Reputational drivers and deterrents of accountability: why are some EU agencies subject to more intense account-holding than others?

ABSTRACT Empirical studies have shown how some (semi-)independent agencies are intensely scrutinised by various public account-holders, while others are overlooked. Such variations in account-holding intensity have also been observed for EU agencies. This article aims to explain variations in the account-holding intensities that EU agencies are subject to. To this end, the account-holding that EU agencies experience from six major institutional EU account-holders has been measured based on a novel conceptualisation of ‘account-holding intensity’ as a combination of both the ‘frequency’ and the ‘diligence’ of account-holding. Based on a reputational perspective on accountability, public visibility and salience is hypothesised to drive account-holding intensity, and reputational strength and reputation-sourced authority is hypothesised to deter account-holding intensity. While reputational strength is not found to deter the overall account-holding intensities that EU agencies are subject to, results provide support for the hypothesis that public visibility and salience drive account-holding intensity.


Introduction
Empirical studies have shown that different public agencies are being held to account with varying intensity by public accountability forums (i.e., accountholders) such as legislatures and ministries.For example, Pollitt (2006) finds the performance management of agencies in four European countries to vary considerably between agencies, as some agencies 'languish in obscurity ' (p. 29).Similarly, Hogwood et al. (2000), note that only a few agencies in the United Kingdom (UK) receive the bulk of written parliamentary questions and experience a lot of informal ministerial involvement, while most agencies have 'minimal direct contact ' (p. 222) with these account-holders.
Another important avenue of research on EU agencies has been their accountability (e.g., Busuioc, 2013;Font & Duran, 2016;Scholten, 2014).This field of research has been fuelled by scholarly concerns about potential accountability deficits, considering the independence and non-majoritarian nature of EU agencies (e.g., Everson, 1995;Shapiro, 1997).Empirically, EU agencies have been found to be 'subject to different sets of accountability obligations with no clear justifications of why the differences (should) exist' (Scholten, 2014, p. 186).In other words, different EU agencies havesimilarly to their national counterpartsbeen found to be held to account with varying intensities.An example hereof is how only three EU agencies have been found to receive approximately half of all written questions from the European Parliament, while 'roughly three-quarters of EU agencies (…) hardly receive question-type oversight attention' (Font & Duran, 2016, p. 1363).This article aims to explain these variations in account-holding intensity, seeking to answer the following research question: Why are some EU agencies subject to more intense account-holding than others?
Account-holding intensity is measured with survey responses from directors and senior managers of 39 EU agencies pertaining to the accountholding that they are subject to from six major institutional EU accountholders: the European Commission, the European Parliament, the Council of the European Union, agency Management Boards, the European Court of Auditors, and the European Ombudsman.The measure is based on a novel conceptualisation of account-holding intensity as a combination of both the frequency and the diligence with which account-holding practices are executed.Together, the answers to the survey questions constitute an original dyadic dataset of 229 accountability relationships between EU agencies and account-holders, which leverages the unique knowledge and insights of directors and senior managers as agency representatives directly involved in external accountability processes and directly in contact with institutional account-holders.
Drawing on the burgeoning reputational perspective on accountability (Busuioc & Lodge, 2016, 2017), this paper takes its theoretical starting point in the idea that reputational considerations function as a 'filtering mechanism' (Busuioc & Lodge, 2017, p. 92) through which EU account-holders decide to prioritise certain EU agencies over others.As such, public visibility and salienceoperationalised with an unobtrusive measure of media attention in 27 major European newspapersis hypothesised to function as a driver of account-holding intensity, providing reputational incentives for account-holders to intensively carry out account-holding activities.Reputational strength and reputation-sourced authority (Bertelli & Busuioc, 2020) operationalised with an unobtrusive measure of the valence of Twitter comments addressed towards the EU agenciesis, in turn, hypothesised to function as a deterrent of account-holding intensity, providing reputational incentives for account-holders to refrain from intense account-holding.As such, this article is the first to test recently theorised relationships between reputational considerations and accountability behaviour in the context of the accountability of EU agencies.Doing so, the article contributes to a research agenda exploring the explanatory promise of bureaucratic reputation theory in the context of the EU regulatory state (Busuioc & Rimkute, 2019), with this theoretical perspective already having shown considerable purchase power in the US (Carpenter, 2001(Carpenter, , 2010) ) and other national contexts (e.g., Gilad et al., 2015;Maor & Sulitzeanu-Kenan, 2016).
Multilevel linear regression analysis is used to test the presented hypotheses.The regression models estimate the relationship between the hypothesised account-holding drivers and deterrents and the cumulative accountholding intensity that EU agencies are subject to from all forums at the focus of this study.Results show that publicly visible and salient agencies are subject to higher levels of overall account-holding intensity, while reputational strength does not register as a deterrent of account-holding intensity.

