Patterns of regulatory heterogeneity in international trade: Intensity, coverage, and structure

With falling tariffs the role of regulatory heterogeneity in international trade has become central in recent debates about regional integration and trade costs. In describing the NTM incidence few studies explicitly take into account the specific nature of underlying regulatory differences. We propose distinguishing regulatory heterogeneity with respect to the intensity, coverage, and structure of regulations, and present indicators reflecting each one of these dimensions. Enabled by detailed product-level regulatory data based on coded reviews of national legislation, we illustrate the different channels of regulatory heterogeneity on the country-and sector-level. The findings motivate a separate treatment of the different heterogeneity dimensions in the assessment of non-tariff measures in international trade.


INTRODUCTION
The past fifty years have seen an unparalleled process of reducing traditional tariff barriers to international trade.With relatively low tariffs in place, the potential welfare gains associated with trade cost reductions have shifted the attention to so-called non-tariff measures (NTMs).Quite broadly, these are defined as policy measures " … that can potentially have an economic effect on international trade in goods, changing quantities traded, or prices or both" (UNC-TAD, 2017c, p. 3).This broad definition includes at-the-border trade policy instruments, as well as behind-the-border policies traditionally not thought of as trade-related measures.Analysis of such an enlarged "trade" policy space requires systematically collected NTM data with wide geographic scope, and a set of indicators highlighting different aspects of countries' regulatory profiles.
The main objective of this paper is to provide a descriptive account of international patterns of NTMs by using a diverse set of indicators.We focus on (standard-like) technical measures complemented by two types of non-technical measures.The majority of these measures is imposed by the importer in a non-discriminatory fashion across origin countries, that is, like most-favored nation (MFN) tariffs these measures are applied equally to all exporters.Technical measures include sanitary and phytosanitary measures (SPS), technical barriers to trade (TBT), and pre-shipment inspections, while the two non-technical measure groups comprise quantity-and price-based measures. 1e differentiate the NTM incidence along three dimensions: (1) intensity, (2) coverage, and (3) structure.First, regulatory intensity describes the stringency of regulation, which can be proxied by the number of measures imposed on a product, or specified by actual requirements related to the product itself (e.g., a maximum residue limit of a pesticide on agricultural or food products) or production process (e.g., sanitation requirements for a factory) implied by the underlying policy.Second, coverage relates to the scope of "what is affected" by a measure or measure group.Typically, this concerns the value of trade, number of trading partners, or number of products.Third, structural regulatory heterogeneity describes differences with respect to what type of measures are imposed on a given product and to what degree these may depend on each other.This requires relatively detailed information on the NTM incidence, which is not necessarily the case for indicators reflecting intensity and coverage.Combining all three heterogeneity dimensions results in a relatively comprehensive display of a country's regulatory footprint.
The contribution of this study is twofold.First, we extend the set of NTM indicators currently used in the literature in accordance with the three heterogeneity dimensions of cross-country regulatory differences (for overviews see Disdier & Fugazza, 2020; Gourdon, 2014; UNCTAD, 2017c).Particularly, we complement the set of indicators related to regulatory structure and provide a principal component analysis (PCA) based variance decomposition of cross-country differences in regulatory stringency.The developed database contains a comprehensive set of indicators addressing the three heterogeneity dimensions for total trade and the following sectoral aggregations: 2-digit Harmonized System (HS), the Global Trade Analysis Project (GTAP) aggregates, Broad Economic Categories (BEC) Rev. 4, a 15 sectors aggregation of the HS provided by the World Customs Organization (WCO), and the ISIC Rev. 3 based classification of the International Trade and Production Database for Estimation (ITPDE, Borchert et al., 2020). 2 Indicators are differentiated by broad measure groups and more detailed aggregates, and with respect to whether they are imposed in an MFN or bilateral fashion.Moreover, most indicators are calculated for the years 2000-2016 or 2012-2016, and cover 155 or 119 reporting countries, depending on whether the underlying source data is retrieved from the WTO notifications or NTMTRAINS, respectively.
14679396, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Second, with the set of indicators at hand, we analyze international patterns of NTMs and derive stylized facts.While the majority of the analysis is carried out on the basis of NTMTRAINS data, we contrast results with WTO notification-based data where applicable.We first conduct the analysis on the country-level and subsequently highlight differences in the NTM incidence across sectors.
The constructed dataset can be used in multiple ways.For example, the different types of indicators provide the basis for a comprehensive descriptive analysis as presented in Section 4. Furthermore, gravity equations can be augmented by one or more of the NTM indicators representing different interpretations of the source of NTM-related trade costs.In addition, indicators for structural regulatory differences can function as determinants for preferential trade agreements (PTA) or specific (sets of) PTA provisions related to the NTMs, or can highlight how NTMs shape global value chains and vice versa.
We proceed as follows: In Section 2 we shortly describe the properties of NTM data, as well as the data used for this paper.Section 3 presents the sets of indicators for each of the three heterogeneity dimensions, while Section 4 illustrates broad patterns of NTMs by country-and sector-level.Section 5 summarizes and concludes the analysis.

