著者

増山 元三郎

出版者

一般社団法人 日本温泉気候物理医学会

雑誌

日本温泉気候学会雑誌 (ISSN:03694240)

巻号頁・発行日

vol.8, no.1, pp.8-12, 1942

---

The present author has critisized the n-method used by the German school of bioclimatologists. The n-method is not a perfect statistical method, for it may give an absurd result. The author gives here an example from the Fisher-Yates' table of random sampling numbers. Von Schelling has given a significant test for the n-method, but his schema of urnes and balls is not applicable in this case without reservations. He considers only the difference between the maximum value anti the minimum value of total frequencies scattered in the R cells in the sample, where these extremes may not correspond to those of the population in the R ordered cells. He does not distinguish the measure of correlation itself from that of the mode of correlation and does not consider the probability law of the qualitative event in question. The author proposes here a new method called “the temporal m-method” for the investigation of the measure of correlation between a temporal qualitative event A (t) and a temporal quantitative or qualitative event B (t) based on the analysis of variance proposed by R. A. Fisher. The main idea is as follows: We introduce a parameter m in B (t) and denote the measure of B (t) by y (m; t), where m means the time interval between the time of the appearance of the event A and the time of observation of the event B. Then we study the probability distribution of y (m; t) considering m as a variable and t as a parameter. We apply the test of significance in two way for the quantitative event B, viz. the test of variation in sub-sets of means y (m) and that of linearity of regression of y on m (not regression of m on y). Here, of course, these two tests must be used with reservations when we can not consider the variance as a random sampling one of the variance of population.