The KEDS Project

The Kansas Event Data Project is a ten-year project focused on the development and application of political event data; it is funded by the National Science Foundation and the University of Kansas. The project has three major research concentrations:

Software development for the machine-coding of political event data

The initial focus of the KEDS project was the development of techniques for converting English-language reports of political events into event data. This replaced labor-intensive and error-prone process of human coding with inexpensive, transparent and reproducible automated coding. The centerpiece of this effort is the KEDS computer program, a Macintosh program with a user-friendly interface and a variety of output options. In support of KEDS, we have also produce several programs for filtering texts and aggregating the resulting event data so that it can be used in statistical analysis.

The KEDS program has now been used in about half-a-dozen NSF-funded event data projects, as well as a number of dissertations and other smaller projects.

Production of event data sets

Using the KEDS program, news reports from the Reuters wire service, and the World Events Interaction Survey (WEIS) event coding scheme, we have produced several event data sets that can be downloaded for use in political studies. Our primary focus has been on the Levant -- Egypt, Israel, Jordan, Lebanon, the Palestinians and SyriaÑand until recently we maintained a near-real time data set on this region. Other data sets available at this site include a Levant data set coded with the BCOW coding scheme, a 1979-1997 data set covering the Gulf region, a 1987-1997 set covering Algeria, and pointers to several other data sets covering an assortment of contemporary crisis areas. KEDS coding dictionaries are available for most of these data sets.

Development of early warning methods

For the past four years, most of our research has focused on the development of early warning techniques for political change, primarily using the Levant as a case study. We have experimented with a number of different methods, including factor analysis, discriminant analysis, an assortment of clustering algorithms, and most recently hidden Markov models.

This research was supported in part by a subcontract to National Science Foundation grant SES90-25130 (Data Development in International Relations), by NSF grant SBR 9410023 and by the University of Kansas General Research Allocation Fund 3500-X0-0038.