DATA
Consistent with the wide definition of NTMs, information about or related to them can be found in multiple places.These include inventories of legislation, notification portals, business surveys, import refusal data, reviews of legislation, or international agreements.The interpretation of the given information differs by type of source.For example, while legislative inventories and notification portals describe the de jure state, complaint registers, data on import refusals, or business surveys are likely to provide more information about enforcement and trade restrictiveness implications of measures.
Given the array of possible sources, the actual properties of NTM information takes several forms: • Binary variables indicating the existence of a measure; • Numerical indicators capturing the main property of a measure (e.g., maximum residue limits, percentage of foreign equity ownership, etc.); • Text of the actual regulation (or description thereof); • Categorical variables that classify measures into predefined categories (e.g., whether a measure is discriminatory or not); • Ordinal variables implying a ranking along a chosen dimension, for example, level of trade restrictiveness, or status of implementation; • Computed indicators processing original information, for example, count or frequency ratios.
For an overview of NTM data and further information on its concepts see Rau and Vogt (2019).Which data is suitable for a given study depends on the underlying research question, as well as the geographic, sectoral/product, and temporal scope of the analysis.Studies using very specific regulatory data are usually constrained to a sector or set of products because collecting such data is resource intensive (e.g., Otsuki et al., 2001; Winchester et al., 2012).
This study analyzes the global NTM incidence globally across multiple sectors, which constrains us to the use of two databases.First, we use WTO notification data obtained from Ghodsi  et al. (2017), who augment the original notifications retrieved from the WTO I-TIP portal by adding missing HS codes based on text-matching techniques.These data are available from 1995 onward, that is, since the notification mechanism has been in place, although particularly developing WTO members require more time to establish the institutional capacity to notify regulatory changes.With respect to time information we prefer the entry-into-force over the notification date.In addition, using WTO document identifiers we cross-check the data of Ghodsi et al. (2017), who retrieve notified NTMs from the I-TIP portal, with notification information obtained from the SPS and TBT Information Management System (IMS).In some cases this leads to adjustments with respect to the partners affected by a measure.
Second, as our main data source we use UNCTAD's NTMTRAINS, which contains information on NTMs based on full regulatory reviews.The base dataset (Stata researcher file v.12 retrieved from trains.unctad.org)includes measures collected between 2012 and 2018. 3or this paper, we consolidate the data to 2016 by taking cross-sections collected for 2016 or the latest year available prior to 2016.If data are only available for 2017 and/or 2018 we retrieve the earlier year and remove those measure types introduced after 2016. 4 Both databases use the above-mentioned MAST NTM classification to categorize regulatory information (UNC-TAD, 2019).Table 1 presents a consolidated version of the classification for import related measures covered by this study with more details and a listing of export-related measures provided in Appendix B. Chapters A to C are generally referred to as technical measures, while all other MAST chapters classify non-technical measures.With the exception of internal non-discriminatory charges the latter are exclusively imposed on imports, or in other words they are imposed "at-the-border".By contrast, this is rarely the case for technical measures, which very likely apply to foreign and domestic firms in a similar fashion. 5Thus, SPS and TBT measures are mostly "behind-the-border" measures, usually designed to address non-trade-related policy objectives such as the protection of human or animal life, or technical regulations that specify product characteristics or requirements related to production processes.
NTMTRAINS categorizes measures at a very detailed level, while WTO notifications are only available at the level of notification requirement corresponding to the MAST chapter level.Consequently, comparing the information contained in the two databases is only possible by aggregate measure groups. 6In absence of a common, unique identifier (e.g., an ID of the national legislation from the official gazettes) merging the two databases in order to for example, increase overall country coverage, is not possible unless the researcher is willing to make numerous assumptions.For example, one needs to assume that WTO notified entry-into-force dates as well as products affected by the measure match those recorded in the legislative reviews.However, oftentimes the entry-into-force date is not available and only the notification date is provided.Another problem that particularly pertains to analyses based on regulatory intensity/stringency is that, even if one successfully merges notifications and NTMTRAINS data on the basis of product codes, dates, and whether a measure is SPS or TBT, a notification may contain multiple measures that would be recorded separately in NTMTRAINS.For example, NTMTRAINS codes differences between labeling and packaging requirements for SPS and TBT measures.A corresponding regulation for a given product that contains both requirements may be notified together, but is coded separately in NTMTRAINS, leading to a count of 2 for NTMTRAINS and 1 for the WTO notifications.This means that in a consolidated database sector-level indicators for regulatory intensity are not comparable across observations.T A B L E 1 MAST classification for import-related measures.

MAST chapter MAST codes Description
A-Sanitary and phytosanitary measures (

NTM INDICATORS
This section reviews and extends the set of descriptive indicators based on binary data found in the literature, which we will use to illustrate international, cross-sectoral patterns of NTMs in Section 4. 7 This contrasts studies with for example, a narrow geographical and/or product scope, which are more likely to incorporate detailed regulatory information.In those cases the underlying policy data used to construct NTM indicators are a relatively accurate reflection of the sector-/product-specific regulatory substance.Given adequate detail in the measures' definition even a dummy variable signalling the presence of a measure is in most cases sufficiently informative. 8However, data on specific policy instruments becomes less comparable and dummy variables become less meaningful the further we aggregate products into sectors.In that case, indicators presented in this section gain relevance and present a more feasible account of cross-country and cross-sector variation than a dummy.We adopt a notation commonly used for gravity models of trade, where o is the origin country (i.e., exporter) and d the destination country (i.e., importer).Consequently, for all import-related measures, destination country d is the reporting/imposing country, while the origin country o is the reporter for measures on exports.Each number of measures M is of type A and levied on a product i defined at the 6-digit level of the HS.When aggregating to product groups or sectors we use index k.Furthermore, each measure enters into force at a year t and is assumed to continue being in force unless a date of withdrawal is provided. 9Lastly, measures M of type A can be aggregated to measure groups (e.g., MAST chapters), which are indexed by g.That is, A g signals measure A being part of group g with G number of different measures A g .For example, a single MFN-type TBT testing requirement imposed by the USA on the product with 6-digit HS code 081020 translates into the following: USA is destination country d, the world is origin o, 081020 is product i, which is part of a higher aggregate k (e.g., vegetable products), M is 1, A g is a TBT testing requirement with MAST code B82, and g is an aggregate measures group (e.g., conformity assessment, TBT, or technical measures).

