CODING MANUAL FOR
VERSION 3.0 OF THE CORRELATES OF WAR
FORMAL INTERSTATE ALLIANCE DATA SET, 1816-2000*
(with update information through v3.03)
By Douglas M. Gibler
Assistant Professor of Political Science
University of Kentucky
And
Meredith Reid Sarkees
Professor and Director, Center for Women’s InterCultural Leadership
Saint Mary’s College
Contact Information:
Email:
Web:
Phone: (859) 257-9677
Facsimile: (859) 257-7034
Department of Political Science
1627 Patterson Tower
Lexington, KY 40506
ABSTRACT: This coding manual serves as a companion to the release of version 3.0 of the Correlates of War Formal Interstate Alliance Data Set, 1816-2000. The manual proceeds in six parts. The first two sections describe the development and major characteristics of the early Correlates of War Alliance datasets. The third section describes the procedures used to both identify and code each alliance, while the fourth section provides descriptions of the variable fields found in the 3.0 data set release. Part five discusses several important changes made from the original Singer and Small (1966; 1968) data set. The final section provides an aggregate list of sources consulted in documenting these alliances; sources documenting individual alliances are available from the author.
* We would like to thank Danette Brickman, John Cotter, Harry Fogler, Russ Hubbard, Neel Natarajan, Ashley Leeds, Jeff Ritter, Mel Small, David Singer, Glenn Palmer, Scott Bennett, and Stuart Bremer. The Departments of Political Science at the University of Kentucky and the Pennsylvania State University also provided institutional support for the project. Any errors in either the data set or coding manual should be reported to the author. Please cite this coding manual when using Version 3.0 of the Correlate of War Formal Interstate Alliance Data Set.
UPDATE INFORMATION
V3.01: V3.01 of the alliance data changes the format of one variable, the state names listed within the data set (in the variable “STATENM”). In version 3.0, some state names contained commas (e.g., “South Korea” was listed as “Korea, South”). Most statistical software reading comma-separated-variable data files can handle this format, but it may cause some problems for custom programs and particular routines. As a result, we modified the state names in v3.01 to remove the commas (and so, e.g., “Korea, South” is now listed in the data set as “South Korea”). NO SUBSTANTIVE CHANGES TO THE ACTUAL CORE DATA (in alliance numbers, alliance types, state membership in alliances, or dates of alliance entry/exit) HAVE BEEN MADE IN THIS RELEASE. This change in naming convention parallels upcoming changes in the Correlates of War system membership list, which will be modified to remove commas in a future version.
V3.03: V3.03 of the alliance data changes the coding of 4 alliances to correct for a typo and minor inconsistencies with the COW state membership list. Changes are as follows:
- Alliance #: 2001
Signature Date: April 22, 1813
Change from V3.01: This alliance should be removed from the CoW formal alliance data set. The end date is before 1/1/1816.
- Alliance #: 2074
Signature Date: August 27, 1891
Change from V3.01: The end year is 1893, not 1993.
- Alliance #: 3523
Signature Date: June 17, 1950
Changes from V3.01: UAE should be listed as a member from 12/2/1971, its CoW membership start date. Tunisia should be listed as a member from 3/20/1956; STNOTE changed to 1.
- Alliance #: 3535
Signature Date: October 20, 1955
Change from V3.01: Syria joins Egypt as UAR from 2/1/1958 to 9/29/1961; the alliance does not exist during these years. The STNOTES and ENDNOTES variables have been changed accordingly.
DEVELOPMENT OF THE CORRELATES OF WAR ALLIANCE DATA
In 1966, J. David Singer and Melvin Small marked the release of their first data set on alliances in their article “Formal Alliances, 1815-1939: A Quantitative Description.” The expansion of this dataset was explained in “Formal Alliances, 1816-1965: An Extension of the Basic Data.” In these two articles, Singer and Small detailed the coding criteria upon which the data sets had been assembled.
First, at least two members of the alliance must be qualified system members; second, the alliance commitment must be a defense pact, neutrality or non-aggression pact, or an entente; and third, the effective dates of alliance have to be identified. Implicit within this definition is the formality of the agreement; a formal alliance is a written agreement that identifies at least the members and the commitment type (Singer and Small, 1968: 1-6).
The original Correlates of War state membership criteria were first developed in Russett, Singer, and Small (1968) and were later expanded in Singer and Small (1972) and Small and Singer (1982). The criteria include states that: (1) Have a population greater than 500,000, and (2) are “sufficiently unencumbered by legal, military, economic, or political constraints to exercise a fair degree of sovereignty and independence” (Singer and Small, 1972: 20). The second criterion was operationalized prior to 1920 as recognition by Britain and France, and after 1920 as formal recognition by any two major states or membership in the League of Nations or United Nations. The latest version of this data is available through cow2.la.psu.edu.
