The UN Dimension Revisited: A Critique and Reassessment of Wilkenfeld and Brecher

Abstract

Wilkenfeld and Brecher (1984: 50, 52) investigate two general hypotheses regarding the circumstances and the consequences of UN activity in international crises: (a) the greater the seriousness of crises, the greater the extent of activity; (b) the greater the extent of the activity, the more satisfactory are certain outcomes. The authors test their hypotheses by cross-tabulating their measure of UN activity with indicators of crisis seriousness (severity) and crisis outcome. They then calculate a X2 statistic for each cross-tabulation. This statistic, however, indicates only that there is some general unspecified relationship in a sample-a point acknowledged by Wilkenfeld and Brecher (1984: 66)-and not that a directional relationship exists. Statistics such as Guttman's lambda for nominal data or Somers' d for ordinal data would have been more appropriate. Wilkenfeld and Brecher (1984: 56-58) then conduct post hoc tests of their hypotheses by seeing if, for instance, the plurality of cases of high-level UN activity coincide with the most severe forms of violence. The authors implicitly test a priori predictions derived from their hypotheses. Simply looking at the conditional probabilities of activity for each level of the indicators of seriousness is inadequate because it does not guard against the natural null hypothesis of a disproportionately large number of cases in the cells of no activity or low-level activity with high severity. Consider, by way of example, the authors' (1984: 56) claim that the statistic of 47 per cent of the cases of preeminent violence leading to high-level activity 'proves' hypothesis (a). They do not account for the 53 percent of the cases of preeminent violence that lead to no activity or low-level activity. The authors cannot show that preeminent violence, no violence or low-level activity can be safely excluded from consideration; they do not show that their prediction error is tolerably low. Thus, Wilkenfeld and Brecher do not truly test their hypotheses. A proper test must take into account that their predictions are made a priori and must also account for prediction error. Prediction logic (Hildebrand et al., 1977) fulfills these requirements.