where i is a city in country_o and j is a city in country country_d. Note that $populatio{n}_o$ and $populatio{n}_d$ reflect only the total population residing in the cities of country_o and country_d that were used in calculation of distance, not the total population of the countries overall. We set $𝜃$, which is the sensitivity of trade flows to bilateral distance, equal to 1, which is the sensitivity used by Mayer and Zignago [2011].¹⁴

The distance between cities is calculated using a greater circle distance formula. Greater circle distance, or geodesic distance, is the shortest distance between two points on a sphere.¹⁵ The formula uses the Spherical Law of Cosines to determine distance, which assumes the Earth is a perfect sphere, with a constant radius of 6,372.795 km [Weisstein, 2017]. One potential downfall of this method is that it takes the inverse of a cosine, which is ill-conditioned and can lose accuracy if the distance is small. The distances between cities are sufficiently large to avoid this issue [Chamberlain, 1996]. The benefits are that the Law of Cosines has the advantage of being computationally efficient and clear. The greater circle distance formula is:

$$distanc{e}_{ij}=Karccos\left[cos{\delta }_{i}cos{\delta }_{j}cos\left({\lambda }_i-{\lambda }_j\right)+sin{\delta }_{i}sin{\delta }_j\right]$$

where $\delta$ is the latitude, $\lambda$ is the longitude, and $K$ is the earth’s radius.

Internal Country Observations: In line with gravity trade theory, if there is only one city level observation in our data, we have assumed that internal distance is one so that the natural log of distance is well defined and equal to zero for internal trade. There are 80 countries with an internal distance of one, primarily islands or small countries (e.g. Andorra, Liechtenstein, Qatar).

4.2 Border Characteristics

The border characteristics variables describe the geographic features of each country’s borders. These features include the countries with which each shares a border, the types of geographic borders of each country, and the general geographic region to which each country belongs.

4.2.1 Data Sources

For the variables that describe geographic features of the country, contiguity, landlocked, island, and region, the information was collected from maps at the United States Library of Congress. We were unable to identify a single publisher or source covering all years in the Dynamic Gravity dataset. Instead, we were forced to segment the years based on the availability of atlases from three prominent publishers. For the years 1948–1972, we used annual C.S. Hammond & Company (Hammond’s Incorporated) world atlases and gazetteers.¹⁶ Hammond’s Incorporated stopped publishing the world atlases in 1972. For the years 1973–1995, we used periodic illustrated atlases of the world published by Rand McNally.¹⁷ Finally, for 1996 to the present, we used National Geographic Society atlases of the world.¹⁸ In many cases—particularly ones involving short, difficult to see borders—we used Google LLC [2017] maps to confirm contiguity.

4.2.2 Variables

contiguity: The variable contiguity is a binary indicator that is equal to 1 if country_o and country_d share a border in year t. A border is defined as a stretch of land or river. Countries jointly bordering to a lake or other large body of water but are otherwise noncontiguous are defined as not sharing a border and contiguity takes the value 0.

landlocked_o: The variable landlocked_o is a binary indicator that is equal to 1 if country_o is landlocked. A country or territory is considered landlocked if it does not border an ocean or a body of water directly connected to an ocean.¹⁹

landlocked_d: The variable landlocked_d is a binary indicator that is equal to 1 if country_d is landlocked. A country or territory is considered landlocked if it does not border an ocean or a body of water directly connected to an ocean.¹⁹

island_o: The variable island_o is a binary indicator that is equal to 1 if country_o is an island. A country or territory is considered an island if it does not share any land borders with another country or territory.

island_d: The variable island_d is a binary indicator that is equal to 1 if country_d it is an island. A country or territory is considered an island if it does not share any land borders with another country or territory.

region_o: The variable region_o defines location region of country_o. The potential regions are: Africa, Caribbean, Central America, Central Asia, East Asia, Eurasia, Europe, Middle East, North America, Pacific, South America, South Asia, Southeast Asia, and Southern Pole.

region_d: The variable region_d defines location region of country_d. The potential regions are: Africa, Caribbean, Central America, Central Asia, East Asia, Eurasia, Europe, Middle East, North America, Pacific, South America, South Asia, Southeast Asia, and Southern Pole.

4.2.3 Variable Construction

The border variables were constructed based on the analysis of a time series of atlases at the United States Library of Congress. For each country or territory in the dataset, we examined its representation in these atlases for at least several different points in time between 1948 and 2014. If a border changed for any country during the time frame of the database, we identified the year in which that change was reflected in the atlases and altered its status in the dataset accordingly.