The commitment type of an alliance defines the level of support that an alliance member pledges to another alliance member. Defense Pacts (Type I) commit states to intervene militarily on the side of any treaty partner that is attacked. Neutrality and Non-Aggression Pacts (Type II) specify that parties remain militarily neutral if any co-signatory is attacked. (The neutrality pact is usually more specific than the more sweeping non-aggression pact.) Finally, Ententes (Type III) pledge consultation and/or cooperation in a crisis, including armed attack.
In determining the effective dates of the alliance, the start date corresponded to the signature date of the treaty, even though ratification may have come much later. Alliances not ratified were excluded from the data set. Termination dates are more difficult to determine, though end dates could be either: specified in the agreement; the result of a formal abrogation; the result of informal abrogation via explicitly recognized violation of the commitments, or via the assumption of new and incompatible obligations by one or more of the signatories (Singer and Small 1966).
CHARACTERISTICS OF THE ORIGIONAL ALLIANCE DATASETS
There were several less well-known characteristics of the COW alliance datasets that were consequences of both Singer and Small’s research goals and of the computer technology of the time. Singer and Small were primarily interested in investigating the relationship between alliances and the onset of inter-state war. Thus, no alliances were included which “were consummated by nations while participating in war or within three months prior to such participation, unless those alliances emerged from the war intact” (Singer and Small 1968, 262). Beginning dates and ending dates of the alliances were also given only in terms of month and year. The commitment type of the alliance was coded as “1” defense pact, “2” neutrality or nonaggression pact, “3.” Though the general assumption was that type 1 alliances imposed a greater degree of commitment than type 2, and that they both imposed greater commitments than type 3 alliances, their numerical coding is more a function of the computing requirements than of a ratio or ordinal ranking. As Singer and Small note, “While a Class I alliance obviously is more serious than a Class II or a Class III, a Class III may be more serious than a Class II. In the nineteenth century, a neutrality pact was generally a more serious commitment than an entente. In the twentieth century, however, the entente seems to be a more serious commitment than the nonaggression pact” (1968, 280 note 10). In other words, an agreement to consult may specify positive commitments not often contained in the restrictive neutrality/non-aggression pacts.
Finally, though the journal articles presented the data in an alliance list format, the data that was more generally released (by ICPSR for instance) was actually in the form of alliance bonds. Singer and Small had a particular interest in dyadic alliance relationships and their impact on war initiation; thus, only the highest level of alliance commitment between a pair of states in a given year was coded. This forced multiple alliance commitments to be ignored, and lower level agreements ended with the signing of a defense pact or other higher commitment. Similarly, ententes signed while a defense pact was in force were not included.
These coding decisions were tied both to a specific research agenda and to the extant computer technology. Therefore, subsequent versions of the data (for example, Sabrosky’s 1976 data set and Sarkees 1995 data set) began the process of removing these strictures and expanding the data. The alliance data was kept in the list format and an attempt was made to add all alliances, including those created during wartime and those reflecting simultaneous alliance commitments. There was also an attempt to add greater specificity to the beginning and ending dates of alliances. This process of revision makes the 3.0 version of the data reflect not only an attempt to update the temporal domain of the data, but also a concerted effort to update some of the coding procedures used in the construction of the original alliance data sets.
PROCEDURES USED TO IDENTIFY VERSION 3.0 ALLIANCE CASES
Defining Formal Alliances
An explicit focus on written agreements is probably the most significant change from earlier coding methods. Though Singer and Small had relied primarily upon the written alliance agreement in categorizing alliance commitments, they also supplemented this research with interpretations by diplomatic historians, particularly if there were any ambiguities in the treaty text (1966, 5 and 1968, 261). In Version 3.0 of the alliance data set, historical opinion is no longer consulted. Instead, only written treaties that meet the Singer and Small definition are included in the data set, with two important exceptions.
First, written treaties that do not pass the ratification process are not included. There were several South American treaties during the 19th and early 20th centuries that failed to win legislative approval following negotiations, and at least one treaty was eliminated because the alliance negotiator intentionally failed to sign the final treaty (see Pribram 1920-21, for a description of the latter treaty). Second, several alliances were identified as meeting the Singer and Small definition, but the actual treaty text could not be located. For example, the terms of a few recent alliances were confirmed by multiple press accounts, but the actual treaty texts have not been deposited with the United Nations or otherwise released to the public. The texts for a few 19th century alliances have not been reprinted, were never cataloged in the various treaty collections, and should probably be considered lost. Alliance cases with no surviving treaty texts that could be confirmed by at least two secondary sources are included in this data set.
