Measuring Strength of Evidence

在科学、法律、商业和日常社会交往中，人们试图将自己的信念建立在现有证据的基础上。标准（贝叶斯）理论断言，人们的信念由各种命题的概率组成，例如，特定科学假设是正确的概率，或者被告像被指控的那样有罪的概率。但是，贝叶斯理论在科学和法律领域的应用遇到了强烈而合理的阻力，主要是因为该理论要求人们即使在没有任何证据（先验概率）的情况下也要表达详细的信念。这项研究将发展非贝叶斯理论，其中证据的感知强度将在与概率完全不同的数值尺度上测量。这个测量系统很好地适应了科学和法律，因为在没有证据的情况下，所有的信念都可以被评为零。一项主要的研究活动将是确定将概率和统计信息转换为证据强度测量尺度的公式，从而将为系统地将这种信息与其他种类的证据结合起来提供基础。另一项研究活动将是制定统计和非统计证据的分级标准，以便对任意证据进行比较和评估，就像米尺提供一系列分级长度一样，用来测量其他物体的长度。证据的分级标准如果能够被分析地应用，将复杂的证据分解成部分，每一部分都可以通过与标准的比较来评估，那么这些标准将是最有用的。主要的研究活动之一将是评价这种分析分解所固有的偏差。这项研究将比较两组人对部分证据的评价，一组人知道全部证据，另一组人不知道。如果可能的话，将发现这两个组评估证据相同的条件，也就是说，在分析过程的任何步骤中都不会由于忽略部分证据而产生明显的偏差。这项研究还将测试在哪些条件下会出现强烈的偏见。

In science, in the law, in business, and in everyday social interaction, people attempt to base their beliefs on the available evidence. The standard (Bayesian) theory asserts that people's beliefs consist of probabilities for various propositions, for example, the probability that a particular scientific hypothesis is correct or the probability that an accused person is guilty as charged. But proposed applications of the Bayesian theory in science and in law have encountered strong and quite justifiable resistance, largely because the theory demands that people express detailed beliefs even in the absence of any evidence (prior probabilities). This research will develop non-Bayesian theories in which perceived strength of evidence will be measured on a numerical scale quite different from probability. This measurement system is well adapted to science and the law since, in the absence of evidence, all beliefs can be rated as zero. One main research activity will be to determine formulae for converting probabilistic and statistical information to the strength-of-evidence measurement scale, and thus will provide a basis for systematic integration of such information with other kinds of evidence. Another research activity will be to develop graded standards of statistical and nonstatistical evidence, against which arbitrary items of evidence can be compared and evaluated, much as a meterstick provides a graded series of lengths, against which the lengths of other objects can be measured. The graded standards for evidence will be most useful if they can be applied analytically, decomposing a complex body of evidence into parts, each of which can be evaluated by comparison with the standards. One of the principal research activities will be to evaluate the bias inherent in such analytic decompositions. That research will compare evaluations of partial evidence by two groups of people, those who are aware of the full body of evidence and those who are not. If possible, conditions will be found for which these two groups evaluate evidence identically, i.e., there is no appreciable bias due to parts of the evidence that are ignored in any step of the analytic process. The research will also test conditions under which strong biases are expected.