Subsidiarity does not win cases: A mixed methods study of the relationship between margin of appreciation language and deference at the European Court of Human Rights

Abstract In August 2021, Protocol 15 inserted the doctrine of the margin of appreciation into the preamble of the European Convention of Human Rights, presumably cementing what President Spano has referred to as the ‘Age of Subsidiarity’, in which the European Court of Human Rights applies the margin of appreciation more often and increases deference to state parties. This insertion was done on the behest of the High Contracting Parties as part of the Interlaken reform process, and there is already a strong narrative in certain member states and parts of the scholarly literature that this focus has prompted the Court to increase the usage of the margin of appreciation and therefore the deference to states, judging more frequently in their favour. This article hypothesizes, however, that the increased usage of the margin of appreciation language which has been taken as proof for this narrative, might not, in fact, indicate higher levels of deference. Rather, the language of the margin of appreciation could be the result of usage by other actors or a marker of complexity for so-called ‘hard cases’. To investigate this relationship, the article applies a mixed legal-doctrinal and quantitative methodology to analyse who in the case law invokes the doctrine, what their purpose is for doing so, and what adjudicative consequences follow. It finds that usage of the margin-language topped well before the Interlaken process began, that governments are not the most frequent invokers and that, statistically speaking, states are no more likely to win margin-cases than other cases.


Introduction
The margin of appreciation has appeared in case law interpreting the European Convention on Human Rights (ECHR) ever since it first came into force. The first usage was by the European Commission of Human Rights 1 and the doctrine was since developed further by the European an increased usage of the margin of appreciation in the Court's interpretation would lead to more elbowroom for states. Prima facie, this claim seems logical, but it has not actually been confirmed empirically, and doctrinal literature has already produced many examples where it has not been the case. 9 The second claim, that increased subsidiarity would lessen the Court's workload, is much less straightforward and would, in any case, be working indirectly. As a result, it has already been challenged in the literature. 10 This article will therefore limit itself to empirically testing the first claim, namely whether an increased usage of the margin of appreciation would result in more 'elbow room' for or deference to states.
The question is an important one because even though the connection has been questioned in some of the doctrinal literature, it has been relied heavily on by actors at the Court in general, in particular in connection with the Interlaken reform process. For example, in 2009 the English Lord of Appeal, Lord Hoffmann, voiced a scathing critique of the ECtHR arguing that the Court should increase its usage of the margin of appreciation, 11 or risk losing its legitimacy. 12 When, in 2014, ECtHR Judge Spanonow President Spanodecided to respond to this legitimacy challenge, he surprisingly did not argue that the critique was misplaced because of the universality of human rights or legitimate foundations of the Court, 13 but rather because 'the Court is [already] in the process of developing a more robust and coherent concept of subsidiarity as well as attempting to reformulate the conditions for allocating deference to the Contracting States'. 14 In 2018, now citing a quantitative study by Mikael Rask Madsen showing that the language of the margin of appreciation had been rising since 2010, he reiterated this claim. 15 This study and the claim that the margin was being used ever more, was central in the political debates surrounding the drafting of the Copenhagen Declaration. Specifically, preprints of Spano's 2018 article were circulated among the team responsible for the preparatory works for the drafting of the Copenhagen Declaration. 16 This particular (qualitatively contested) understanding of the margin was thus part of the foundation of knowledge for the politicians and civil servants drafting the Draft Copenhagen Declaration. 17 European Consensus', (2015) 22 iCourts Working Paper Series 1, at 12; Lord Lester of Herne Hill, 'Universality versus subsidiarity: a reply', (1998) 3 European Human Rights Law Review 73, at 75; P. Mahoney, 'Universality versus subsidiarity in the Strasbourg case law on free speech: Explaining some recent judgments', (1997) 2(4) European Human Rights Law Review, at 368. 9 See, for example, Kratochvil, supra note 2, 336-9, at 42; J. Gerards, 'Margin of Appreciation and Incrementalism in the Case Law of the European Court of Human Rights', (2018) 18 Human Rights Law Review 495, at 498-506. 10 See the debate between G. Van der Schyff, 'Interpreting the protection guaranteed by two-stage rights in the European Convention on Human Rights: The case for wide interpretation', in E. Brems and J. Gerards, Shaping Rights in the ECHR: The Role of the European Court of Human Rights in Determining the Scope of Human Rights (2014), at 65-83 and J. Gerards, 'The scope of ECHR rights and institutional concerns: The relationship between proliferation of rights and the case load of the ECtHR', ibid., at 97.  16 Documentation made available to the author by the Danish Ministry of Justice following a freedom of information request filed on 9 September 2020. Available upon request. The claim was also reiterated repeatedly at the High Level Conference in Kokkedal which served as a venue for knowledge transfer between state agents, Court representatives, academics, and nongovernmental organizations. See Conference report High-Level Expert Conference 2019 and Beyond: Taking Stock and Moving Forward from the Interlaken Process Chairmanship of Denmark: Council of Europe (22-24 November 2017); List of Participants for High-Level Expert Conference, Chairmanship of Denmark: Council of Europe (22-24 November 2017) 1. 17 And likely for the other Declarations as well, but I have only been able to trace the argument in the Copenhagen Declaration because the process was so relatively transparent.
