HomePersonalDemocratic PeaceDemocide20th C. DemocideMegamurderersLesser MurderersWhy DemocideDimensionsConflictMethodsTheoryPolicyLinks

Volume 4

Expanded Contents | Figures | Tables


1. Perspective And Summary
2. International Relations
3. The International Actors
4. International Behavior Space-Time
5. International Expectations And Dispositions
6. International Actor And Situation
7. International Sociocultural Space-Time
8. Interests, Capabilities, And Wills
9. The Social Field Of International Relations
10. Latent International Conflict
11. International Conflict: Trigger, Will, And Preparations
12. The Balancing Of Power
13. Comparative Dynamics Of International Conflict
15. Empirical Dynamics Of International Conflict
16. Causes And Conditions Of International Conflict And War
17. Ending Conflict And War: The Balance Of Powers
18. The International Conflict Helix
19. Theoretical And Empirical Conclusions On Conflict And War 20. Principles Of Peace And Conflict


15A. Phasing Propositions and Their Evidence on International Conflict
16A. On Causes of International Conflict
16B. Propositions and Their Evidence on the Causes and Conditions of International Conflict Behavior
16C. Evidence on the Causes and Conditions of International Conflict Behavior
17A. Propositions and Evidence on the Causes and Conditions of Ending International Conflict Behavior
18A. Descriptive Propositions on International Conflict
19A. Overall Evidence on 54 Social Field Propositions on International Conflict
19B. Primary Propositions on Social Conflict
I. Unpublished Research and Results on International Relations
III. Characteristics of Published Quantitative International Relations Studies

Other Volumes

Vol. 1: The Dynamic Psychological Field
Vol. 2: The Conflict Helix
Vol. 3: Conflict In Perspective
Vol. 5: The Just Peace

Other Related Work

Conflict And Violence page

Democratic Peace page


Appendix II

Event Data:
Bases Of Empirical
Conflict Analysis*

By R.J. Rummel


Any assessment of the political significance and consequences of, say, the July, 1977, Egyptian-Libyan mini-border war requires information on

Any good journalist or analyst can uncover information on the first two. It is the comparison that becomes difficult. How does the war compare to others fought? To other border wars? To other conflicts in Africa or the Middle East? Have similar wars occurred recently? If so, have they been preceded by the same events? Was the sequence of events leading up to the July war atypical, a product of President Col. Muammar el-Qaddafi and President Sadat? Or, was progression of events similar to that preceding other wars, suggesting certain events foreshadow war, as particular economic events warn of recession? And so on.

It is the complexity of comparison that often defeats journalists and experts alike. For comparison requires mentally scanning similar events, similar periods, similar cases; and here the mind is overcome with detail, memory misleads, and hopes and wishes, biases and suppositions, have free play.

To ease comparison, the mind will always find patterns in events, somehow. These patterns will involve an explanation of why the events occur, and provide examples of similar events. They are part of one's political and social culture. Thus, some may see the July war as Sadat's complicity in a Western plot to overthrow Quaddafi, and compare it to Chile; some may see it as Sadat's reaction to Moscow's massive arming of Libya, support for Quaddafi, and encouragement of the Libyan ruler's subversive plots in Egypt and Sudan; some may see it as simply one more indication of Quaddafi's instability and imprudence, another President Idi Amin of Uganda.

Whatever explanation applied, usually the underlying comparisons select events to buttress the explanation and ignore contrary events. Such is natural, the way the mind, unaided, patterns and explains reality. But there are aids to the mind which can increase the validity and reliability of such comparisons. The aid is science.

Science, in a nutshell, is a method for making comparisons that provide a reliable view of reality; it is a means of aiding the mind in assessing patterns, in comparing events, whether in nature or man-made. The essence of science is comparison, and it now appears almost inevitable that scientific methods would eventually have been applied to political events to improve their understanding, to add the comparative third link to knowledge of an event and of those that precede it.

The stuff of comparison is data, and the heart of science is data that have been collected according to consistent rules and operations. This insures that all relevant data have been included (whether for or against an hypothesis or belief), and that others can check or criticize the data underlying results or conclusions. The purpose is to arm the critic of one's results with the best ammunition. For results that then withstand a critical onslaught have credibility worthy of notice.

