4. Opportunity, Willingness and the Diffusion of War, 1816–1965

We acknowledge with thanks the helpful research assistance of Sherry Lutz, Drew Froeliger, and Juliann Emmons in the generation and analysis of the data and the timely programming assistance of Brenda Gunn. The collection of the data was supported by the Institute for Governmental Affairs, University of California, Davis.

While diffusion has been given several meanings, and has a number of dimensions, we will be using it essentially to mean the growth of an ongoing conflict, the process by which states join an ongoing war and the scope of the conflict becomes enlarged. The concept of diffusion and its theoretical relevance to the study of international conflict is reviewed extensively in Most et al. (1989). For other reviews of the diffusion concept, see Most/Starr (1981), O’Loughlin (1984) and Welsh (1984). For an overview of the renewed interest in geopolitical approaches (including diffusion) to war, see Diehl (1988). For a classic statement on the contagion of political conflict, see Schattschneider (1960).

We use the term treatment to emphasize the epidemiological nature of our research, not to suggest that we are performing an experiment in which we have the ability to randomize and control exposure. However, what we report is in many respects a historical experiment, similar in method to the procedures described by Singer (1974) and Holsti/North (1965).

Although research on this point is not clear (Garnham 1988), it is reasonable to suspect that under some circumstances geographical proximity to, or shared borders with, some other nation will in themselves influence a nation’s willingness to become involved in conflict. If so, opportunity and willingness are not independent, as we noted. Indeed, borders themselves can be seen in terms of environmental possibilism by providing possibilities for interaction. The warring border nation model, however, can be seen in terms of cognitive behaviorism or willingness, as activities on a state’s borders effect the perceptions of threat, uncertainty, and opportunity held by that state’s decision makers. Clearly, most nations are likely to be sensitive to the security concerns of neighboring nations or, when the capabilities are present, are likely to see their neighbors as the greatest potential threats to their own security.

Of course, states may attempt to manipulate their immediate political geography by creating neighbors (e.g., Belgium in 1830) or eliminating them (e.g., Poland in 1939), but the costs of these activities are likely to be quite high.

Alliances, of course, are not a perfect indicator of willingness. For example, while there is general tendency for alliances to be reliable, it is clear that reliability is less than complete (Sabrosky 1980). In addition, some alliances are specific to certain issues or geographic areas (e.g., NATO), and their existence does not mean that one ally will join another no matter where the conflict. Also, alliances signed at one time may deteriorate while continuing to exist formally. In short, we do not construe alliances as indicating a general willingness to fight. However, alliances have long been recognized as the key means that nations have chosen to indicate their political position in the international system. Formal alliances are matters of serious concern to decision makers and in addition to their presumed rewards of enhanced security, they impose costs and risks both domestically and internationally (Sullivan 1974).

In fact, Altfeld/Bueno de Mesquita (1979) successfully apply expected utility theory to the problem of explaining which side nations will join in a war. While this is a problem similar to the one pursued here, it does not directly address the question of diffusion. For some other differences, see n. 16.

Following Small/Singer (1982), the major powers and the years of their inclusion in that group are Austria-Hungary (1816–1918), Italy (1861–1943), the United Kingdom (1816–1965), Russia* USSR (1816–1965), Japan (1895–1945), Prussia-Germany (1816–1945), the United States (1989–1965), France (1816–1940, 1945–65) and China (1949–65).

This third category of data was the most difficult to obtain because of the number and almost unbelievable obscurity of some colonies. We were aided very considerably by Henige’s (1970) comprehensive list of colonial governors. This list, which begins with the fifteenth century, identifies all political units that were the colonial possession of some other nation. Finding them in this list then permitted relatively easy location in one of the historical atlases.