Conceptualising account-holding intensity
Account-holding intensity as a concept is under-theorised.This stems from the accountability literature generally assuming forums (i.e., accountholders) to be motivated and willing to hold actors (i.e., account-givers) accountable in order to prevent them from 'drifting' away from the interests of forums (usually political principals).However, contrary to such assumptions, empirical studies have shown that there are considerable variations in the intensity with which different forums discharge their accountholding tasks towards different actors, and there are many identified instances in which forums have been found to neglect their accountholding responsibilities (see Schillemans & Busuioc, 2015 for an overview).
Such empirical studies, concerned with (a potential lack of) account-holding intensity, are often implicitly focused on (one of) two dimensions of accountholding intensity: (1).Frequency (i.e., the quantity of account-holding activities) and (2).Diligence (i.e., the quality with which account-holding activities are executed).
With regards to the frequency of account-holding activities, accountability studies in various contexts have noted how various account-holding practices such as written parliamentary questions (e.g., Font & Duran, 2016;Hogwood et al., 2000), hearings (e.g., Busuioc, 2013), debates and dialogues (e.g., Auel et al., 2015;Walgrave et al., 2017), and usage of performance measurement information (e.g., Pollitt, 2006)are executed with inconsistent frequency.Oftentimes, a few accountability actors are found to be the subject of most account-holding activity, while the majority of actors receive very limited attention.In more general terms, account-holding frequency thus refers to how often forums ask for reports or other information on agency operations, how often forums ask (follow-up) questions and provide feedback on information received about agency operations, and how often forums attend account-holding meetings and hearings.
As pertains to the diligence of account-holding activities, some studies describe how account-holding activities are indeed executed, but not with care and interest.This is for example expressed in illustrations of forums that fail to carefully review accountability information (e.g., Dubnick & Frederickson, 2010;Heinrich et al., 2010), or depictions of accountability forums that do not ask oversight questions that are relevant to the issue at hand (e.g., Kluger Dionigi, 2020;Maricut-Akbik, 2020).Sometimes such lack of diligence is described as a result of a lack of capacities (e.g., time and expertise).For example, some management board members of EU agencies have been found to be poorly prepared and to remain passive during meetings, among other things due to 'resource deficiencies' (Busuioc, 2013, pp. 247-248).But there are also examples in which forums do not carefully execute their account-holding responsibilities due to what seems more like strategic considerations.Diligent account-holding, exposing poor performance and mismanagement of actors, could namely reflect poorly also on forums.Especially in cases where there are interdependencies between forums and actors, or the forum has been responsible for delegating power, responsibilities, and resources to the actor in question (Busuioc, 2013, p. 107;Dicke, 2002, pp. 461-462).Some forums, especially if operationally dependent on an actor, may furthermore be unwilling to antagonise the actor with diligently critical account-holding as they attempt to foster cooperative relationships (e.g., Schillemans, 2008).In more general terms, account-holding diligence thus refers to how thoroughly forums read the information that they receive, how actively forums participate in account-holding meetings and hearings, and the extent to which forums' questions and feedback are relevant to agency operations and pertinent to the accountability processes in question.
In sum, common sense dictates that in order for an actor to be intensively held to account, account-holding activities have to actually take place (frequency).If forums very rarely execute account-holding activities vis-à-vis an actor, that actor will (all things equal) simply experience less intense account-holding than an actor that is often subject to a variety of accountholding activities.However, 'simply measuring the level of activity reveals little about the impact' of account-holding (Auel et al., 2015, p. 284).As Busuioc (2013) writes: 'An obligation to supply an annual report and an actor that complies with this obligation will have little impact if the forum does not have the resources (time, expertise, administrative capacity), the interest to read the report, prepare for the meetings, or engage the actor in discussions ' (pp. 265-266).In other words, intense account-holding extends beyond mere execution, implying that account-holding activities are also discharged with diligence.
Importantly, account-holding intensity should not be taken as informative of the purpose and intent of account-holding activities.No matter if formal or informal, taking place either to steer agency actions or to evaluate agency performance, being either collaborative and learning-focused or adversarial and blame-focused, account-holding activities can take place more or less often (i.e., frequency), with the account-holder more or less thoroughly processing information and providing feedback relevant to agency operations and the accountability process at hand (diligence).As such, the accountholding intensity concept has the necessary flexibility to be applied to a variety of different forums with different account-holding tasks and roles.
With account-holding intensity being conceptualised as both the frequency and the diligence with which account-holding practices are executed, the question however remains why some accountability actors are being held to account with high intensity, while others 'operate at the periphery of political attention' (Schillemans, 2011, p. 10).