Table II.1

The application of scientific methods to political events, then, has been aided by the development of event data. That is, data for which the rules of inclusion and exclusion of political events are clear and consistently applied to all events. For example, Table II.1 above presents event data covering July 1, 1962, to June 30, 1963, for several internal conflict events on a sample of states (this is from a collection for all states given in Rummel, 1965). To give a flavor for the rules governing the collection or riot data, a riot was defined as:

Any violent demonstration of at least one hundred people. A mob or crowd of people clashing with police or troops or attacking private property is counted as a riot, as long as such violence appears to be spontaneous. Riotous clashes between rival political groups, racial clashes, and the like, are categorized under "clashes." Riots of a distinct antiforeign nature are not counted here. The term "violence" refers to the use of physical force, and the existence of a riot is generally evidenced by the destruction of property, people being wounded or killed, or by the use of police or riot control equipment such as clubs, guns, or water cannons. Arrests per se do not indicate a riot.
---- Rummel, 1965: 205

Of course, not all may agree on this definition, but at least all riot data will have been collected in adherence to it, and others can determine with what they disagree, exactly.

The value of the event data in Table II.1 is immediately evident. It can be noted easily that Peru was the most unstable among the six states, having many demonstrations, riots, plots, and attempted or successful coups; that instability in Argentina was at the top, involving plots and coups, and not mass violence; and that India by comparison had its instability in the street, with numerous riots and demonstrations.

Event data, then, are collected according to some consistent and explicit definition of behavioral political events within or between states. Through such data collected across a number of possible events, an internal or external conflict profile of a state (or directed dyad, for external conflict) can be developed, as that for Argentina in Table II.1. Moreover, if such data are collected for a number of states, then comparisons among them in the number of riots, coups, and so on, can be made.

Of course, several problems immediately arise. How useful is a numerical count of such events, of which significance, intensity, and even meaning will vary across states and years? Moreover, how can five riots of about 100 people each in Cleveland, Ohio, be considered five times more significant than one riot of 10,000 people in Washington, D.C.?

How reliable can such counts be; how valid the source of data? How can one reduce such diverse events and multidimensional data to clear patterns? The investigation of such questions has occupied many since the early 1960s and a background on this will be given shortly.

First, however, event data, should be discriminated from three other kinds of data on nations. One is of behavioral flows, that is, statistical aggregates measuring many kinds of uniformly occurring transactions, such as trade, economic aid, tourists, migrants, and the like. The signing of a treaty or agreement establishing such flows is event data, the resulting daily, monthly, or annual exchanges usually are not considered events.

Second, there are behavioral structures. These are existing, formal behavioral relationships, such as a treaty, alliance, or common membership in an international organization. The signing of a treaty or alliance, or the act of joining an international organization is an event. But the outcome is a formal structural relationship which then remains constant for a period of time. Thus, if one counts the number of treaties a state is a party to, then behavioral structures are being measured. The results, as should be clear, would not be necessarily correlated with the number of treaties signed during the same period. State A, for example, may be a party to few treaties, but then sign many new ones in a year (perhaps consequent on a new government), while state B which is a party to many treaties may sign fewer treaties than State A in the same period.

And third, there are attribute data, which measure or define the magnitude of a nation on some characteristic. Such data are widely used, as of gross national product, population, area, military expenditures, and so on. Table 4.1 presents the results of factor analyzing mainly flow and structural data; Table 7.1 presents similar results for attributes. For both analyses, event data were employed for the conflict variables.

To summarize, we can generally distinguish between


As part of the attempt to study more systematically conflict and war, two scientific projects in the early 1960s began to collect and analyze event data. One was directed by Charles McClelland at the University of Southern California, who was interested in quantitatively mapping behavioral events between states, especially in crises. He developed a coding system for tracking the tempo and magnitude of cooperative and conflictful events in a political crisis (McClelland, et al., 1965) that involved 63 types of events organized into the 22 broad categories shown in Table II.2.

