(Page 9 of 26) - Balancing and War authored by Levy, Jack. and Thompson, William.

reflect a serious intent to invoke the alliance in the event of attack, the former are much better
than the latter as general indicators of likely commitment over a large number of cases, in part
because a formal, written commitment involves a "costly signal" of a state's willingness to honor
the treaty (Morrow 2000). Recent evidence suggests that states live up to their alliance
commitments 75% of the time in the last two centuries (Leeds et al., 2000), with selection effects
explaining why we do not always observe this (Smith, 1995).

By focusing on written alliance treaties, we undoubtedly exclude some cases of

alignments that were never formalized but that contributed to joint responses against an
aggressor. This is another respect in which our coding rules constitute a conservative bias in our
test of balancing propositions. Another way in which we err on the conservative side is that we
exclude military interventions that do not involve a formal, written treaty. Note, however, that in
cases of intervention in ongoing wars the bias concerns only the size of a wartime balancing
coalition, not whether balancing occurs.

To construct an inventory of formal alliances for the last five centuries we utilized several

previously constructed data sets (Small and Singer 1969; Correlates of War Alliance List 1993;
Gibler 1999; Leeds et al, 2000) and 29 general or specific diplomatic histories.

The multiple-

source focus is necessary given the absence of agreement across these sources as to the specifics
of various alliances or even their existence. Each source has a story to tell and a theme to
emphasize, and some selectivity is to be expected. We consulted multiple sources until we
concluded that further sources were unlikely to yield any new information. The most serious
problem concerned the duration of alliances in very early modern Europe, given the tendency of
diplomatic historians to give far more attention to the formation of alliances than to the
termination. While we have information on the duration of many of these alliances, and can
fairly assume that other alliances terminated with the occurrence of a war, in the remaining cases
early in the system we adopted the following rule of thumb: in the absence of information we
assumed that alliances lasted only for the year in which they were created. We hasten to add,
however, that this rule primarily applies to 16

century alliance behavior, and much less so to

subsequent activity.

These procedures led to the identification of 223 alliances, though this number is

sensitive to exactly how one counts subsequent joiners and renewals, which we handled on a
case-by-case basis. Not all of these alliances are directly relevant for this study, however,
because our hypotheses call for a focus on alliances that are explicitly targeted against the
leading military power.

This leaves us with 84 relevant targeted alliances.

While there is undoubtedly some measurement error in the data, we want to

emphasize that we have defined balancing narrowly in terms of alliances and narrower still in
terms of written alliance treaties specifying as a target the system’s most powerful state, so that
most of the biases in the data work against the confirmation of balancing hypotheses for the
European system.

Our unit of analysis is the individual great power and its alliance behavior (alliance/no

8 A complete list of sources is available from the authors.
9 We excluded alliances for which we could not identify a specific target that was associated
with the alliance by the pertinent historical literature.