Intensity
Indicators of regulatory intensity reflect the stringency with which policy makers regulate products.Similar to previous studies we assume that the number of measures, of the same or different type, constitutes a suitable proxy for stringency (Cadot & Malouche, 2012; Gourdon, 2014;  UNCTAD, 2017a).This assumes that a combination of measures increases the likelihood that corresponding policy objectives (e.g., consumer health and safety) are achieved-stringency regarding policy objectives-that each additional measure increases regulatory compliance costs-stringency regarding costs 10 -and for the subset of quality-related technical measures, a higher number of measures reflects increasing constraints on endogenous quality choices of firms-stringency regarding product quality. 11he NTM count C dkg is the total number of measure-product combinations imposed by destination country d, for products i in sector k, and measures M di in group g. (1) A measure can affect multiple products and a product can be affected by multiple measures.Thus, C dkg is interpreted as the total NTM incidence.However, C dkg is an increasing function of the number of products i in sector k and consequently can be misleading when comparing NTM footprints across different sectors.This problem is addressed by the prevalence score PS dkg , which is the average number of measure per product in a given aggregate k.It is calculated by dividing the NTM count by the total number of 6-digit products i in a given sectoral aggregation k.
Both indicators can be bilateralized by adding subscript o, which would further differentiate between MFN-type and bilaterally imposed measures.For example, C odkg would then be the number of measure-product combinations imposed by country d on imports from o in sector k. 12

Coverage
In contrast to indicators reflecting regulatory intensity, indicators capturing the coverage, or scope, of NTMs are (a) the share of products covered by at least one measure (frequency index), and (b) the share of trade covered by at least one NTM (coverage ratio).Both coverage indicators are invariant to regulatory intensity.The frequency index FI dkg is defined as the number of products affected by at least one measure of group g, divided by the total number of products in aggregate k-that is, the share of products i in aggregate k.
This implies that the wider the measure group g is defined, the more likely a product is affected by at least one measure (i.e., 1(M dig > 0) equals 1).This means that FI dkg increases with wider definitions of g.In addition to the frequency index, the coverage ratio CR dkg defines the volume of trade affected by at least one NTM divided by the total volume of trade in sector k.
Similar to the frequency index, CR dkg increases with wider definition of measure groups.Furthermore, as it is usually the case with trade-weighted indexes the coverage ratio is highly sensitive to measures that are trade restrictive or even prohibitive like an import ban that would render the nominator to zero. 13

Structure
Indicators representing regulatory structure require relatively detailed information on NTMs because variation in the indicator is caused by differences in types of measures rather than number or coverage thereof.A basic indicator of regulatory structure is the unique number of measures U dkg defined by the average number of unique measures of a certain type per product i in sector k for a given measure group g.
Dividing U dkg by the corresponding prevalence score results in the share of unique measures vis-a-vis all measures.Thus, a value of one means that on average all imposed measures are different, while lower values translate to a regulatory profile characterized by many measures of the same kind.Lesot et al., 2009).From this we derive the number of measures types: Similar to standard gravity distance variables simple matching and Jaccard distance measures are symmetric.However, firms with relatively high compliance capacity operating in a complex regulatory environment may find it easier to export to a country with a lower regulatory footprint.To capture this asymmetry, we define the distance measures above as a decreasing function of b, that is, measures only imposed by the exporter o.This is similar to what UNCTAD (2017b) defines as regulatory overlap.Such overlap measures can be based on simple matching and Jaccard distance measures. 15Furthermore, we define the asymmetric a-c difference to relate the number of measures imposed by o and d to measures only imposed by the destination country.The indicator decreases in the number of measures imposed by country d that are additional to measures imposed on the home market of the exporter.
Distance measures can further incorporate regulatory stringency and be weighed by the average number of measures between the origin and destination country, as well as total number of measures at the destination country.The combined indices increase in the number of NTMs but at a lower rate for country pairs with a similar regulatory structure.

Interdependence
A driver of regulatory similarity is by design the co-occurrence of specific measures across different countries.In order to identify how meaningful co-occurrences of two measures are, we employ indicators used in association analysis (see for example Hastie et al., 2017).By this we aim to identify patterns of co-occurrences that point towards particular regulatory designs (e.g., are measures restricting the use of certain substances accompanied by testing requirements).
As a basis we determine for each product i and different measures A 1 and A 2 the number of countries that impose both measures, given that at least 2 countries impose either A 1 or A 2 .The share of these countries among countries that impose at least 1 measure is referred to as support, or P(A 1 ∩ A 2 ). 16The degree to which one measure is implied by the other is the confidence defined by P(A 1 ∩ A 2 )∕P(A 1 ).The confidence indicator adjusts the probability with which A 1 and A 2 jointly occur by the probability of A 1 .Consequently, confidence is an estimate of P(A 2 |A 1 ).It decreases in high occurrence of A 1 and takes into account that co-occurrence may simply be a function of A 1 's high incidence.Thus, a statement such as that A 1 implies A 2 is further qualified.To what degree A 1 and A 2 are associated is referred to as lift (P(A 2 |A 1 )∕P(A 2 )), which adjusts the conditional probability of A 2 on A 1 by the probability of measure A 2 being present.Any value of the lift higher than 1, implying that P(A 2 |A 1 ) > P(A 2 ), signals a relatively high association of the measures.For example, if 20% of countries impose measures A 1 and A 2 at the same time the support is 0.2.If A 1 is imposed by 30% of the countries, the corresponding confidence index will be 0.67, which means that in 67% of the cases when A 1 was present A 2 was imposed, as well.Adjusting for the unconditional probability of A 2 (e.g., 0.25) the lift index is then 2.7.
We pool products in a given sector k to derive association measures for sectoral or total aggregates.In order to adjust the association indicators to their relevance within a sector we multiply them with the share of products they apply to, that is, we additionally provide a version of the support, confidence, and lift that takes into account the number of unregulated products.In case of the support, the derived index (Sup kA 1 A 2 ) is comparable to the sector-level frequency index presented above, averaged over all countries that have at least on measure in place (FI kg ).While Sup kA 1 A 2 is measure-specific (e.g., support of B81 and B33), the share of products affected by any measure is defined over measure group g, in this case defined on the MAST chapter level.As a result, the relevance of the support vis-a-vis the frequency index FI kg is determined by comparing the average share of products to which the rule applies to the share of products affected by at least one measure within the MAST chapter.This is captured by Sup kA 1 A 2 ∕FI kg .By construction this ratio is defined for the interval [0,1] with 1 meaning that A 1 and A 2 are always imposed when any measure (incl.A 1 and A 2 ) of the MAST chapter in question is present.Analogously, a value of 0.5 implies that A 1 and A 2 are imposed in 50% of the cases when at least on NTM of group g is imposed.
In summary, we differentiate between three sets of NTM indicators corresponding to different regulatory heterogeneity dimensions: intensity, coverage, and structure.Measures of regulatory intensity reflect the degree to which multiple measures, including many of the same type, are imposed on a product, while coverage indicators highlight the pervasiveness of NTMs across different products and trade values.In addition, indicators describing regulatory structure focus on regulatory heterogeneity with respect to the types of measures imposed and to what degree they may be complementary.We present these indicators acknowledging that the underlying binary data provides little information about the actual policy substance.This is a general constraint for NTM analyses with a broad sectoral and geographic scope.Due to the sparseness of NTM data the usefulness of the indicators increases with higher aggregation of the data, while on the product-level, dummy variable research designs are likely to be preferable.

