Can the management school explain noncompliance with international agreements? ABSTRACT Although the management school has been highly influential in the international cooperation literature, the explanatory power of Chayes and Chayes three explanations of noncompliance with international treaties remains severely understudied. This paper develops a framework for examining the explanatory power of treaty ambiguity, lack of state capacity, and unexpected social or economic changes, and conducts a rigorous empirical test in the context of a well-suited case the 1999 Gothenburg Protocol. A careful reading shows that the language of the protocol is clear and unambiguous; indeed, there has been no disagreement over the treaty s content. Furthermore, statistical analyses show no positive correlation between political capacity and compliance. Finally, parties have had adequate time to meet their obligations, and unexpected developments explain only a small part of the observed noncompliance. These findings pose a serious challenge to Chayes and Chayes three explanations of noncompliance at least as far as the Gothenburg Protocol is concerned. PAPER PRESENTED AT ISAs ANNUAL CONFERENCE 2017, BALTIMORE. PLEASE DO NOT CIRCULATE OR QUOTE WITHOUT AUTHOR S PERMISSION Word count: 10,371 (not including my approximately 500-word appendix)
1 Introduction How can we account for noncompliance with international agreements? In their seminal article On Compliance, Chayes and Chayes (1993) formulate their version of the management school. They argue that compliance problems often do not reflect a deliberate decision to violate an international undertaking on the basis of a calculation of interests (Chayes and Chayes 1993, 176). Rather, noncompliance is usually caused by (one or more of) three factors beyond the control of national authorities: Treaty ambiguity, lack of state capacity, and what Chayes and Chayes refer to as the temporal dimension unexpected changes of conditions for compliance following social and economic developments between commitment and implementation. Regarding citations and other attention from fellow scholars, Chayes and Chayes works (notably Chayes and Chayes 1993; Chayes and Chayes 1995; and Chayes, Chayes, and Mitchell 1995) arguably rank among the most influential in the literature about compliance with and the effectiveness of international agreements. 1 The compliance debate gained momentum during the 1990s and early 2000s; however, this progress was driven more by theoretical contributions than by empirical advances. Raustiala and Slaughter argue that compliance remains a relatively young field and that empirical testing of compliance theories is limited (Raustiala and Slaughter, 2002: 548). Raustiala and Slaughter s statements still ring true. 2 Although the relatively few existing empirical studies on compliance with international environmental agreements (IEAs) have improved scholarly knowledge substantially, this literature has a number of gaps. I aim to close some of these gaps. In particular, this paper 1 As of November 2017, Google Scholar counts 1309 citations of Chayes and Chayes 1993 article. 2 As my literature review shows, empirical studies of international environmental agreements have grown substantially in number, but scholars have focused more on effectiveness than on compliance. 1
develops a framework for examining the explanatory power of treaty ambiguity, lack of state capacity, and unexpected social or economic changes. Based on this framework, I conduct a set of rigorous empirical tests in the context of a well-suited case the 1999 Gothenburg Protocol. Crucially, the present paper differs from previous studies in that I statistically control for the ambitiousness of the participating countries commitments (i.e., the size of the required emissions reductions). Failing to control for ambitiousness entails a risk of biased results a risk that is often overlooked (Raustiala 2005, Downs et al. 1996). 3 Unlike both Jacobson and Brown Weiss (1998) and Breitmeier et al. (2006), my dependent variable (compliance) is measured on the ratio level. My study further differs from Breitmeier et al. in that I measure each state s compliance level (rather than the general compliance with a regime at large). Moreover, my explanatory variables (for instance, capacity) are at the state level, rather than at the regime level. Finally, Breitmeier et al. (2006) leave many important judgements concerning causality to their coders, thereby limiting transparency. The case of the 1999 Gothenburg Protocol is well suited for the development of a framework to empirically assess the explanatory power of the management school. First, since the protocol includes national emissions targets for four pollutants, 4 compliance can be measured precisely. Precise measurement of the dependent variable is a prerequisite for the kind of statistical analyses I conduct. Second, assessing whether compliance has been affected by unexpected social and/or economic changes between commitment and implementation (Chayes and Chayes third explanation) is certainly challenging: Ideally, it requires data on how states believed the future would look like when they entered the protocol. However, Gothenburg is part of an international cooperative effort with a strong scientific basis (Castells and Ravetz 2001; Rensvik 2017; Tuinstra 2008). Much energy has been devoted to modeling past and future environmental quality, emissions, and emissions drivers. Such projections were important when Gothenburg s emissions targets were agreed (see Kelly et al. 2010). Therefore, I use projections of future emissions drivers to assess if compliance proved be more difficult to reach than the member states expected when entering Gothenburg. To my 3 My statistical analyses demonstrate the importance of controlling for ambitiousness, since adding the ambitionlevel variable changes the relationship between capacity and compliance, and increases R 2 substantially. 4 See the next section for further elaboration on the Gothenburg Protocol. 2
