How governments strategically time welfare state reform legislation: empirical evidence from five European countries

Abstract

Building on studies on the political business cycle, the literature on welfare state retrenchment has argued that governments which cut the welfare state try to avoid blame by implementing painful measures in the beginning of the mandate and expanding benefits as elections approach. In contrast to this linear relationship, this article argues that governments often feel pressured to fulfil (mostly expansionary) campaign promises during the first months in office. Consequently, cutting right away is not what should be expected. Instead, a more nuanced, U-shaped timing trajectory is probable with a period in the beginning characterised by both cuts and fulfilment of expansionary pledges, followed by a period of cutbacks, and finally an expansive phase towards the end of a mandate. This argument is tested on our new original dataset of legislative changes in five European countries – Britain, Denmark, Finland, France and Germany – during the last four decades.

Keywords:

• Electoral pledges
• electioneering
• political business cycle
• welfare state

Notes

Acknowledgements

We want to thank the two reviewers and the editors who helped improve the original manuscript substantially. We also wish to thank the student assistants who helped in collecting and coding the data. They are Malin Grüninger, Kathrin Hartmann, Anja Konanec, Lasse Leipziger, Sami Mustikkamaa, Fabienne Müller, Kristian Nicolaisen and Stefanie Thurm.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 The only study with more mitigated findings is the paper by Tepe and Vanhuysse, who find different effect sizes depending on their model specification and the operationalization of the timing variable (Tepe and Vanhuysse 2010: 1228–9). This is partly due to the fact that the authors cannot pinpoint whether a cutback occurred before or after the election date in an election year (which always takes the value of 0, see footnote 11 in their paper). Our approach using monthly data does not suffer from this problem.

2 The works of Armingeon and Giger (2008), Giger and Nelson (2011), and Schumacher et al. (2013), for instance, code cutbacks as ‘1’ and status quo and expansions as ‘0’. When both status quo and expansion are coded together as ‘0’, years (or election cycles) without any reforms are treated as identical to years (or election cycles) with expansions, making it impossible to assess the independent effect of expansions.

3 Admittedly, legislation also has to pass the legislative process, which may take some time; but politicians who care about their re-election might know these delays and will therefore already include this in their calculation of when the proposal has to be adopted. In any case, government spending is much more difficult to time because it is not only affected by the delays of legislative processes, but also by many other factors, such as demand for benefits (e.g. because of rising unemployment or population ageing), economic growth, or the possibility to actually spend the money that is allocated in a budget.

4 By ‘net change’, we refer to the overall direction of changes in welfare legislations made by a government in a given period. For example, if there were more expansionary changes than cutbacks, the net change would be expansionary and this overall net position, rather than the sheer extent cutbacks or expansions, would inform the voters the general image of the government (on the matter of welfare policies).

5 These are the Danish governments starting in 1973 and 1982, the French election in 1986, the German election in 1983, and the British election in 1983. Hence, electoral pledges seem to reflect the common notion of parties’ positions on social policies.

6 Due to data availability, the time series start in 1970 for UK, 1971 for Denmark, and 1974 for Germany, Finland, and France.

7 The individual events were identified by a team of trained research assistants and a senior researcher on the basis of as many secondary sources as possible and supplementing these with searches in legislative databases when necessary (see Online Appendix, pp. 9–23, for a list of sources). The coding of the instruments and the direction (cutback, no change, expansion) were done by the assistants and subsequently controlled by a senior researcher. In the event the senior researcher did not agree with the original coding, the relevant research assistant and senior researcher discussed the coding decision in detail to reach agreement; however, there were very few such instances. We have double-checked our selection of reforms for 4 countries by contrasting it with studies based on a full search of legislative databases (due to the lack of data, we had to exclude Finland). To do so, we have randomly chosen one year from our sample – the year 1992 – and compared, whether the 9 reforms identified by us (out of 703) were similar to those identified by other researchers in this year. For 1992, we can show that our selection is very close to what other researchers starting from the full selection of legislative databases have found: 6 of the 9 reforms were completely matched, 2 reforms identified by us were part of budget bills (in Germany) and therefore not listed in the respective search category, and only 1 reform (in Denmark) was not identified by us, but in the database of the Comparative Agenda Project (Baumgartner and Jones 1993) (for more details, see Online Appendix, pp. 24–6).

