2. The System of Notifications in the WTO

Notification obligations as the central issue addressed in this paper represent one of the key means to increasing transparency of trade policy. In general terms, transparency in a political environment is usually understood as the state of affairs when entitled or interested actors dispose with relevant information about a respective political action (e.g. Forssbaeck and Oxelheim, Reference Forssbaeck and Oxelheim2014: 5–10). In the field of trade, policy transparency is concerned with the availability of information on the provisions and processes related to national trade policies (Plummer and Tafti, Reference Plummer, Tafti, Forssbaeck and Oxelheim2014). The WTO as such relates transparency to the degree to which national trade policies are open and predictable (World Trade Organization, 2018).

Transparency constitutes a factor that significantly affects the behavior of both firms and governments. Concerning firms, a key precondition of their involvement in international trade is their knowledge of the conditions in foreign markets, particularly of foreign tariffs and the rules applying to their products. Consequently, transparency in national trade policies decreases the resources that firms need to invest to obtain information about foreign markets (Handley and Limão, Reference Handley and Limão2012). In this way, transparency facilitates their entry into those markets and, therefore, also the growth of international trade. Furthermore, policy transparency reduces concerns that governments may have regarding respective compliance with the agreed terms of cooperation (Keohane, Reference Keohane1984: 92–96). When governments maintain their markets open, they do so with the precondition that their counterparts reciprocate. Hence, a government will normally view having information about the trade policies of other states as a precondition for its own cooperative behavior. This problem is exemplified by the current high-profile conflict between China, the USA, and the EU concerning the steel industry, and the lack of transparency of the various subsidy schemes and anti-dumping measures that states apply in this industry (World Trade Organization, 2017b, 2017d).

Let us describe the notification system that underpins the activity of Council committees in more detail, as it lies at the very core of our interest. WTO member states have a duty to notify the organization about the laws and regulations they adopt, and which either have something to do with the organization's treaties or affect other member states (Collins-Williams and Wolfe, Reference Collins-Williams and Wolfe2010; Wolfe, Reference Wolfe2013; World Trade Organization, 2017c). The principle of notifying about trade policy changes was introduced already by Article X of the GATT in 1947 (GATT, 1947). This principle was re-emphasized in 1979 by the ‘Understanding Regarding Notification, Consultation, Dispute Settlement, And Surveillance’ (GATT, 1979). Article 3 of the Understanding stipulates that member states should notify their counterparts about the adoption of trade measures affecting the operation of the GATT. Notification should occur before the respective decisions are taken. In case this is not possible, states are expected to provide notification of their measures ex-post. The contemporary system of notification obligations was further elaborated by the WTO decision ‘On Notification Procedures’ adopted in 1994 (World Trade Organization, 1994). The decision delimited a list of particular measures for which notification should be given. In addition, it also established a central registry of notifications.

The overall number of notification obligations currently reaches approximately 170, although for only around 120 of them it is possible to locate at least some submitted notification documents in the WTO online database system. Three basic types of notifications exist: one-time, regular, and ad-hoc. One-time notifications are submitted by member states only at the beginning of the notification process; practically speaking, they are due shortly after WTO accession. They provide an overview of the legislative measures that apply to a specific area of national trade policy, but also, for example, identify the relevant administrative contact points for business inquiries. Concerning some aspects of trade policy, states need to announce any changes made on a regular basis, by means of regular notifications. These are delivered semi-annually, once a year, or every other year. Ad-hoc notifications are used for the provision of information about changes in governments’ trade legislation and policies in selected areas. The rare hybrid form of ‘one-time and ad-hoc’ notifications then combines the initial one-time and the later ad-hoc component. The set-up of notification obligations reflects the WTO's general principle of special and differentiated treatment (SDT). The duty of individual countries to notify depends on their development status. While the majority of notifications concerns all members, some notification obligations are concerned only with the trade policies of developed countries and some others are related only to developing countries.