Countries and territories are defined as being contiguous if they share a land or river border of any length. We do not, however, recognize water borders beyond rivers, such as lakes or seas. For example, we do not considered Turkmenistan and Azerbaijan to be contiguous despite both being situated on opposite sides of the Caspian Sea. As a result of this, the variable contiguity can reasonably be used to reflect two countries’ ability to trade directly with one another via land transport.

We defined a country or territory as being landlocked if it did not border an ocean or other major body of water with direct access to the ocean or major sea. For example, this definition does not consider countries on the Mediterranean Sea as being landlocked. However, countries or territories for whom the only access to the ocean is through a river controlled by other parties are considered landlocked. Similarly, countries bordering only a large inland sea such as the Caspian Sea, are considered landlocked because such a sea is not reflective of significant access to other countries via water. As a result of this definition, the variables landlocked_o and landlocked_d reflect the ability of a country to widely and directly import or export using water transport.

We define a country or territory as an island if it does not share a common land or river border with any other country or territory. This definition implies that even countries such as Indonesia, which intuitively seem like island nations, are not technically considered islands in the dataset. As a result, this variable definition reflects the inability of a country to trade with any foreign parties via land transport.

The region variables were defined according to general, consistent categorizations in the atlases consulted during the construction of the other border variables. The variables can reasonably be used as a general measure of regional proximity. However, no official source for regional location was consulted so users should be careful about the types of inference drawn from these variables as they may not accurately reflect other, more substantial definitions of any of the regions into which each country or territory was placed.

Special Considerations for Contiguity: In general, we have chosen not to extend common border across colonial relationships. That is, countries or territories do not share borders with any parties that are contiguous to their colonies or hegemons unless they, themselves, are contiguous with the party. For example, Mexico does not share a border with Great Britain during the period when British Honduras exists as a British territory.²⁰

Western Sahara: (iso3 : ESH, dynamic_code : ESH) is a disputed territory under Moroccan (iso3 : MAR, dynamic_code : MAR) control. Because Comtrade features trade data for Western Sahara, we have chosen to recognize this territory. If you do not wish to treat Western Sahara separately from Morocco, set common border of Morocco and Mauritania (iso3 : MRT, dynamic_code : MRT) to 1.

Byelorussian SSR and Soviet Union: Despite being a member of the Soviet Union (iso3 : SVU, dynamic_code : SVU), Byelorussian SSR (iso3 : BYS, dynamic_code : BYS) features trade reported independently while the Soviet Union existed. For this reason, we have chosen to include independent observations for it as well. Contiguity for Byelorussian SSR reflects not only the countries or territories to which it is contiguous but also any to which the Soviet Union is contiguous. Similarly, despite being itself landlocked, we have defined it as not being landlocked because the Soviet Union is not landlocked.

Special Considerations for Landlocked: Serbia is technically landlocked, but has access to a sea through Montenegros Port of Bar.²¹ Observations in the dataset assume it is landlocked. Anyone wishing to recognize this access to sea could set landlocked_o/d = 1 for Serbia (iso3 : SRB)

Internal Country Observations: We have assumed that countries or territories are not contiguous to themselves. The variable contiguity reflects this assumption by setting the variable equal to zero whenever country_o is equal to country_d.

5 Cultural Variables

This section describes the set of variables that characterize cultural and historical relationships between country pairs. The measures included are those reflecting common spoken languages and former or current colonial ties. The current release contains seven cultural indicators described below.

5.1 Common Language

Common spoken languages are thought to be trade facilitating. The variable defined below, common_language, is an indicator of whether a language is spoken by at least some residents of a country pair. In order to develop this indicator, we identify “commonly spoken” languages by residents of each country and territory using the CIA World Factbook definition of languages:

5.1.1 Data Sources

The data used for construction of the common_language variable comes from the CIA World Factbook [CIA, 2017].²³ The current edition of the CIA World Factbook lists 375 languages, one or more of which are commonly spoken in 276 countries and territories.²⁴

5.1.2 Variables

common_language: The variable common_language takes value equal to 1 if residents of both country_o and country_d speak at least one common language, as listed in the CIA World Factbook [CIA, 2017].

5.1.3 Variable Construction

The common_language variable was constructed by cross-referencing languages listed for each country in the CIA World Factbook and setting the binary indicator for common language equal to 1 for all country pairs that share at least one listed common spoken language. A considerable limitation of this approach is that the source does not always list a proportion of the population of each country that speak a given language, implying that it is not generally possible to identify the extent to which two countries speak similar languages. Nonetheless, it was the most complete data source that we were able to identify.

