统计模型的“不确定性”问题与倾向值方法
发布时间：2018-09-10 19:51
【摘要】：量化社会学研究往往基于特定的统计模型展开。近十几年来日益流行的倾向值方法也不例外,其在实施过程中需要同时拟合估计倾向值得分的"倾向值模型"与估计因果关系的"结果模型"。然而,无论是其模型形式还是系数估计,统计模型本身都具有不可忽视的"不确定性"问题。本研究在倾向值分析方法的框架下,系统梳理和阐释了模型形式不确定性与模型系数不确定性的内涵及其处理方法。通过分析"蒙特卡洛模拟"数据与经验调查数据,本文展示了在使用倾向值方法进行因果估计的过程中,研究者如何通过"贝叶斯平均法"进行多个备选倾向值模型的选择,以及如何通过联合估计解决倾向值模型与估计模型中的系数不确定性问题。本文的研究也表明,在考虑倾向值估计过程的不确定性之后,结果模型中对于因果关系的估计呈现更小的置信区间和更高的统计效率。
[Abstract]:Quantitative sociological studies are often based on specific statistical models. The tendency value method, which has become more and more popular in recent years, is no exception. In the process of implementation, it is necessary to fit the "tendency value model" and the "result model" of estimating causality. However, no matter its model form or coefficient estimation, the statistical model itself has the "uncertainty" problem which can not be ignored. Under the framework of the tendency value analysis method, the connotation and treatment methods of the model formal uncertainty and the model coefficient uncertainty are systematically summarized and explained in this study. Based on the analysis of Monte Carlo Simulation data and empirical survey data, this paper shows how to select multiple alternative tendency value models by Bayesian average method in the process of causality estimation using tendency value method. And how to solve the problem of coefficient uncertainty in the model and estimation model by joint estimation. It is also shown that the estimation of causality in the result model presents a smaller confidence interval and a higher statistical efficiency after considering the uncertainty of the estimation process of the tendency value.
【作者单位】： 复旦大学社会学系;
【基金】：国家社科基金青年项目(15CSH030) 上海市教育委员会科研创新项目(15ZS001) 复旦大学“卓学人才计划”项目的支持~~
【分类号】：C8
本文编号：2235443
[Abstract]:Quantitative sociological studies are often based on specific statistical models. The tendency value method, which has become more and more popular in recent years, is no exception. In the process of implementation, it is necessary to fit the "tendency value model" and the "result model" of estimating causality. However, no matter its model form or coefficient estimation, the statistical model itself has the "uncertainty" problem which can not be ignored. Under the framework of the tendency value analysis method, the connotation and treatment methods of the model formal uncertainty and the model coefficient uncertainty are systematically summarized and explained in this study. Based on the analysis of Monte Carlo Simulation data and empirical survey data, this paper shows how to select multiple alternative tendency value models by Bayesian average method in the process of causality estimation using tendency value method. And how to solve the problem of coefficient uncertainty in the model and estimation model by joint estimation. It is also shown that the estimation of causality in the result model presents a smaller confidence interval and a higher statistical efficiency after considering the uncertainty of the estimation process of the tendency value.
【作者单位】： 复旦大学社会学系;
【基金】：国家社科基金青年项目(15CSH030) 上海市教育委员会科研创新项目(15ZS001) 复旦大学“卓学人才计划”项目的支持~~
【分类号】：C8
【相似文献】
相关期刊论文 前5条
1 林小涵;;测测看什么职业最适合你[J];农村百事通;2009年11期
2 ;测试适应你的最佳职业[J];科海故事博览(智慧文摘);2009年01期
3 周振;牛立腾;孔祥智;;户籍歧视与城乡劳动力工资差异——基于倾向值的匹配分析[J];区域经济评论;2014年04期
4 二十一;;你最适合哪种职业?[J];时代青年(月读);2008年04期
5 ;[J];;年期
相关硕士学位论文 前3条
1 申晓晔;基于语义理解的Web新闻事件倾向性分析[D];西安电子科技大学;2009年
2 王飞帆;基于倾向值方法的数据分析技术及在医疗服务中的应用[D];浙江大学;2016年
3 唐果;BBS主观倾向分析[D];西南大学;2010年
，本文编号：2235443
本文链接：https://www.wllwen.com/shekelunwen/shgj/2235443.html
教材专著