When literature attempting to test the connection empirically is scarce, it is because studying it is an exceedingly difficult endeavour. Madsen's oft-cited study was innovative, especially at the time, but the abducted corpus-linguistic methodology it applied; counting the occurrence of the expression 'margin of appreciation' per million words, 18 tells us nothing about whether these mentions are accompanied by increased deference to states. One methodological problem with discovering the link between the margin and deference is that there is little agreement in the academic literature about both sides of that equation, namely what exactly the margin of appreciation does and how it can be recognized, and what exactly 'elbow room' or deference to High Contracting parties might look like. The Court itself has also not provided definitions of either. As a result, the language of deference may take many different forms. There also appear to be instances in which the Court uses the margin language but without applying deference, in the doctrinal literature referred to as 'empty' uses of the margin. 19 Similarly, because adjudication is a task which includes a multitude of factors which are not always explicated, there may also be instances where deference is applied in the initial determination of the scope of the right, 20 or where it is applied implicitly without including a mention of deference. This article will attempt to create a bridge between the empirical/quantitative studies of the margin and the doctrinal studies. This will be done by conducting a transparent mixed-methods study of the margin of appreciation, which applies both high-abstraction level statistics and lower abstraction level doctrinal analysis as a sample study to address some of the concerns raised when attempting to study the margin statistically. This article aims to identify some sources of false negatives, that is, cases where deference is applied but which are unlikely to be picked up by statistical methods, but its main analytical focus is on false positives, i.e., cases where the margin of appreciation language appears but where it is not followed by increased deference to the High Contracting parties.
The article will proceed as follows: Section 2 will situate the study in the existing literature and explicate how theories on the margin of appreciation have contributed to forming the research questions. This will be followed by Section 3 on the methodology applied and the proxies used. Hereafter, Section 4 will follow, consisting of a high-abstraction level mapping of the relative prevalence of margin of appreciation language in caselaw from 1960 until 2021, and an overview of a high abstraction-level deference determination in the form of case results in admissibility or violation/non-violation judgments. Section 5 presents a lower abstraction level study opening the black box of the individual cases reviewing which actors are using the margin of appreciation language and howresulting in an estimate of the amount of 'empty' margin of appreciation mentions we might expect to see. Together these two analyses give a better insight into what a linguistic mention of the margin means statistically, laying the ground for Section 6, which tests whether the use of the language of the margin of appreciation is a predictor of the amount of deference afforded to states.

Theories and research questions informed by the state of the art
: : : State authorities are in principle in a better position than the international judge to give an opinion : : : [this] leaves to the Contracting States a margin of appreciation : : : The domestic margin of appreciation : : : goes hand in hand with a European supervision : : : It is in no way the Court's task to take the place of the competent national courts but rather to review : : : the decisions they delivered. 21 18 The above quote is from the Handyside case from 1976, which is widely recognized in the literature as a key case in which the margin of appreciation gained a more explicit articulation building on previous expressions of the margin of appreciation in Article 15 cases on declarations of emergency and previous discussions on subsidiarity in for example the Belgian Linguistics case from 1968. 22 This is, however, more or less where the general agreement ends. Normative discussions on the topic have been particularly rich, encompassing whether the margin is 'the "natural product" of the distribution of powers between the Convention instruments and the national authorities', 23 or rather a 'principled recognition of moral relativism [that] is at odds with the concept of the universality of human rights'. 24 As have definitional discussions on whether the margin has a clear albeit complex doctrinal usage, 25 or should rather be considered a strategic instrument that the Court can put to use as a 'convenient way to dispose of cases', 26 or to please difficult states. 27 These types of normative discussions are to be expected for a doctrine that delimits the boundaries of international control with national legislation and administration, but literature on the margin is also not in agreement about foundational descriptive facts. There are, thus, examples of literature that operates with a margin which is applied as part of the proportionality assessment, whether before, 28 incorporated, 29 or as a final weighing of all things considered. 30 While another body of literature works with a margin determined foremost by the right it is applied to, 31 the type of applicant, 32 or whether cases include economic and social policies. 33 Yet another body investigates the relationship between the margin and the level of interference with the essence of the right, 34 with European consensus, 35 or with procedural protections 22 P. Mahoney, 'Judicial activism and judicial self-restraint in the European court of human rights: Two sides of the same coin', (1990)    Surveillance: Ethical issues, legal limitations, and efficiency: Synthesis report from WP4, merging the ethics and law analysis and discussing their outcomes, FP7 -SEC) 2011)284725 (2015) 9-10, at 9-10. 31 Lavrysen, supra note 29, at 309, 316; Macdonald, supra note 2, at 121-2; Yourow, supra note 1, at 113-17. 32 Dembour makes the case that the Court's many standard of review instruments, not limited to the margin but including also the differentiation between positive and negative obligations are often applied differently in cases concerning nonnationals than in cases concerning citizens, see M.-B. Dembour, When humans become migrants: study of the European Court of Human Rights with an Inter-American counterpoint (2015), at 110-15. at the national level, 36 or the quality of the democratic debate nationally. 37 Together the literature presents a wealth of different influences on the width and application of the margin, going some way to explaining both why the doctrine is difficult to quantify, having often been described as unclear 38 or elusive, 39 and why works presenting the doctrine as not one but several concepts have been so illuminating. 40 As a whole, the literature operates with some connection between deference and the margin of appreciation, but also demonstrates a high level of complexity affecting this connection. This is contrasted by the clear political insistence on a 'more robust' application of the doctrine in the political Declarations in the Interlaken Process. The political Declarations issued by the High Contracting Parties were not without political and soft-power costs, and their negotiations took months. 41 The reason for this insistence was, without any doubt, that central political actors expected an increased usage of the margin to result in more deference to statesor put more simply, in more state wins. 42 This expectation was likely reinforced by the president of the Court's notion of the Court having entered an 'Age of Subsidiarity'. 43 There is, thus, a gap between the qualitative studies of the margin of appreciation and the political assumptions about the margin which are partially supported by quantitative studies. It is this gap that this article will venture to fill.