Dissolved Countries: The most recent edition of the CIA World Factbook at the time of data construction does not list countries that have ceased to exist such as the Soviet Union or Yugoslavia. For those countries, the list of languages was supplemented by the earlier editions of the CIA World Factbook, taking the languages in the last year of each country’s existence as commonly spoken in that country. Note that in the current version of the Dynamic Gravity dataset, the common_language does not change over time. It is current as of 2015 for the countries that exist in that year. For the countries that do not exist, it is current as of the last year of that country’s existence.

Internal Country Observations: We have assumed that for observations in which the origin and destination country are the same, there is at least one spoken language common to residents of that country. The variable common_language reflects this assumption by setting the value of the variable equal to one whenever country_o is equal to country_d.

5.2 Colonial Relationships

The set of variables describing colonial relationships consists of six bilateral variables that describe historical colonial relationships between countries. These variables exclusively reflect relationships in which one country was a “colony” of its trading partner and, therefore, does not account for “protectorates”, “possessions”, or other types of possible entities lacking full sovereignty. These six variables are all directional. Three of them indicate that the origin country in a country pair was the colony of the destination country, implying that the destination country was the hegemon or colonizer. The remaining three variables reflect the reversal of that relationship so that the destination country is the colony. Additionally, these six variables also refer to three potential time frames. The first indicates whether the country pair was currently in a colonial relationship during the year of the record, the second indicates whether they were ever in a colonial relationship, and the third indicates whether they were in a colonial relationship at any point after 1945. Below are definitions for each of these six variables, along with the methods and data sources used to construct them.

5.2.1 Data Sources

The colonial relationship variables are based primarily on two sources, which were necessarily supplemented with several other sources in order to fill in some relationships missing from the two main sources. These combined sources were used to extract data for 325 unique colonial relationships (i.e., colonies with a combination of a distinct colonizer and distinct period of colonization), 263 unique entities that have been colonies, and 23 unique colonizers. The start and end dates for each of these colonial relationships were also extracted from these data sources.

The primary data source is Colonial Contiguity Data, 1816-2016, Version 3.1 dataset provided by the Correlates of War (CoW) Project [Correlates of War Project, 2017]. The dataset provides an extensive historical list of dependency relations between hegemons and their respective dependencies between 1816 and 2016. While the Dynamic Gravity dataset utilizes only those dependencies considered “colonies”, the source dataset contains other types of relationships such as protectorates and occupations. Each record provides the names of the two entities in a dependency relationship, the start and end year of the relationship, and the type of dependency. In total, the CoW data describes 2,697 such records.

In order to accurately include colonies that have gained independence from their colonizer prior to 1816, and thus are not listed in CoW dataset, we incorporate some additional data from WorldStatemen.org Cahoon [2001–2017], Kammen [1996], Johnson [1915], and Weber [1992]. Specifically, these sources were used to identify the colonial histories of Haiti, Morocco, Oman, Paraguay, Tunisia, and the United States.

5.2.2 Variables

For each variable, a country or territory is considered a “colony” if it is defined as a colony (i.e., lacks any sovereignty from its hegemon) but is not an occupied territory in the CoW Entities dataset.²⁵ Given that most of the source data provides a start year and end year for colonial relationships, but not the exact day and month on which the relationship began and ended, all of the variables below were set equal to 1 for a specific year of the panel if the relationship existed at any time during that year, even if only for a day.

colony_of_destination_current: The binary variable colony_of_destination_current denotes whether country_o was a colony of country_d in year t.

colony_of_origin_current: The binary variable colony_of_origin_current denotes whether country_d was a colony of country_o in year t.

colony_of_destination_ever: The binary variable colony_of_destination_ever denotes whether country_o has ever been a colony of country_d.

colony_of_origin_ever: The binary variable colony_of_origin_ever denotes whether country_d has ever been a colony of country_o.

colony_of_destination_after45: The binary variable colony_of_destination_after45 denotes whether country_o was a colony of country_d for at least one year after 1945.

colony_of_origin_after45: The binary variable colony_of_origin_after45 denotes whether country_d was a colony of country_o for at least one year after 1945.

5.2.3 Data Construction

Considerable care was taken to align the 3-letter CoW country codes with ISO 3-alpha codes to line up bilateral country pairs with the rest of our dataset. In many cases, the codes used by the CoW Project did not match those used by Dynamic Gravity dataset or the International Organization of Standards, requiring the development of an extensive concordance.