Hypotheses: reputational drivers and deterrents of accountholding intensity
Seminal accountability literature has characterised accountability as a process of managing and prioritising between different accountability expectations (e.g., Day & Klein, 1987;Romzek & Dubnick, 1987).While this literature has traditionally focused on accountability from the perspective of account-givers (i.e., actors managing the different expectations of different forums), this description of accountability has bearing for account-holders as well.For accountability forums, accountability is a process of managing and prioritising between different account-holding responsibilities vis-à-vis different accountability actors.EU account-holders are clear examples hereof, as they have account-holding responsibilities for a large number of EU agencies, as well as other actors, dealing with a large variety of issues.As such, EU account-holders have to engage in prioritisation processes in order to decide which agencies that will 'become the focus of limited organisational energy and resources' (Busuioc & Lodge, 2017, p. 92).Consistently, the theoretical starting point of this article is that reputational considerations help forums 'make sense of and prioritise among the seemingly endless plethora of accountability responsibilities' (ibid).As such, public visibility and salience is hypothesised to be a driver of account-holding intensity, while reputational strength and reputation-sourced authority is hypothesised to function as a deterrent.

Drivers: public visibility and salience
In their conceptual article on accountability and reputation, Busuioc and Lodge (2017) ascribe variations in forums' investments in account-holding to reputational concerns.In this article, they argue that -'because of increased salience or heightened controversy' (p.96)visibility makes it increasingly reputationally risky for a forum to neglect its account-holding responsibilities, while simultaneously providing opportunities for a forum to 'cast the organization as an active and diligent account-holder' (p.96).As such, forums can hardly afford to not execute account-holding activities, or to execute them without diligence, as pertains to visible issues and actors.Due to increased audience attention, the reputational stakes of account-holding are simply higher when the actor and the issue in question is salient.Ignoring account-holding responsibilities in such instances will be more reputationally damaging than otherwise, while being a frequent and diligent account-holder will return a higher reputational reward.Consistent with this notion, empirical studies have found account-holders to limit their attention to a few agencies dealing with high-profile issues (e.g., Hogwood et al., 2000), prioritise visible misconduct (e.g., Etienne, 2015), and only act in the face of scandals and disaster (e.g., Pollitt, 2006).
Salience is well-recognised as an influential factor also in principal-agent models on the political oversight of bureaucratic actors.In such models, political principals are assumed to be seeking electoral benefits.For such account-holders, oversight comes with opportunity costs (i.e., time spent on oversight cannot be spent on other, more electorally beneficial activities), and while political principals are likely to seek to implement various costefficient control mechanismse.g., 'deck-stacking' and 'fire alarms' (McCubbins et al., 1987;McCubbins & Schwartz, 1984) the costs of account-holding might still outweigh the benefits for issues that lack salience (e.g., Calvert et al., 1989).
A reputational perspective on accountability extends this logic, opening up for the possibility that the visibility and salience of an actor or issue affects the behaviour of forums beyond political principals, this time in the name of reputational benefits, rather than electoral gains.In other words, reputational theory 'put[s] forward a unifying logic of action' (Busuioc & Lodge, 2017, p. 94), in which salience can be expected to drive the account-holding intensity of accountability forums beyond political principals, also including forums with account-holding as its core task such as the European Court of Auditors and the European Ombudsman, as all account-holders are generally theorised to make cost-benefit analyses with the purpose of maximising broader reputational benefits critical to their legitimacy and authority.It is thus hypothesised that: Hypothesis 1: More publicly visible and salient EU agencies are subject to more account-holding intensity.