knowledge, no previous studies have systematically and rigorously assessed the explanatory power of Chayes and Chayes third explanation. The remainder of this article consists of four parts. First, I briefly present the Gothenburg Protocol and the environmental problems it seeks to alleviate. Second, I elaborate on the debate between Chayes and Chayes management school and its counterpart, the enforcement school. I also review previous attempts at testing these two schools hypotheses against empirical evidence. Finally, focusing on the Gothenburg Protocol, I conduct rigorous empirical tests of Chayes and Chayes three explanations. I show that their ambiguity explanation clearly is incapable of explaining noncompliance with the Gothenburg Protocol, since states have mutually consistent interpretations of the contents of the agreement. A series of regressions show no or even a negative relationship between state capacity and compliance, thereby suggesting that the capacity explanation cannot explain noncompliance with the Gothenburg Protocol. Finally, although Chayes and Chayes third explanation can account for some noncompliance, in only four cases can (unexpected) social and economic changes explain all of the gap between targets and observed emissions in 2010. Thus, Chayes and Chayes theory can account for only a small fraction of the noncompliance with the Gothenburg Protocol. 2 Compliance, effectiveness, and the Gothenburg Protocol The Gothenburg Protocol 5 was adopted in 1999 under the Convention on Long-range Transboundary Air Pollution (CLRTAP). Emissions of four different pollutants are regulated by the protocol sulphur dioxide (SO2), nitrogen oxides (NOX), non-methane volatile organic compounds (NMVOC), and ammonia. Together, these pollutants cause the three interconnected environmental problems that the protocol seeks to solve: Acidification, 6 eutrophication, 7 and ground-level ozone 8. 5 I refer to the Gothenburg Protocol of 1999, not the revised protocol of 2012. 6 Largely caused by sulphur and nitrogen oxides emissions. Affects life in water and soil negatively (Miljødirektoratet 2015). 7 Often resulting from ammonia emissions. Increases algae growth and thereby harms organisms. 8 The result of nitrogen dioxides reacting with volatile organic compounds (VOCs). 3
As these are regional environmental problems, most participants in the cooperation under CLRTAP are European states. The protocol came into force in 2005, and 2010 was set as the year by which targets in the protocol must be reached. Table 1 presents the parties, the emissions targets, and the compliance rates. The criterion for being compliant is straightforward: As my analysis below shows, Gothenburg includes emissions targets for four substances for each party, and no provisions that can relieve a party of its obligation to reach its target by 2010 except by withdrawing from the protocol. Thus, a state complied with a target if and only if its emissions of the relevant substance in 2010 were below or equal to this target. 9 Hence, Table 1 does not distinguish between Mitchell s (2010: 147) two kinds of noncompliant behavior ( good-faith noncompliance and intentional noncompliance ) or between his two kinds of compliant behavior ( coincidental compliance and treaty-induced compliance ). Compliance thereby differs from effectiveness, since measuring IEA effectiveness typically involves measuring an IEA s ability to improve environmental quality or state behavior compared to a no-agreement counterfactual. However, establishing this counterfactual is a notoriously challenging task (Helm and Sprinz 2000; Hovi et al. 2003; Young 1999, 2003). Econometric studies have tried to convincingly estimate the effects of CLRTAP protocol participation, producing mixed results (Vollenweider 2013; Ringquist and Kostadinova 2005; Bratberg et al. 2005). As demonstrated by Table 1, there is substantial noncompliance, although the overall compliance is substantial, too. In total, 21 targets were not reached by the deadline in 2010. Ten of them were targets for NOX emissions, eight for ammonia, and three for NMVOC. All SO2 targets were reached by 2010. In seven instances of noncompliance, the target was exceeded by 10 percent or less. Six targets were exceeded by between ten and 20 percent, while another six were exceeded by 20 40 percent. Table 1 also shows that several targets were reached by a wide margin, indicating that compliance could be achieved without much effort. I discuss the potential implications in the 9 In accordance with Young s (1979) definition. See also Fisher s (1981) notion of first-order compliance, as opposed to second-order compliance. 4
analysis section. Table 1: Compliance with targets (deadline year 2010) in the Gothenburg Protocol Party NOX NMVOC Sulphur dioxide Ammonia Belgium 139 107.9 57.1 88 Bulgaria 52.1 55.8 45.2 38.4 Croatia 74 61 49.6 129.4 Cyprus 80 71.3 56.3 62.2 Czech Rep. 77 78.4 56.6 67.1 Denmark 114 147.5 27.9 115.9 Finland 97.6 89.4 57.6 123.4 France 127.5 79.5 71.3 93.4 Germany 123.4 124.5 79 116.8 Hungary 77.8 91.3 5.7 86 Latvia 45.7 65.6 2.4 32.7 Lithuania 45 77.7 14.3 51.4 Luxembourg 358.7 94.3 43.9 67.5 Netherlands 103.1 82.7 68.2 112.3 Norway 113.6 71.6 89.5 119.3 Portugal 68.1 89 31.2 42.8 Romania 53.1 66.5 38.1 80 Slovakia 68.2 45.6 63.1 63.9 Slovenia 104.7 96 36.5 95 Spain 113 97.4 54.6 111 Sweden 101.1 79.5 47.7 90.6 Switzerland 98.3 62.6 46.7 101 United Kingdom 95.1 71.3 68.4 93.9 2010 emissions (as reported in 2015) in % of targets. Targets that were not reached shown in grey (emissions data from CEIP 2015). 3 Theory, previous research, and hypotheses In this section, I present the main theoretical perspectives in the compliance debate and some 5
empirical contributions, and develop a set of hypotheses. 3.1 Enforcement or management? Sources of noncompliance, treaty design The enforcement school 10 (see for example Downs et al. 1996; Barrett 2003; Aakre et al. 2016) argues that states comply only if their expected marginal cost of complying are lower (or equal to) expected marginal revenue. The enforcement school thus views noncompliance as a rational, self-interested actor s reaction to a given material incentive structure. The generally high compliance with international agreements (Henkin, 1968) is attributed to the shallowness of commitments (Downs et al., 1996: 382). 11 Agreements are shallow if their commitments only codify what would happen even if the agreement did not exist. 