On the basis of the discussions referenced above, this article will investigate the following research questions: R1: Which actors are the most frequent users of which kind of margin of appreciation language? The question is an important one mainly because it determines limitations on to what extent margin-language is likely to be an indicator that the Court considers the questions of deference and subsidiarity relevant in the present case. Since if the respondent states or even intervening third-parties are the most frequent users, this might be seen solely as an adjudicative strategy, whereas if the Court is the most frequent applier, it would be more reasonable to assume that the margin-language would be connected to the application of the doctrine in legal reasoning. This question also includes whether margins are wide, narrow, or used without an indicator of the width. R2: How frequent are 'empty' mentions of the margin? This question is connected with the previous one. The presumed connection between margin language and actual application of the margin-doctrine (which is one assumption on which qualitative and quantitative studies currently clash), can be established by reviewing what kinds of arguments the margin is connected with, how often a narrow margin is advocated as opposed to a wide one, and whether the margin is applied as part of the scope, the proportionality assessment, or the procedural review. Together these elements will result in an assessment of the relative prevalence of empty margin uses which one would not expect to impact the whether a judgment results in a violation or non-violation result. R3: Are 'margin cases' more likely to result in state wins than non-margin cases? This is the main research question, the investigation of which is tempered and informed by the answers to R1 and R2. In this case, 'state win' is used as a proxy for 'state deference'. While the two are most certainly not the samethe Court may well find a violation in a case even in an area where there is usually a high level of deference if the inference with the right is intense. There are also many cases that states can win without invoking a deference-argument (though these will often be declared clearly inadmissible)but in the aggregate, for the group of cases where a delicate balancing is undertaken, if the margin is connected with increased deference, it should also result in higher numbers of state wins. On the basis of the 'delicate balancing' element in this argument, this article furthers a hypothesis, namely that both margin-mentions and state wins are likely to be more common in 'hard cases' than in other cases. Hard cases here are defined as those that are neither obviously inadmissible nor obvious violations in line with the existing case law. Proxies for these hard cases may be the higher importance-level, key-and level one and, to a lesser extent level two, the presence of a separate opinion, or treatment by the Grand Chamber. These variables are in turn somewhat colinear. Almost all Grand Chamber cases are key-or level one, while on the flipside only about a quarter of Key-and Level 1 cases have been treated in the Grand Chamber.

Methodology
The analysis in this article relies on a mixed qualitative-quantitative approach engaging with several different types of data. It contains several different levels of abstraction. At the highest abstraction level there are statistical analyses of an almost complete body of caselaw, while at the middle-lower level I have analysed a sample of 202 key-and level one cases containing at least one mention of the margin of appreciation. The reason for conducting the analysis in this way is to avoid conscious or unconscious cherry-picking of a particular type of case, whether those with a particular usage of the margin or those against a particular country or similar. The point is also to get a complete overview of the caselaw since the very different experiences of the degree of deference afforded to states by different political actors, different scholars, applicants, and national authorities may well be due to the different populations of cases that reach these groups. This will be explained in detail in Section 4.
The highest abstraction level statistical analysis will take place using the yearly reports of caselaw statistics published by the Court, 44 followed by case law searches conducted in the HUDOC database, 45 and analyses on an existing dataset made public by Voeten and Stiansen. 46 The benefit of this triad is that each set brings something different. The Court's statistical reports include numbers on the inadmissibility cases and administratively rejected cases which are not published in HUDOC, but no detail on the individual cases, while HUDOC has every judgment and decision made in a Chamber or Committee. The Voeten-Stiansen Dataset is a scraping of metadata on almost all (21,954) judgments (but no decisions) published in 44 Available at www.echr.coe.int/Pages/home.aspx?p=reports&c=. Compiled for ease of replication in the supporting material for this article entitled 'ECtHR own statistics'. 45 The European Court of Human Rights, HUDOC (1960->), available at hudoc.echr.coe.int/. Compiled for ease of replication in the supporting material for this article entitled 'HUDOC searches dataset'. English between 1960 and 2019, and it has the benefit of being malleable to a greater extent, allowing for more targeted statistical analysis. In both the study of the HUDOC database and the Voeten-Stiansen Dataset, the study looks for simple margin-mentions; 'margin of appreciation' or 'margin of discretion' rather than 'wide' or 'narrow' margins, which will instead be studied at the lower abstraction level to inform the high abstraction level study.
