A new approach to the study of partisan effects on social policy

ABSTRACT

Much research has analysed the influence of government partisanship on social policies. This contribution addresses three challenges in that literature. First, it advocates a dynamic linear modelling (DLM) approach well suited to address time-varying partisan effects. Second, the DLM makes it possible to examine partisan effects country-by-country, thereby avoiding the cross-national pooling approach dominating this literature. Third, it offers a new approach to the challenge of choosing the right control variables. Empirically, we demonstrate the approach using data from 1972 to 2011 from 14 Western European countries. The main conclusion is that the relationship between government partisanship and welfare generosity does not appear to vary over time according to any pattern predicted by existing theories. More generally, findings indicate that in the countries and years covered by this study, government partisanship has very little or no association with the five different social policy measures included in the analyses.

KEYWORDS:

• dynamic linear model
• Europe
• government partisanship
• public budgeting
• social welfare benefits

Acknowledgements

The authors would like to thank the three anonymous reviewers for their very helpful comments and input. Many thanks also go to the participants in the conference on Political Budgeting across Europe at Texas A&M University organized by Christian Breunig, Christine Lipsmeyer and Guy Whitten. Finally, we are especially grateful for the help and comments from Carsten Jensen and Jonas Kraft.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Matt W. Loftis is assistant professor of political science at Aarhus University, Denmark.

Peter B. Mortensen is professor of political science at Aarhus University, Denmark.

Notes

1. Schmitt (2015) herself actually reports persistent partisan effects. We return to her study in the methods section in discussing the proper unit of analysis.

2. Similar difficulties with motivating a priori information apply to the Chow test, which can reveal if shifts in effect parameters occur at particular points (Beck 1983; Wood 2000).

3. These have been shown to be algebraically equivalent approaches to modelling dynamic coefficients (Montana et al. 2009).

4. A random process varying over time according to a certain pattern: the probability distribution over next period’s realization is only a function of the current period’s realization.

5. The process we describe informally here is detailed in Online Appendix B. Our estimating equations follow from the derivations in Shumway and Stoffer (2010).

6. Our selection of Western European countries and time frames in individual regressions is limited only by data availability. Ideally we would impute missing values to broaden our data. However, we are doubtful that the assumption of data missing at random applies in these cases. Therefore, we opt to use only existing data.

7. Definitions of right and left parties come from the CWS data (Brady et al. 2014).

Additional information

Funding

This work was supported by Det Frie Forskningsråd [grant number 1327-00091].