Records for dependency relationships in which the dependent entity is not a colony have been dropped from the CoW Entities dataset (e.g., protectorates, possessions, etc.). This partly explains why there are 1,142 unique dependencies in the raw, unfiltered data from the CoW Entities dataset, which exceeds the 300 total countries in our gravity dataset. Furthermore, the entities that are not currently a nation-state or have never been a nation-state (e.g., U.S. states such as Alaska or cities such as Aleppo) were excluded when constructing colonial relationship variables for this gravity dataset from the CoW Entities dataset. Therefore all records in the CoW Entities dataset that include an entity that is not part of our gravity dataset were excluded.

The list of CoW countries was supplemented by data from WorldStatesmen.org and three books (described in section 5.2.1) for six countries that gained independence prior to 1816 and were not covered by the CoW Project dataset. Although WorldStatemen.org includes information on protectorates and other forms of dependencies beyond colonies, the only information extracted for this gravity dataset was for entities that were missing from the CoW data. These countries and their hegemons are Tunisia, independent from Turkey in 1591; Morocco, independent from France in 1666; Oman, independent from Portugal in 1741; the United States of America, independent from Great Britain in 1783; Haiti, independent from France in 1804; and Paraguay, independent from Spain in 1811.

Internal Country Observations: We have assumed that countries cannot be their own colonies or hegemons. The colonial variables reflect this assumption by setting each of the variables equal to zero whenever country_o is equal to country_d.

6 Trade Facilitation Variables

The trade facilitation variables are those that reflect policies put in place by nations for the sake (at least partially) of influencing aspects of international trade. The current data release includes seven variables describing preferential trade agreements between countries and territories and three sets of three variables, each set describing membership in the General Agreement on Tariffs and Trade (GATT), the World Trade Organization (WTO), and the European Union (EU), respectively. These variables, the data sources they were based on, and the details of their construction are discussed in the following subsections.

6.1 Preferential Trade Agreements

6.1.1 Data Sources

The variables describing active preferential trade agreements (agree_pta,
agree_pta_goods, agree_pta_services, agree_cu, agree_eia, agree_fta, and agree_psa) between country pairs are derived from the list provided by the WTO’s (2017) “Regional Trade Agreements Information System (RTA-IS)”.²⁶ The RTA-IS provides information “cards” for each of 484 recognized trade agreements, which include the name of the agreement, the original signatory members; the current signatories; the date of entry into force and inactivity date (if applicable); the coverage of the agreement (goods and/or services), and the type of agreement (free, partial scope, customs union, etc.). While the WTO refers to these agreements as “regional trade agreements (RTAs)”, we have chosen to refer to them as “preferential trade agreements” instead as they need not be regionally based.²⁷ The source data was compiled between February 26, 2017 and March 3, 2017, and the variables discussed in this section reflect the information that was listed during that period. A full list of the agreements recognized by the RTA-IS can be found in appendix E, table E6.

In some cases, the information provided by the WTO RTA-IS is supplemented with additional sources to better account for changes in agreement membership overtime. These cases are detailed below in section 6.1.3 and in appendix E, table E6.

6.1.2 Variables

agree_pta: The variable agree_pta takes a value equal to 1 if country_o and country_d are engaged in a preferential trade agreement of any type within the given year. The WTO defines such an agreement as one that grants more favorable conditions to the agreement members than those faced by other WTO members [WTO User Guide]. Specifically, it is equivalent to the maximum of agree_pta_goods, agree_pta_services, agree_cu, agree_eia, agree_fta, and agree_psa.

agree_pta_goods: The variable agree_pta_goods takes a value equal to 1 if country_o and country_d are engaged in a preferential trade agreement that covers goods within the given year.

agree_pta_services: The variable agree_pta_services takes a value equal to 1 if country_o and country_d are engaged in a preferential trade agreement that covers services within the given year. Additional information about services provisions in trade agreements can be found in Article 5 of GATS [General Agreement on Trade in Services (GATS), 1994].²⁸

---

---

agree_cu: The variable agree_cu takes a value equal to one if country_o and country_d are engaged in a customs union within the given year. Paragraph 8(a) of Article XXIV of GATT 1994 defines a customs union as territory in which (i) duties and other regulations of commerce are eliminated for substantially all trade between members and (ii) substantially the same duties and regulations are are extended by all members to non member territories [General Agreement on Tariffs and Trade (GATT 1994), 1994].²⁹

agree_eia: The variable agree_eia takes a value equal to one if country_o and country_d are engaged in an economic integration agreement within the given year. Article V of GATS defines an EIA agreement pertaining to services that (a) features substantial sectoral coverage and (b) eliminates substantially all discrimination for these sectors [General Agreement on Trade in Services (GATS), 1994].³⁰

agree_fta: The variable agree_fta takes a value equal to one if country_o and country_d are engaged in a free trade agreement within the given year. The RTA-IS defines a “free trade agreement” using Paragraph 8(b) of Article XXIV of GATT 1994, which defines a “free-trade area”. This definition describes an FTA as a group of customs territories where duties and regulations of commerce for all products originating within those territories are suspended for all members of the agreement [General Agreement on Tariffs and Trade (GATT 1994), 1994].³¹

agree_psa: The variable agree_psa takes a value equal to one if country_o and country_d are engaged in a partial scope agreement within the given year. A partial scope agreement is defined as one in which only certain products are covered [WTO User Guide].