Table II.2

He found that crises were clearly indicated by a rise in their intensity of events and the range of types of events. Indeed he developed a threshold coefficient for events above which crises clearly were underway (McClelland, 1968).

McClelland's coding scheme, called WEIS (for World Event/Interaction Survey), has been seminal and the bases for much scientific and applied research. Besides his continuing research on international crises (McClelland, 1972; McClelland and Hoggard, 1969), either his data, data collected by others using his scheme, or a variant thereof, have been applied to assess the Berlin crises of 1948-1949 and 1961 (Tanter, 1974); the patterns of foreign cooperation and conflictful events (Kegley, et al., 1974); the dependency of cooperative and conflictful events on national development, size, political system, and stability (Wilkenfeld, et al., 1977; Rosenau and Hoggard, 1974; and Powel, et al., 1974).

Moreover, the WEIS data have been useful simply to map the nature and frequency of cooperative and conflictful events among nations. For example, McClelland and Hoggard (1969) claim that for 1966 there were 5,550 foreign cooperative and conflictful events for all states recorded in the New York Times Index (I will get to the reliability and validity of such sources subsequently). Twenty states accounted for 70% of the events; five (the United States, USSR, China, United Kingdom, and France) accounted for 4wo. Conflict events comprised 31.5% of the total.

The WEIS approach has been used by corporations doing government contract research on internal and external conflict, such as by General Electric (Rubin, 1969), Consolidated Analysis Centers Inc. (Rubin, 1973), and Decisions Designs, Inc. (Andriole, 1976). An overview of events data research ("The Utilization . . .", 1971), especially that supported by The Department of Defense (Advance Research Projects Agency), was written under the guidance of Robert Young when he was at Consolidated Analysis Centers Inc. Both Young and Andriole have been active in interesting defense and military analysts and decision makers to use the WEIS system to assess and forecast foreign conflict and crises. McClelland and colleagues (1971) and Andriole (1976) have proposed systems for forecasting crises using events data.

Parallel to McClelland's early work, the Dimensionality of Nations Project (DON) which I directed from 1962 until it ended in 1975 (for an overview of DON, see Rummel, 1976; for critical assessments, see Hazelwood, 1976; Hilton, 1973, 1976; Seidelmann, 1973; Van Atta and Robertson, 1976), developed an approach for collecting and analyzing event data on the internal and foreign conflict of nations. This amounted to a frequency count of particular events for nations, such as riots, threats, or military clashes, and their component analysis (Rummel, 1970, Section 5.3), to reduce the data to the basic event-patterns. From this early effort a foreign conflict event code sheet was developed (Rummel, 1966), and the resulting data have been much used by others.

The DON data or approach to coding has helped determine the dimensions of international crises (Phillips and Hainline, 1972), crises dynamics (Phillips and Lorimore, n.d.), the reciprocity in conflict behavior between nations (Phillips, 1973), as well as conflict patterns (Rummel, 1963; Hall and Rummel, 1970; Hazelwood, 1973), and dynamics (Wilkenfeld, et al, 1972), and simply for mapping conflict (Taylor and Hudson, 1972). Numerous studies have been done to replicate my finding (Rummel, 1963) that foreign and internal conflict have little general relationship (Tanter, 1966; Wilkenfeld, 1973; Burrowes and Spector, 1973; Collins, 1973; Wilkenfeld and Zinnes, 1973). The most important use of the DON event data, however, has been in determining its interrelationship with the behavioral flows and structures between states as shown in Table 4.1, and dependency upon differences between states in wealth, power, and political system (Rummel, 1972, 1977, 1979), which results are consolidated in Appendix 9A.

The WEIS and DON event data approaches encouraged and influenced a number of large scale event data projects that began in the late 1960s and early 1970s, and a number of comparisons and discussion of such projects have now been published (Azar, 1970; Sigler, et al, 1972; Kegley, et al., 1975; "The Utilization . . .", 1972). For concise summary and comparison of eleven such projects see Kegley's 1975 study. The largest of these projects are currently the Conflict and Peace Data Bank (COPDAB) under Edward Azar (1970, 1972, 1975) at the University of North Carolina, and the Comparative Research on the Events of Nations (CREON) (Hermann, et al, 1973; Hermann, 1975) project under Charles Hermann at Ohio State University. Both these projects have focused on foreign events.