Sources Used
The generation of the updated formal alliance data set began with the various data sources that are currently available – the original Correlates of War data set and the various independent updates. These data sources were augmented with new United Nations data and several new anthologies that were released after the last revision of the Correlates of War data. In addition, Lexis-Nexis, Keesings, and Facts on File were used for every year available (the years vary according to each source) to update the 20th century alliances.
Secondary historical sources were used to collect background information on the duration of each alliance (exact signature and abrogation or termination dates), and as mentioned above, secondary sources were also used to identify the existence of a formal alliance when no treaty texts could be identified. Once again emulating the Singer and Small data collection procedures, secondary sources were chosen carefully to ensure global coverage during the entire time period. All sources – both secondary and primary – used for this data set are listed in the bibliography of this coding manual.
Alliance Sources and Avoiding Missed Alliances
Using a multiplicity of sources is a way to try to identify the entire list of alliances. Unfortunately, it is not easy to avoid omissions in this type of research because alliances can always exist that were never registered, were never included in anthologies, and were never mentioned by historians. The prospect of these missing alliances is troubling, but great efforts were taken to ensure that the data collection procedures gave an accurate representation of the universe of alliances during the years since 1816.
As a reliability test of the methods used to collect these alliances, 20 state-year periods of 10 years were chosen randomly and were then independently investigated by several students not involved in the original coding of alliances. These in-depth investigations also focused on secondary sources not originally used in the construction of the data set. In all but one of these 20 investigations, the alliance portfolio produced by these more intense investigations matched the original alliance portfolio. In the only aberrant investigation, a single alliance was added to the current data set. Given these reliability results, one can be fairly confident that the prospects for Type I errors in this data set are low. However, if the data user detects errors of omission (or any other errors), please contact the data set author.
System Membership
System Membership affects the alliance data in several ways. Alliances created by one system member with one or more extra-systemic groups are excluded from this data set. If one of these groups gains system membership during the tenure of the alliance, then the alliance begins for that state on the date of entry into the international state system. These left-censored cases are noted by a separate “start notes” variable. When a state loses system membership – through civil war, loss in war, or political union – the alliance ceases to exist for that state. If two or more system members are no longer present in the alliance, then the entire alliance is coded as ending. Alliance cases that end due to Correlates of War System Membership rules (right-censored cases) are noted in the data set by a separate “end notes” variable.
Alliance Terminations
Unfortunately, termination dates for alliances are not as easy to code. Many treaty texts argue that the commitments will last “in perpetuity” or will be renewed every 10 or 20 years after the date of signature, provided both states agree. Although original alliances are often registered with international bodies (at least since the League of Nations), renewals are not often registered, and terminations rarely are. Given the complexity of determining when an alliance relationship ends, a clear set of procedures was developed that provided for greater consistency of application. First, the coding of the termination dates began with the text of the treaty and its stated end date. Second, primary and secondary sources were consulted around this renewal time period to determine if a renewal took place. Third, primary and secondary sources were examined to determine if abrogation could have occurred prior to the expected end of the alliance. Typical events include certain levels of conflict like intense crises or wars and regime changes such as the disintegration of the Soviet Union and its satellite states beginning in the late 1980s. If it is clear that the treaty was abrogated, the termination date for a formal alliance then becomes the stated end date, given renewals, adjusted for possible abrogation by one or more members.
VARIABLES INCLUDED IN VERSION 3.0 OF THE DATA SET
Listed below are the descriptions of the variables included in the Version 3.0 release of the Correlates of War Formal Interstate Alliance Treaty Data Set, 1816 to 2000. The data set is organized by alliance member, so each alliance contains at least two member state entries. Multilateral alliances have entries for each member state and can include multiple entries for a member state if it suffers loss of system membership or leaves and then re-enters the alliance. The alliance number remains the same as states enter and leave an alliance unless the purpose behind the alliance changes. The variables are listed in the order that they appear in the data set.
ALLYNUM – The assigned number for each alliance case. Note that this number is new in the v3.0 data, and does not correspond to alliance numbering in prior versions of the data set.
CCODE – The Correlates of War System Membership Code for the member state.
STATENM – The state name that corresponds to each System Membership Code number; note that several states change names through the data set even though their CCODE remains constant (“Prussia” becomes “Germany” and “Russia” becomes “USSR” for example).
STARTDAY – The day the alliance commitment began for the member state.
STARTMON – The month the alliance commitment began for the member state.