Aggregation
Aggregation in an NTM-context relates to the weight assigned to product i in sector k, or measure M A g in an index for measure group g, or both.Indicators presented above assign equal weights to products and measures with the exception of the coverage ratio, which introduces trade weights to the frequency index.In this section we present how trade weights are used to aggregate products into sectors, as well as how a principal component analysis (PCA) can be used to define variance-based weights to aggregate sub-indexes of specific measure groups. 17

Trade-weighted aggregation
Count and prevalence indexes presented above weigh products equally when aggregating to k.Such an aggregation is likely to give too much weight to products that may not be relevant as imports for destination country d.In order to address this problem we follow an approach from the tariff literature and weigh NTM indicators by trade.The approach differentiates between measures applied on an MFN-basis and those imposed bilaterally.Thus, the total NTM incidence between two countries in a given indicator Z and sector k is captured by: Here, w di and w odi refer to the share of product i in sector k's imports of country d, from all countries or specific origin country o, respectively.While w di is used with NTMs that are applied in an MFN-fashion, w odi is used with bilaterally applied measures.This avoids "bilateralizing" MFN measures via trade weights. 18imilar to atheoretical tariff aggregation, trade-weighted aggregations of NTM indicators suffer problems of endogeneity when measures are very trade restrictive or promoting (Anderson  & Peter Neary, 2005).To alleviate this problem weights can be constructed on the basis of world trade.However, this leads to the loss of country-specific information with respect to the structure of trade.Alternatively, Bouët et al. (2008) create reference group-based weights to aggregate tariffs, that is, weights based on average trade of a reference group of countries.The idea is to determine what a country typically should import given the trade profile of a group of similar countries, for example, determined by their GPD per capita.Thus, assuming that not all countries of the reference group impose trade restrictive/prohibitive measures on product i, C dig and C odig still receive positive weights even if one of the reference group's countries imposes prohibitive measures.

3.4.2
Variance-based measure aggregation In the absence of expert opinion based weighting schemes, contribution-to-variance-based weighing offers an alternative to aggregate single NTMs to higher level measure groups (see Nicoletti  et al., 2000). 19The calculated weights contain valuable information about where cross-country regulatory differences are most prevalent, that is, they help to identify key measures in particular sectors across countries.
To obtain weights we perform a principal component analysis (PCA) on the covariance matrix of NTM sub-indexes and retrieve the contribution of each component to the overall variance in the data, as well as the contribution of each sub-index to the variance of each component. 20More specifically, the PCA is based on an eigenvalue decomposition of the square covariance matrix Σ, that is, ΣV = ΛV with Λ the diagonal matrix of eigenvalues and V the matrix of corresponding eigenvectors.The eigenvalues  a captures the contribution of each component to the overall variance (i.e., C w a =  a ∕ ∑ i  i ), while the Hadamard product matrix W = ((ΛV) 2 •1∕( T ) 2 ) * 100 gives us the contribution of each variable to the respective variation in the components (see e.g., Husson et al., 2011.Here, w C a lists the contribution of each variable to component a, and a row vector w V b lists the contribution of a variable b to each of the components.The weights for composite NTM indicators are then calculated by: The covariance matrix Σ is calculated on the basis of centered prevalence scores-that is, if A is the data matrix with centered prevalence scores then Σ = A T A. Usually, prior to performing a PCA, vectors of data matrix A (i.e., variables) are standardized.This procedure is applicable when variables are measured in different units.By using prevalence scores we already work with variables measured by the same units (average number of measures) and thus do not need to standardize.Furthermore, the advantage of prevalence scores over simple counts is that we adjust for the number of products in a given aggregation.Thus, we avoid that high counts are a function of the sectoral aggregation. 21 possible downside of this approach is that the calculation of the weighting scheme is sample dependent.Thus, adding or removing a country from the sample changes aggregation weights, which contrasts for example, expert opinion based approaches with constant weights per measure.

PATTERNS OF NTMS
In the following we make use of the indicators described above to highlight patterns of NTMs in international trade and summarize the main findings in stylized facts.We focus on overall country-and sector-level patterns using NTMTRAINS and WTO notification data consolidated by Ghodsi et al. (2017).In Section 4.1, we provide aggregate and country-level comparisons of WTO notifications based on the mapping described in Section 2. Due to notification requirements under the SPS and TBT Agreements the analysis focuses on SPS and TBTs when WTO notification data is presented and is expanded to other import-related measures otherwise (see Table 1).Section 4.2 illustrates patterns of NTMs across sectors.