Figure 1 shows the distribution of the notification requirements, by types and main thematic areas of the WTO agenda. With regard to notification types, ad-hoc notifications account for 56.2% of all notification requirements. One-time and one-time and ad-hoc notifications follow with a share of 28.4%. The remaining 15.4% of obligations are represented by regular notifications. However, these figures do not correspond to the actual volume of submitted notifications, as for example regular notifications are submitted typically every year, and thus constitute a much larger share of all notification activity. Two thematic areas characterized by the highest number of notification obligations are trade in textiles (17%) and the protection of intellectual property rights (12%). These are followed by safeguards and services (9% each), agriculture (7%), and subsidies and countervailing measures (also 7%). Nevertheless, the use of the three types of notification requirements also varies across the individual thematic fields. To mention the most outstanding examples: whereas a large majority of notifications related to administrative border measures (such as customs valuation) are one-time notifications, the prevailing part of obligations connected with textiles, services, and intellectual property rights are ad-hoc notifications. Agriculture is the only area in which the majority of obligations are regular (periodic).

Notes: Ad-hoc obligations are depicted in the lightest color (bottom), one-time and ‘one-time and ad-hoc’ obligations in light color (middle) and regular obligations in dark color (top). The following thematic abbreviations are used: ADP – anti-dumping, AG – agriculture, BOP – balance-of-payments, GATS – services, LIC – licensing, MA – market access, PSI – preshipment inspection, RO – rules of origin, RTAs – regional trade agreements, SCM – subsidies and countervailing measures, SG – safeguards, SPS – sanitary and phytosanitary measures, STR – state trading enterprises, TBT – technical and safety regulations, TEX – textile goods, TPR – Trade Policy Review Mechanism, TRIMS – investment measures, TRIPS – intellectual property rights, VAL – valuation

Figure 1. All notification obligations – types and thematic areas

3. Theoretical Explanations

Let us now move to the theoretical core of our article, namely the identification of factors that we expect to influence states’ overall levels of compliance with their notification obligations. As we outlined in the introduction, since this is the first study to systematically quantitatively map compliance with one-time and regular notifications, we primarily seek to identify and explain the variation in the submission of notifications across WTO members. Our theoretical framework is thus designed with this purpose in mind. In contrast, we do not systematically explore the possible variation across policy areas of the WTO agenda. This task would require a rather different theoretical framework and a separate analysis.

We base our synthetic theoretical framework on explanations from the literature on the provision of information by states to international institutions. Our first two hypotheses are concerned with the trade policy preferences of states, whereby we consider the degree of their interest in the liberalization of trade to be particularly decisive for the level of their compliance with notification obligations (in line with the more general reasoning put forward by, e.g., Mitchell, Reference Mitchell1998). The relationship between interest in trade liberalization and compliance with notification obligations is relatively straightforward and simple. It stems from the fact that the WTO represents the basic institutional framework for the reduction of trade barriers on a global scale. As a result, those states that are characterized by a stronger interest in trade liberalization should be particularly determined to support the proper functioning of the WTO. For this reason, they should also display greater discipline in the fulfillment of notification obligations.

Of course, it is notoriously difficult to assess empirically the preferences of actors (e.g. Zürn, Reference Zürn1997), especially when the need to do so covers a large number of states – in our case, all WTO members. Hence, we resort to two key long-term, structural conditions of states’ involvement in the global trade regime to approximate their interests and preferences. These are trade dependence, or the trade–GDP ratio, and trade openness, reflecting the degree of tariff protection. Trade dependent countries should be particularly interested in trade liberalization as their economic welfare considerably depends on trade (Simmons, Reference Simmons1997: 188–189). Concerning trade openness, a low level of trade barriers indicates that a state has a strong interest in open trade.Footnote ³ After all, it has been argued that open trade policy leads to the strengthening of liberal preferences of key domestic constituencies, and thus strengthens the overall liberal orientations of states’ foreign economic policies (Milner, Reference Milner1987). Neither of these two proxies measures directly states’ preference for trade liberalization, but in combination they provide a very solid account of which states have direct material interests in the maintenance of a relatively open global trade.

