Duncan Greg J
University of California, Irvine, Department of Education, 2001 Berkeley Place, Irvine, CA 92697-5500, USA.
Demography. 2008 Nov;45(4):763-84. doi: 10.1353/dem.0.0031.
Demography's population perspective, and the sampling methods that help produce it, are powerful but underutilized research tools. The first half of this article makes the case for more vigorous promotion of a population perspective throughout the sciences. It briefly reviews the basic elements of population sampling and then provides examples from both developed and developing countries of how population sampling can enrich random-assignment policy experiments, multisite studies, and qualitative research. At the same time, an ill-considered application of a population perspective to the problem of causal inference can hinder social and behavioral science. The second half of the article describes the "slippery slope" by which some demographic studies slide from providing a highly useful description about the population to using regressions to estimate causal models for that population. It then suggests that causal modeling is sometimes well served by a highly selective look at small subsets of a population with interesting variability in independent variables of interest. A robust understanding of causal effects, however, rests on convergence between selective and population-wide perspectives.
人口统计学的人口视角以及有助于得出该视角的抽样方法，是强大但未得到充分利用的研究工具。本文前半部分阐述了在整个科学界更积极地推广人口视角的理由。它简要回顾了人口抽样的基本要素，然后列举了发达国家和发展中国家的例子，说明人口抽样如何丰富随机分配政策实验、多地点研究和定性研究。与此同时，对人口视角在因果推断问题上的不当应用可能会阻碍社会和行为科学的发展。文章后半部分描述了一些人口统计学研究从提供关于总体的非常有用的描述滑向使用回归来估计该总体因果模型的“滑坡”现象。然后它指出，有时通过高度选择性地审视总体中在感兴趣的自变量方面具有有趣变异性的小子集，因果建模能得到很好的效果。然而，对因果效应的稳健理解依赖于选择性视角和总体视角之间的趋同。