---

---

6.1.3 Variable Construction

Each agreement listed by the WTO was expanded across its listed members based primarily on those listed as “original signatories” to the agreement. For each country pair in which both countries were listed as signatories, the agreement was considered active in a given year so long as the agreement was active and the members were signatories for at least one day during that year. An agreement was considered active based on the listed “Date of entry into force” and, if applicable, the “Inactive date”.

A considerable limitation of the WTO RTA database is that it often lacks clear information about member countries that enter or exit trade agreements after they have been signed and entered into force. In the database, active agreements report member countries at only two points in time: the point the agreement was signed (original signatories) and the current period (current signatories). Agreements that have become inactive report only original signatories. Thus, from the dataset alone, it can be difficult to identify changes in member countries over time. In many cases, the addition of new member countries is accounted for through the addition of new trade agreements that reflect accessions, such as the “Central American Common Market (CACM) - Accession of Panama” of 2013 that followed the original “Central American Common Market (CACM)” agreement of 1961. In some other cases, the accession or exit of countries can only be identified by comparing the original and current signatories. Of the 289 active trade agreements recognized by the WTO, 48 featured differences in original and current signatories that needed to be addressed. Because agreements that have become inactive do not report a second list of members, we are unable to identify changes in members for those agreements using the RTA-IS. For the active agreements that we are able to identify differences, we used additional sources to determine the years in which countries entered or exited and edited agreements or added accession agreements to capture this variation. A full list of the agreements that required changes to membership can be found in Table E6 in appendix E.

Of the 48 agreements that required special attention, each one was addressed in order to improve its accuracy to the best of our ability. In 31 of these cases, the difference in original and current members was a result of European Union (EU) accessions, requiring only the introduction of new accession agreements that reflect the progressive growth in EU membership. In these cases, new EU members were added to trade agreements in the year in which the entered the EU. Other agreements required additional information to establish the timeline of countries exiting and entering. This process has corrected many of these issues but it is likely that there still exist situations in which some members are either missing from agreements or included in agreements erroneously.

In addition to these issues, several other special circumstances arose that required additional considerations and special treatment:

Partially Missing Trade Agreements: The WTO lists trade agreements in two ways. The first is through a table of trade agreements that provides some information such as the date of entry into force and the type of coverage. This list does not, however, list signatories. Thus, we must turn to a second collection of “cards” that correspond to each of the trade agreements. The cards provide much of the same information as the aforementioned list but also include information on member countries. The construction of the data required the extraction of member countries from these cards to be combined with the list of trade agreements.

In three cases, there are trade agreements listed that do not have corresponding cards. These agreements are “EU - Colombia and Peru”, “Eurasian Economic Community (EARC)”, and “Russian Federation - Tajikistan”. Because there was no listing of member countries, we have not included these agreements in the data set. Nonetheless, in the first two cases, there do exist similarly scoped trade agreements. Specifically, “EU - Colombia and Peru and Ecuador” and several “Eurasian Economic Union” agreements are included.

Issues With Listed Member Countries: In several cases, the WTO recognizes agreements that do not list member countries or list member countries that did not exist when the card suggests they do. These situations are described below.

The “Borneo Free Trade Area” agreement does not specify any member countries. The original text of the agreement states that it existed between North Borneo and Sarawak between 1962 and 1969. We do not include this agreement in the data because neither member country was independent prior to their joining Malaysia in 1963 and are not included in our collection of countries.

The WTO does not appear to list Czechoslovakia as a member country during years in which it existed but does include Slovakia and the Czech Republic prior to their existence. For example, the “EFTA - Czechoslovakia” agreement, which lasted from 1992-1993, does not list Czechoslovakia as a member but does list the Czech Republic, despite it not yet existing. To address this, we have classified both Slovakia and and the Czech Republic as Czechoslovakia between 1948 and 1992.

Several countries appear to be listed with inappropriate country names for certain periods of time. In each case, we have attempted to correct this where it appears to be appropriate. Between the years 1949 and 1990, we have assumed that “Germany”, as listed by the WTO, is “West Germany”. “Saint Kitts and Nevis” was renamed to “Saint Christopher-Nevis-Anguilla” for 1967–1982. “Burkina Faso” was renamed “Upper Volta” for 1960-1983. “Congo, Democratic Republic” was renamed “Zaire” for 1971–1997.

