Trial of earthquake prediction in Japan and a statistical test of time-shift

Earthquake prediction was practiced in Japan to examine the hypothesis that “a pair of earthquakes with similar magnitudes may be a signal of an impending larger earthquake”. In the present study, predictions were announced with expected probabilities of 20–30% (rank A) or 10–20% (rank B). In 2001–2002, excepting the Ogasawara region, 26 and 6 cases among 61 and 30 predictions of ranks A and B, respectively, were successful. Based on a statistical test of time-shift, i.e., one-year shift in this paper, and averaged activity in 1990–1999, the success rate of 43% for rank A was shown to be greater than that expected by chance with a confidence level more than 99%. The success rate of 20% for rank B gave a corresponding confidence level of only about 40%, suggesting that the predictions of rank B were not confident in this period. According to the results, a statistical test of time-shift was found to be useful to evaluate the significance of prediction methods of this type.