12 Positive and negative incentives are enforcement scholars main solution to malign collective action problems. In such cases, however, scholars in the enforcement camp are skeptical of the prospects of international cooperation. Since sanctions and rewards may entail high costs for the sender state, promises of carrots or threats of sticks are usually not credible. Unless material incentives are credibly altered, compliance beyond a business-as-usual scenario cannot be expected. In contrast, Chayes and Chayes (1993: 178) claim that sanctions are costly, inefficient, hard to sustain, and unnecessary. Their managerial strategy the management school s prescription for curing the problem of noncompliance consists of softer measures: Monitoring and knowledge sharing, effective dispute settlement, building state capacity, and adjusting treaties in light of economic, technological, social, and political changes. Managerialists (See also Young 1979; Mitchell 2010) argue that international society s anarchical structure is not as detrimental to cooperation as their opponents believe. The main reason is states general propensity to comply a tendency to sincerely try to act in accordance with their international obligations. 10 Other labels are used: The enforcement school is called the incentive perspective by Breitmeier et al. 2006, 110. 11 However, Bernauer et al. (2013) find no support for the enforcement school s hypothesisis of a trade-of between depth and participation. 12 See also Mitchell s (2010) term coincidental compliance and generally his classification of noncompliant and compliant behavior. 6
Three arguments support their claim about states general propensity to comply, according to Chayes and Chayes: First, constant recalculation of interests is inefficient often you are better off doing as the commitment dictates. 13 Second, states negotiate treaties to solve problems. What they commit themselves to do should therefore be in accordance with each state s interests. The final argument concerns states relationship to norms: In common experience, people, whether as a result of socialization or otherwise, accept that they are obligated to obey the law. So it is with states. In other words, states are not only selfinterested, utility-maximizing actors. They are largely also norm-driven actors. And, in international relations, the norm is to do as agreed (Chayes and Chayes, 1993: 178 185. See also Henkin, 1968; Finnemore and Sikkink, 1998; Simmons 1998, 85 88; 2013). Thus, whenever noncompliance occurs, the cause is usually not that cheating maximizes the individual state s private net benefit. Rather, the sources of noncompliance lie beyond the state s reach. First, ambiguity may cause noncompliance. Chayes and Chayes (1993: 188 189) state that Treaties ( ) frequently do not provide determinate answers to specific disputed questions. The fact that economic, technological, scientific, and even political circumstances change creates a zone of ambiguity within which it is difficult to say with precision what is permitted and what is forbidden. Second, compliance might require more than parties can deliver. Scientific and technical competence, bureaucratic resources, and economy are the three constraining factors specified by Chayes and Chayes. Lack of bureaucratic resources can be especially problematic whenever an effective domestic apparatus is needed to regulate individuals and companies behavior, as in the case of long-range transboundary air pollution (Chayes and Chayes 1995, 14, 194). Third, the temporal dimension might explain noncompliance. Chayes and Chayes (1993: 195) argue that Significant changes in social or economic systems mandated by regulatory 13 This may be seen as a version of Keohane s (1984, 13) argument concerning the usefulness of rules of thumb for states in their behavior under regimes. 7
treaties take time to accomplish. Thus, a cross section at any particular moment in time may give a misleading picture of the state of compliance. The moment in time when compliance is assessed should therefore be chosen carefully. Furthermore, conditions for compliance may change between the moment when a commitment is made and the deadline for reaching the targets. If these changes are substantial, unexpected and difficult to control, they may affect states compliance considerably. 3.2 Empirical studies of compliance A series of large empirical studies have shed light on the compliance debate. The score seems, however, unsettled. Previous studies main findings As mentioned in the introduction, Chayes and Chayes three explanations of noncompliance remain somewhat understudied instead, other implications of IEA compliance theories have been empirically assessed (Miles et al. 2002). Breitmeier et al. (2006: 67 68) 14 argue that their findings show that IEAs do have a causal impact on compliant behavior. This finding runs contrary to the expectations of the enforcement school, which does not expect regimes to significantly influence state behavior absent enforcement. Moreover, Breitmeier et al. (2006: 110) support the enforcement school when the authors state that most successful regimes rely on compliance mechanisms involving horizontal sanctioning and institutionalized verification procedures. Summarizing their analyses concerning compliance, Breitmeier et al. (2006: 110-111) state that neither the shallowness argument of Downs, Rocke, and Barsoom (1996) nor the 14 Breitmeier et al. (2006) analyze data from the International Regime Database (IRD). Although 23 international environmental regimes forms the basis for the coded data, the number of units in Breitmeier et al. s analyses is higher, since even units (agreements, protocols) under the regimes are coded and included in the IRD (for instance, the N in Breitmeier et al. s Table 3, showing how various mechanisms correlate with compliance, is 186). The IRD data is coded by 46 experts who qualified for the coding because they were well-known in the scientific community for their knowledge on these regimes. 8