Our second line of theorizing is concerned with the institutional characteristics of the WTO members. First, we draw on an argument that generally links the level of democracy with greater policy transparency (Hollyer et al., Reference Hollyer, Rosendorff and Vreeland2011). In democratic systems, the survival of the government depends on the election result and, therefore, on the welfare of voters. For this reason, political representatives in democracies are more likely to make their policies transparent in order to provide the voters with information about the relationship between policy decisions and their welfare. In line with this, we would expect countries with a higher level of democracy to be more active in the WTO notification system:

Third, we introduce two other variables that are also connected with the institutional characteristics of states, but enhance the ability of states to respond to their procedural obligations in international institutions. The first of these variables is domestic administrative capacity. As was underlined by one theoretical stream in the research on compliance with international commitments, this compliance can be shaped considerably by the (lack of) bureaucratic capacity of the state (Chayes and Chayes, Reference Chayes and Chayes1993). To be sure, a limited administrative capacity decreases the ability of a state to implement policy significantly more than its ability to report on this implementation. However, since poorly functioning administrative systems are characterized by general deficiencies, they are likely to also have shortcomings related to information obligations. In fact, the idea that deficiencies in bureaucratic capacity can influence the submission of notifications has already been pointed out by Wolfe in his analysis of notifications delivery (Wolfe, Reference Wolfe2013: 18). As he notes, some member states fail to deliver their notifications due to a lack of knowledge or data, or to an inability to engage in cross-departmental cooperation. This reasoning leads to the next hypothesis:

As notification obligations documents in the WTO are not prepared by a country's Geneva diplomatic mission, but rather by various domestic agencies, domestic administrative capacity is especially relevant concerning compliance with notification obligations.

In addition, we hypothesize that a state's compliance with notification obligations is also likely to increase with the extent to which it is exposed to institutionalized multilateral policy-making through its membership in IOs. This expectation is grounded in organizational theory, which claims that state behavior is shaped by its long-term institutional adaptation (Powell and DiMaggio, Reference Powell and DiMaggio1991). Membership in IOs exposes states to this type of adaptation. It entails for a state a rather extensive range of policy and administrative tasks and opportunities. These tasks involve procedural obligations that a state has to fulfill, e.g. by the formulation of negotiation positions or the provision of reports. However, participation in multilateral policy-making also provides a state with institutional opportunities to advance its interests, or to seek to resolve some policy problems. States that are relatively more exposed to multilateral policy-making need to react more frequently and intensively to tasks and opportunities related to this policy-making. As a result, participation in various forms of multilateral policy-making becomes part of the standard operating procedures of their foreign policy-making systems (Allison, Reference Allison1971). This implies that they regard various activities associated with IO membership, including the preparation and supply of reports and notifications, as rather standard and routine operations.Footnote ⁴ Consequently, they are also likely to participate more in such activities:Footnote ⁵

These are the five factors that we deem theoretically most relevant and interesting. Nonetheless, in our explanatory analysis we also control for several other potentially relevant factors. The first of these factors is the overall size of the economy (GDP). This factor reflects the fact that states differ in the size of their economies by (several) degrees of magnitude, thus distinguishing between the principal economic powers, mid-sized economies, and smaller states. We find it more suitable to treat economic size as a control variable, rather than one of the key explanatory variables, since it can be immediately associated with various lines of reasoning. First, it is plausible to view it as a factor that captures state support for the existence and preservation of the trade regime. Countries with larger markets and bigger volumes of exports could be more interested in constructing trade institutions (Baccini et al., Reference Baccini, Dür and Elsig2015: 768; Krasner, Reference Krasner1976). This interest may also manifest itself through their better discipline in the submission of notifications. Second, the exact opposite reasoning can be put forward, whereby states with larger markets are, ceteris paribus, less concerned with open trade and access to other markets.Footnote ⁶ Finally, the size of the economies of states captures not only their economic interests but, equally importantly, also their power in the international system. Hence, large economies might also display a stronger discipline in their submission patterns, simply due to their possession of greater economic resources. Given its theoretical indeterminacy, we do not include economy size as a well-theorized predictor. Nevertheless, in our empirical analysis we of course control for it as a variable of major potential relevance.

The second factor we will control for is the overall level of the country's development and the level of GDP per capita, as its conventional proxy. Highly economically developed countries generally have better capacity to fulfill the transparency obligations connected with their participation in international institutions, as stated in hypothesis H4. In this vein, we use GDP per capita in one of the models as an alternative measure of state administrative capacity. However, the catch-all variable of GDP per capita captures much more than that, and its inclusion enables us to control for a broader range of phenomena connected with economic development. The third and fourth factors we control for are the length of a state's membership in the WTO and the number of obligations that apply to each of the states. For the former, an intuitive expectation could be that new members might have difficulties with the notifications system at the beginning, and that a certain learning period might be necessary for them to comply fully. On the other hand, latecomers may be required to compile relevant notification information in their accession process, and hence find it unproblematic to comply at least with their one-time notifications to start with.Footnote ⁷ In the face of this indeterminacy, we again resort to treating this variable as a control rather than a well-theorized predictor. For the latter, we simply want to control for the overall magnitude of the notifications submissions task.

