative state. Specifically, we argue that information provided by the ONS provided crucial grounding for semi-independent press coverage.
From May 2009, the ONS produced quarterly packages of data that included estimates of the numbers of people entering and leaving the United Kingdom. These were potentially helpful raw materials for journalists framing immigration in terms of scale or pace. To specifically look at the relationship between press coverage and ONS quarterly reports, we filtered the 3,227 mentions of net migration to include only those instances where the this phrased appeared within five words to the left or right of the terms ONS, statistic(s), or figure(s). This occurred 270 times. Figure 5, which plots these observations annually, shows a clear and substantial increase in how frequently the press cited ONS migration data in 2014 and 2015.
Furthermore, by subdividing the 201 occurrences in 2014 and 2015 by newspaper as done in Figure 6, we see that a variety of publications mentioned the ONS migration figures. These include left-(The Guardian) as well as right-leaning broadsheets (The Daily Telegraph) and tabloids (Daily Mail, The Sun). As noted earlier, British newspapers are known to represent political viewpoints and often cover immigration very differently as a result. However, the trends documented here appear to transcend the ideological orientation of particular newspapers.
Concordances of net migration drawn from these two years provide qualitative evidence of the anchoring role played by the ONS, and, further, of coverage that was critical of government performance-even from right-leaning publications such as the Daily Mail. Often, these usages juxtaposed the ONS figures with the government's promises on net migration, as seen in the following examples: ONS figures show that net migration rose from 154,000 in 2012 to 212,000 in 2013, which suggests we are a long way off David Cameron's target of 100,000 in 2015. (Daily Mail, June 28, 2014) The figure, the highest ever, makes a mockery of David Cameron's pledge to cut net migration to below 100,000. (Daily Mirror, November 27, 2015) The quantitative and qualitative linguistic evidence converge to suggest that British newspapers produced coverage critical of government policy using quarterly ONS data reports as a source. These data releases, and the accompanying interactions between media and civil servants, epitomize routinized operations of the administrative state rather than the exceptional events seemingly required in the indexing account.
Of course, the mere presence of ONS data does not explain the rise of net migration coverage. Press releases of government statistics do not automatically translate into press coverage, and myriad governmental and nongovernmental organizations in the migration sector were actively aiming to attract media attention at the same time (Balch 2009;Somerville and Goodman 2010). Between 2009 and 2013, data releases were not associated with increased mentions of net migration, suggesting that journalists were not monitoring the migration statistical releases with any particular zeal. From 2013 onward, however, the regular production and release of net migration data by ONS provided an anchor for routinized press coverage of immigration using this frame. The resulting critical coverage perpetuated the net migration language that the political branches of government were rapidly trying to shed, as illustrated by changing communications from the more political Home Office.

Other Factors: External Events, Public Opinion, Political Opposition
Routine reports on net migration may have helped drive mentions of immigration's quantitative aspects. But as noted above, they were not the sole cause. Other factors cited as influencing media coverage include exogenous events, public opinion, and political actors beyond government officials. Although the full role of these factors is beyond our present scope, we discuss each briefly here. We demonstrate that, even if each of these factors did help drive coverage, our argument about semi-independence still holds.
First, could external events have direct responsibility for changing coverage? As net migration numbers rose, it is possible that the intrinsic newsworthiness of the ONS data did, too. Statistical releases were painting a substantially different picture from the stated Conservative pledge to meet a target for net migration. This explanation fits Boydstun's (2013) "alarm/patrol" mode, where she argues media produce a burst of coverage around an event (the "alarm"), and then sustain a degree of attention afterward (the "patrol"). However, the case for this relationship is not straightforward: the use of net migration language did not track with net migration figures prior to 2010. It seems likely that not just the raw numbers but also the contrast between the migration statistics and the stated aims of the government made the ONS reports a useful hook for patrol-type coverage from 2013. More importantly, to the extent that migration flows affected news content, this would still illustrate a divergence from the indexing of coverage to power by showing the ability of data produced by the administrative state to sustain such divergent coverage.
Second, perhaps media coverage of immigration over this period was a proxy for public interest. British newspapers operate within a competitive marketplace: If there were no public interest in immigration, then we would not see much reporting on it. However, our study finds patterns that cannot easily be reconciled with a view of the press as merely following public concerns. Immigration is highly salient in British public opinion, but these concerns changed little during the period of our study (Duffy and Frere-Smith 2014). As a result, they cannot explain the changes in volume and content of news coverage from 2006 onward. However, from a broader perspective on public opinion, we suggest that public concern played a less obvious role in shaping immigration coverage. The net migration target appears to have been designed to resonate with media and public opinion, unlike other policy targets undertaken with different aims, such as providing internal discipline to policy-making processes (Boswell et al. 2015). So, political elites considered the extent and perceived nature of public concern when they generated this frame. Moreover, British public opinion is commonly assessed by polling on preferred quantitative levels of immigration, a natural fit with the language of scale. The numbers frame, then, can be viewed as jointly created by the media (and its pollsters), the public, and the Conservative party, rather than as a purely elite-generated construct.