Deterrents: reputational strength and reputation-sourced authority
Account-holders are organisations with limited time and resources (Busuioc & Lodge, 2017) that are tasked with conducting time and resource intensive account-holding activities, which might not yield results of misconduct and poor performance.Account-holders can therefore be expected to prioritise actors and issues where they suspect to find evidence of misconduct or poor performance, using less of their scarce time and resources on actors where they do not expect to uncover substantive accountability findings (this is largely the basis of McCubbins and Schwartz (1984) now seminal argument for why legislators prefer so-called 'fire alarm' over 'police patrol' oversight).As such, the reputation of account-givers could function as important 'signals' in the environment (Jones & Baumgartner, 2005) of account-holders, informing the frequency and diligence with which forums decide to execute account-holding activities vis-a-vis different actors.
But while good reputation may very well be well-deservedafter all 'reputation accumulates in response to sustained, high-quality performance' (Mailath & Samuelson, 2001, p. 416)reputation remains based on perceptions, and can as such also be undeserved (Bertelli & Busuioc, 2020).In such situations, account-holders couldeven if alerted to potential problems remain hesitant to engage in account-holding activities vis-à-vis the wellreputed actors in question, while also shying away from diligent scrutiny for the account-holding activities that are executed.The reason therefor lies in forums' formal powers to hold actors accountable not being the sole sources of authority in accountability relationships.
For both actors and forums, informal 'reputation-sourced authority' (Bertelli & Busuioc, 2020) is largely based on audience support among 'coalitions of esteem' (Carpenter, 2001, p. 4).With both actors and forums being situated in often overlapping audience networks on whose support they depend, the power balance between a forum and an actor can swing in the actor's advantage if the audiences 'throw their weight' behind the actor (Bertelli & Busuioc, 2020, p. 41).In these instances, '[c]ertain control mechanisms or courses of action (…) while formally available, can become reputationally barred' (Busuioc & Lodge, 2017, p. 94).In other words, a strong reputation with broad audiences can serve as a 'force that offsets formal controls' (Busuioc & Rimkute, 2019, p. 1265).It is thus hypothesised that: Hypothesis 2: EU agencies with a strong reputation with broad audiences are subject to less account-holding intensity.

Method
This study focuses on the account-holding intensity that EU agencies are subject to from six major institutional EU account-holders.The European Parliament, the Council of the European Union, and the European Commission are co-principals of EU agencies (Dehousse, 2008), agency management boards have been described as 'the main and most direct confines on the grant of authority of the agencies and their directors' (Busuioc, 2012, p. 719), and the European Court of Auditors and the European Ombudsman are the EU's 'watchdog institutions' (Wille & Bovens, 2020).The analysis leverages an original dyadic dataset of agency-forum accountability relationships based on survey responses from EU agency directors and senior managers on the account-holding intensity that their agencies are subject to.

The dependent variable: account-holding intensity
The dependent variable is based on survey responses from EU agency toplevel managers.The survey was disseminated via email to 221 executive directors, deputy directors, administrative directors, and operational directors 1 of all 45 members of the EU Agencies Network. 2 The purposive sampling strategy of distributing the survey exclusively to top-level managers was chosen since top-level managers are commonly the most involved in accountability processes as 'organizational reporting is usually a decision made by a collective of top managers or the leadership team' (Yang, 2009, p. 84).75 respondents (33.9 per cent response rate), from 39 agencies (86.7 per cent), answered the survey.
The survey answers form a dyadic dataset with a complex structure in which each observation ( 421) is cross-classified in both respondents (75) and forums (6), with respondents being further nested in agencies (39).Considering the small agency group sizethere is only one respondent for a majority of agenciesbuilding regression models that reflect this data structure is potentially problematic, as models whose complexity is not supported by the data are prone to a loss of power as a result of being overfitted (Matuschek et al., 2017).To simplify the data structure, analyses are run only with the answers of the highest-ranking respondent for each agency.This leaves us with a dyadic dataset of 229 observations, in which each observation provides information about an accountability relationship between an agency (39) and a forum (6).This data selection strategy is informed by the notion that since the most senior officials are the most involved in accountability processes (see Bovens, 2005, p. 203), they are also the most knowledgeable about these processes.Averaging respondent answers to agencylevel was also considered an alternative to simplify the data structure, but was opted against as group-level means based on low group sizes have been shown unreliable (Bliese, 1998).
In line with the novel conceptualisation of account-holding intensity described earlier in the article, survey questions were designed to capture both the frequency and the diligence of account-holding.Differentiating accountability from related concepts such as transparency, accountability is considered to require not only the provision of information by actors, but also, at a minimum, dialogue between actors and forums in which information is discussed and actions and decisions are explained and justified (Bovens, 2007, p. 451).Consistently, the account-holding intensity measure is based on answers to four survey questions, with one question for frequency and one question for diligence having been asked in relation both to the provision of accountability information and to accountability discussions, respectively (Online Appendix, p. 1). 3  The dimensionality and internal consistency of the dataset was explored (Online Appendix, pp.2-3) in order to check if these four account-holding intensity questions reflect a common construct (i.e., account-holding intensity) and should be used in the analysis as a composite measure.Dimensionality was explored through an exploratory factor analysis, and the Cronbach alpha coefficient was calculated to measure internal consistency.The results of the factor analysis indicate unidimensionality (only one factor has an eigenvalue > 1.00) and internal consistency (α = 0.809).For complementary information, analyses have still been run also for the frequency and diligence questions separately (Online Appendix pp.9-11).

