![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
This paper focuses on the analysis of determinants of public education spending in OECD countries. It starts out by reviewing and replicating the model presented by Castles (Citation1989, Citation1998), finding that only a few of his explanatory variables remain significant in a pooled time-series framework. I present an alternative model that contains socio-economic, political, and institutional variables: the level of economic development, the magnitude of demographic demand, the constitutional veto structure, the level of public social expenditures, the degree of tax-revenue decentralization as well as government participation of conservative parties.
Notes
1. The research for this article was supported by funding from the German Research Foundation (DFG) and conducted as part of the research project ‘Education Spending in International Comparison’ under the direction of Professor Manfred G. Schmidt of the University of Heidelberg, Germany. I am very grateful for comments and support from Professor Schmidt and from my colleagues Rita Nikolai and Frieder Wolf. I would also like to thank Uwe Wagschal, Herbert Obinger, Nico Siegel, Gunther Hega, Bernhard Kittel, Jens Hainmüller and Frank Castles and the participants of the research workshop at the Institute for Political Science in Heidelberg, and the participants and audience of the panel on ‘Determinants of Human Capital Investment’ at the MPSA Conference, Chicago, 2006 for helpful comments and suggestions.
2. The (cross-sectional) variation coefficient of education spending is only slightly lower (0.17) than the coefficient for social spending (0.2).
3. Of course, the relationship between education spending and economic development is reciprocal. As studies have shown, investment in education is beneficial for economic development (Nijkamp and Poot Citation2004).
4. In this regard, I support Castles' approach of using cross-sectional regressions with long-term average values for partisan government participation.
5. (The following is based on Fox Citation1997: ch.16.1.) The idea of the bootstrap is to treat the sample as if it were the whole population. For n number of times (in our case: 1,000), re-samples are drawn with replacement from the original sample. This means that a given country can appear more than once in these new samples. The statistical procedure in question, in our case simple linear regression, is performed on each of the new samples and the estimates obtained are saved in a matrix. The final bootstrapped estimates are average values of all the saved estimates in the matrix.
6. Austria, Australia, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Italy, Ireland, Japan, New Zealand, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, the UK, the USA.
7. The inclusion of country fixed-effects in model specifications with a lagged dependent variable introduces a further bias on the coefficient estimates via the LDV (Kittel and Winner Citation2002: 18).
8. Of course, this is not true for public spending on higher education in Sweden, which has increased since the 1990s.
9. Predictions are based on models using levels of public spending instead of predicted change and cannot be derived from the results in directly. More detailed regression results on request and in Busemeyer Citation(2006).
10. Regression results can be provided on demand.
11. Alternatively, regressions were run with the Schmidt veto index (Schmidt Citation2000: 353). In the end, I preferred the Huber, Ragin and Stephens index because it focuses on the institutional context, while Schmidt's index also includes indicators of political or societal veto power (for example, coalition government or consociationalism).
12. When country fixed-effects are included, the veto index loses statistical significance because it does not change much over time and the country fixed-effects dominate its impact. In the sub-period models (models 7 and 8), the cross-sectional variation of the veto index is not large enough to achieve statistical significance at conventional levels because of the shorter time frame.
13. Owing to non-stationarity problems, I included change in public social spending as an independent variable instead of levels of spending. Furthermore, the association between education and social spending suffers from problems of endogeneity, therefore social spending is not included in the base model (model 1).
14. This is done by including the veto index in the regression model.
15. When using moving average values for government participation of rightist parties, the z-value of the coefficient estimate remains essentially the same (−1.18).
16. However, we might be observing a statistical artefact: the veto index changes little over time and the brief time period of the 1990s does not contain enough cross-sectional variation to yield statistically significant estimates.
17. A similar graph, with reversed signs, can be drawn for the conservative party family.
18. The results are similar when I use an interactive term between Christian democratic government participation and a dummy variable for Germany, Italy, and Austria instead of the general government participation variable.