8 In the 2554 coded months, we find 254 months with a positive value of net changes (i.e., more expansion than cutbacks) and 163 months with a negative value of net changes (i.e., more cutbacks than expansions). The data include 2134 months with a value of zero, which indicates either the case where an equal number of cutbacks and expansions (a total of 26 months) or the case where there was no relevant legislative action (a total of 2108 months).

9 We have also tested the models using two other timing measures as independent variable: 1) time (months) until the next scheduled election and 2) time (months) since the last election. The substantive results do not change in both cases (see Online Appendix, pp. 6–7). We believe that the percentage of time elapsed since the last election is a better proxy of our theoretical argument as at least two of the countries (Denmark and the UK) often experience early elections.

10 We always used the electoral term as basic unit of coding and did not code a change of Prime Ministers within an electoral term as a new start of the term, unless the change of Prime Minister did result in a major change of the political camp of the government that can be seen as a ‘new government’. However, there were only two of such ‘changes of government’ without election: Denmark in 1993 and Germany in 1982. Data source for elections and government changes is Parlgov (Döring and Manow 2018).

11 Empirically, the pledge variable varies from -0.48 to 0.32, and there is no case in which the pledge variable takes the value of zero in the dataset.

12 The fact that the distribution of legislative events has a high peak around 0 is a classic finding of studies on legislative behaviour (Baumgartner and Jones 1993; Baumgartner et al. 2006).

13 Hurdle models (or selection models) are not a useful option as we do not have strong theoretical assumptions about the underlying process that leads to legislative events as such. Using hurdle models would, however, imply that we have an argument about the first step, e.g. the probability of an event. Moreover, our dependent variable is not a pure count of events (i.e., the net change variable varies from negative to positive values as it is calculated by taking the difference between the two event counts) and all our theoretical arguments are linked to the overall direction of the legislative changes.

14 A Wooldridge-test (xtserial-command in Stata) does not reject the null hypothesis of no serial correlation. The lagged residuals are not correlated with the residuals.

Additional information

Funding

The project has been generously supported by the Independent Research Fund Denmark (grant no. 4003-00013).

Notes on contributors

Georg Wenzelburger

Georg Wenzelburger is Professor for Policy Analysis and Political Economy in the Department of Political Science at the University of Kaiserslautern. His research focuses on comparative public policy analysis with an emphasis on welfare state reforms, law and order policies and the influence of political parties. He has published in journals including the British Journal of Political Science, the European Journal of Political Research and the Journal of European Public Policy. [[email protected]]

Carsten Jensen

Carsten Jensen is Professor at the Department of Political Science at Aarhus University. His research interests are focussed with comparative politics and political behavior, mostly with a focus on the welfare state. He has published in the American Journal of Political Science, the British Journal of Political Science, Comparative Political Studies and the European Journal of Political Research. [[email protected]]

Seonghui Lee

Seonghui Lee is Lecturer in Comparative Politics at the University of Essex. Her research on comparative politics and political behaviour has appeared in journals such as British Journal of Political Science, Comparative Political Studies, Electoral Studies and Political Science Research and Methods. [[email protected]]

Christoph Arndt

Christoph Arndt is Lecturer in Comparative Politics in the Department of Politics and IR at the University of Reading. His main research interests are voting behaviour, party politics and the political sociology of the welfare state. His research has been published in the British Journal of Political Science, the European Journal of Political Research, the European Sociological Review and the Journal of European Social Policy among others. [[email protected]]