ぎょーむ日誌 2003-08-24

2003 年 08 月 24 日 (日)

• 0830 起床． 本日は北大全学停電日． 朝飯． コーヒー．
• また 35 年前の縞枯山論文の数字の森にさまよいこむ …… 数値のケタに関してはいま使ってるのは それほど悪くはないように見える． しかし2 倍以上のずれならあちこちにある． このあたりは同化モデルあたりから再検討したほうがよさそうだ． ここまではいいとして， やっかいなのは水利用効率なるものがますますわからなくなる． 単純計算からの憶測でいくと， あの標高 2000m の寒いところでしょぼしょぼと生きてる針葉樹どもは， なぜかしら水をひどく効率的に使っている， ということになってしまいそうでですね …… じつは呼吸の見積り (CO₂ 吸収させる何らかの方法を使ったらしい ……詳細書いてない) が過大なんではなかろうか? うーむ．
• というような試行錯誤で日曜日は消費されていくのであった．
• 1800 自宅発北大構内走． 曇天． 日没． 北大内電力回復してるな． 1850 帰宅 体重 70.0kg． おお， 瞬間最低体重とはいえここまで下がるか? ぱいぷ樹木やせ， とか．
• 2040 自宅発． 2050 研究室着．
• サーヴァー機など再起動． ぱいぷ樹木計算再開．
• 2150 研究室発． 2200 帰宅． 晩飯．
• [今日の素読]
  • Salsburg, D. 2001. `` The Lady Tasting Tea -- How statistics revolutionized science in the twentieth century''. Owl Book.
  • Chapter 10. Testing The Goodness Of Fit (continuation)
    • Pearson's goodness of fit test
    • Testing whether the lady can taste a difference in the tea
    • Fisher's use of p-value
  • These problem was one that Karl Pearson recognized early in his career. One of Pearson's great achievements was the creation of the first ``goodness of fit test.'' By comparing the observed to the predicted values, Pearson was able to produce a statistic that tested the goodness of fit. He called his test statistic a ``chi square goodness of fit test.'' He used the Greek chi (χ), since the distribution of this test statistic belonged to a group of his skew distribution that he had designed the chi family. Actually, the test statistic behaved like the square of chi, thus the name ``chi squared.'' Since this is a statistic in Fisher's sense, it has a probability distribution. Pearson proved that the chi square goodness of fit test has a distribution that is the same, regardless of the type of data used. That is, he could tabulate the probability distribution of this statistic and use that same set of tables for every test. The chi square goodness of fit test has a single parameter, which Fisher was to call the ``degrees of freedom.'' In the 1922 paper in which he first criticized Pearson's work, Fisher showed that, for the case comparing two proportions, Pearson had gotten the value of that parameter wrong.
  • Note the expression ``knows how to design an experiment ... that ... will rarely fail to give a significant result.'' This lies at the heart of Fisher's use of significance test. To Fisher, the significance test makes sense only in the context of a sequence of experiments, all aimed at elucidating the effects of specific treatments. Reading through Fisher's applied papers, one is led to believe that he used significance test to come to one of three possible conclusions. If the p-value is very small (usually less than .01), he declares that an effect has been shown. If the p-value is large (usually greater than .20), he declares that, if there is an effect, it is so small that no experiment of this size will be able to detect it. If the p-value lies in between, he discusses how the next experiment should be designed to get a better idea of the effect. Except for the above statement, Fisher was never explicit about how the scientist should interpret a p-value. What seemed to be intuitively clear to Fisher may not be clear to the reader.
  • We will come back to reexamine Fisher's attitude toward significance testing in Chapter 18. It lies at the heart of Fisher's great blunders, his insistence that smoking had not been shown to be harmful to health. But let us leave Fisher's trenchant analysis of the evidence involving smoking and health for the later and turn to 35-year-old Jerzy Neyman in the year 1928.
• [今日の運動]
  • 北大構内走 1800-1845． ストレッチング．
• [今日の食卓]
  • 朝 (0920): パンケイキ．
  • 昼 (1300): パンケイキ．
  • 晩 (2220): 米麦 0.7 合． タマネギ・ジャガイモ・ニンジンのカレー．