Identifying Latent Policy Dimensions from Public Opinion Data: An Inductive Approach

Abstract

A common way of classifying policy dimensions is by means of a two-dimensional model, based on one economic (Left/Right) dimension and one liberal–conservative dimension that Marks et al. (2006) refer to as Tan/Gal. However, as Lipet and Rokkan (1967) note, voter alignments developed in different ways in different European countries, so the dimensions that define the respective policy spaces may vary from country to country. In this paper, I test the two-dimensional model on opinion data from four European territories using confirmatory factor analysis (CFA) and go on to develop alternative “better fit” models using a combination of exploratory factor analysis and CFA. I find that the two-dimensional model is appropriate in some but not all the cases and suggest that a more context-sensitive approach is needed to identify the dimensionality of the policy space.

Notes

1 Datasets generated by both Greek VAAs (i.e. May and June) are included in the analysis. This is because issue statements relating to the economic crisis were rather over-represented in the May questionnaire at the expense of cultural items. While the June questionnaire contained a better balance of items, the relatively low number of users filling in the questionnaire meant that rather few supporters of the smallest parties (approximately 100) could be identified for mapping (see below). Inclusion of analyses from both datasets provides an extra test of validity.

2 See Codebook: 2010 Chapel Hill Expert Survey at http://www.unc.edu/~gwmarks/assets/data/pp/2010_CHES_expert_level_codebook.pdf, accessed 19 November 2013.

3 The fact that the number of policy categories is equal to the number of issue statements within the VAA is coincidental. Each category embraces a policy theme, into which typically several issue statements from different VAAs will fall.

4 Those garnering more than 3% of the vote. However, supporters of the mainly ethnic Turkish party in Bulgaria, the Movement for Rights and Freedoms, which gained more than 3% of the vote, is not included because insufficient respondents from their mainly rural heartlands in the South and Northeast completed the VAA.

5 Density maps are generated in such a way that each contour line encloses areas in which the density of users is greater than outside. The 50% contour line is identified using the two-dimensional Kernel density estimation function in the r-package MASS (Venables and Ripley 2002).

6 The 2010 Chapel Hill Survey did not include codings of three new or small parties: the Independent Greeks (ANEL), the Democratic Left (DIMAR) and Golden dawn (XA).