Studying Determinants of Social Network Size: Comparison of different statistical models

在对社会网规模的估算中,0LS方法的适当性和实际效果目前还缺乏较为系统的考察。本文在对现有统计学文献进行梳理的基础上,初步探讨了在估算社会网规模时OLS方法可能存在的问题及可供选择的替代方案。通过对CGSS 2003数据集的统计分析及计算机模拟分析,本文在经验层面发现,OLS估计结果确实与替代方案(特别是负二项回归模型)的估计结果存在明显的差异。最后,本文建议研究者采用可供替代的建模方式研究社会网络规模的影响因素。

Whether OLS is an appropriate and effective model for studying determinants of social network size has not been scrutinized systematically. Based on review of statistical literature，this article preliminarily explores potential problems of 244 OLS in studying determinants of network size and its alternative—negative binomial regression model. Based on analysis of CGSS 2003 data set as well as simulation， this article finds that there is revealing discrepancy between OLS and negative binomial regression model when they are used to estimate the effects of independent variables on network size. It concludes that negative binomial regression model might be more appropriate for estimating the effects of determinants of network size.

黄荣贵； 桂勇

国家社会科学基金项目(05CSH17);教育部人文社会科学研究规划基金项目(05JA840004)的支持

欢迎阅读《社会学研究》！您是该文第2538位读者！