Several agreements list member countries prior to their gaining independence. In these cases, we have not included countries as members until they become independent and appear in our data set.

The “Arab Common Market” agreement lists Yemen as a member prior to the unification of North and South Yemen. The original text of the agreement lists the Yemenite Arab Republic as an original member so we include North Yemen as a member and exclude South Yemen during the relevant years (1948–1989).

Internal Country Observations: Observations in which the origin and destination country are the same are included in the data set. We have assumed that countries cannot be in a trade agreement with themselves by definition. The trade agreement variables reflect this assumption by setting each of the variables equal to zero whenever country_o is equal to country_d.

6.2 EU, WTO, & GATT Membership

6.2.1 Data Sources

Data describing membership in the European Union was based on information made available directly from the European Union. The European Union provides a list of the 28 members and the dates at which they joined Union between 1958 and 2016.³²

Data describing membership in the World Trade Organization and the General Agreement on Tariffs and Trade was based on information made available by the World Trade Organization.³³ For both groups, the WTO provides a comprehensive list of member countries and their date of membership [The World Trade Organization, a].³⁴ Table E7 in appendix E lists the members and accession dates of the GATT and WTO, respectively.

6.2.2 Variables

member_eu_o: The variable member_eu_o takes the value one if country_o is a member of the European Union in the given year.

member_eu_d: The variable member_eu_d takes the value one if country_d is a member of the European Union in the given year.

member_wto_o: The variable wto_o takes the value one if country_o is a member of the World Trade Organization in the given year.

member_wto_d: The variable member_wto_d takes the value one if country_d is a member of the World Trade Organization in the given year.

member_gatt_o: The variable member_gatt_o takes the value one if country_o is a member of the General Agreement on Tariffs and Trade in the given year.

member_gatt_d: The variable member_gatt_d takes the value one if country_d is a member of the General Agreement on Tariffs and Trade in the given year.

member_eu_joint: The variable member_eu_joint takes the value one if both country_o and country_d are members of the European Union in the given year.

member_wto_joint: The variable member_wto_joint takes the value one if both country_o and country_d are members of the World Trade Organization in the given year.

member_gatt_joint: The variable member_gatt_joint takes the value one if both country_o and country_d are members of the General Agreement on Tariffs and Trade in the given year.

Some summary information for these variables is depicted in figures 5 and 6

---

---

---

---

6.2.3 Variable Construction

Using the sources described above, countries were assigned membership into each of these groups based on the dates on which they joined. Countries are considered members in a given year so long as they belonged to the organization for at least one day during that year.

Internal Country Observations: For observations in which the origin and destination countries are the same, the “joint” variables member_eu_joint, member_wto_joint, and member_gatt_joint are assumed to take the value of 1 if the country is a member of the respective organization and 0 otherwise. In doing so, we have implicitly assumed that internal, domestic trade qualifies as within organization trade.

7 Measures of Institutional Stability

The institutional stability variables are those that measure various events a country may be involved in or characteristics of a country that may influence its propensity or desire to conduct international trade with a given trading partner due to different aspects of the country’s stability. The current data release includes three sets of variables, each describing a different form of institutional stability in either a bilateral (hostility level towards another country or sanctions with another country) or unilateral (level of political stability) manner. These variables, the data sources they were based on, and the details of their construction are discussed in the following subsections.

7.1 Hostility

7.1.1 Data Sources

Source data for the variables reflecting hostility (hostility_level) comes from The Correlates of War Project’s (CoW) Militarized Interstate Disputes dataset (MIDB v4.01) [Palmer et al., 2015].³⁵ The dataset is organized by dispute with each observation identifying a country that participated in a particular dispute. A variable denoting what “side” of the conflict a given participant existed indicates a given country’s allies and enemies in the conflict. From that we are able to derive bilateral lists of country pairs exhibiting some level of conflict within a given year as well as the level of hostility of each conflict.

7.1.2 Variables

hostility_level_o: The variable hostility_level_o is coded 1–5 and denotes the level of hostility of country_o towards country_d in year t.³⁶

hostility_level_d: The variable hostility_level_d is coded 1–5 and denotes the level of hostility of country_d towards country_o in year t.

Figure ?? shows the number of conflicts by hostility_level by year.

---

---

7.1.3 Variable Construction

Data was constructed by expanding the information provided by the source data into a bilateral format over time to fit the structure of the Dynamic Gravity dataset. Care was then given to properly concord the country-identifying codes used by the CoW Project to those used in the Dynamic Gravity dataset.