For the lower abstraction level, I have created a sample of all 202 key-and level one 'margin of appreciation cases' 47 issued between 23 February 2012 48 and 1 January 2020. These cases are coded for when the margin of appreciation language is used and whether it is 'wide margin', 'margin' or 'narrow margin', what it is used for, and whether submissions came from the applicants, governments, third parties, the Court's own assessment, or separate opinions. 49 When coding what the margin is used for, I have attempted to keep this as simple as possible to avoid situations where the determination is unclear. The categories are (i) empty mentions, 50 (ii) anti-margin mentions, (iii) the margin as part of the scope, (iv) the margin as something that should be granted to certain domestic institutions, (v) the margin as part of the proportionality assessment, (vi) the margin as a 'better placed' argument, and (vii) 'other'. For each of these codes, except the last, there are subcategories. 51 The coding is done so that each margin-argument has only one code. Still, a single case may contain several mentions of the margin by several different actors, each of which can be coded differently. The 202 cases act as a sample, though not a random one, to indicate how often margin-mentions are, in fact, narrow margins, and how often a case might appear as a 'margincase' in the statistical analysis while it is actually only non-court actors that are engaging with the margin or where the main argument is that the margin is not relevant in the present case. It is important to keep in mind, that the 202 level one and key-cases are not representative of the total population of applications to the Court or even the total population of judgments. Cases at the different importance levelswhich are curated by the Bureau of the Court in collaboration with the Jurisconsult 52are likely to be different because they treat different types of cases. In this article, key cases and case reports will be treated as one category. The Court used the terminology 'Case Reports' between 1998 and 2015 when the most important cases were published in the ECHR Series of Judgments and Decisions while from 2007 these cases started being named 'Key cases' in the HUDOC database, and since 2015 'Key cases' has been used to label both types in the database. 53 The Bureau classifies level one cases as 'high-importance judgments, decisions and advisory opinions not selected for the Case Reports but which nevertheless make a significant contribution to the development, clarification or modification of the Court's case-law'. It was also the highest level before the introduction of Case Reports. Levels two and three, on the other hand, are considered of lesser importance, but where level three cases only apply well-established caselaw, level two cases can go beyond this and may therefore still be relevant precedents for specific cases. There are big differences in the prevalence of each type of case. HUDOC lists, as of December 2021 in English, 1,364 key cases, 1,560 level one cases, 5,796 level two cases and a total of 44,677 level three cases. The reason for using high importance-level cases for 47 These are cases where the margin of appreciation is mentioned at least once. Searching the HUDOC database with a textual search for 'margin of appreciation' yielded 298 search-word cases, while using the database's keyword search yielded 165 cases. After removing any doubles from the list, the final list ended up with just 202 cases. 48 The 'Brighton Moment' where the draft declaration was leaked. 49 See Annex for full explanation. 50 The empty mentions have recieved particular attention and are well defined in Kratochvil, supra note 2, at 336-42 and Gerards supra note 9, at 503-4. The definition used for this articlesee Annexis based on these, but does not assume as a matter of course that a 'certain margin' is an empty mention, but rather digs slightly deeper to investigate what this certain margin is being used to argue. 51 Please see Annex for full codebook. 52 The Bureau is composed of the Court President and Vice-President assisted by the Registry and the Jurisconsult. 2019 Rules of Court, Plenary Court 9A, HUDOC user manual, European Court of Human Rights (2017), at 8. 53 European Court of Human Rights: Selection of Cases, available at www.echr.coe.int/Pages/home.aspx?p=caselaw/ reports&c=. the sample is that these are more likely to contain a wealth of different or innovative uses of the margin, whereas lower-level cases are more likely to contain margin-mentions that refer to older higher-level cases. For the purpose of understanding the many different types of uses that might be found in the population of cases as a whole, this sample is therefore superior to a random sample which would be likely to consist of at least 80 per cent of level three cases. The sample will therefore form the foundation first and foremost for the answering of R1 and R2.
Attempting to count how often different actors use the margin comes with several important caveats, and the results carry with them some uncertainty. The caveats include that the arguments published in the judgments and decisions which form the corpus of this study, are not the original arguments and allegations submitted by the parties, but the Court's summary of them. Therefore, it is not inconceivable that governments or applicants may in some cases, have mentioned the margin of appreciation, but this has not been reflected in the final written version of the proceedings. Presumably, however, this should exclude only less likely cases, that is cases where the margin did not play a large role; otherwise, one would expect either the Court to mention the doctrine itself or the parties to protest its absence in deliberations. Another important caveat for testing whether the usage of the margin of appreciation language results in more state deference in the form of state wins is that written judgments are not optimal data for quantitative analysis. Judicial reasoning relies on a range of factors related to the case at hand including the right in question, the degree of interference, the legitimate aim, vulnerability of the applicant, and many more, of which the margin of appreciation only encapsulates some.
Furthermore, judgments do not always result clearly in one result or another, oftentimes some elements of the case are inadmissible, whilst others are violations and non-violations. It is not uncommon for a case to result in both violation-and non-violation decisions regarding the same article for different parts of the course of events. 54 In some social science studies of the ECtHR, this problem has been solved by creating a unit of analysis of 'at least one violation-decision', 55 and while this can work for some studies, it does not give a full picture of the relative tendency of deference which this study is attempting to capture. It also misses the fact that often from Chamber to Grand Chamber the number and types of articles that are found to be in violation or non-violation may change in ways that absolutely affect the perception of how stringent the Court is. A recent example of this is Savran v. Denmark where an Article 3 violation in the Chamber was translated into an Article 8 violation in the Grand Chamber, a change interpreted as a win by the Danish authorities. 56 In this article, the problem of several decisions in a single judgment will instead be addressed to some extent by making the individual decision the unit of analysis for some of the calculations, rather than the case. 57 The most important way this article attempts to address these shortcomings, however, is by not abandoning the qualitative, doctrinal method altogether. It relies on a strict differentiation between linguistic use, which can be measured quantitatively, and adjudicative importance, which demands a legal analysis, and utilizes the methodologies together to reach robust conclusions.