Country level
Stylized Fact I.The majority of technical measures are formally applied in a non-discriminatory fashion across trading partners.
Overall, in the period from 2000 to 2016 countries consistently notified new or changes to existing SPS and TBT measures to the WTO, with the overwhelming majority of measures being imposed in a non-discriminatory fashion across all trading partners.This is captured by Figure 1, which shows the stock of WTO-notified SPS and TBT measures-expressed as the average number of measures per product-carried over time in total (left) and differentiated by whether measures applied bilaterally (i.e., with partner countries specifically targeted) and on a MFN-basis A comparison between the two global NTM datasets on the country-level reveals that the NTM incidence of SPS and TBT measures notified to the WTO and identified in the legislative reviews positively correlates, despite noticeable differences between the two databases for some countries.This is shown by Figure 2, which maps countries' SPS and TBT counts contained in NTMTRAINS and WTO notifications against each other.We observe the following: First, for each measure group there are a number of countries, which do not notify SPS and TBT measures at all.Among them are relatively large countries like Ethiopia, Algeria, Cote d'Ivoire, or Belarus.However, NTM-TRAINS regulatory reviews indicate that many of these countries should have a relatively high NTM incidence (e.g., Cambodia has a similar count index as Canada).Second, SPS counts are generally higher for review-based data, which suggests that either countries under-notify, or that the more detailed coding of NTMTRAINS data results in a higher count per se.Third, TBT counts are similar for some countries in both databases (e.g., Switzerland, Australia, or China), while others are relatively far away from the 45 degree line (e.g., Israel, Morocco, Pakistan).This significantly changes the ranking of countries in terms of their implied regulatory stringency.For example, Israel is one of the more stringent countries based on WTO notification data but in the lower third of countries based on the legislative reviews.Fourth, in terms of income level, clear patterns emerge in terms of (a) higher income countries notify more actively to the WTO, (b) low income countries have a low regulatory footprint, and for TBTs tend to severely under-notify, and (c) a clear contrast of the regulatory footprint is difficult to establish between high, upper-middle, and lower-middle income countries.Overall, the apparent differences should be kept in mind when comparing sector-or aggregate-level analyses of SPS and TBT measures using different data-sources.
The independent legislative reviews in NTMTRAINS are a useful source to assess WTO members' notification behavior, particularly for those countries not notifying at all.20 years after the notification mechanism was put in place some countries seem to have addressed early concerns regarding notification compliance and quality of WTO members' notifications (see e.g., Cadot & Malouche, 2012), while others are still not participating in, or struggling with for example, the institutional capacity requirements of this transparency mechanism.Especially low and lower middle income countries tend to under-notify, which would support the lacking-institution hypothesis.
Stylized Fact II.Regulatory stringency positively correlates with income levelshigh and middle income countries impose more measures per product than low income countries.However, broad measure groups comprise similar shares in the total NTM incidence across income groups.
On average regulatory stringency increases with income, while the composition of different groups of NTMs is relatively similar across income groups (see Figure 3).High and middle income countries impose approximately twice as many measures per product compared to low income countries-3.4to 4.1 for lower middle to high income versus 1.7 for low income countries.The majority of measures across all income groups are SPS and TBT measures comprising ca.75% of all measures per product.Quantity-and price-based measures (14% to 18%) as well as pre-shipment inspections (3% to 13%) constitute only a small share of total measures, but are the types of policies exclusively targeted at international trade.Trade-related charges, licensing requirements, and prohibitions for economic reasons are the most prominent policies among non-technical measures.In contrast to licensing requirements for economic reasons, some non-technical measures  are trade-restrictively designed with a legitimate policy objective in mind.For example, many import prohibitions are imposed for non-economic reasons (e.g., an import ban of alcohol or print media with pornographic content for religious or moral reasons) and are sometimes even tied to international agreements.This concerns for example international conventions on wild life, arms or drug trade, dual use goods, or chemicals that can act as precursors.
Stylized Fact III.Structural regulatory differences follow regional patterns with countries from the same region showing a more similar regulatory structure.
Similarity in regulatory stringency does not necessarily translate into an equal regulatory structure in terms of the types of measures imposed.Figure 4 plots the Jaccard distance for technical measures, as well as the average number of uniquely imposed measures mapped against the average number of all measures differentiated by region.Countries that are geographically close or in the same region tend to impose similar types of technical measures.For example, we can identify a Latin American bloc (e.g., Uruguay, Brazil, Chile, Nicaragua, Argentina, or Jamaica) in the lower part of the figure, East Asian & Pacific countries that are located close to each other, as well as a cluster of countries that share a Soviet past (Russia, Belarus, or Kazakhstan) and a cluster of countries that impose few regulations (Sub-Sahara African countries).Additionally, the right-hand side of Figure 4 highlights that countries' regulatory structure differs in terms of whether they impose the same types of measures multiple times-countries further away from the 45 degree line-or impose a unique set of measure types-countries closer to the 45 degree line.By combining the two sides of Figure 4 we can for example, infer that Korea and Brazil are relatively similar in terms of uniqueness and prevalence, but exhibit a comparably high regulatory distance.Thus, they impose different types of measures.This example emphasizes that the countries' NTM profile is complex and multifaceted, which requires a nuanced set of indicators to properly account for regulatory differences.
Stylized Fact IV.Among technical measures, there are distinguishable joint occurrences of specific measures suggesting regulatory complementarity.Specific pairs of technical measure types occur jointly relatively more often than others, which indicates measures interdependence and resembles a form of regulatory system.Figure 5 transforms the association measures developed in Section 3 into heatmaps and identifies frequent measure associations via the confidence index (P(A 2 |A 1 ), the lift (P(A 2 |A 1 )∕P(A 2 )), and a weighted version of the confidence index.First, we observe multiple relationships of measure pairs where the presence of A 1 (y-axis) implies the presence of A 2 (x-axis) in 60% to 80% of the cases.SPS process control measures often imply the presence of SPS substance tolerance and use limits (A2), hygiene (A4), as well as SPS certification requirements (A83).Similarly, SPS certification and inspection requirements (A83, A84) come with post-production treatment obligations (A5), and TBT substance tolerance and use limits imply with a high likelihood TBT labeling and marking, product performance, and testing requirements.Second, we find a generally high association between two measures (P(A 2 |A 1 )∕P(A 2 )) for SPS testing and packaging requirements (A82 and A33), SPS process control and TBT product identity (A6 and B6), as well as SPS and TBT registration and approval requirements (A81 and B81).Third, comparing the confidence index with its weighted counterpart shows that the distinctive pattern of joint occurrences of SPS measures is not visible anymore for the weighted confidence index.This coincides with the strong proliferation of SPS measures for agricultural products, which represent a small share in the number of total products.By contrast, patterns of joint occurrences of TBT measures still hold for the weighted confidence index, which is consistent with the widespread use of TBTs across all sectors.Thus, the identified associations between different measure types and implied regulatory systems are likely to be more pronounced on the sectoral level and across different country groups, as suggested by the geographic clusters of regulatory distances in Figure 4.