Some pairs of countries feature more than one conflict in a given year. In these cases, hostility_level takes maximum of the levels of hostility a country has towards another country in that given year. For example, Israel has three distinct conflicts with Syria in 1950; two of which are coded as having a hostility level of 4 while the other reflects a hostility level of 1. Therefore, hostility_level_o = 4 for the observation where Israel is the origin country, Syria is the destination country, and the year is 1950.

It is worth noting that hostility_level is not necessarily symmetrical with respect to a country pair, reflecting the possibility that the level of hostility one country extends towards another in a given year is not the same is it experiences from that country. For example, in 1948 the CoW data observes the hostility level of France towards Thailand as a 4, while the hostility level of Thailand towards France is a 1.

Assigning Conflicts to Countries: Considerable care was taken to align CoW country codes with ISO3 codes to line up bilateral country pairs with the rest of our dataset, and then various observations needed additional special attention.³⁷ In total, 122 countries had CoW and ISO codes that differed. In these cases manual changes needed to be made to the CoW data for merging purposes.

Additionally, since the CoW data is intended to list participants in a given dispute, it often lists participants that do not immediately concord to a country. For example, participants can take part in a conflict before or after their country exists (e.g. civil wars), or the country can change identity during conflicts (USSR conflicts post 1992). For example, North Vietnam and South Vietnam were separately “at conflict” with various other members of the Vietnam War even after the country officially reunited as Vietnam in the CoW data. Similarly, there were still conflicts involving the Soviet Union after it dissolved. For instances where a participant in a conflict logically concorded to a different country that we observe in that particular year due to changes in the geopolitical landscape, we have accounted for these and assigned them appropriately. However, conflicts where a participant did not logically concord to an existing country in a given year that we observe as existing, we chose to exclude those participant-year pairs from our dataset.

Internal Country Observations: The CoW MIDB dataset does not include civil wars, explicitly describing the disputes it catalogs as, “historical cases of conflict in which the threat, display or use of military force short of war by one member state is explicitly directed towards the government, official representatives, official forces, property, or territory of another state.”³⁸ As a result, we do not observe civil wars, domestic unrest, or other such situations and set internal hostility levels equal to zero.

7.2 Polity

7.2.1 Data Sources

The variables describing political stability (polity and polity_absolute) are based on data from the Polity IV Project [Marshall et al., 2016]. The Polity data scores countries annually on an ordinal scale ranging from -10 to 10 denoting their level of democracy or autocracy.³⁹ A score of -10 denotes a country that is strongly autocratic, a 10 denotes a country that is strongly democratic, and a 0 denotes a country neither autocratic nor democratic (i.e. lacking in government) with intermediate values describing situations within that spectrum.⁴⁰ In total, the current dataset includes 194 countries, although not all countries exist each year. For example, countries like Germany, Japan, and East Germany are all omitted in certain years, typically around with years following conflicts or changes in the state (e.g. West Germany and East Germany unifying). In their most recent release, Polity scores cover 1948 to 2016. Appendix tables F8 and F9 list countries that exist in our dataset but are not tracked by the Polity IV project for any year, and countries that are tracked in some years but not the entire span of our data.

7.2.2 Variables

polity_o: The variable polity_o is the polity score in year t of country_o.

polity_d: The variable polity_d is the polity score in year t of country_d.

polity_absolute_o: The variable polity_absolute_o is the absolute value of the polity score in year t of country_o.

polity_absolute_d: The variable polity_absolute_d is the absolute value of the polity score in year t of country_d.

7.2.3 Variable Construction

In addition to reporting the Polity scores reported by the Polity Project, we also provide an alternative measure (polity_absolute) that reflects a slightly different interpretation of the information. A considerable limitation to the way in which Polity indexes stability is a lack of a real understanding about the relative impact of moving from one value to the next (i.e. -10 to 1 means that the country is now more democratic but less stable). In light of this, we have also provided a secnd measure polity_absolute that is the absolute value of these scores. The motivation for this measure is the notion that a higher nominal value corresponds to a higher level of stability within a country regardless of its form of governance.

There are also a few instances in the data where we observe a country twice within a single year. These cases are due to Polity identifying the country under multiple different regimes within a year that concord to the same country in the Dynamic Gravity dataset. In these situations, we had to make special decisions about which of two polity scores to assign them. For example, The Polity project assigns polity scores to Sudan, South Sudan, and North Sudan in 2011, while we only observe Sudan from 1956-2016 and South Sudan from 2011-2016. We therefore used North Sudan’s polity scores for 2011 and onwards for “Sudan” coinciding with the birth of South Sudan. Similar issues arise with Ethiopia in 1993, West Germany/Germany in 1990, and Yugoslavia/Serbia and Montenegro in 1991.