Focusing on internal events have been the influential projects of Ivo Feierabend (Feierabend and Feierabend, 1966; Feierabend, et aL, 1972) at San Diego State College, and Arthur Banks (1971), Director of The Center of Comparative Research, State University of New York at Binghamton. Moreover, the work of Ted Gurr (1969) at Northwestern University on internal conflict is especially within the event data movement (for the utilization of a broad range of internal event data studies for evidence on internal conflict propositions, see Chapter 35 in Vol. 2: The Conflict Helix).

The generation and analyses of events data has become an academic and scientific subdiscipline. (See Azar, 1970; Hermann, 1972; and Brady 1971 for a description of the events data movement and extensive bibliographies. A bibliography devoted exclusively to event data studies has also been published by Wynn, 1973). Its research can be found in diverse professional journals and has focused many panels of the annual American Political Science Convention. A Foreign Policy and Events Data Section has been created within the International Studies Association. Books, monographs, and numerous articles have been written on or employing event data, a portion on which are referenced here and in Appendix III. Schools within the subdiscipline have developed, utilizing different events data approaches (WEIS, DON, COPDAB, CREON, and so on) and associated with different universities.

The scientific interest in events data was not wholly for basic scientific research, even at the beginning. A major purpose was to eventually provide a global monitoring system which, much like weather reports, would map, track, and forecast internal and, particularly for many, international conflict, crises, and violence. It was and is with this potential in mind that the Department of Defense has given so much support to event data research, and the Department of State has undertaken its own event data based The Foreign Relations Indicator Project (Burgess and Lawton, 1975: 116).

Towards an eventual monitoring system, events data and methods have been developed to analyze and track Middle Eastern conflict (Azar, 1972; Milstein, 1972; J. McCormick, 1975; Burrowes and Garriga-Pico, 1974); to predict the future behavior of the Soviet Union to South Korea (Choi, 1976); to analyze the assumptions of détente and the implications of U.S.-Soviet arms trends (Rummel, 1976a); to weigh the effect of Israeli reprisals on Arab violence (Blechman, 1972); to determine the causes and conditions of instability in Africa (Copson, 1973; Collins, 1973), Asia (Schubert, 1975), Latin America (Bwy, 1972), and globally (Gurr, 1969); to delineate the patterns of U.S. naval responses to instability and conflict abroad ("United States Naval . . .", 1968); to predict foreign conflict (Azar, 1975; Azar, et al, 1974); to provide an operational crises forecasting system (Andriole, 1976); and to assess political risks for investors (Haendel, et al., 1975).


All the event data projects have collected their data from public sources. WEIS has been mainly from the New York Times Index, DON data from the daily New York Times, COPDAB from several sources, including the Times Index, and CREON form Deadline Data. Projects have also used Keesings Contemporary Archives, Facts on File, as well as regional and other national sources, such as the Times of London.

Of course, two questions immediately follow. How good are these sources? Does one get a different picture from different sources? These questions have been the subject of much research (Azar, 1970; Gamson and Modigliani, 1971; Burrowes, 1974; Hazlewood and West, 1974; Hoggard, 1974; Smith, 1969). In sum, research and analyses have shown that the daily New York Times, or its Index, suffice to capture most events, particularly the more intense and significant ones. It tends to omit events in the middle or lower range, which can be found in regional sources, such as the Middle East Journal or Asian Recorder.

Moreover, assessments relying on frequency counts (such as a comparison of the number of threats by India and South Africa) or employing percentages of events (such as in observing that 30% of international threats are made by the United States) are most prone to error when the data are taken from only a few sources.

The most valid approach to event data--the one least susceptible to source bias--is to assess patterns of events (Hazelwood and West, 1974), especially from multiple sources, including the New York Times (Hoggard, 1974). This is the approach used in some of DON's early studies (Rummel, 1963) and especially in subsequent work I have collected data from a wide variety of media, including regional sources (Project 15). Reliance on many sources and event data patterns is the best possible guarantee against source bias.