Sectoral level
This section highlights sectoral heterogeneity of NTM patterns.It includes a PCA-based variance decomposition of regulatory stringency to illustrate which specific measure groups contribute most to cross-country differences in NTMs.Overall, the results suggest that there is significant sectoral heterogeneity across all indicators embedded in the total averages presented in the last section.
Stylized Fact V. Agri-food sectors are across almost all measure groups consistently the most regulated sectors in terms of regulatory intensity and coverage.
SPS and TBTs are the most prevalent NTMs, with agri-food sectors the most regulated in terms of regulatory stringency and coverage.Table 3 depicts the average regulatory intensity and coverage by sector for five import-related measure groups, as well as export-related measures.By measure group we can identify the following main patterns: First, SPS measures cover 16% of all products and 24% of total trade value.The high total incidence is driven by a high prevalence for animal, vegetable and food products with approximately 7 to 15 measures per product.By contrast, SPS measures play a limited role in manufacturing industries and extractive sectors.Second, TBTs cover significantly more products (30%) and trade (44%) than SPS measures.They cover products more evenly across sectors, while the number of measures is highest for agri-food products, chemicals, textiles and clothing, transport, and machinery and electronics.
Third, pre-shipment inspections are the least used technical measure covering only 10% of all products and ca.14% of trade, with the highest number of measures imposed in agri-food sectors, textiles and clothing, and hides and skins.Fourth, among non-technical import measures quantity-control policies are the most prevalent and cover more products and trade than the more concentrated SPS measures.Besides the highly regulated agri-food sectors, quantity controls are imposed on ca.30% of all chemical, hides and skin, machinery and electronic, and transport products, affecting approximately 32% to 42% of imports in these sectors.In addition, price-based measures are the least used import measure besides the aforementioned pre-shipment inspections.Considering that Guinea makes up ca.25% of all counts, the relative importance of these types of measures is even less than Table 3 suggests. 22ifth, measures on exports are mainly composed of technical measures (e.g., authorization requirements or conformity assessments) with quantity control measures for non-technical reasons, as well as measures on re-exports playing a smaller role.Thus it is not surprising to see a high incidence for those sectors that are also heavily affected by technical import measures, for example, agricultural and food products, chemicals, and the transport sector.Overall, Table 3 shows that SPS, TBT, and quantity-control measures are the most relevant import measures across all sectors, with agri-food sectors clearly displaying the highest NTM incidence in terms of intensity and coverage.