The independent variables (and controls)
Media attention is used to measure public visibility and salience, as major traditional media outlets have a wide reach and more extensive spread than most other forms of communications (including social media content such as Twitter comments).Several empirical studies have provided evidence suggesting that media attention provides important 'salience signals' (Carpenter, 2002, p. 490), bringing agencies to the forefront of the political agenda (e.g., Bertelli & Sinclair, 2015).The media attention measurement used in this article (Online Appendix, p. 5) is based on a search in the Factiva database of 26 newspapers in five languages (English, French, Spanish, German, Italian), from seven EU member states (UK, 4 Ireland, France, Spain, Germany, Italy, Austria,), over the five-year period before the start of the data collection (October 2014-October 2019).Since the survey questions ask about the forums' account-holding intensity in general, rather than the intensity that agencies were subject to at the specific time of the study, measuring media attention over a five-year time period was considered preferable to a measure spanning a shorter timeframe (for example one year), as specific events over the short-term might cause spikes in agency media attention that is not representative of the general salience of the agency.The selection of newspapers is building on the salience measure of Reh et al. (2013), who selected newspapers with a particularly strong focus on EU affairs.In their study of European Parliament written questions to EU agencies, Font and Duran (2016) added three Spanish newspapers to this list, which have been kept for this study.Including a large number of newspapers from several member states lowers the risk for national media biases (e.g., EU agencies being more in the media in the countries where they are based).The search terms used were the full agency names in all languages of the different newspapers (as according to their founding regulations).Acronyms of the agency names were not used in the search terms, as this was often found to return a large number of irrelevant search results.Exceptions were made for Frontex, Europol and Eurojust, which are often exclusively referred to by their acronyms.A manual review of the results furthermore showed that searching on the acronyms of these agencies did not return irrelevant results.The media attention variable was found to be heavily right skewed and have a non-linear relationship with the dependent variable, and was therefore transformed with a logarithmic transformation following Mosteller and Tukey's (1977) 'bulging rule' (p.84).
The valence of Twitter comments towards agencies in 2018 and 2019 (up until the start of data collection in October) 5 has been used to measure agencies' reputational strength.The Twitter valence data is adopted from de Boer et al. ( 2021), who conducted a sentiment analysis of Twitter comments directed at EU agencies using an, on Twitter-data pre-trained version of a BERT machine learning language model to assess the reputational consequences of agencies' Twitter interactions.The model produced a sentiment classification of each Twitter comment as either positive, negative, or neutral.Subsequently, de Boer et al. calculated 6 an overall score, ranging from −1 (maximal negative reputational consequences) and +1 (maximal positive reputational consequences), for each agency.The assumption for this paper is that agencies that have recently been subject to more positive Twitter comments enjoy a stronger reputation (and vice versa).Relying on sentiment analysis is an established approach in reputation literature.Prior reputational measurements relying on sentiment analysese.g., measuring reputational threats (Maor & Sulitzeanu-Kenan, 2016) and reputational histories (Salomonsen et al., 2021) has focused on the valence of traditional media coverage.However, traditional media valence has recently been criticised as a measure of reputation, as 'media coverage is mediated by editorial concerns' (Bellodi, 2022, p. 3).In other words, media can be seen as more of an intermediary of audiences, filtering out what aspects of agency operations and what audience opinions to report on.In contrast, Twitter is a public arena where a variety of audiences, including traditional media, public organisations, politicians, interest groups, industry organisations, and citizens, all have the opportunity to openly critique, or praise, the EU agencies.Twitter comments thus directly taps into the unfiltered sentiments of different audiences in a way that newspaper articles do not.
Five control variables have also been added to the analysis.These control variables are: agency type (Joint Undertakings), regulatory tasks, budget size, staff size, and organisational age (Online Appendix pp.3-4).The EU agencies in this study include decentralised agencies, Joint Undertakings (JUs), Common Security and Defence agencies and the European Institute of Innovation and Technology.As public-private partnerships, the governance of JUs differ from those of other agencies, and the account-holding intensities that they are subject to might be affected by this.
In addition, some agencies have regulatory tasks, which makes them more powerful than non-regulatory agencies.All agencies were coded as regulatory or non-regulatory, following the categorisation of Busuioc and Jevnaker (2022) between agencies that have the decision-making powers to adopt binding acts (i.e., regulatory) and those who do not (i.e., non-regulatory).
As per budget and staff size, these variables have been shown to matter in previous accountability research.For example, Font and Perez Duran (2016) have found that EU agencies with more staff receive more written questions from the EP, and Pollitt (2006) found that national agencies with larger budgets receive more attention from their parliaments and ministries.Data on both budget and staff size was collected from the European Commission's annual financial reports for the years 2015-2019.Budget size is reported as the annual average budget in millions of Euros (based on total payment appropriations available during the year), and staff size is reported as the annual average number of staff members employed by the agency (including EU officials, EU temporary agents, EU contract staff, and seconded national experts; excluding interim workers and consultants).
Organisational age (at the time of data collection) was calculated from the year of establishment, as specified in agency founding regulations.
By collecting independent variables and control variables that are based on unobtrusive dataunrelated to the survey used to collect the accountholding intensity data (the dependent variable)common source bias is avoided (Favero & Bullock, 2015).All continuous variables (both independent and control variables) have been centred, providing a more interpretable intercept.Since this study focuses on the effects of agency-level (level 2) predictors, the variables were centred at the grand mean (CGM) (Enders & Tofighi, 2007).

