A METHOD OF MATCHING GROUPS FOR EXPERIMENT WITH NO LOSS OF POPULATION
A METHOD OF MATCHING GROUPS FOR EXPERIMENT WITH NO LOSS OF POPULATION
复制标题
DOI：
10.1080/00220671.1941.10881036
复制
发表时间：
1941-04-01
影响因子：
2.1
通讯作者：
Peters, Charles C.
中科院分区：
文献类型：
--
作者：
Peters, Charles C.
在所有良好的对照实验中，平行组都是匹配的，以便在可行的情况下，除了实验因素的差异效应之外，确保匹配组的进展前景相同。但是，这样的群体平衡几乎肯定会迫使一个或两个群体中的一些成员被淘汰，从而减少实验群体，从而降低可靠性。当我们试图同时根据几个标准来匹配群体时，人口的损失可能特别大，特别是如果我们坚持相当接近的平等。由于实验通常是在通常的教室规模的小组中进行的，配对后剩下的人数很可能是灾难性的小。在本文中，我将阐述和说明一种回归方法，该方法给出了完美配对的效果，而不会损失实验组或控制组的任何成员（当然，除非存在这样的异常偏离，因为它们不属于抽样的总体，所以无论如何都会被消除）。这里所提出的方法并不难应用，尽管读者最初可能会得到相反的印象。该方法同样适用于单一标准或一组标准的匹配。我将用我的一个研究生欧文·J·贝勒（Irvin J. Bearer）的实验数据来说明这种方法，他的实验是关于在一年级拉丁语教学中引入丰富的”通用语言”材料的效果。实验组的28名成员在正常的拉丁语课程中按时学习这些材料，而控制组的22名成员则将所有时间用于拉丁语的直接学习。Bearer先生从四个方面进行了测量
IN ALL good controlled experimentation parallel groups are matched so as to secure equal prospect of progress for the matched groups apart from the differential effect of the experimental factor, so far as that is feasible. But such balancing of groups is nearly certain to compel the elimination of some members from one or both groups and thus to reduce the experimental population, with consequent decrease of reliability. The loss of population is likely to be especially great when we attempt to match groups simultaneously on several criteria, particularly if we insist upon pretty close equality. Since experiments are often conducted with groups of the usual classroom size, the numbers remaining after matching are likely to be disastrously small.In this paper I shall set forth and illustrate a regression method which gives the effect of perfect pairing without the loss of any members from either the experimental or the control groups (unless, of course, there are such abnormal deviates as would be eliminated anyway on the ground that they do not belong to the population sampled). The method as here presented is not difficult to apply, in spite of a contrary impression which the reader might at first receive. l The method is equally applicable to matching on a single criterion or on a battery of criteria. I shall illustrate the technique with data from an experiment by Irvin J. Bearer, one of my graduate students, on the effect of introducing enriching" general language" material into the teaching of first-year Latin. The 28 members of the experimental group were taught such material on time taken off of the regular Latin period, while the 22 members of the control group gave all of their time to the straight study of Latin. Mr. Bearer made measurements of four aspects