Third, did political opposition contribute to the rise of net migration in the press? Certainly, the Conservatives lost public confidence on immigration during this period. Entering the May 2010 election, 28 percent of the British public thought Conservatives had the "best policies" on immigration, and on the eve of the following election in April 2015, this figure had fallen to 17 percent (Duffy and Frere-Smith 2014). If other political parties were responsible for pushing the net migration terminology from late 2013 onward, perhaps the elite-based model explains more than we have suggested.
However, there are limitations to this objection. First, the leading political opposition to the government came from the Labour party, which made little use of the language of net migration. Labour's approach avoided numerical targets. Instead, the party framed migration in terms of "mak[ing] immigration work for Britain" (Hanson 2014) through reinforcing minimum wages and mandating apprenticeships for "local workers." More importantly, even if we underestimate Labour's role in promoting net migration, our case would still not fit the indexing model. Indexing predicts a broader set of media frames when political elites disagree but relies firmly on the notion of press coverage being connected to power. Yet, while Conservatives had power over immigration policy during 2013-2015, the dominant change in media coverage was away from their preferred message.
Nonetheless, a limitation of our study is that it isolates the role of two of the many actors whose language may have informed media coverage. Further research could more fully explore the nature and impact of communications produced by Labour and other intermediaries comprising the immigration policy community (Statham and Geddes 2006), especially when compared with media coverage. In addition, future work could explore how other actors-with distinct political aims-spread information generated by bodies such as the ONS within and beyond immigration policy networks. This would shed light on whether journalists' use of civil service data may stem partially from agenda-shaping efforts by intermediary actors.

Conclusion
Our empirical study provides two sets of key results. First, we found that British press coverage of immigration between 2006 and 2015 departed from the preferred language of powerful governmental actors. Our quantitative analysis revealed that the share of modifiers related to the scale or pace of migration increased sharply over the 2006-2015 period, comprising nearly 60 percent of all modifiers by 2015. The phrase net migration became particularly visible from 2013 onward. At the same time, however, Home Office documents increasingly avoided this phrase. This divergence suggests that rising use of net migration and related "scale" terms was not indexed to power. Second, we have argued that civil servants' publication of credible data sustained semi-independent media coverage. Media coverage drew on statistical reports featuring net migration, providing information that citizens could use to hold politicians accountable for a salient campaign promise.
Our study makes several contributions to literatures on press-state relations and political communication. Substantively, we provide evidence of media semi-independence in the less-explored terrain of non-U.S. domestic policy, finding that seemingly mundane activities such as publishing statistical bulletins can support a degree of independent media coverage. Methodologically, we illustrate how corpus linguistic techniques offer efficient ways of detecting patterns in large amounts of text, while preserving enough direct access to documents thatenables researchers to interpret these patterns.
Our results also have normative relevance. We underscore the role of civil servants as sources of credible, politically independent information who provide journalists with alternatives to the language of governing elites. In our political moment of resurgent populism, leading political figures increasingly attack journalists and civil servants as independent sources of information. In the 2016 U.S. Presidential election and the United Kingdom's EU Referendum, for example, the Trump and Leave campaigns respectively derided experts and analyses they produced. Our study shows that credible data-generated by competent, politically neutral government bureaucrats-can support media coverage that, at times, can help citizens hold politicians accountable for campaign promises while in office.
We have linked our narrative to Schudson's notion of political observatories, revived from Lippmann's classic works on journalism and democracy. As these writers have argued, in large democratic societies it is impossible for journalists and citizens to have expertise in more than a small fraction of matters of public concern. British news media-and media operating in similar systems-face further challenges in the forms of "media management" (Kuhn 2007), political spin, and heightened market pressures accelerated by the Internet age (Davis 2014). These kinds of challenges can make independent, investigative reporting all but impossible to initiate and maintain. Therefore, we echo calls for incorporating independent sources of expert knowledge into public debate as a partial solution to this dilemma. Observatories "offer promise for developing the kind of public information that makes democracy possible" (Schudson 2010: 107).
Of course, these developments invite and require critical discussion about what constitutes reliable public information, and about the limitations of quantitative data for informing policy making (see Allen 2017; Jerrim and de Vries 2017). Nevertheless, observatories are now plentiful both within and outside government (Schudson 2010).
In times when sources of expertise and knowledge are under sustained, politically motivated attack, it seems more important than ever for journalists to routinely incorporate real expertise into public debate, rather than wait for an exceptional case to justify departing from the indexing of news to power.