A PRIMER ON SEQUENCE METHODS
A PRIMER ON SEQUENCE METHODS
复制标题
DOI：
10.1287/orsc.1.4.375
复制
发表时间：
1990-11-01
影响因子：
4.1
通讯作者：
Abbott, Andrew
中科院分区：
文献类型：
--
作者：
Abbott, Andrew
本文考虑分析社会事件序列的技术问题。组织行为中的这种序列的例子包括组织生命周期、创新发展模式和个人的职业轨迹。这里考虑的方法使分析师能够在这样的序列中找到特征模式。形成这些模式的力量可以通过更传统的方法找到。在简短的定义部分之后，本文首先讨论了三种类型的序列问题：（1）关于是否存在典型序列的问题，（2）关于为什么可能存在这种模式的问题，以及（3）关于这种模式的后果的问题。第一类问题的理论基础，这实际上是最重要的，然后考虑。在确定了本文所采取方法的合法性后，本文然后介绍了两个示例性数据集，以供重点讨论。这就提出了序列数据的概念化和测量问题。说明性的情况下，显示的重要性，极端谨慎的构思一个序列来衡量，然后选择它的指标。然后，本文转向适当的方法，考虑他们在几个类别。它首先简要地提到了不采用事件之间的“距离测量”的方法：置换技术，随机（例如，马尔可夫）模型和持续时间的方法。大多数这些并不直接解决序列问题，但可以用来这样做，如果必要的。转向基于事件距离的方法，本文首先考虑测量事件之间的距离的问题：（1）根据经过的时间，（2）根据事件的类别，（3）根据观察到的序列。然后，它认为独特的事件序列（序列中没有事件重复）的方法，提出使用多维尺度，并说明它与医疗机构的数据分析。对于重复事件序列的单独情况，本文讨论了最优匹配方法，该方法计算将一个序列改变为另一个序列所需的单个变换的数量。这些方法说明了音乐家的职业生涯的数据分析。然后，本文简要地考虑了问题的调查结果共同的几个较长的序列（或重复在一个较长的序列）。最后讨论了使用这些类型的方法时所作的假设和需要注意的问题。
This paper considers the technical problem of analyzing sequences of social events. Examples of such sequences from organizational behavior include organizational life cycles, patterns of innovation development, and career tracks of individuals. The methods considered here enable the analyst to find characteristic patterns in such sequences. Forces shaping those patterns can then be found by more conventional methods. After a brief definitional section, the paper begins by discussing three types of sequence questions: (1) questions about whether a typical sequence or sequences exist, (2) questions about why such patterns might exist, and (3) questions about the consequences of such patterns. The theoretical foundations of the first type of question, which is in fact the most important, are then considered. Having established the legitimacy of the approach here taken, the paper then introduces two exemplary datasets with which to focus discussion. These raise the issue of conceptualization and measurement of sequence data. Illustrative cases are presented to show the importance of extreme care in conceiving a sequence to measure and then choosing indicators for it. The paper then turns to methods proper, considering them in several categories. It first briefly mentions methods not employing "distance measures" between events: permutational techniques, stochastic (e.g., Markov) models, and durational methods. Most of these do not directly address sequence questions but can be used to do so if necessary. Turning to the methods based on event distance, the paper first considers the problem of measuring distance between events (1) in terms of elapsed time, (2) in terms of categories of events, and (3) in terms of observed successions. It then considers methods for unique event sequences (sequences in which no events repeat), proposing the use of multidimensional scaling and illustrating it with an analysis of data on medical organizations. For the separate case of repeating event sequences, the paper discusses optimal matching methods, which count the number of individual transformations required to change one sequence into another. These methods are illustrated by an analysis of data on musicians' careers. The paper then briefly considers the problem of findings subsequences common to several longer sequences (or repeated in one longer sequence). It closes with a discussion of assumptions made and caveats required when these types of methods are used.