One final question. Even if event patterns (components) are focused upon and many sources used, will not censorship, a nation's lack of development, low international interest in a country, and such other possible influences affect the news? I investigated this question systematically for all nations for event data from the New York Times Index, Keesings Contemporary Archives, Facts on Files, New International Yearbook, and Britannica Book of the Year (Rummel, 1963, 1972: 9.3.3), and concluded that patterns in internal and foreign conflict events were independent of such possible effects. Undeniably, censorship, development, and so on, influence and distort the news. But their impact on the patterns of events observed in the news appears random. Conclusions based on these patterns should be valid.

The straight forward result of all these validity studies is that the internal and foreign conflict patterns presented in these volumes (such as those in Table 11.2), or underlying the analysis of manifest conflict, are most likely valid--surely more so than alternative ways (as by frequency counts or percentages) of systematically or subjectively assessing conflict events.


Aside from whether event data collected from the media really reflect current events is the question of data reliability. Can different coders produce the same data? The real world is messy and ambiguous and news reports about them often capture their ambiguity and color them with the particular interpretations of a newspaper, journal, or newsman. After seven years of experience coding events daily, I humbly admit that cold and hard definitions and classifications, no matter how often revised, often do not fit reported events and some thought and analyses by a coder are required before what is reported can be defined as, say a riot, political clash, or nonpolitical clash; as a threat, warning, or protest; as terrorism, guerrilla warfare, or rebellion.

Much research has also been done to assess reliability (Burgess and Lawton, 1975; Taylor and Hudson, 1972; Hermann, 1971; Azar, 1970; Sigler, 1972; Rummel , 1966). The results have shown event data coding, in general, to be highly reliable; different coders can be trained to produce the same data from the same sources, with about 80%-90% accuracy. The stress is on training. Reliability increases sharply as coders develop a feel for their sources and an understanding of the rules and definitions applied.

The most stringent test of reliability and validity combined have been systematic comparisons of event data sets collected by different projects employing different coders and using somewhat different definitions and overlapping sources. Phillips (1972) systematically compared the WEIS and DON event data for 1967 on the conflict between 182 pairs of states and found similar conflict patterns in each (see also Project 5). That is, for either set the conclusions about conflict event patterns would have been much the same.

However, when one works at the more detailed level of correlations between events, different event data sets might yield different conclusions. Chan (n.d.), for example, found that the CREON and WEIS sets had lack of empirical correspondence for U.S., China, Vietnam, and USSR interactions, both in terms of event magnitude and composition.

Again, these assessments of reliability point to the importance of relying on an analysis of event patterns.


There are two major ways of coding political events. One can code events by type, such as done in Table II.1. The event data then consists of the frequency of events of each type, such as the number of coups for Argentina. Such was the WEIS and DON approach, and it is the approach of the CREON project. These codings may differ only in their definitions, their scope (WEIS and CREON cover both cooperative and conflictful external events; DON covers conflictful internal and external events), level of classification, and sources.

The second approach is to classify events and scale (weight and accumulate) them in some manner to reflect their importance or intensity. The event data is then scaled Such is the approach in Azar's COPDAB set, the data developed by Gamson and Modigliani (1971) to analyze the cold war, and in my own later work (Project 15).

There are advantages and problems in both frequency and scaled event data. Frequencies stay close to events, minimize assumptions, and allow the data themselves to show their patterns. But, frequencies give all events of a certain type equal weight and their quantitative analysis is a confession of substantive ignorance. Scaled events enable qualitative insight to play a role and weight events of the same type by their political significance. However, this enables subjective bias and a priori beliefs to influence the results.

Event data analysis has matured scientifically. As a scientific subdiscipline, it is vigorous and its products are remaking our knowledge of internal and external politics, conflict, and violence. As an approach it has proven its worth and established its credibility.


* Scanned from Appendix II in R.J. Rummel, War, Power, Peace, 1979. For full reference to the book and the list of its contents in hypertext, click book. Typographical errors have been corrected, clarifications added, and style updated.

For citations see the Vol. 4: War, Power, Peace REFERENCES

Go to top of document