Does Blocking Reduce Attrition Bias?
Does Blocking Reduce Attrition Bias?
复制标题
阻止会减少磨损偏差吗？
DOI：
--
复制
发表时间：
2011
期刊：
影响因子：
--
通讯作者：
T. Dunning
中科院分区：
文献类型：
--
作者：
T. Dunning
一些政治学家最近在实验研究中引起了人们对封锁优点的关注。只要不出现磨损和其他一些对内部有效性的威胁，实验就可以对实验研究组的平均因果效应进行公正的估计。然而，估计器可能或多或少精确。这就是分组的用武之地。在这里，单位被分组为层或块，然后随机分配到这些块内的处理和控制条件。例如，实验单位可以被分组为具有相似收入、过去投票历史或可以预测结果的其他变量值的对。在这种“配对”设计中，每对中的一个成员被随机分配到治疗组，而另一个则被随机分配到对照组。正如 Moore (2010) 和其他人指出的那样，阻塞可能是一种有价值的策略，尤其是在较小的实验中。当块内单元相对同质（相对于结果）且块间单元相对异构时，块是最有益的。因此，如果研究人员能够识别出对结果有良好预测作用的变量，那么在随机化之前封锁单位可能会提高治疗效果估计器的精度。
Several political scientists have recently drawn attention to the merits of blocking in experimental studies. Experiments allow for unbiased estimation of average causal effects for the experimental study group, as long as attrition and some other threats to internal validity do not arise. However, estimators may be more or less precise. This is where blocking comes in. Here, units are grouped into strata, or blocks, and then randomized to treatment and control conditions within these blocks. For instance, experimental units may be grouped into pairs with similar incomes, past voting histories, or values of other variables that may predict an outcome. In such “matched pair” designs, one member of each pair is randomly assigned to treatment, while the other is randomly assigned to control. As Moore (2010) and others have pointed out, blocking can be a worthwhile strategy, especially in smaller experiments. Blocking is most beneficial when units are relatively homogeneous (with respect to the outcome) within blocks and heterogeneous across blocks. Thus, if investigators can identify variables that are good predictors of the outcome, blocking units before randomization may increase the precision of treatment effect estimators.