4. Sectoral and Cross-National Variation in Compliance Rates, 1995–2014

In this section, we proceed to the presentation of descriptive empirical evidence with regard to states’ compliance levels. As we mentioned, there are in total around 170 obligations. Out of these, for 121 obligations it is possible to identify positively at least some submitted documents in the WTO database. Since we focus in this paper on a quantitative analysis of states’ compliance with their obligations, we only include in this analysis one-time and regular notifications. For these, establishing compliance is possible and fairly straightforward, as we can rely on the information provided by the Secretariat itself. There are in total 39 notification obligations for which compliance is actually systematically traced by the Secretariat of the WTO.Footnote ⁸ The authoritative information on compliance or failure to comply with these 39 obligations is made available via the regular publication of annexes to the G/L/223 documents series. These documents, published yearly, each contain between 40 to 80 pages of tables detailing, for each of the tracked obligations, whether or not the obligation was applicable to a concrete WTO member in the given year and whether the member fulfilled it. The tables also trace compliance over time, so it is possible to establish whether a certain obligation was fulfilled with a delay. Hence, our measure of compliance is a direct translation and aggregation of the data provided by the WTO itself.

The source documents of the G/L223 series, available in pdf and doc formats, were manually translated into table format by the authors of this article, in cooperation with a group of research assistants. The validity of the procedure was verified by double-checking our records from the Secretariat reports with the information available in the WTO Documents Online database. The resulting dataset totals 22,121 country-year obligations, each of which is marked as either unfulfilled, fulfilled with a delay, or fulfilled. From that information we calculate directly the compliance rate, which takes on the values 0 (unfulfilled), 0.5 (fulfilled with delay), and 1 (fulfilled as prescribed).Footnote ⁹ For regular notifications, the aggregate compliance value is calculated as the arithmetic mean over all the years in which notification applied. For one-time notifications, this value of 0/0.5/1 is simply the final compliance rate.

The strength of this approach is that we use for our calculation the authoritative information by the WTO secretariat, we do not need to rely on own estimates. There are caveats, however. First, we need to stress that this information does not portray the substantive quality of the content of the submitted documents. Even if a state submits a notification, its content may not be factually accurate or complete. The member may either fail to provide accurate and exhaustive information, or intentionally choose not to do so. After all, the ‘quality’ of the submitted notification information constitutes an important part of the discussions between the US and China in the high profile case of the steel industry and its subsidies, mentioned earlier.Footnote ¹⁰

As a second caveat, it should be noted that the G/L/223 series only covers notifications pertinent to trade in goods.Footnote ¹¹ Hence, to complement these data, we also collected information on the number of submissions pertinent to all the other regular and one-time obligations not covered in the G/L/223 series. There are further 24 such obligations. For these, we can assess the number of individual notification submissions, but not, strictly speaking, the states’ compliance levels (as we cannot judge with certainty if a specific notification applied to a specific state in a given year). For the sake of conceptual clarity, we focus in the explanatory part on the 39 obligations for which positive authoritative information on state compliance is available. However, we also run a separate analysis including these additional 24 obligations and, as we show later, the results are virtually identical.Footnote ¹²

Finally and importantly, as already indicated, our analysis of compliance omits the sizable group of ad-hoc notifications. The assessment of members’ compliance with these notifications is not compatible with our quantitative focus. For one-time and regular notifications, there is a clear specification available of when they should be delivered. This specification is provided by the very nature of these obligations and by the authoritative information published by the WTO Secretariat, as elaborated above. This paves the way for the analysis of these two types of obligations on a quantitative, large-scale basis. By contrast, fulfillment of the ad-hoc notifications requirements is conditioned by a deep, qualitative analysis of individual cases. To judge the submission of ad-hoc notifications, one would first need to identify, by looking in a detailed way into the changes in the policies, regulations, or legislation of the given state, the situations in which these notifications were actually due.Footnote ¹³ Such an approach cannot be apparently applied on a large scale. Hence, we concentrate on the quantitative analysis of one-time and regular notifications and leave the analysis of compliance with ad-hoc notifications obligations for further research.