management school of Chayes and Chayes can explain patterns of compliance with international environmental regimes. No similarly clear summary of which theoretical perspective receives more support from their study is provided by Jacobson and Brown Weiss (1998). 15 Nonetheless, both schools receive support from different findings in the editors summary. Figure 15.2 summarizes the findings of their case studies, and lists 32 variables that the authors (1998: 534 536) believe are the most important factors that affect compliance. Among them are sanctions, in keeping with the enforcement school s expectations. On the other hand, scholars of the enforcement camp would not expect most of these factors (for instance equity, administrative capacity, reporting requirements, NGOs) to increase compliance with deep commitments absent enforcement. Depth is, however, not systematically assessed by Jacobson and Brown Weiss. Implementation and effectiveness are the main dependent variables in the 14 regime studies of Victor et al. (eds.), but their findings shed light on the compliance debate. In the summary, the editors state (1998: x): Many analysts believe that failures to implement international commitments are largely unintentional and the best responses are entirely soft measures such as dialogue and financial assistance. We find that some implementation failures are intentional and that harder measures, such as sanctions, are available and sometimes necessary. As cooperation deepens, the need for hard measures may increase. Intentional noncompliance is precisely the kind of state behavior that Chayes and Chayes claim to be rare. Another finding supports the enforcement school s shallowness claim that a high degree of compliance does not necessarily imply behavioral change: our cases suggest that compliance often simply reflects that countries negotiate and join agreements with which they know they can comply ( ) In short, legally binding agreements often codify what is already under way (Raustiala and Victor 1998, 662). 15 This anthology includes studies of eight states (and the EU s) compliance with five international environmental treaties. 9
What explanatory power have previous studies attributed to the three factors that, according to Chayes and Chayes, cause noncompliance? Concerning ambiguity and compliance, Breitmeier et al. (2006: 90 93, see also their Table 3.11) find that the association between the precision of rules and compliance rates is positive but not strong. Jacobson and Brown Weiss (1998) conclude similarly. Jacobson and Brown Weiss (1998) and Breitmeier et al. (2006) offer divergent findings concerning capacity. The latter conclude that [our data] do not confirm expectations about the role of capacity building, while the former find that administrative capacity is important. 16 Studying previous protocols under CLRTAP, Underdal and Hanf (eds., 2000) find that a model of parties as unitary rational actors a model corresponding roughly to the enforcement school s predicts patterns of negotiation position, implementation, and compliance (operationalized as emissions reductions) well (Underdal, 2000: 351 353). Effectiveness has been the main focus of previous studies of cooperation to reduce long-range transboundary air pollution (Levy 1993). According to Wettestad (2012: 34) efforts to mitigate this environmental problem have achieved significant results. However, to a considerable degree, these results are due to other factors than the regime itself. Helm and Sprinz (2000) find that CLRTAP cooperation has reduced emissions compared to the counterfactual situation, although it falls short of reaching the collective optimum emissions levels. Böhmelt and Vollenweider (2015) find that states centrality in the CLRTAP network explains effectiveness (operationalized as emissions reductions). Bratberg et al. (2005) conclude that the 1988 Sofia Protocol had a positive effect on emissions reductions, while Ringquist and Kostadinova (2005) find no such effect of the 1985 Helsinki Protocol. 16 Their differing findings may be explained by the differences of research design and units of observation: As notet above, Breitmeier et al. s (2006) dependent variable is the general (not state-level) compliance with a treaty or regime, Brown Weiss and Jakobson s (1998, eds.) dependent variable is individual states compliance. Although four of the five treaties studied by Brown Weiss and Jakobson (1998, eds.) are among the 23 regimes under scrutiny by Breitmeier et al. (2006), the latter s empirical focus is certainly the broadest of the two. 10
An in-depth study concludes that neither the enforcement school nor the management school can account for Norway s noncompliance with its Gothenburg Protocol NOX target (author, 2016). NOX emissions across Northern and Western European countries fit well with what I call the deadline pressure theory. This theory suggests that although states increase their efforts to comply with an IEA when the deadline nears, they are reluctant to implement costly policies when the deadline is more distant. Prospects for compliance are thereby weakened, since the increased efforts may well come too late to reach compliance. How my study differs from previous studies The IEA compliance literature suffers from a number of gaps and limitations which I aim to overcome. This is not to say that my research design necessarily is preferable to those of previous studies, but that different designs yield mutually informative insights. For instance, international regime scholars should be interested in explanatory and outcome variables at both the state and at the regime level. My explanatory variables are (whenever appropriate) at the state level rather than at the regime level. In contrast, when Breitmeier et al. (2006) study the capacity-compliance relationship, they compare regimes with mechanisms for increasing states capacity to regimes without such mechanisms. Since the dependent variable (compliance) and several of my independent variables are measured at the ratio level, I use far more information than, for example, Jacobson and Brown Weiss (1998), who classify compliance as either low, medium, or substantial. Granted, my continuous operationalization of compliance has both advantages and drawbacks. However, the same can be said about ordinal-scale and binary compliance variables. Hence, scholarly knowledge on compliance becomes more solid when scholars study compliance measured on various levels. While I measure each state s compliance with several commitments, Breitmeier et al. s coders assess the overall level of compliance with each regime and assign a value between 1 and 5 (See for instance Breitmeier et al. 2006, 75.) Thus, the level of compliance is not measured in the strict sense of the word, but is subject to the judgement of each coder. In contrast, I use publicly available statistics of states emissions. 11