Copies of the coding scheme employed in this study are available from Siverson. It should be noted that in general we followed the data given by Small/Singer (1982) with respect to war expansion. We did, however, depart from their delineation of what constituted war joining in a few cases. For example, they show Japan as joining World War II on 7 December 1941. While it is true that Japan began fighting the United States on that day, it is also true that the Japanese decision had relatively less to do with the European war that had been in progress since 1939 than it did Japan’s war with China that had been ongoing since 1937. Hence, we do not treat Japan as joining the war. Also, we do not treat as war joiners the nations that left World War II (albeit briefly) and then changed sides (i.e., Italy, Romania, and Bulgaria) to reenter it. It makes sense to count the initial participation only. It should be noted that including these cases would have favored the opportunity and willingness hypotheses.

Recall that Most/Starr (1980) distinguished between reinforcement, or addiction (where a state’s war behavior affects the probability of its own subsequent behavior) and diffusion (where a state’s war behavior affects the probability of the subsequent war behavior of other states). They also recognized that either could have positive or negative effects (either increasing or decreasing the probability of war behavior). The reader must be alerted that the matrix employed in the present analyses follows that utilized in Most et al. (1987) and is not the same as the matrix used in Most/Starr (1981) or Starr/Most (1983, 110–11).

The single case of war involvement by a nation with no bordering nations of any kind and no alliance involvement is New Zealand’s entry into World War II in 1939.

The weak results from A2 may be partly due to the fact that there were far fewer of this type of alliance in the data set than either of the other two types. When they are analyzed together, the more numerous alliances may overwhelm A2. In the analysis that follows it may be seen that A2 does not apply to a large number of cases but that when it does, a clear effect is evident.

Recall the treatment matrix from Table 4.1. In the single treatment tables that follow, the numbers being reported are those from cell d of that figure. Reporting all the tables would be cumbersome, so only the effects of the treatments are given. As an example of what a table looks like, we offer a full table for the treatment involving B1. Note that the numbers of cases reported in the tables refer to the total of the b-d column and that the percentages reported are for d of that total.

What we mean by percentage increase is the magnitude by which involvement in a treatment situation is larger than the base. Thus, the first three columns of Table 4.6, while the B1 treatment percentage of 8.7 is more than three times the size of the baseline percentage of 2.4, the relevant measurement is the amount of increase from 2.4 to 8.7. Thus, the appropriate calculation is the treatment percentage minus the base percentage with the remainder divided by the base percentage.

The tests of significance used in Tables 4.6 and 4.7 require a brief comment. Because the percentage changes between the no-treatment and treatment columns are often based on relatively small numbers of cases, the skeptical reader may be curious as to (1) whether the observed changes are statistically significant and (2) whether there is any way that the method could fail to produce a change, i.e., whether the results are predetermined. With respect to the first question, we have tested the differences for statistical significance in two ways. First, when the number of cases in the treatment column is greater than 20, the p-value is derived from the Z-score in a test of a difference of proportions. When the number of cases is less than 20, we have determined the p-value from Fisher’s exact test. (Recall that the cases in the treatment column are also present in the baseline no-treatment column). Because the two results are therefore not independent, before we could test for statistical significance between the no-treatment baseline and the treatment, it was necessary to remove the treatment cases from their respective baseline. A full set of the resulting tables is available from Siverson. More important than each test of significance is the correspondence between our theoretical structure and the overall pattern of the data. With respect to the second question (whether the results are predetermined), if one suspects that the method will always produce an increase in the observed cases of war diffusion under the treatment condition, see the last three columns of Table 4.7, where failure does occur.

Altfeld/Bueno de Mesquita (1979) are interested only in those cases of war expansion that took place within the first two months of a war. Hence, their main data set contains only 40 cases of war joining. In addition, they exclude from their data set the participation of the four Commonwealth nations (i.e., Canada, Australia, New Zealand and South Africa) that joined World War II in 1939, on the grounds of missing data (that is, they had no alliance memberships). Those cases are included here. We also include the cases from the Korean War, which were totally excluded by Altfeld/Bueno de Mesquita (p. 94, n. 7).