Results
Table 1 presents the descriptive statistics of all variables (raw) used in this study.Table 2 showcases the bivariate correlations between these variables (centred and transformed).Strong correlations can be observed between media attention and twitter valence (Pearson's r = −0.72),and media attention and staff size (Pearson's r = 0.62).Strong correlations between independent variables could cause collinearity issues.However, Variance Inflation Factor (VIF) scores do not indicate such issues.Additional diagnostic tests were run on the presented regression models (Table 3), checking: linearity (plotting residuals vs. fitted values), homogeneity of variance across groups (Levene's test), and normality of residuals (Q-Q plots) (Online Appendix, pp.6-9).Major issues were not identified in these diagnostic tests.The models were also checked for influential observations (Cook's D).As influential observations were identified, regression models with robust standard errors were run as a robustness check (Online Appendix pp.12-14).Discrepancies between results in the presented models and the regression models with robust standard errors are mentioned in text.
The multilevel linear regression models presented in this article examine the relationships between media attention and twitter valence, and the cumulative account-holding intensity that EU agencies are subject to across forums.Fixed effects have been specified to account for variation across the six forums, The results from the regression analysis (Table 3) provides support for hypothesis one.Across all models that include media attention, there is a statistically significant relationship (p < 0.1) showing that agencies which receive more media attention are subject to higher overall account-holding intensity.When robust standard errors are applied, or influential observations are removed, this relationship is however not statistically significant for models 3 and 4 (Online Appendix pp.12-14), indicating that the relationship might be driven by a few high-influence observations.Three-level models based on the full dataset with all respondents (including a model with individual level control variables: age, gender, organisational position, and length of tenure), as well as models where individual respondents' answers have been aggregated to agency level, were also specified as additional robustness checks (Online Appendix pp.17-20).These models show similar results to the models presented in the paper.
No statistically significant relationship was found between more positive twitter comments and lower account-holding intensity.In other words, the regression analysis provides no support for hypothesis two.