Moreover, my study differs from previous studies concerning how conclusions are drawn. For example, while Jacobson and Brown Weiss often rely on a rather informal mode of inference, 17 I use statistical tools when I study the relationship between political capacity and compliance. The procedure used by Breitmeier et al. s coders to score units variable values implies that many inferences concerning causality are drawn by the individual coders. Although leaving such important inferences to coders may be necessary in larger-n studies, it reduces transparency. 18 Finally, but most importantly, my regression analyses control for the commitments ambition level how large emissions reductions were required by each country. In contrast, Breitmeier et al. only implicitly take this potentially crucial factor into account, while Jacobson and Brown Weiss do not systematically consider it. Failing to control for the depth of IEA commitments may lead to biased results. 3.3 Hypotheses and research design In the face of noncompliance, the empirical expectations of Chayes and Chayes management school are clear. From the ambiguity explanation, the following hypotheses may be derived: H1a: The contents of the Gothenburg Protocol are open to interpretation. H1b: The parties have divergent views of their obligations under the protocol. Even though capacity limitations are more common among developing countries, Chayes and Chayes (1993: 194) argue that they may be a problem for more well-to-do countries as well. Thus, if the lack of state capacity explains the observed 17 See for instance Jacobson and Brown Weiss 1998, 530 531, where the authors state that Our analyses confirmed our expectation that richer and more democratic countries would in general do better than poorer and less democratic countries. 18 International Regime Database (IRD) coders were asked to judge regimes causal influence on various actors behavior (see for example pages 75, 84 85, 270). Thus, many of Breitmeier et al. s (2006) inferences depend on IRD coders ability to construct counterfactual no-regime scenarios and compare them to the observed outcomes. As is well known, counterfactual judgements are challenging to conduct. 12
noncompliance, we should find a positive correlation between political capacity and compliance: H2: The higher a state s capacity, the higher the (chance of) compliance. Chayes and Chayes temporal dimension suggests that time may have been too short to reach the targets not complied with. I examine how conditions crucial for compliance have developed. If these conditions have developed differently than the parties expected when the agreement was adopted for instance if consumption of energy in 2010 was higher than projected in 1999 the temporal dimension may account for noncompliance. However, the difference between projections and what actually happened must be substantial, and large enough to account for the gap between the 2010 emissions and the 2010 target. 4 Analysis: Can treaty ambiguity explain noncompliance? Gothenburg s Article 3, Paragraph 1, states that Each party shall, as a minimum, control its annual emissions of polluting compounds in accordance with the obligations in annex II. Annex II specifies emissions ceilings for ammonia, NOX, sulphur dioxide, and NMVOC in thousand (metric) tonnes per year for 36 states (as well as for the EU). The deadline year is 2010. Gothenburg includes no provision that could exempt parties from being obliged to reach the emissions ceilings by 2010 unless they withdraw from the agreement. Thus, Gothenburg s language is clear and unequivocal: Parties to the agreement shall in 2010 and thereafter not exceed their designated annual emission ceilings. Essentially, Gothenburg consists of emissions targets and timetables. 19 Germany s environmental agency, the Umweltbundesamt (2017), writes that After 2010, NOX emissions above 1,081 thousand tonnes are not allowed [by the Gothenburg Protocol]. 20 19 Like many other UN agreements, CLRTAP protocols have no significant enforcement mechanisms (Wettestad 2012). 20 The original text is as follows: Seit dem Jahr 2010 dürfen 1.081 Tausend Tonnen NOx nicht mehr 13
Similarly, the government of Norway, a party that did not reach its 2010 NOX target, seems to have had no doubts about its obligations. When I asked former Norwegian Minister of Development and Environment Erik Solheim (2013) directly if there had been any disagreements over what Norway had committed itself to, he replied: I cannot remember that anyone ever suggested that the content of the protocol was unclear. Our discussions concerned two key questions: What can we achieve, and how fast can we do what we are obliged to, knowing that we are behind schedule? Statements from Danish (Miljøstyrelsen 2002), Swedish (Naturvårdsverket 2016), and British (Department of Environment, Food and Rural Affairs 2015) authorities express views fully consistent with the statements from former Minister Solheim and Germany s Umweltbundesamt. It seems clear that ambiguity did not cause the noncompliance with the Gothenburg Protocol. The fact that compliant and noncompliant parties alike have mutually consistent interpretations of their obligations strengthens this conclusion. 5 Analysis: Can lack of capacity explain noncompliance? In this section I examine the relationship between political capacity and compliance. 5.1 Operationalization According to Chayes and Chayes (1993: 194), economic wealth increases states capacity for compliance. Thus, gross domestic product (GDP) per capita, logtransformed, 21 should be a reasonably valid operationalization of state capacity: GDP per capita, a common indicator of economic prosperity, allows comparisons across states. The literature seems to agree that a state s bureaucratic resources and capabilities depend considerably on its general level of überschritten werden. 21 I have logtransformed GDP per capita because its relationship to political capacity is likely nonlinear: I expect that the positive relationship gets weaker with increasing GDP per capita. 14