Stylized Fact VI.
For each sector, the majority of cross-country variation in regulatory stringency is captured by a small subset of measures.
The specific drivers of cross-country differences in regulatory intensity widely differ across sectors, with SPS authorization and registration, tolerance and use restrictions, certification, and inspection requirements, and TBT labels and marking obligations dominant in agri-food sectors, and with TBT product performance, labels and marking, and certification requirements prominent in manufacturing sectors.This is summarized by Table 4, which shows the percentage contribution of each measure subgroup to the variance in regulatory stringency of all import measures (see Appendix B for a mapping of subgroups). 23This means for example, that 9.4% of cross-country variation of animal products' regulatory stringency is caused by SPS measures that define tolerance limits and/or restrict the use of certain materials, and thus contribute most to regulatory differences in this sector.The weights correlate with the underlying intensity indicators.However, if all countries had the same underlying prevalence score, irrespective of the level, the weight in Table 4 would be zero.
Overall, technical measures cause most of cross-country variation of import measures (ca.66% in total), which is in line with the descriptive indicators presented in Table 3 and the country-level analysis of the previous section.However, sectoral differences are significant.Particularly, for agricultural and food products differences in the intensity of import measures across countries are primarily caused by technical measures.By contrast, non-technical measures are relatively more relevant for stones and metals and chemical and plastics products, albeit that technical measures are still responsible for the majority of cross-country variation.Furthermore, in terms of technical versus non-technical measures there are little differences between intermediate and consumption products.This also holds for most measure subgroups, with the exception of more heterogeneity in TBT labeling and marking, and product performance measures for consumption goods.
Importantly, a handful of measures explains more than half of the variation in regulatory intensity across manufacturing sectors, while the set of measures imposed on agri-food sectors is more diverse.For total trade, TBT labeling and marking requirements (16%), charges related to trade (24%), differences in licensing (4.8%), TBT certification (4.5%) and product performance (4%) requirements account for ca.half of total variation.However, on the sectoral level, for agri-food sectors, we observe relatively more variation in SPS authorization and registration .4 1 2 .1 13.2 1 6 .  .requirements, tolerance and use restrictions, labeling and marking, and inspection requirements.
Whereas for machinery and electronics, and transport products sectors, TBT product performance and certification requirements, and licensing measures are the most relevant ones.Moreover, charges related to trade contribute to cross-country differences in regulatory stringency across all manufacturing sectors.The results presented in Table 4 illustrate that the relevance of different measures types varies relatively strongly across sectors.Thus, any sector-level trade cost estimates for more aggregate measures groups are likely to be driven by different measures depending on the sector at hand.In terms of structural regulatory heterogeneity, the patterns of relatively higher intra-regional regulatory similarity identified by Figure 4 hold across all sectors.This is highlighted by Table 5, which shows the degree to which countries differ in terms of the structure of sectoral regulations.For this, we use the Jaccard distance for all technical measures (SPS, TBT, and pre-shipment inspections), which means that regulatory distance decreases only with joint presences of measures. 24We identify across sectors similar regional patterns in terms of within versus between regional differences because the average intra-regional distance is generally lower than the average between-regional distance.This difference is relatively high for stone and metal, transport, and agri-food sectors.Moreover, the lowest between-regional distances across most sectors can be observed for Europe & Central Asia, Asia & Pacific, and North America.By contrast, Latin America & Caribbean countries impose a heterogeneous set of technical measures, which not only differs from other regions but also results in the highest intra-regional regulatory distance.This potentially leads to relatively higher NTM-related trade costs for exporters of this region, for example, to geographically close and large markets of the USA and Canada.
On the sectoral level, we observe that the big manufacturing blocs (Asia, Europe, North America) impose more similar regulation in manufacturing sectors compared to other regions.As a consequence, manufacturers that export within or between these regions are less likely to face different regulations in export markets compared to their home market.By contrast, firms in Africa & Middle East operate in a low regulatory environment at home and may face unfamiliar compliance requirements in these export markets.Furthermore, regulatory distances are largest for chemicals and plastics, stone and metals, as well as transport products.To a large degree this is caused by lower shares of minerals and fuels, and stone and metal products being covered by technical measures-compare frequency ratios presented in Table 3. Lastly, agri-food products are consumed and/or produced by more countries than manufactures, and additionally contain relatively more consumer-sensitive products.Both circumstances require governments to either regulate production processes or impose regulation that specifies final product quality.This leads to a higher incidence and variety of technical measures.However, regulatory differences presented in Table 5 illustrate that the types of measures imposed in these sectors are more similar across all regions compared to manufacturing sectors.
The identified patterns of structural regulatory differences across sectors and regions lend further support to including indicators of structural differences when describing patterns of NTMs and their potential effect on trade.Structural differences are likely to represent an impact channel that is distinct from intensity indicators such as prevalence, count or indicators related to regulatory coverage.In an extreme case two countries imposing for example, five different measures would have the same count or prevalence score, suggesting a similar regulatory profile, but would also be separated by the highest regulatory distance, suggesting a very different regulatory profile.Thus, in isolation neither indicator is sufficient to describe cross-country differences or similarities of NTMs.
T A B L E 5 Regulatory heterogeneity between regions, per sector, 2016.

CONCLUSION
The paper presents the most commonly used NTM indicators to describe international patterns of NTMs across countries and sectors.We organize indicators into three categories-stringency, coverage, structure-and illustrate that each of these categories describes a distinct dimension of a country's NTM profile.Particularly, for standard-like, quality-increasing NTMs, which increase trade cost and potentially imply positive demand-side effects, too, this categorization may lead to new insights into the empirical assessment of NTM trade effects.Furthermore, we extend the set of existing indicators by introducing metrics from association analysis to demonstrate joint occurrences of specific measures and by applying a standard PCA to highlight which groups of measures drive cross-country variation in regulatory stringency.All indicators presented and used in this study are publicly available for multiple sectoral classifications and ready to use in descriptive and/or empirical work.
The descriptive analysis identifies a set of stylized facts about international patterns of NTMs.Overall, countries continuously legislate, which leads to a constantly changing regulatory environment.The overwhelming majority of measures is imposed in a non-discriminatory fashion across all trading partners.In addition to classical border measures imposed only foreign firms, regulatory differences in standard-like measures imply bilateral trade costs dimension that complexity to Thus, imposing MFN-type regulations not only results in different trade cost effects across foreign exporters, but also changes the position of domestic firms vis-a-vis export markets.These effects are likely to be heterogeneous across sectors.
The concepts and indicators presented in this paper are in part determined by the constraints of global NTM databases.In contrast to binary data points, more detailed information about the regulatory burden implied by NTMs (e.g., specific certification requirements, actual tolerance limits, etc.) would allow us to construct more accurate indicators (see e.g., Winchester et al., 2012).However, the combination of geographic scope, diversity of products, and complexity of regulation pose an almost insurmountable challenge to consistently collect more detailed regulatory data.Furthermore, we only focus on de jure measures while private and international standards play an increasing role in international trade (see e.g., Schmidt & Steingress, 2019).Private standards and/or standards set by public organizations can enter official regulations by reference.However, a record of the extent with which policy makers introduce such standards in legislation is a question for future research and data collection efforts.• We use the WTO Notifications database by Ghodsi et al., 2017, who retrieve the original notifications data from WTO's I-TIP portal and impute missing HS codes.