Having said that, for the sake of completeness of the empirical overview of notifications submissions we also collected the data on the numbers of all submitted ad-hoc notifications. There are in total more than 43,000 such submissions, pertaining to 63 obligations.Footnote ¹⁴ The submissions were extracted from the WTO Documents Online database. While we cannot meaningfully include them in the analysis of compliance, we at least provide, in our additional models, a basic comparison of the ad-hoc notifications submissions with the identified patterns of compliance with regular and one-time obligations. By providing this overview, we hope to facilitate any further (most likely qualitative) research on ad-hoc notifications, which is certainly a worthwhile agenda. Yet, to be clear, these are not included in our measure of compliance.

Having defined our core concept and the dependent variable, let us proceed to basic descriptive results. The average compliance rate of WTO members with the regular and one-time notifications obligations, over the entire period and across all states, is equal to 0.52 (median 0.54, standard deviation (sd) 0.24). This value is calculated as an unweighted average across obligations, not obligation-years, hence it gives equal weight to one-time and to regular notifications. This procedure is also applied in the calculation of compliance rates used in the explanatory analysis. As an alternative, we can also calculate the average across obligation-years, and thus count multiple values for regular obligations. If we adopt this procedure, the value of the overall compliance rate would drop to 0.39 (median 0.33, sd 0.23). The difference is due to the fact that regular obligations have systematically lower compliance values (mean 0.37) than one-time obligations (mean 0.59), a matter we return to later on in our explanatory analysis. In spite of the significant difference in the overall values obtained with this procedure, the actual cross-state variation is very similar, and the two procedures produce highly correlated results (r = 0.91***).

The temporal perspective on the compliance rates is provided in Figure 2. Figure 2(a) shows that there is virtually no development in the overall compliance rate over a period of 20 years. While we do see some fluctuation in average compliance rates, no trend can be discerned. This is equally the case when we only consider regular obligations, as in Figure 2(b). Having said that, the de facto composition of compliance obligations has changed over time, as most one-time notifications had been due in late 1990s, right after the creation of the WTO. In the more recent periods, the non-periodic obligations covered in our dataset only concern the newly acceding members. For regular obligations, this is of course not the case, as they are supposed to be submitted continuously.

Figure 2. Overall compliance rates (a) and compliance rate across regular obligations only (b)

Table 1 depicts the sectoral variation in compliance rates. The first column captures the overall compliance across both regular and one-time obligations falling within a certain area, combined. Columns 2 and 3 then do so individually for regular and onetime notifications. The column “No. of data points” gives, for each area, the number of instances in which an obligation existed in a given year for a given state. This is the number of situations in which a notification should have been submitted, indicating the overall magnitude of the obligation. The sectors with the highest compliance in regular and one-time notifications are sanitary and phytosanitary measures (SPS) and safeguards, both with compliance rates above 0.85. In addition, a high compliance of above 0.6 is shown in trade-related investment measures, agriculture, rules of origin, and anti-dumping. In contrast, a low compliance of below 0.5 is obtained in customs valuation, import licensing, and state trading enterprises. The one regular obligation falling into the broad category of market access (the obligation is on the reporting of quantitative restrictions) shows a compliance rate of less than 10%. This extremely low value, but also low compliance scores in other sectors, has recently been highlighted by some members as a sign of problems in the transparency system in the WTO (cf. document (World Trade Organization, 2017e: 1). We should stress again, though, that these compliance scores do not cover ad-hoc notifications, which in some sectors (particularly SPS and TBT) are quantitatively dominant.

Table 1. Compliance rates and obligations (regular and onetime) across sectors

It is not our goal to explain this cross-sectoral variation in detail in this paper, as a very different theoretical framework and data set would be needed for that. However, we find that a part of this variation can be explained by considering the volume of the individual notification obligations. On the level of the 39 individual obligations, there is a clear negative relationship between the number of obligation instances and compliance (r = −0.47**), i.e. primarily yearly regular obligations have, on average, lower compliance.Footnote ¹⁵ A good deal of variation across the 13 sectors of obligations can then be explained with the types of obligations within each sector. Having said that, one example of a sector that clearly does not fit this pattern is agriculture, as it has at the same time by far the highest volume of notification obligation instances, and one of the highest compliance rates with one-time and regular notifications obligations.Footnote ¹⁶ Somewhat similarly, obligations in areas of anti-dumping and subsidies and countervailing measures also show relatively high compliance rates of 0.6 or more, for regular and one-time notifications, while at the same time also imposing on the states a relatively high burden.

