Abstract
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.
Original language | English |
---|---|
Pages (from-to) | 923-938 |
Number of pages | 16 |
Journal | Quality & Quantity |
Volume | 48 |
Issue number | 2 |
DOIs | |
Publication status | Published - Mar 2014 |
Keywords
- Abstention
- Ecological inference
- Linear regression model
- Multicollinearity
- Parisian agglomeration
- Political geography
- RIDGE-REGRESSION