4. Margin and deference over time: The highest abstraction level analysis As indicated in the introduction, the claim that the Court is applying the margin of appreciation more often in its case law has been central in much of the debate surrounding the reform. Many scholars relying on this premise cite the same source, a quantitative study by Mikael Rask Madsen from 2018, and certainly, Madsen's data suggest a clear narrative (see Figure 1 This was also the solution adopted by Madsen, supra note 15, at 218. margin of appreciation cases appears to be rising steadily while the output of the Court is falling. With the use of a textual search 58 for the 'margin of appreciation' in the HUDOC database, the data is generally replicable. 59 Madsen's study, for reasons of scope and comparability, covered a narrow timeframe of just six years. Taking instead a temporally broader view, and with the benefit of another five years passing, the trend appears much less clear. Furthermore, whereas Madsen included only judgments in his study, his work also showed that the margin is mentioned most often with Article 35 on admissibility. Therefore, it seems more appropriate to study the phenomenon in both judgments and decisions as is donebelow in Figure 2.
Because of the dramatic increase in the number of cases processed by the Court, the absolute number of cases using the wording 'margin of appreciation' is also rising. 60 As highlighted above, however, the trend is nowhere near as clear in relative terms as much existing scholarship would have us believe, and the most frequent usage of the margin took place in the 1980s and 1990s rather than after the initiation of the Interlaken reform. When dividing the numbers between judgments and decisions, the picture is even more murky.
There is an indicator that the percentage of margin-mentions in judgments (but not in decisions) has been rising slightly since 2010 (Figure 2.1). The percentage of margin-cases was, however, higher ten years earlier (Figure 2), and in any year the cases mentioning the margin still account for less than 15 per cent of all cases (judgments and decisions) from the ECtHR, which is not a lot for a doctrine that could in principle be invoked in the treatment of almost any article. 61 The left-hand axis denotes the number of judgments per year while the right-hand indicates the percentage of judgments containing a mention of the margin of appreciation.

58
On differentiating between textual (search-word) searches and keyword searches: the HUDOC database has coded some cases with the keyword 'margin of appreciation' but this picks up only roughly 1/3 of the cases that a textual search for the phrasing 'margin of appreciation' picks up. 59 Madsen's data uses corpus-linguistic methodology of counting 'appearances per million words' as the unit of analysis whereas my replication uses the individual case as the unit of analysis. While I would argue that it makes more sense to use the individual case as a unit of analysis since cases vary tremendously in length without there being a clear indicater of what this means, the colinearity of the two suggest that both methods work in the same wayin other words the relative lenght of the individual judgment does not appear to have a great impact on the data. Another often neglected detail, which is relevant to include when considering these numbers, are the cases that were never made public in HUDOC and which are therefore absent from both this study and most if not all other qualitative and quantitative studies. These include the cases that are rejected before they reach a judicial formation, as well as the cases declared inadmissible by the single judge formations. The single judge formations were established by Protocol 14 in 2010 and allow for a single judge to declare applications inadmissible. 63 Most of these are rejected due to the non-exhaustion of domestic remedies or non-adherence to the six-month deadline (from August 2021: four-month deadline), but as the formation works under Article 35, a case could also be declared inadmissible for being manifestly ill-founded, treating the Court as a fourth instance, or because the applicant is deemed not to have suffered a significant disadvantageall The left-hand scale shows the percentage of margin cases compared to the yearly total of cases, while the right-hand scale indicates the number of judgments and decisions published each year. This figure includes all cases with a HUDOC entry in English.  When illustrating the judgments and the decisions apart in this way it becomes obvious that while the relative use of the margin of appreciation in judgments has risen since the initiation of the new Court in 1998, the usage in decisions has fallen. determinations which might well include considering the margin of appreciation. 64 As was established by Madsen in his study of cases that were made public, decisions under Article 35 did indeed often include a margin of appreciation consideration. 65 Another category of cases which are not made public is those rejected under Rule 47, which came into force with the 2014 changes to the Rules of the Court and the introduction of an online application form. This change enabled the Registry to reject cases at an earlier stage if any information or documents were missing. 66 Together these types of cases might be considered a kind of adjudicative 'dark matter' since it remains invisible to most scholars and political actors with opinions on the Court. The cases do exist, however, and the Court and its registry keep track of their number. Since 2006 the Court has published yearly statistical information about the number of cases dismissed administratively in addition to statistics on judgments and decisions which have been around for longer (see Figure 3). Those numbers show unequivocally that a judgment on the merits is a rare outcome for an application, but there does not appear to be a clear trend that this has changed over time.