Let us now turn to the variation in compliance across WTO members, as mapping and explaining this variation is at the core of our interest. In Table 2, we provide the compliance rates for all WTO members as a cross-section over the entire period 1995–2014. Once again, the measure is directly translated from the compliance evaluations provided by the Secretariat. What we see is an enormous level of variation across WTO membership, in fact with an almost uniform distribution of states in the range of between approximately 0.2 and 0.8.

Table 2. Overall compliance rates across all WTO members, 1995–2014

Notes: The column “Compl.” indicates the overall compliance rate, across all one-time and regular obligations. State name abbreviations are the standard ISO 3166-1 alpha-3 codes.

For the list of countries and their abbreviations, see: https://unstats.un.org/unsd/tradekb/knowledgebase/country-code.

The top of the rankings is occupied by the most export-oriented WTO members, namely Taiwan and Hong Kong, with Singapore close behind them. Very high compliance scores are also obtained by some of the largest members, most notably the EU, Canada, Japan, and later also Australia and the USA. From the major ‘non-Western’ trading powers, China is absent from the top of the rankings, while India ranks only 35th, with a compliance score of just over 0.7 (for a discussion on China and transparency in the WTO, see Wolfe, Reference Wolfe2017a). In contrast, Brazil and especially Russia score much higher. The top of the ranking also includes several smaller Latin American countries, while others, such as South Korea or the wealthy oil-exporting United Arab Emirates and Saudi Arabia, score surprisingly low.

5. Explanatory Findings

We now proceed to the empirical test of the hypotheses regarding states’ compliance. The core tests we present are based on a cross-sectional OLS regression analysis. This choice is justified by three facts. First and most importantly, only regular notifications have a meaningful time-dimension, as one-time notifications were mostly due in 1990s. Hence, there is no obvious way to aggregate regular and one-time notifications in a panel-data structure. Second, we do not observe any major changes in compliance rates over time that would indicate a need to systematically study the time-dimension of the phenomenon. Third, most of the variation in our predictors is across states, not over time. In spite of long-term changes in, for example, administrative capacity or in trade dependence in some countries, the systematic stable differences across countries are empirically much more prominent. Hence, a fixed effects model specification, for example, would be rather unsuitable for modeling the effects of our predictors of interest. Yet, as robustness assessments, and to bring in the time dimension at least for regular obligations, we also include in our results two pooled OLS and random effects panel regression models.

How do we operationalize our predictors of interest? The degree of trade dependence (H1) is operationalized as the share of exports and imports, combined, in GDP (variable Trade dependence). Trade openness (H2) is operationalized as the applied tariff rate (variable Trade openness, all data are from the World Bank World Bank, 2015). Of course, in the interpretation, the variable needs to be reversed, as high tariffs indicate low openness. The level of democracy (variable Democracy) is measured with the Polity IV democracy score (data from Polity IV Project, 2010) and, as a robustness test, with the transparency index of Hollyer et al. (Reference Hollyer, Rosendorff and Vreeland2011). Domestic administrative capacities are measured using the Worldwide Governance Indicators of the World Bank (2015), specifically the indicator measuring government effectiveness (variable Administrative capacity). The connection of states to the network of IOs is calculated as the number of IOs the state is a member of (variable IOs membership), based on data from the Correlates of War dataset (Pevehouse et al., Reference Pevehouse, Nordstrom and Warnke2004). Finally, data on GDP and GDP per capita (variables Log (GDP) and Log (GDPpc), respectively) are taken, again, from the World Bank (2015). For all the cross-sectional OLS models, we use country averages across the full 20 years period. The exception is IO membership, where we only use the data for year 2005.Footnote ¹⁷ For the estimation of the panel regression models, we of course use the full panel data.

So do our hypotheses find support in the empirical evidence? In Table 3, we present the results of six OLS regression models testing the hypotheses, with various specifications. All variables have been standardized to allow for a direct comparison of the magnitude of their effects on compliance. In all the models, we show (in brackets) robust standard errors and the models have been subjected to standard post-regression diagnostics. All the models show solid explanatory power, accounting for around two thirds of variation in states’ compliance.

