Grusky D B, Charles M
Department of Sociology, Stanford University, CA 94305, USA.
Demography. 1998 Nov;35(4):497-504.
We review the logic underlying margin-free analyses of sex segregation arrays. In the course of our review, we show that the Karmel-MacLachlan decomposition does not live up to its margin-free billing, as the index upon which it rests, Ip, is itself margin-sensitive. Moreover, because the implicit individualism of D is necessarily inconsistent with margin-free analysis, the field would do well to abandon not merely the Karmel-MacLachlan decomposition but all related efforts to purge marginal dependencies from D-inspired measures. The criticisms that Watts (1998) levels against our log-multiplicative approach are likewise unconvincing. We demonstrate that our preferred models pass the test of organizational equivalence, that the "problem" of zero cells can be solved by applying well-developed methods for ransacking incomplete or sparse tables, and that simple log-multiplicative models can be readily devised to analyze disaggregate arrays. We illustrate these conclusions by analyzing a new cross-national archive of detailed segregation data.
我们回顾了性别隔离阵列无边缘分析背后的逻辑。在回顾过程中，我们表明卡梅尔 - 麦克拉克伦分解并不符合其无边缘的说法，因为它所依赖的指数Ip本身对边缘敏感。此外，由于D隐含的个人主义必然与无边缘分析不一致，该领域不仅应摒弃卡梅尔 - 麦克拉克伦分解，还应放弃所有相关的努力，即从受D启发的度量中消除边缘依赖性。瓦茨（1998年）对我们的对数乘法方法提出的批评同样缺乏说服力。我们证明了我们首选的模型通过了组织等效性测试，零单元格的“问题”可以通过应用成熟的方法来梳理不完整或稀疏表格来解决，并且可以很容易地设计简单的对数乘法模型来分析分类阵列。我们通过分析一个新的详细隔离数据跨国档案来说明这些结论。