A closer look at the relationship between media attention and account-holding intensity
Considering the log-transformation, it should be noted that the identified positive relationship between media attention and account-holding intensity is a relationship of 'diminishing returns'.More concretely, the results of model  Note: Regression models were built using the lme4 R-package (Bates et. al., 2015).The R-package sjPlot was used to create the regression table output (Lüdecke, 2021).Standard errors are reported in parentheses.
4 predicts that a 100 per cent increase in media attention leads to an 0.10 point (2 per cent of the account-holding intensity scale) increase in cumulative account-holding intensity.Figure 1 illustrates the relationship between media attention and account-holding intensity identified in model 4, with two partial residual plots.The first plot showcases what this relationship looks like (i.e., linear) when the media attention variable is transformed and centred, while the second plot illustrates the 'actual' relationship (i.e., logarithmic) between account-holding intensity and media attention (not centred or transformed).Furthermore, it should be remembered that the models test the extent to which media attention is positively related to the cumulative account-holding intensity that EU agencies experience from six account-holders.However, these account-holders might be affected by media attention to different extents.Figure 2 explores this through an interaction plot illustrating the relationship between media attention and account-holding intensity by forum when control variables are included in the model (see the results of the interaction models in the online appendix, p. 21).The interaction plot illustrates that, while there is a positive relationship between media attention and the account-holding of all forums, the positive relationship between media attention and cumulative account-holding intensity seems to be The positive relationship between media attention and account-holding intensity is particularly large as pertains to the European Ombudsman.What is more, additional analyses indicate that the identified relationship between media attention and cumulative account-holding intensity is to a large extent a product of the European Ombudsman's account-holding.More specifically, models excluding the European Ombudsman results in a considerably smaller, and no longer statistically significant, relationship between media attention and cumulative account-holding intensity.Furthermore, the relationship between media attention and the account-holding intensity of the European Ombudsman is considerably larger than the relationship between media attention and the account-holding intensity of the other forums, and the difference is statistically significant (online appendix pp.14-16).
As such, the proposed reputational causal mechanism behind the relationship between media attention and the account-holding intensity is particularly deserving of additional examination as pertains to the European Ombudsman.While an in-depth examination is largely outside the scope of this article, available data on European Ombudsman investigations allows at least for a quick exploration of the frequency of European Ombudsman account-holding activities, which could serve to at least further guide future research.While the European Ombudsman only opened investigations into a small proportion of complaints received against EU agencies 2015-2019 (18.3 per cent), there is a strong correlation (Spearman's rho = 0.74, p < 0.01) between the complaints received and the investigations opened against agencies.Furthermore, while there are also strong correlations between agency media attention (log) and both complaints received (Spearman's rho = 0.72, p < 0.01) and investigations opened (Spearman's rho = 0.71, p < 0.01), there is no correlation between agency media attention (log) and the proportion of complaints that lead to opened investigations (Pearson's r = −0.06,p = 0.449).However, in addition to its complaints-based accountholding, the European Ombudsman also has the possibility to on its own initiative open strategic inquiries, strategic investigations, and public consultations, and there is also a strong correlation between the frequency of this type of account-holding activities and agency media attention (Spearman's rho = 0.62, p < 0.01).