Table 3. Regression analysis results

Notes: *p < 0.1; **p < 0.05; ***p < 0.01; †p < 0.001; robust standard errors in brackets.

Model 1 includes the five core variables capturing the five hypotheses and the control for economic size (Log (GDP)). The model supports all our five hypotheses, although variables Trade dependence, Administrative capacity, and Democracy only show a significant effect at the 10% level of confidence. That means compliance rates increase with a state's Trade dependence and Trade openness, with Administrative capacity, with Democracy, and with IOs membership. Regression residuals suggest that the effect of the variable Trade dependence may be largely due to two extreme observations, Hong Kong and Singapore, a matter we will return to shortly.

In Model 2, we add the two additional control variables: the length of country membership in the WTO and the overall number of its notification obligations. Neither of these variables shows a significant association with compliance. However, their introduction leads all our hypothesized predictors except Trade dependence to show a fully statistically significant association with compliance at the standard 5% confidence level (Trade dependence maintains significance at the 10% confidence level). These results remain stable across the further model specifications presented below.

Model 3 replicates Model 2 but here we proxy the overall capacities of a state with its level of economic development (Log (GDPpc)). With this variable, we seek to assess two sets of problems. First, as mentioned, we use it as a broader proxy for state capacities. In addition, though, it helps us control for other phenomena connected to the overall levels of states’ economic development, beyond the factors already included in our previous models. The variable shows the expected positive effect on compliance, but it fails to reach the standard significance level. This suggests that while administrative capacities, narrowly defined, are systematically significant for states’ compliance, a broad level of economic development is not, all other things held equal. The same result holds also when we replace GDP per capita with the World Bank classification of countries by income group; no association of the level of economic development and compliance is discernible. In addition to this substitution, we have also replaced our measure of democracy – the Democracy score based on the Polity IV data – with the Transparency index introduced earlier (results are not reported in the table). This variable also shows a significant association with compliance, like the variable Democracy, but it has a significantly lower coverage of countries, so we prefer to stick to the Democracy score.

Model 4 again replicates model 2, but with the exclusion of Hong Kong and Singapore, with extremely high values of Trade dependence. As expected, the variable Trade dependence shows no effect in this model, as its power has indeed been driven to a large extent by these extreme observations. This testifies to the fact that trade dependence may be relevant in some specific cases, but does not seem to play a systematic role.

Comparing the relative magnitudes of the substantive effects, we see that the strongest predictor of compliance (in Model 4) is the economic size. If the (logged) size of the economy increases by one standard deviation, there is an associated rise in compliance of 0.45 standard deviations. The variable IOs membership has approximately half of the explanatory power, with a coefficient size of 0.23; Administrative capacity is somewhat weaker, with a coefficient size of just below 0.2. Finally, the effect of a change in compliance for the variables Trade openness and Democracy is just above 0.1 standard deviation, for a change in the predictor of one standard deviation. That means, from our five hypothesized predictors, IOs involvement and Administrative capacity have the strongest effect, while Trade openness and Democracy have weaker, yet still significant association with compliance.

In Models 5 and 6, we show the results separately for compliance with regular obligations only (Model 5) and with one-time obligations only (Model 6). While the previous insights mostly hold, across the models we identify two differences. First, while for regular obligations Administrative capacity is a major predictive factor, for non-regular obligations no systematic association of compliance with this variable is discernible. On the other hand, for non-regular obligations we see an association with states’ IOs membership, a factor that is not linked to compliance with regular obligations.

To provide greater insight into the observed compliance patterns, we present, in Table 4, a series of robustness tests and additional analyses. They mostly confirm the main findings but also point at some limitations and avenues for further research.

Table 4. Additional multiple regression tests

Notes: *p < 0.1; **p < 0.05; ***p < 0.01; †p < 0.001; robust standard errors in brackets.