economic development (Jänicke 1997, 2; see also Jacobson and Brown Weiss 1998, 531, as well as their preface, xv). Previous studies, such as Börzel et al. s (2010) assessment of compliance with EU law, also operationalize capacity as GDP per capita. 22 Nonetheless, state capacity is a highly contested concept which is challenging to measure validly (Hanson and Sigman 2013; Jänicke 1997). I therefore present regressions using two alternative operationalizations as robustness tests. Ambition level, a potentially crucial control in compliance studies, is operationalized as a function of the difference between the 2010 emissions target and the corresponding emissions of the year the protocol was adopted, 1999. The idea is that the larger the emissions reductions required, the harder it is to reach compliance. This variable is calculated as follows: Each country s observed emissions of a given substance in 1999 are divided by the corresponding emissions target. To illustrate, the United Kingdom s NOX target was 1,181,000 tonnes, 23 and its 1999 NOX emissions were 1,866,700 tonnes. Thus, because the emissions in 1999 were 58 percent above the target, the UK s NOX unit is assigned the value 1.58 on ambition level. Compliance is operationalized as a continuous variable, 24 measuring the 2010 emissions deviance from the 2010 target. Values above 0 indicate that emissions were below the target (the state thus complying), while emissions targets that were not reached score values below 0. For instance, the UK s NOX target unit scores 0.049 on the compliance variable, because 2010 emissions were 4.9 percent below target (see also Table 1). 22 Similarly to Börzel etal. (2010) I also operationalize capacity as states scores on a government effectiveness index (see my robustness tests). 23 Unless I state otherwise, all emissions are in metric tonnes, and as reported to UNECE in 2015. 24 Arguably, a continuous compliance variable may include irrelevant information, since several states have over-complied significantly. My robustness tests therefore include analyses that use a dichotomous compliance variable. 15
5.2 Data and estimation The observational unit of my regressions is a given emission limit for a given party. All emissions targets shown in Table 1 thus correspond to a unit in my data set. Emissions data are from states reports to UNECE (CEIP 2015). Every party has four obligations, one for each regulated substance. Consequently, all standard errors are clustered on states. With 23 European states being a party to the protocol, the data set consists of 92 observational units. Because all my variables are measured on the interval scale or are dichotomous, I use OLS regression to estimate the causal effects of my independent variables. 25 5.3 Results Model 1, including capacity as the only independent variable, shows a strongly negative and significant effect of capacity on compliance. When I control for ambition level (Model 2), the effect of capacity on compliance is still somewhat negative, but no longer statistically significant. Thus, the importance of controlling for ambition level is demonstrated. Model 3 also includes a dummy variable that controls for geographical, historical, political, and economic ties. When this control is added, the estimates for capacity and ambition level are similar to those of Model 2, except that the effect of capacity is once again significant. Thus, I do not find the positive and significant relationship between capacity and compliance expected by the management school. The robustness tests of section 5.4 show that this conclusion holds under a number of conditions. Hence, the conclusion that the lack of capacity cannot explain (non)compliance with the Gothenburg Protocol seems highly robust. 25 A multilevel model is infeasible because of few (4) units on the state level. Likewise, estimating causal effects by using instrumental variables (Angrist and Pischke 2009) is infeasible since it is highly doubtful that any valid instrument Z exists for my variables (see Angrist and Pischke s (2009, 117) discussion of criteria for valid instrumental variables). Several studies (Vollenweider 2013; Ringquist and Kostadinova 2005; Bratberg et al. 2005) estimate the effect of participation in CLRTAP agreements on emissions by employing the difference-indifferences (DID) estimator, thus comparing participants to non-participants. The DID technique is, however, less feasible when compliance is the dependent variable, since only states that participate in the agreement may comply (or defect). 16
Table 2: OLS regressions. Dependent: Compliance Model 1 Model 2 Model 3 Constant 4.709 1.985 3.552 (log) GDP/cap. -1.008*** -0.263-0.590** Ambition level -0.525*** -0.553*** Eastern Europe -0,189* R 2 0,229 0,508 0,526 N 92 92 92 * Coefficient is significant at the 10 percent level. ** Coefficient is significant at the 5 percent level. *** Coefficient is significant at the 1 percent level Standard errors are clustered on states. Even though the models explained variance is not crucial for the purpose of this paper, it is interesting to note that R 2 increases substantially when ambition level is included. 5.4 Robustness test Operationalizing compliance continuously enables scholars to study degrees of (non)compliance. The continuous compliance variable might, however, include irrelevant information. As Table 1 shows, several 2010 emissions were far lower than required by the protocol. Significant overcompliance may indicate that the emissions level was not a result of efforts to reach the target. Yet, available data do not indicate to what extent the continuous compliance variable includes irrelevant information. Thus, running regressions using a dichotomous dependent variable provides an important robustness test. Units with 2010 emissions above the protocol s target now score 0, as they have not been complied with. Units with emissions below or equal to the target score 1. 17
Table 3: Logistic regressions. Dependent: Compliance Model 1 Model 2 Mod. 3 Constant 22.83 18.13 13.38 (log) GDP/cap. -4.789*** -3.217** -2.248 Ambition level -1.883-1.711** Eastern Europe 0.658 Nagelkerke R 2 0.176 0.283 0.288 N 92 92 92 * Coefficient is significant at the 10 percent level. ** Coefficient is significant at the 5 percent level. *** Coefficient is significant at the 1 percent level. Standard errors are clustered on states. Table 3 shows the results of three regressions corresponding to the models shown in Table 2, except that the dependent variable of Table 3 s models is dichotomous. The relationship between capacity and compliance is negative in all three analyses, and statistically significant in Models 1 and 2. As I argue above, GDP per capita may be a somewhat rough operationalization of political capacity. Hence, I test the robustness of my findings using alternative operationalizations. Table 4 shows the results of three OLS regressions in which capacity has been operationalized as all parties average score between 2000 and 2010 on the Government Effectiveness indicator, one of the World Bank s Worldwide Governance Indicators (WGI) (World Bank 2017). According to the World Bank, this indicator Reflects perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies. All three models show a negative relationship between capacity and compliance; hence, these results support the findings reported in Tables 2 and 3. 18
