Introduction: Research and Methodologybeen developed. Alliances in the traditional literature have been
revered as sources of peace when they have acted as balancing
agents and reviled as sources of war when the same alliance sys-
tem breaks down. Based mostly on logic and historical anecdote
and often mired in post hoc balance of power explanations, the
traditional literature has not produced a convincing, consistent,
theoretical explanation of the relationship between alliances
and war. Not until the investigatory process turned empirical
did researchers begin to start a process of accumulation that has
provided many of the answers to the alliance-war puzzle.
Importance of Good Data
Part of the problem leading to such differences across theories
and results was the relative dearth of data on interstate alliances.
For most of the early studies, no consistent agreement existed
on what constituted a military alliance, and certainly no reposi-
tory of data existed with which to test the various theories. That
all changed in 1966 when J. David Singer and Melvin Small
released the first data set on international alliances. This release,
along with the release of an expanded data set three years later,
established detailed coding criteria upon which all Correlates of
War formal international alliance data sets have since been
assembled: First, at least two members of the alliance must be
qualified system members; second, the alliance 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 obligations of each alliance member
(Singer and Small 1966, 1–6; Singer and Small 1968).
There were also several less well-known characteristics of the
Correlates of War alliance data sets that were consequences of
both Singer and Small’s research goals and of the computer
technology of the time. Singer and Small were primarily inter-
ested in investigating the relationship between alliances and the
onset of interstate war. Thus, no alliances were included that
“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 alliance type
was coded as “1” defense pact, “2” neutrality or non-aggression
pact, or “3” entente. Type 1 alliances imposed a higher level of
obligation on the signatories than the Type 2 alliances, and both
Types 1 and 2 imposed greater obligations than Type 3 alliances.
As Singer and Small (1968, 280n10) note, however, Type 3
ententes and Type 2 non-aggression or neutrality pacts often
had a greater historical impact on the international system than
Type 1 defense pacts, depending on the states involved in the
alliance.
Finally, although the journal articles presented the data in an
alliance list format, the data were more generally released in the
form of alliance bonds (by the Inter-University Consortium for
Political and Social Research, for instance). Singer and Small
had a particular interest in dyadic alliance relationships and
their impact on war initiation; thus, only the highest level of
alliance obligation between a pair of states in a given year was
coded. This forced multiple alliance types to be ignored, and
lower-level agreements ended with the signing of a defense pact.
Similarly, ententes signed while a defense pact was in force were
not included.
These coding decisions were tied to both a specific research
agenda and the extant computer technology. As technology has
changed, the data have been improved and expanded. Revisions
of the data also began to include alliances created during
wartime and alliances that reflected simultaneous alliance obli-
gations. There was also an attempt to add greater specificity to
the beginning and ending dates of alliances. Thus, the data con-
tained in these volumes represent a much larger sample of the
interstate alliances formed since 1816 than did earlier works.
Also contained in these volumes is an extension of the origi-
nal data to include the years between 1648 and 1815. This
extended data set was developed for several reasons, but most
important, after the outstanding early work of J. David Singer,
Melvin Small, and the Correlates of War project, most studies
limited themselves to analyses from the post-Napoleonic period
to 1965, 1980, or to whatever year the data had been updated.
This was unfortunate because often many inferences were
drawn from temporally limited data sets, which did not include
even the largest wars of the nineteenth century. This was proba-
bly why most alliance researchers noted that the nineteenth
century was much more peaceful for alliance formation than
the twentieth century.
The year 1648 was chosen as the beginning of the extended
data set for several reasons. First, the Peace of Westphalia has
often been considered the beginning of the international state
system as it was the first peace to recognize both territorial sov-
ereignty and autonomy—the liberties ofstates(Holsti 1991).
Second, it represented a period during which alliance formation
was quite frequent (because of the states’ newfound freedom,
presumably) and varied (Parry 1978). Third, the century and a
half after Westphalia provided an opportunity to accumulate a
data set that was large enough to be theoretically interesting yet
still remain small enough to be manageable. In other words, an
additional 160 years of data could answer questions regarding
the uniqueness of the nineteenth century but was limited
enough in scope to be completed in a timely fashion. Fourth,
the manageability of a data set extension to Westphalia was
greatly aided by new treaty sources accumulated during the past
twenty years—sources that were not available to previous
alliance scholars. Parry (1978, index volume I, ix), for example,
provided a compilation “comprehending all treaties from the
Peace of Westphalia of 1648” up to the year 1851.
Combined, the two data sets represent a comparable set of
more than 450 cases of alliances that span more than four cen-
turies. These cases, especially the early Correlates of War data
sets, were of great importance for answering many of the ques-
tions in the traditional literature on alliances. Instead of relyingli