中華心理學刊 民 69, 22卷, 一期, 33-40
Chinese Journal of Psychology 1980, Vol. 22, No. 1, 33-40
黃榮村（國立臺灣大學）；黃明宏（國立臺灣大學）
摘要
在不確定的狀況下，人對事物的判斷往往趨於保守。 W.Edwards 及其同事
首開此類研究先河，並認為貝氏統計 (Bayesian Statistics) 是處理人類判斷行爲的最佳工具。依據貝氏理論，事後機率祗受兩個参數所決定，一爲群體中兩類不同事件的比值，一篇樣本中兩類不同事件的差值（個數差），以往的實驗即以該事後機率的理論值與人類的實際判斷值，求其差距，以定其保守性的高低。本文認爲在以往文獻中一向未予探討的樣本中兩類不同事件的比率性，在人類保守性的消減過程中，也具有影響力，因此提出部份數據以驗證該一想法。實驗結果驗證了以往文獻中一再强調的個數差效果，亦即樣本中兩類不同事件的個數差愈大則保守性愈高。
同時也顯示出在相同的個敷差下，不同的個數比會顯著的影響保守性的大小，亦即個數比愈大則保守性的消減愈大。具氏理論一向被認為是處理人類判斷行爲中，很有效的一個理論模型，但因個數比在事後機率判斷中所具有的數值特性及在本實驗中所表現的顯著效果，具氏理論應考慮個數比所造成的影響以求完備。本研究同時探討團體的互相參與決策，是否有助於消減保守性，但實驗結果並不支持該一想法，本文也提出一可能的解釋。
Parametric Proporties and Reduction in Human Conservatism
Jong-Tsun Huang(National Taiwan University); Ming-Horn Huang(National Taiwan University)
Abstract
Men are conservative information processors. Whatever the merits and demerits of a built-in tendency to conservatism in information processing in daily life, such a tendency is clearly a hindrance to human effectiveness in information-processing systems. Many attempts have been focused to delineate the parameters inherent in human conservatism in a Bayesian theory context. Bayesian statistics has long been considered powerful in predicting probability revision processes.
Previous experiments in tackling subjective probability phenomena, when compared with theoretical Bayesian posterior probability, have consistently found the increasing underestimation of the occurrence likelihood of theoretically highly probable events.
The present experiment investigates a long-ignored parameter, ratio of the two different events in the presented datum or sample, and tries to assess its effect on the reduction of human conservatism. Experimental results indicate that the amount of human conservatism increases as a function of increased difference between the two different events in the sample, which are in line with those of the previous experiments. The effect of ratio parameter is first demonstrated in the present experiment. In the same difference level, the larger the ratio in the sample the more reduction in human conservatism. The clearly demonstrated effect of ratio parameter in the human probability revision experiment offers a way to reevaluate the comprehensiveness of Bayesian theory in describing human inference behavior.