Table 4: OLS regressions. Dependent: Compliance Model 1 Model 2 Model 3 Constant 0.509 0.886 1.074 WGI_GovtEff -0.238*** -0.024-0.107 Ambition level -0.567*** -0.588*** Eastern Europe -0.147 R 2 0.148 0.498 0.505 N 92 92 92 * Coefficient is significant at the 10 percent level. ** Coefficient is significant at the 5 percent level. *** Coefficient is significant at the 1 percent level. Standard errors are clustered on states. Table 5 shows the results of additional OLS regressions 26 using another WGI indicator to operationalize capacity. According to the World Bank s description, Regulatory Quality reflects perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development. Table 5: OLS regressions. Dependent: Compliance Model 1 Model 2 Model 3 Constant 0.612 0.889 1.058 WGI_RegulatoryQuality -0.329*** -0.03-0.106 Ambition level -0.572*** -0.595*** Eastern Europe -0.098 R 2 0.113 0.498 0.502 N 92 92 92 * Coefficient is significant at the 10 percent level ** Coefficient is significant at the 5 percent level *** Coefficient is significant at the 1 percent level Standard errors are clustered on states. 26 In the models presented in Tables 4 and 5, I operationalize compliance continuously. I have, however, also run regressions with the dichotomous compliance variable and with the WGI indicators as operationalization of capacity. These regressions show the same results as the models in Tables 4 and 5 do. 19
Once again, no positive relationship between capacity and compliance is found. Thus, all my regressions show the opposite relationship between capacity and compliance to the one expected by the management school (H2). Hence, the result seems highly robust. 6 Can the temporal dimension explain noncompliance? I now turn to Chayes and Chayes third explanation, the temporal dimension. A starting point for this analysis is the fact that the emissions targets were not binding until 2010. Considering that Gothenburg was adopted in 1999, the kind of premature judgement of compliance that Chayes and Chayes warn against (1993: 195) cannot be a problem. Still, it is too early to conclude that the temporal dimension cannot explain noncompliance. Chayes and Chayes general proposition if a state is noncompliant, the reason is not lack of good intentions can be extended as follows: Changes in costs, knowledge, technology, or other factors may fundamentally alter the conditions for compliance. If these conditions prove considerably worse than expected when the obligations were accepted, these changes can, fully or partly, explain breaches of obligations. This section reviews compliance-relevant changes that occurred between 1999 and 2010, and asks if they were large enough to explain instances of noncompliance with the Gothenburg Protocol. 6.1 NOX: Unexpected developments of conditions for compliance I begin by showing how parties energy consumption and other major NOX emissions drivers developed over the period 1999 2010 compared to projections. Next, I show how sciencebased knowledge on NOX emissions has evolved and discuss how it might have affected states success in reaching their emissions targets. Unexpected development of energy consumption 20
Amann et al. (1999) 27 identify population, GDP per capita, the number of vehicles, and energy consumption as determinants of NOX emissions. However, as GDP per capita, population, and number of vehicles hardly can influence emissions otherwise than through energy consumption, I examine only energy consumption developments. 28 Table 6 presents analyses of the difference between projected (Amann et al. 1999) and observed 29 2010 energy consumption of all Gothenburg parties that did not comply with their 2010 NOX targets. The analyses show the extent to which the difference between projected and observed consumption of energy in 2010 can account for the noncompliance. The numbers in the column Emissions attributable to the difference between projected and observed 2010 energy consumption are calculated by using the following formula: (Observed energy consumption in 2010 (Petajoule) - Projected energy consumption in 2010 (Petajoule)) * NOX emissions per Petajoule consumed in 2010 If the product of this formula is positive, the state involved consumed more energy in 2010 than was projected in 1999. If the product is negative, the observed energy consumption in 2010 was lower than was projected in 1999. In all states whose 2010 energy consumption was lower than projected Belgium, Germany, France, and the Netherlands the development of energy consumption made compliance easier to reach than was projected in the 1999 scenario. Thus, unexpected developments did not make compliance less attainable. We may therefore conclude that the analysis of development of energy consumption suggests that the temporal dimension cannot explain these states noncompliance. On the contrary: As Table 6 shows, Belgium s 2010 NOX emissions were 39,700 tonnes above the target for 2010. When we consider the fact that energy consumption in 2010 was lower than expected, one might argue that there is more deviance between target and behavior to account for. As the far-right column of Table 6 shows, Belgium s 2010 noncompliance plus the emissions attributable to the difference 27 This report was written by scientists at the International Institute for Applied Systems Analysis (IIASA) to make the scientific background for Gothenburg s commitments available to the wider public. Projections and other analyses from IIASA are considered as important inputs in the process deciding emissions targets (Castells and Ravetz 2001; Rensvik 2017; Tuinstra 2008). 28 As emission coefficients vary considerably among sources of energy, aggregate energy consumption is not my first best data. However, this is the only projection on energy consumption included by Amann et al. 1999. 29 Based on data from Eurostat 2017. 21