For most legal-doctrinal scholars, the non-published cases are of very little interest. They are unlikely to contain any innovative interpretations, and the Court itself does not refer to these cases in its reasoning in other cases. For practitioners and potential applicants, however, it is crucial to keep these cases in mind, as more than 90 per cent of all applications will be rejected, struck off the list or declared inadmissible for the same reasons as accorded in these cases. 67 For those interested in preserving the Court's legitimacy, the fact that the vast majority of individuals who turn to the Court for help experience having their cases rejected with little or no explanation might also be a cause for concern. 68 For this article, the unpublished inadmissible cases are relevant when addressing claims in the literature concerning whether the Court is increasingly activist or restrained or, in other words, whether it has entered what president Spano has referred to as an 'Age of Subsidiarity'. On the one side of this debate there are claims that the Court is expansive  in its interpretations and is increasingly finding violations, whether accompanied by a critique of this or not. 69 On the other side, there are the claims that subsidiarity, judicial restraint, and state deference are on the rise. 70 Logically, it should follow that both sides cannot be right, even though there are large bodies of literature supporting both. While some of the discrepancy, especially in politically charged literature, might be explained by bias or cherry-picking on the part of the author, we can also explain the discrepancy by taking the adjudicative dark matter into account. Considering that subsidiarity concerns can play a role both in the determination of admissibility and in the substantive interpretation, an increase in the number of inadmissible cases may be as much an indicator of increased deference to states as a decrease in the number of non-violation judgments. The evidence for this argument can be seen in the two figures below. The first (Figure 4) charts the percentage of state wins over time. This figure was made by dividing the number of non-violation and inadmissibility decisions by the total number of decisions in 21,954 judgements, using the conclusions meta-data scraped from the HUDOC database; 72 it, therefore, does not include the inadmissible cases that were not made public. The second figure (Figure 4.1) shows instead the number of judgments in relation to inadmissible cases as reported in the Court's yearly statistical reports. While we should not overstate the importance of these data; both the specifics in interpretation, as well as the importance and general interest of a case, mean that a single violation-judgment could represent more adjudicative innovation and receive infinitely more public scrutiny than the next ten cases regardless of their outcome, it is still worth noting that where Figure 4 clearly supports the claim that the Court has become more likely over time to find violations against states, and by proxy that it is less deferential with a slight adjustment after 2012, Figure 4.1 suggests the opposite, namely that more cases are rejected before even reaching the Court, indicating greater state deference.

Empty mentions and the use by different actors
Section 4 aimed to nuance and explain discrepancies in existing claims that the Court is either more activist or increasingly deferential to state parties and whether it is using the margin of appreciation language more or less often over time. It also showed the limits of high abstraction level descriptive statistics on the Court's case law, which cannot tell us whether usage of the margin of appreciation language is associated with increased deference, nor how it impacts decision-making overall. The first step in getting closer to these answers is to move down an abstraction level and determine which actors at the Court invoke the margin of appreciation language at which width (R1) and for what purpose (R2).
On R1, my coding of the 202 key-and level one cases showed that governments did mention the margin more often than applicants, but they were not by far the most active users of the doctrine, which were instead the Court itself, whether as majority or through judges' separate opinions. This first result suggests that there will not be many cases in which the government has attempted to invoke the margin of appreciation, without that invocation being taken up by the Court (see Figure 5). Furthermore, the coding showed that about two-thirds of margin-mentions did not include an indicator of the width, while one-third concerned a wide-or wider margin. The number of narrow-or narrower margin mentions was minuscule, about three for every hundred mentions (see Figure 5.1).
This second result can be interpreted in different ways. While some authors have in some cases, lamented the tendency of the Court not to indicate the width of the margin when it invokes it, 73 other contributions, including from judges and former judges at the Court, simply treat a 'certain' margin as a mid-level margin. 74 Below is a bit more detail about whether some of these are empty mentions. In any case, however, the result seems to suggest that a quantitative analysis based on the margin of appreciation language should not contain many false positives where the marginlanguage is applied to argue in favour of a narrow margin.
On the other hand, however, the analysis also showed that the number of margin-mentions was widely different between cases with one case sporting a full 50 uses of the margin 75 while in 55 other cases the margin was mentioned only once, and in 22 cases (11 per cent) the Court itself did not engage with the margin of appreciation or with any argument of subsidiarity (see Figure 5.2). In 13 of these cases the margin was instead used in the separate opinions of the dissenting and  The figure illustrates the total number of 'margin of appreciation'-mentions in the sample of 202 key-and level one cases divided by whether it was accompanied by an indicator of width and which actor has used the margin language.

73
Kratochvil, supra note 2, at 341; Gerards, supra note 9, at 503. 74 Spielmann, supra note 33, at 389, 98, 402; Spano, supra note 15, at 481-2; Greer, supra note 2, at 29-30. concurring judges, in some cases decrying that the majority has not engaged with the margin. 76 While in another eight cases the margin was used only by the government or its national courts; in two of these also by the applicant and in one only by the applicant stating that the margin was not relevant. 77 In the final case where the Court does not engage with the margin, an intervening third state was responsible. 78 These cases are the most obvious 'empty' instances of margin-cases where the margin-language is not followed up by an engagement with the doctrine in the determination of the majority opinion. It is not automatic, however, that cases where the Court does not use the margin are empty in this way. In an additional seven cases the Court did not use the margin of appreciation language but referred to a similar subsidiarity-based argument in another way, for example, by stating that it would not substitute its view for that of the national courts. 79 In addition to these 11 per cent of cases, there may be cases where the Court mentions the margin, but where it still has no influence on adjudication (see Figure 6). The literature on empty mentions includes as contestants for this type of cases: standardized quotes that the Court includes on its general principles of interpretation without necessarily engaging with the margin in its adjudication, 80 and where the margin is mentioned only at the very end of the Court's considerations as a concluding note. 81 In the following, the first type of empty mention is 1.c. when the Court applies a standardized formulation to explain its interpretive tools including the margin. The second kind, here labelled 1.a., is where the margin is used as a concluding remark. The 1.b. in this coding refers to situations in which the Court mentions the margin as a principle for balancing between the individual and society but does not engage with it further.