The Polity Project group has both increased its country coverage overtime and updated past polity scores to reflect changes in their institutions. In general, these updates are indicative of countries trending towards the “strongly democratic” end of the spectrum. Table 2 shows the spread of polity and polity_absolute scores across several years where the Polity Project made substantial updates.⁴¹ As shown in the table, countries appear to be becoming more stable—and more democratically stable in particular—over time.

As was the case with the hostility_level variables, the Polity Project uses CoW codes for its country classification system, so the same careful considerations had to be taken here as noted previously. Along with aligning differences in labeling conventions, if a polity score logically belonged to a country we observe under a different name in a given year, those adjustments have been made. In the case that a country received a polity score before we observe its existence, its score in those years have not been included.

7.3 Economic Sanctions

7.3.1 Data Sources

There are four variables describing economic sanctions directed towards partners. These variables are derived from the Threat and Impositions of Sanctions (TIES) dataset [Morgan et al., 2014]. The TIES dataset records economic sanctions imposed between 1945–2005 and provides end dates for these sanctions through 2012 (when the dataset was last updated). For sanctions still in place in 2012, it provides an “as of” date for the last time they verified the sanction remained in place. We used these “as of” dates as the end year for sanctions that TIES does not feature a confirmed end date.

7.3.2 Variables

sanction_threat: The binary variable sanction_threat denotes whether or not a threat to impose any sort of sanction existed between country_o and country_d in year t.

sanction_threat_trade: The binary variable trade_sanction_threat denotes whether or not a threat to impose a trade sanction existed between country_o and country_d in year t.

sanction_imposition: The binary variable sanction_imposition denotes whether or not any sort of sanction existed between country_o and country_d in year t.

sanction_imposition_trade: The binary variable trade_sanction_imposition denotes whether or not any sort of trade sanction existed between country_o and country_d in year t.

---

---

7.3.3 Variable Construction

The variables sanction_threat and sanction_imposition correspond to dummy variables in the TIES dataset and indicate whether or not a pair of countries has a sanction threatened or in place between them in a given year. The trade specific sanction variables, sanction_threat_trade and sanction_imposition_trade, are a refinement of this information, taking a value of one one if the TIES variables for type of sanctioned threatened and/or imposed correspond to trade specific policy decisions. Specifically we coded sanction_threat_trade = 1 if the “sanctioned type threatened” variable in TIES was 2, 3, 4, 5, 6, 9, or 10. For sanction_imposition_trade, we coded that as 1 if “sanction type” in TIES was 1, 2, 3, 4, 5, 8, or 9.⁴² Figure 8 demonstrates the frequency of sanction threats and impositions over time. It’s worth noting here that the threat of sanction and imposition of sanction variables are constructed independently of one another and that threat is not a requirement for eventual imposition. In total, the TIES dataset has 1,412 unique cases of a sanction either being threatened or imposed (or both). In 567 cases a sanction was only threatened, in 359 cases a sanction was only imposed, and in 486 cases a sanction was both threatened and imposed.

Assigning Sanctions to Countries: Careful attention had to be paid to lining up counties in TIES to countries in the Dynamic Gravity dataset due to differences in the naming conventions between the two. Conveniently, the TIES dataset identifies countries using the same three digit code as the Correlates of War datasets. This permitted us to use the same concordance developed and described in section 7.1.3.

Many sanctions in the TIES database feature not only a principal sender or senders, but also sanctioning institutions. The Dynamic Gravity dataset currently only pairs the senders specifically observed in the TIES dataset to the target of a given instance of economic sanctions.⁴³

TIES includes three instances in which a sender sanctions a single country twice within the same year. In these cases, we assigned the maximum as we did with hostility_level above.

Additionally, there were 38 instances in which a country involved in a sanction in a given year did not exist in the Dynamic Gravity dataset. An example of this is the United States sanctioning East Germany in 1948. In these cases, the observations were not included.

Internal Country Observations: The source data does not report sanctions threatened or imposed by parties on themselves for the countries we have identified in the Dynamic Gravity dataset. Therefore, all sanctions variables for observations where country_o is equal to country_d are set to zero.

References

Appendices

A List of Variables

B List of Unique dynamic_code’s

C Matching Dynamic Gravity to Comtrade and WITS

D List Supplemental Cities for Geographic Variables

E List of Trade Agreements and International Organization Members

F List of Countries Without Polity Scores