Aggregation level matters: evidence from french electoral data

作     者：Russo, Luana Beauguitte, Laurent 

作者机构：98 Rue Univ F-75007 Paris France 

出 版 物：《QUALITY & QUANTITY》 (质与量)

年 卷 期：2014年第48卷第2期

页      面：923-938页

核心收录：

学科分类：0303[法学-社会学] 03[法学] 0714[理学-统计学（可授理学、经济学学位）] 

主　　题：Abstention Ecological inference Linear regression model Multicollinearity Parisian agglomeration Political geography 

摘      要：This paper, issued from the Cartelec project, raises a methodological issue regarding ecological inference. Studying abstention in two recent French elections in the Parisian agglomeration, a linear regression is launched on three different levels of aggregation: from the finest one (polling station), through cities, to electoral districts (circonscriptions). Tests on multicollinearity clearly indicate that employing the finest level of aggregation allows more variables to be significant, enhances the explanatory power of the model, and improves the collinearity amongst the dependent variables.