The coding shows that the Court uses the margin more often than any other actor, mentioning the margin 413 times in 172 out of 202 cases. Roughly a quarter of these are standardized reiterations of how the margin works which may or may not impact the case. Another quarter connects the margin with the scope of the right or situation rather than with a proportionality argument while the remainder engages with proportionality, 'better placed' arguments, and procedural questions. In about half of the cases in the sample, the Court has divided its assessment into a section on general principles and a section applying those principles to the present case. Margin mentions in the sections on general principles account for about Illustration of by whom the margin was mentioned in 202 cases. In 29 cases, the Court doesn't mention the margin directly, but in seven of these (including the four where neither court, government, applicant nor a judge in a separate opinion mentioned the margin) the Court referred to the margin using other types of language. Kratochvil, supra note 2, at 336-9. 81 Ibid., at 342; Gerards, supra note 9, at 501.  one-quarter of all margin mentions and a lot of them are simple explanations for how the margin is usually applied in connection with a particular right, 82 often directly citing definitions used in earlier cases. 83 This does not mean, however, that these are necessarily empty mentions since in three out of four cases they are followed up with a proportionality 84 or procedural 85 type arguments in the following section on the application in the present case. The same applies in cases where the empty margin-mention is present in the concluding remarks. Here as well it is in most cases preceded by other types of margin mentions. In most cases the Court's use of the margin is therefore related to balancing arguments, though there are a few examples of truly empty mentions of the margin without follow-up in legal argumentation in the sample. One of these is M.N and others v. San Marino, a case on the privacy of banking information, where the Court in its treatment of Article 8, included a standardized margin-explainer in paragraph 73 under 'general principles' but relied instead on an either-or analysis of exhaustion of domestic remedies as the main theme in its treatment of the present case. 86 Another is Piechowicz v. Poland, a case concerning the right to family life for a detained person. Here the Court too included a standardized explainer of the margin of appreciation in the application of Article 8, 87 but the following balancing was conducted by considering the degree of interference and potential other solutions the state could have applied rather than a complete ban on family visits, and not whether the state was well placed politically or pragmatically to make that distinction. 88 With regards to the antimargin mentions there are no cases in the sample where the Court uses one of these without also having engaged with the margin in relation to scope, procedure or proportionality.
Altogether, there are 14 cases (7 per cent) in the sample of 202 cases where the Court has engaged with an empty margin mention without also having applied another type of margin mention, whether proportionality-based, procedural, or related to scope. These types of empty mentions with no followup are therefore relatively rare but put together with the 22 cases where the margin is mentioned, but not by the Court, this still leaves us with 17 per cent of margin of appreciation cases in this sample which can be characterized as 'empty' mentions of the margin without impact on adjudicationstatistically speaking, false positives. This rate of false positives is not so high that statistical analysis is impossible, but it does place limitations on the conclusions we can draw from statistical analysesboth in this article and in analyses that have been done on this topic before. 89 Knowing the frequency of empty mentions informs the interpretation of the quantitative datasomething which has been missing from quantitative analyses of the margin of appreciation until now.

Adjudicative consequences of invoking the margin of appreciation
Turning to the question of whether the margin of appreciation mentions generally are connected with state deference through the proxy of state wins, let us start at the highest abstraction level with descriptive statistics on the 21,954 judgments between 1960 and 2019, the relative prevalence of the different outcomes in cases that mention the margin. At first glance, margin-cases appear quite different from cases with no mention (see Figure 7).
Among the cases that do not mention the margin, 80 per cent of decisions result in a violation judgment, in the margin of appreciation cases this is just 56 per cent. However, even though these results Ibid., paras. 214-16. 89 This study unfortunately cannot reveal whether the portion of empty mentions has increased or decreased over time, that question will be for future studies.
prima facie appear significantly different, we should not draw any conclusions yet. When conducting a simple regression analysis (see Table 1) on the relationship between the chance of a state win 91 and the appearance of the margin of appreciation in the case, there does not appear to be a very strong correlation. The result is significant (p-value is less than 0,001) and the estimate is in the direction that a marginmention increases the chance of a non-violation or inadmissibility conclusion, but the model can explain only roughly four per cent of the variance (the R-squared), meaning that 96 per cent of what determines the result of a case is something else. Furthermore, taking into account that 17 per cent of marginmentions are false positives we really cannot explain much variance at all.  This model is on the decision-level, in the sense that the variable it aims to explain is the percentage of state wins out of the total number of decisions in each case. Using the percentage rather than the total number of state wins prevents too great impact from cases that address many different articles. 91 The state win combines the non-violation and inadmissibility decisions into one score.
The explanation for this oddity is the rare usage of the margin of appreciation. Only 1,868 of our cases, representing 3,261 individual violation, non-violation, or inadmissibility decisions out of 35,115 decisions in the 21,954 cases, mention the margin of appreciation (see Figure 7.1).
The model in Table 1 qualifies cases as 'margin cases' where the margin of appreciation appears at least once in the judgment rather than the cases containing the margin of appreciation keyword. This has been done since one might argue that the keywords are not the most accurate portrayal of a case, because they are allocated to fewer cases than the ones in which the phrase appears somewhere in the text. 92 Furthermore, it is not public knowledge how the HUDOC administrators determine which cases receive the keyword and which do not. Second, the model combines non-violation judgments and inadmissibility decisions into one value, 'state wins' since both may indicate state deference. However, still given these conditions, the analysis suggests rejecting the assumption that the use of the margin increases the chance of state wins to a significant degree. It may still be the case that while other characteristics have a greater impact on the variance in the chance of state wins, the margin still has an impact within these categories. Comprehensive works on the use of the margin of appreciation such as Arai-Takahashi and Greer contain chapters on the use of the margin in different articles because the use is different in different contexts. 93 In order to check for this possibility, this analysis presents the model in Table 2, which in addition to regressing the margin of appreciation on the percentage of state wins also checks for interactions with the 14 most commonly used Articles (Table 2).