Discussion and conclusion
This study has tested two hypotheses about what drives and deters accountholding intensity, both derived from a reputational perspective on accountability.More concretely, public visibility and salience was hypothesised to function as a reputational driver of account-holding intensity (H1), and a strong reputation was hypothesised to deter account-holding intensity.The results of this study have provided support for H1, as EU agencies subject to more media attention are found to also be subject to more accountholding intensity.The results do not support H2, as EU agencies subject to more positive comments on Twitter are not found to be subject to less account-holding intensity.
A closer look at the relationship between media attention and accountholding intensity show that while media attention is positively related to the account-holding intensity of all six EU forums, there are differences in the extent to which media attention seems to drive the account-holding intensity of the different forums.The largest positive relationship between media attention and account-holding intensity pertains to the European Ombudsman (one of the EU watchdog institutions).This finding is interesting, as it indicates that the relationship between visibility/salience and accountholding intensity might extend beyond political principals (i.e., the Council of the European Union and the European Parliament), whose accountholding previous research has already been noted to be sensitive to salience (e.g., Busuioc, 2013;Egeberg & Trondal, 2011;Font & Duran, 2016).However, when interpreting these results, limitations of the data and the analysis should be recognised.
Firstly, it should be noted that the identified relationship between media attention and account-holding intensity might largely be driven by a few high-influence observations, considering that the relationship is statistically significant only in the bivariate regression model (model 1) when robust standard errors are applied or influential observations are removed (online appendix pp.12-14).
Secondly, the precise causal mechanism behind the identified relationship between media attention and account-holding intensity cannot be established with the method used in this study.To do this, additional (qualitative) research that traces and test the (reputational) causal mechanisms behind the relationship (for the different forums) is required.This limitation is particularly important to keep in mind as pertains to the relationship between media attention and the European Ombudsman's account-holding intensity, since the identified relationship between media attention and the overall cumulative account-holding intensity in the EU accountability landscape is found to be largely a product of this forum-specific relationship (online appendix pp.14-16).Based on extant literature on reputation and accountability (Busuioc & Lodge, 2016, 2017), it is theorised that reputational concerns underlie the identified relationship, as the European Ombudsman (just like the other account-holders) has an interest to invest more time and effort in diligently investigating highly visible issues and account-givers, having more to gain (loose) reputationally by being seen to (fail) doing its job well in highly visible cases, compared with (failing) to do its job well in low-visibility situations that fly under the radar of most audiences.However, an exploration into European Ombudsman investigations seems to suggest that other alternative explanations might also be at play.In particular, the number of complaints that the European Ombudsman receives about agency misconduct could drive both the frequency of the European Ombudsman's complaint-based investigations and the media attention that an agency is subject to.In addition, the analysis in this study does not allow for ruling out the possibility that account-holding by the European Ombudsman (i.e., opened and ongoing investigations) also drive media attention, and the two might mutually reinforce each other.Reputational considerations might be more likely to be an influential factor driving the positive relationship between media attention and the European Ombudsman's accountholding intensity as pertains to its frequency of own-initiative investigations as well as its account-holding diligence (pertaining to both complaintbased and own-initiative investigations).In any case, a forum such as the European Ombudsman, whichconsistent with a reputational perspective explicitly links organisational visibility to organisational relevance and impact in its multi-annual strategies (European Ombudsman, 2014;2019), is certainly worthy of receiving more attention from reputational researchers.
Thirdly, for the European Commission, the European Court of Auditors, and the agency management boards, the identified relationship between media attention and account-holding intensity is marginal.This could be a consequence of the small number of observations precluding the identification of smaller effects with statistical certainty.Considering that the European Commission, the European Court of Auditors, and agency management boards consistently score as high intensity account-holders across all agencies (including agencies with low visibility/salience), the lack of a substantial effect for these forums could also be the result of a so-called 'ceiling effect' (Cramer & Howitt, 2005, p. 21).In other words, these forums' formal account-holding responsibilities and obligations set a very high baseline for their account-holding intensities, and it could be that additional (potentially reputation-driven) account-holding beyond this high baseline cannot be captured with the survey instrument that this study relies on.Such a ceiling effect could also explain the logarithmic nature of the identified relationship between media attention and account-holding intensity.
Notwithstanding these limitations, this article has constituted a first test of theorised relationships between reputational considerations and accountholding behaviour (Bertelli & Busuioc, 2020;Busuioc & Lodge, 2016;2017) in the EU (Busuioc & Rimkute, 2019), which future research could build upon further.Additionally, the conceptualisation of account-holding intensity as both the frequency and the diligence of account-holding, which is presented and applied in this paper, provides a framework for taking into consideration not only the quantity but also the quality of account-holding practices and mechanisms.Future research focused on investigating account-holding behaviour could with benefit adopt (and further develop) this conceptual framework.Notes 1.Following invitations, one agency decided to provide a centralised agency response, and four respondents chose to refer us to other staff members.These responses were also included in the analysis.2. At the start of data collection (October 2019), the EU agencies network (https:// euagencies.eu/),included 34 decentralised agencies, 7 Joint Undertakings, 3 Common Security and Defense agencies, and the European Institute of Innovation and Technology.ELA, EPPO, and EuroHPC were not yet members of the EU agencies network.3.In accordance with Bovens (2007) three-phase definition of accountability, a question about the likelihood of sanctions and rewards (i.e.possible consequences) was also asked in the survey.However, including this question reduces the Cronbach alpha coefficient slightly (α = 0.799).As the alpha coefficient tends to increase as more items are added, this is an indication that this question does not reflect the same underlying construct as the other questions (about information and discussion) (Taber, 2018).This speaks to broader debates in accountability literature about whether (possible) consequences is a constitutive element of accountability (see Bovens, 2007, p. 451).While not to say that (possible) consequences are irrelevant to accountability, these findings suggest that 'answerability' (through information and discussion) constitutes a different aspect of accountability than 'enforcement' (Schedler, 1999).
4. The UK was still an EU member state during this period (2014-2019).5. Ideally, the twitter valence measure would span a longer time period for the same reasons outlined with regards to the media attention variable.This would also be consistent with how Carpenter (2001) emphasise that reputation is built incrementally over time.However, twitter valence data before 2018 is unfortunately not currently available for EU agencies.6. (Σ positive comments-Σ negative comments) / (Σ positive comments, neutral comments, negative comments).

Figure 1 .
Figure1.Partial Residual Plots -The effect of media attention on account-holding intensity.Note: The partial residual plots were created using the R-package jtools(Long, 2020).The plot points are jittered in order to avoid overlap.

Figure 2 .
Figure 2. Interaction Plots -The effect of media attention by forum.Note: The interaction plot was created using the R-package interactions (Long, 2019).

Table 1 .
Descriptive statistics (raw).Random effects (random intercept, fixed slopes) have been specified to account for variation between agencies.Table3displays the results of the regression analysis.Models 1-2 include only the variables that test the corresponding hypotheses (model 1media attention; model 2twitter valence).Model 3 includes both independent variables.Model 4 includes all control variables.P-values <0.1 have been highlighted.