A.2 Descriptive indicators, WTO notifications
• Countries affected by specific notifications are cross checked against the information in the SPS and TBT Information Management System (IMS) of the WTO, and corrected where necessary.
• The dataset is available in a bilateralized and reporter-based (i.e., aggregated over affected countries) version.
• In the bilateralized version we map into a grid of all reporting countries and 240 possibly affected countries.
• Indicators are calculated for the following NTM categories: Technical measures, non-technical measures, MAST chapters, and Notification requirement.
• EU member-states are split out as reporter and affected country, depending on their entry date.Thus, notifications in the data submitted by individual member-states are included.

A.3 Structural heterogeneity
• Distance indicators are calculated for the following NTM categories: All import measures, technical measures, SPS, TBT, and non-technical measures.
• For technical measures the set of measures excludes non-specific categories like broad chapters (e.g., A000) or "not elsewhere specific (nes)" coded measures.• Jaccard distance assumed to be 1 when no measures present, that is, assumption that is can only decreases in joint presence of measures.
• EU member-states are split out and intra-EU distance is set to zero.

A.4 Co-occurrences of measures
• We retrieve the full set of pair-wise measure combinations for each 6-digit product using the Apriori algorithm and average by sectoral classification.
• The Jaccard distance is based on the transpose of the underlying country-measure matrix used for the distance indicators presented in Table A3.Thus, two measures are "closer" the more common countries use them jointly.

F
Comparison of measures by income group.(1, Data-NTMTRAINS; 2, Guinea was removed from price-based measures.It exerts disproportional influence on the average low-income country prevalence score, which drops from 0.7 to 0.1 price-control measures per product when excluding Guinea.)[Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 4 Structural heterogeneity of technical measures, 2016.(1, Data-NTMTRAINS; 2, The LSH figure is based on the first two dimensions derived from multi-dimensional scaling (MDS) of the Jaccard distance matrix for technical measures.)[Colour figure can be viewed at wileyonlinelibrary.com]

14679396, 0 ,
Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons LicenseF I G U R E 5 Association of SPS and TBT measures, 2016.(1,Data-NTMTRAINS; 2, y-axis and x-axis represent measures A 1 and A 2 , respectively' 3, For the confidence index P(A 2 |A 1 ) the interpretation is as follows: A 1 → A 2 , that is, the degree to which A 2 measures come with A 1 measures; 4, Measures were averaged to the sub-categories presented in Appendix B; 5, RHS figure is weighted by the share of products to which joint occurrence applies; 6, Measure legend based on the MAST classification presented in Table1and Appendix B: A2 SPS tolerance and use, A31-2 SPS labels and marking, A33 SPS packaging, A4 SPS hygiene, A5 SPS post-production treatment, A6 SPS process control, A81 SPS registration and approval, A82 SPS testing, A83 SPS certification, A84 SPS inspection, A85 SPS documentation, B2 TBT tolerance and use, B31-2 TBT labels and marking, B33 TBT packaging, B4 TBT process control, B6 TBT Product identity, B7 TBT product performance, B81 TBT registration and approval, B82 TBT testing, B83 TBT certification, B84 TBT inspection, B85 TBT product documentation.)[Colour figure can be viewed at wileyonlinelibrary.com]

T A B L E 2 Distance measures.
While indicators a, b, c, and d are informative in their own right, they also provide the basis for constructing the regulatory distance indicators presented in Table2.The application base for each indicator depends on the underlying definition of regulatory distance.While Sokal and Michener-based (or simple matching) measures decrease with the joint presence and absence of measures (also used by UNCTAD, 2017a), Jaccard distances only decrease with two countries having actual measures in common, that is, joint presences. 1414679396, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License In contrast to indicators of intensity and coverage, distance measure are defined on the product-level. Note: 14679396, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License versus WTO notifications by country, 2016.(1, Data retrieved from Ghodsi et al. (2017); 2, Average number of notifications per product.)[Colour figure can be viewed at wileyonlinelibrary.com] 14679396, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

Pre-shipment inspections Sector Count Prevalence Frequency Coverage Count Prevalence Frequency Coverage Count Prevalence Frequency Coverage
Regulatory intensity and coverage per MAST chapter and sector, 2016.Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E 3 T A B L E

3 Continued Quantity-control (E) Price-based (F) Export-related (P) Sector Count Prevalence Frequency Coverage Count Prevalence Frequency Coverage Count Prevalence Frequency Coverage
Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E 4 Continued Weights for the following sectors are averaged: MinFuel, Chemicals, and PlastiRub to Chemicals & Plastics; HidesSkin and Wood to Skins & Wood, StoneGlas and Metals to Stone & Metals, and TextCloth and Footwear to Footwear & Clothing; 3, For a more detailed description of measures groups see Appendix B; 4, We use the BEC classification to identify consumption and intermediate products.14679396, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Jaccard distance measures used; 3, Regions Sub-Saharan African and Middle East & North Africa were combined to Africa & Middle East, and regions South Asia and East Asia & Pacific were combined to Asia & Pacific; 4, Regulatory distances for the following sectors are averaged: MinFuel, Chemicals, and PlastiRub to Chemicals & Plastics; HidesSkin and Wood to Skins & Wood, StoneGlas and Metals to Stone & Metals, and TextCloth and Footwear to Footwear & Clothing; 5, Colors are differentiated based on within-sector differences, except for average intra-and between-regional distances for which colors are based on cross-sector differences.Color coding for relatively low (green) to relatively high (brown) regulatory distance.14679396,0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

•
The underlying basis of the PCA are the reporter-based prevalence scores for the more detailed measure categories presented in Appendix B. Using those scores, the sample groups for the PCA are: All, import, and export measures, technical and non-technical measures, and MAST chapters A, B, C, E, F. That is, the variance is decomposed for these groups and weights add up to one for each group g and sector k.Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E A5 Dataset: PCA-based variance decomposition, 2016.Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/roie.12736 by Universitat Bern, Wiley Online Library on [14/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E A4 Dataset: co-occurrences of measures, 2016.