First, Model 7 simply replicates the analysis from Model 4 but with unstandardized coefficients, allowing for the assessment of the substantive significance of the effects. A lowering of the applied tariff rate by 10 percentage points (i.e. a rise in Trade openness) is associated with an increase in compliance of 0.06, i.e. 6 percentage points.Footnote ¹⁸ An increase in the IOs membership of 10 is associated with a compliance increase of 3 percentage points. Since the variable IOs membership has an average of 61 and a standard deviation of 17, this is highly relevant. A unit rise in the Administrative capacity score (ranging from around –2 to around 2) is associated with a 6-percentage-points increase in compliance, again a relatively large difference. A unit rise in the Democracy score (ranging from –10 to 10) is associated with a 0.5-percentage-point increase in compliance.

Second, models 8 and 9 are panel regression estimates replicating model 5. Within our framework, panel estimation is only meaningful for regular notifications, as one-time notifications do not have a substantial time dimension. While Model 8 is a pooled OLS, Model 9 is based on random effects estimation. Due to the dominant variation across countries, rather than over time, in most of our predictors and certainly the dependent variable, a fixed effects estimation would not be feasible (e.g. Bell and Jones, Reference Bell and Jones2015). Both panel regression models support the results presented earlier in the cross-sectional Model 5, showing a positive association between compliance with regular notification obligations and Administrative capacity and Democracy. The only difference is that the panel regression models do not show an association with Trade openness, which in the cross-sectional model was significant at a 10% level. As in Model 5, these are very intuitive results, where for the more demanding regular notifications the key predictors are Administrative capacity and Democracy.

Finally, Models 10 and 11 are substantively very different from all the other models in the analysis. Here our dependent variable is not compliance as such – the core interest in our paper. Instead, the dependent variable here is the aggregate number of submitted notifications by the respective state. In Model 10, it is the submission of all its one-time and regular notifications. In Model 11, we explore the submissions of ad-hoc notifications.

First, the purpose of Model 10 on one-time and regular notifications is to verify whether the results that we obtain when analyzing compliance as such are similar for the broader set of all one-time and regular notifications, including those that are not tracked by the Secretariat in the G/L/223 series (see our discussion in Section 4). That is, we also include the one-time and regular obligations for which we do not have authoritative information on compliance. In spite of this principal difference in the nature of the dependent variable in this model, the associations with the hypothesized explanatory factors are almost identical to the other models.Footnote ¹⁹ This suggests that also in the areas not covered by the G/L/223 series the patterns of regular and one-time notifications submissions are similar.

Second, for Model 11, the situation is different. The dependent variable in this model – the number of submissions of ad-hoc notifications – cannot be quantitatively associated with compliance, for reasons we discussed earlier. Hence, we only provide the model to offer a comparison with the previous results for regular and one-time notifications. The results indicate that indeed the patterns behind the submissions of ad-hoc notifications (not their compliance, as we cannot assess it) are significantly different. The only predictors that show an association with submissions are Trade openness and the size of the economy (Log (GDP)). Democracy, IOs membership, and Administrative capacity do not show a significant association with submissions. In addition to stressing the different nature of the dependent variable, we should highlight that the results may be also driven by the fact that the vast majority of the submitted ad-hoc notifications are in the SPS and TBT areas, and hence the patterns may be driven by the specificities of these sectors.

This is related to a more general point regarding the sectors. Our results, as presented in Tables 3 and 4, explain overall compliance patterns for the aggregates across all the sectors combined. The identified patterns, however, do not all hold in each individual sector and the relative explanatory power of the predictors varies across sectors. An explanatory framework that would seek to address compliance patterns within individual sectors would need to consider their specificities, the types of obligations in them, as well as the substantive demands of the individual notifications. These would then, most likely, also need to be modeled empirically with the use of additional sector-specific predictors.

Overall, the multiple regression results presented above support the validity of the theoretical reasoning presented in Section 3. There we hypothesized that compliance is likely to be connected with Trade dependence (H1), Trade openness (H2), Democracy (H3), Administrative capacity (H4), as well as IOs membership (H5). The subsequent empirical analysis supports all but the first hypotheses. It holds that Trade openness, Democracy, Administrative capacity, and Involvement in IOs are statistically significant factors in the majority of the four basic OLS models. The analysis of their substantive effects reveals that the level of compliance is relatively more strongly determined by Administrative capacity and IOs membership. In addition, we also identify a pattern whereby administrative capacities matter particularly for compliance with the more administratively demanding regular notifications, while adaptation to membership in IOs has a relatively greater effect on the compliance with one-time notifications. Finally, the fulfillment of notification obligations is connected with the simple economic size of the countries.