Indexed on: 01 Mar '73Published on: 01 Mar '73Published in: Quality & quantity
Much of statistical theory and most of statistical inference has been developed to meet situations which fundamentally differ from the problems one has to face in international conflict data analysis. True scores, error scores, and observed scores of the dependent variable are likely to be asymmetrically distributed. Independent variables may be asymmetrically distributed as well. There are outlayers in true scores, in error scores, in observed scores. Measurement error may produce an observed outlayer or suppress a true outlayer.We should expect few and large measurement errors instead of, or in addition to, many small errors. Often discontinuous indicators are supposed to represent continuous theoretical variables. Measurement errors are likely to be correlated with true scores or other variables of theoretical interest. There is little reason to expect the random measurement error assumption to hold. In addition, lack of theory makes for gross conceptual errors. Some actual data are supplied in order to demonstrate the undesirable properties of international conflict data and the near-impossibility of testing even simple propositions in a rigorous fashion. International conflict data analysis often comes dangerously close to testing probabilistic propositions with a handful of events. Rarely, if ever, can conflict data be regarded as a random sample from a well-defined universe of data.For all these reasons, standard procedures of statistical inference are generally not applicable in international conflict data analysis. We do not know how to go beyond descriptive statistics with this kind of data. The only way to generate confidence in findings may be extensive and systematic replication. As results of replication studies may be somewhat inconsistent, even replication may be more efficient in reminding us of our lack of knowledge than in establishing it beyond reasonable doubt.