between projected and observed energy consumption is 42,960 tonnes of NOX. I now turn to the parties whose 2010 energy consumption was higher than was projected in 1999 Spain, Luxembourg, Sweden, Norway, and Denmark. For these states, energy consumption developments made the NOx target more difficult to reach than was expected when the protocol was adopted. Is this unexpected development large enough to account fully for the noncompliance? The answer is provided by the numbers in the far-right column in Table 6. They are calculated by subtracting the emissions attributable to the unexpectedly high energy consumption from the state s noncompliance 30. For instance, Norway s 2010 energy consumption was substantially higher than projected, and can account for 30,800 tonnes of NOX. Since Norway s noncompliance was only 28,300 tonnes, this analysis suggests that Norway s noncompliance may be explained by unexpected developments. The same conclusion holds for Denmark, whose (rather small) noncompliance of 1,800 tonnes is outweighed by the emissions attributable to the unexpectedly high energy consumption. The opposite conclusion may be drawn in the other three cases where 2010 energy consumption was higher than projected. Spain s 2010 NOX emissions were 136,900 tonnes above its target, but only 1,600 tonnes can be attributed to the energy consumption factor. Similarly, the energy consumption factor can account for only 1,300 of Sweden s 13,400 tonnes noncompliance. Given Luxembourg s 35,200 tonnes noncompliance, almost 26,000 tonnes are left unaccounted for after 9,300 tonnes are subtracted. 30 The gap between the 2010 NOx emissions and the emissions target, as shown in the column Noncompliance. 22
Table 6: Energy consumption and NOx noncompliance (all numbers in thousand tonnes of NO X) Party Emissions attributable to the difference between projected and observed 2010 energy consumption Noncompliance Noncompliance minus emissions attributable to the difference between projected and observed 2010 energy consumption Belgium -3.26 39.7 42.96 Denmark 5.6 1.8-3.8 Germany -57.8 241.9 299.7 Spain 1.6 136.9 135.3 France -43.6 220.3 263.9 Luxembourg 9.3 35.2 25.9 Netherlands -14.1 9.9 23.99 Sweden 1.3 13.4 12.1 Norway 30.8 28.3-2.5 A state s noncompliance is calculated by subtracting its 2010 emissions target in the Gothenburg protocol from its 2010 emissions (as reported to UNECE in 2015). Thus, this analysis suggests that only two of nine cases of NOX noncompliance may be explained by unexpected developments of energy consumption, which is a major determinant of emissions. Can previous underestimation of NOX emissions account for noncompliance? A number of sources (UNECE 2003; European Commission 2016) point out that over the last 10 to 15 years, scientists have discovered that diesel vehicles emit more NOX than previously thought. Estimates of states NOX emissions have consequently been adjusted several times, potentially making the quantified targets of the protocol harder to reach. Two crucial questions arise: First, when did the parties acquire this new information? Second, how large was the gap between the real NOX emissions and the emissions levels estimated using previous knowledge? A study presented in 2003, carried out in Germany, Switzerland, and Austria, found that projected German NOX emissions were 10.2 percent higher than anticipated (UNECE 2003). This study was presented at a meeting in the science and advisory body TFIAM 31 in June 31 The Task Force on Integrated Assessment Modelling (TFIAM) was established in 1986 as a body under the 23
2003. National experts from 13 states 32 plus the EU were present (UNECE 2003). Except Spain, all parties that did not comply with their 2010 NOx targets were represented. Additionally, according to minutes from the meeting, the Task Force had noted that several countries were reviewing NOX emission data from heavy-duty vehicles (HDVs). 33 The findings seemed to suggest that NOX emissions from HDVs following the EURO 2 and 3 specifications were in reality higher than assumed in previous estimates. Thus, it seems like the parties did have good reasons to increase their efforts to reach compliance already in June 2003, more than 6.5 years before the 2010 deadline. Thus, there was time aplenty to implement measures to compensate for the increased emissions estimates. This is, however, only part of the story. The EURO 2 and 3 standards mentioned by the minutes above are only two of a series of EU emissions standards for road vehicles put in force since 1991 (European Commission 2016. Hence, a number of sources may have led to mis-estimation of emissions between 1999 and 2010. It is challenging to calculate exactly how much mis-estimation can be attributed to deviations from each EU emissions standard. For the purpose of the present paper, however, publicly available emissions data may provide answers. Table 7 shows how estimates of 1999 34 NOx emissions (as reported to UNECE) have varied over time. 35 These data allow comparison of what states believed were their NOX emissions in a given year to what the real 36 emissions were. Table 7 shows that for several of the studied states, estimates of 1999 NOX emissions have been fairly stable. In 2015, Belgium s 1999 NOx emissions were estimated at 312,700 tonnes. In 2001 2007, the Belgian 1999 emissions were reported at 292,000 tonnes thus, 20,700 LRTAP Convention. The Task Force s main focus [is] ( ) to combine information gathered from the Parties and from other Convention bodies and through computer models assist in the development of legal instruments. 32 Belgium, the Czech Republic, Denmark, Finland, France, Germany, Italy, the Netherlands, Norway, Slovenia, Sweden, Switzerland, and the United Kingdom. 33 Vehicles in the HDV category use diesel fuel. 34 The appendix includes tables showing estimates of 2003, 2006, 2008, and 2010 emissions. 35 I include only those states that were noncompliant with their 2010 NOx Gothenburg target. 36 Defined as the estimates reported in 2015. Obviously, there is an artificiality to this classification, since even recent emissions estimates are subject to change because of scientific evidence. However, since estimates from 2015 are derived from presently best available scientific knowledge, I use 2015 estimates as baseline. 24