Several interesting connections and data points become visible through the model presented in Table 2. The first detail to notice is that with both the margin and the articles we can explain 11.6 per cent of the variation (R-squared), which although higher is still not impressive considering the high number of variables. Furthermore, several of the variables no longer yield significant results (pvalues are too high). This is not in itself surprising since including the articles slices the data set into many smaller sets each with a smaller chance of yielding significant results. The second result of interest is that when the regression checks for interactions between the margin and the individual articles, in several cases the chance of a state win appears to decrease rather than increase. This is the case for Articles 2, 3, 8 and 14 (see Table 2). Considering that the margin in general is associated with a greater chance of a state win, this is a surprising result. There is no obvious theoretical explanation Furthermore, the keyword regression could explain no more than 3.6% of variation. 93 Greer, supra note 2; Arai-Takahashi, supra note 62.
for why these articles should be different from other articles adjudicated at the Court. In fact, Article 8 is among the list of articles where the Member States in the Copenhagen Declaration welcomed the strict and consistent application of the margin of appreciation. 95 For ease of reading, the non-significant results have been greyed out. 95 Copenhagen Declaration, Art. 28(c).
As a result, the study must conclude that statistically there is no clear relationship between the mention of the margin and a state win. Member states pouring a lot of energy into, and potentially overstepping the independence of the Court in order to make the Court use the margin more, may be disappointed to learn that it does not significantly improve their chances of winning cases.
Turning then to the hypothesis in this article that the hidden variable that ties the margin of appreciation language to the adjudicative results, might be the degree to which a case can be considered hard, in terms of complexity, novelty, and importance. Implicitly if this hypothesis should be able to help explain the discrepancy in Figure 7, then hard cases should be more likely to result in state wins than less complex cases. As explained briefly in Section 4, the reason why this might be the case can be found in the practice of declaring applications admissible/inadmissible before considering the merits. If a case is similar to well-established case law, and is unlikely to result in a violation judgment, it will be declared inadmissible and therefore rarely be published in HUDOC. 96 If, conversely, the established case law suggest that it is likely to result in a violation judgment, a violation will most often be found by a Committee of three judges, or, before the entrance into force of Protocol 14, by a Chamber, and in both cases be labelled as an importance level three case. Therefore, more low-importance level judgments result in violation-judgements. If, however, a case deals with a problem the Court has not encountered before, or if case law on the topic is contested, there is no clear tendency that the case will resolve in one direction or the other; rather, it will be dealt with by a Chamber or the Grand Chamber with a more equal chance of a violation or non-violation judgment.
The most important proxy for 'hard case' is, therefore, the 'Importance Level' assigned by the Bureau. Other indicators of the hardness of a case may also be whether a case has been heard by the Grand Chamber (or previously, the Plenary Court). The Grand Chamber adds complexity initially because there are simply more voices in the analysis when more judges are considering the case. Furthermore, a case only makes it to the Grand Chamber if there is or it is predicted that there will be disagreement about the case in the Chamber, suggesting that the Grand Chamber cases are more likely to be hard cases. Another is whether any separate opinions have been issued to the case, the cases carrying a separate opinion are similarly assumed to be cases with no straightforward answer resulting in enough disagreement on the bench that at least one judge felt the need to either dissent or clarify their concurrence. As mentioned in the introduction these indicators are somewhat colinear, in that all Grand Chamber cases are key-or level one, but many key-and level one cases are not Grand Chamber cases. Furthermore, very few level three cases contain any separate opinions. Due to the co-linearity, we cannot simply add these variables to one multiple regression table, but we can have a look at them individually (see Table 3).
In each case, the result points in the direction assumed by the hypothesis, namely that higher level cases, Grand Chamber, and cases containing a separate opinion are more likely than other cases to include a margin of appreciation mention. The R-squared remains quite low, with importance level explaining 10 per cent of the variation as the highest. This is still a more robust connection than between the margin and state wins given the lower number of variables.
Testing this indirectly as well, by attempting to predict the likelihood of state wins with whether a case is a hard one, we see that the margin of appreciation has a similar effect to the other complexity variables for predicting state wins (see Table 4). It becomes obvious that 'hard cases' is a better predictor of state wins than the margin was on its own. Models one through four together show that the relationship between the usage of the margin and the chance of state wins is not straightforward and that there is a better case for claiming that the margin of appreciation is one of several indicators of 'hard cases' at the Court. Furthermore, in line with theories on incrementalism, 97 and because of how the Court works in terms of declaring most cases inadmissible, hard cases are associated with greater levels of non-violation judgments 96 And therefore not be part of any of the datasets available for this study. 97 Gerards, supra note 9, at 512. than other cases. To this end, the rise in inadmissible cases may be a greater indicator that the Court has become more deferential to states than anything happening in published cases. Unfortunately, this practice along with the non-publication of most inadmissibility cases, means that there is less transparency regarding how, when, and where this increased deference is applied, and which kinds of cases are most likely to be impacted by it.