¹ For data on war, see Richardson, Lewis F., Statistics of Deadly Quarrels (Pittsburgh 1960)Google Scholar; Wright, Quincy, A Study of War, I (Chicago 1942)Google Scholar; and Sorokin, Pitirim, Social and Cultural Dynamics, III (New York 1937).Google Scholar

² Rummel, Rudolph J., “Dimensions of Conflict Behavior Within and Between Nations,” General Systems: Yearbook of the Society for General Systems Research, VIII (1963), 1–50Google Scholar, and “A Field Theory of Social Action With Application to Conflict Within Nations,” ibid., X (forthcoming 1966). For similar data, see Tanter, Raymond, “Dimensions of Conflict Behavior Within and Between Nations, 1955–60,” Journal of Conflict Resolution (forthcoming 1966).Google Scholar

³ Data generated by the code sheet are being filed with the Inter-University Consortium for Political Research, University of Michigan, and the Peace Research Institute, Oslo, Norway.

⁴ The definitions given for the categories, subcategories, and divisions are those the author has used in collecting his data. They should, however, be considered only suggestions. In the light of their experience, other researchers may want to impose their own definitions on the categories. Regardless of the definitions employed, however, it is crucial that those adopted by the researcher be published with his data.

⁵ The author has used .02 percent as the criterion.

⁶ The author has coded reports or allegations about specific actions contained in such communications as though they occurred. Thus, for example, a 1964 report by Hanoi radio that American jets have attacked a North Vietnamese border village is coded as discrete military action by the U.S. against North Vietnam. The source's (say, the New York Times's) source (Hanoi radio) of this data, and its reliability, can be coded as described under “Revisions.” The inclusion of all such reports or allegations as though they happened reduces the a priori elimination of such data, and enables several analytical cuts to be made of all the collected data as, for example, by analyzing the data by source and by level of reliability.

⁷ A correlation coefficient itself could not be calculated for these codings. Only a small proportion of total squares would be used on any one code sheet. Thus, there always will be a high overall correlation because of the high number of similarly uncoded squares between criterion and test codings. The E-score does not have this problem.

⁸ Since E roughly approximates correlation squared, the correlation would be about .