The Causal Flow between Public Opinion and Policy: Government Responsiveness, Leadership, or Counter Movement?

Abstract

This article examines the causal relationship between public opinion and policy. Does opinion affect policy or is it the other way around? Three hypotheses take centre stage. The responsiveness hypothesis postulates that changes in public opinion lead to subsequent changes in policy in the same direction. The leadership hypothesis reverses the causal arrow and states that a change in policy results in a subsequent change in opinion in the same direction. Finally, the counter hypothesis argues that policy change leads to a subsequent change in opinion in the opposite direction. These propositions are tested with time-series data from the United Kingdom from 1973 to 2006. Strong evidence is presented in support of policy responsiveness to public opinion. However, only conditional results were found for the other two hypotheses. Policy pushes public opinion in the same direction for popular incumbents (leadership), but in the opposite direction for unpopular incumbents (counter movement).

Acknowledgements

A previous version of this paper was presented at the 2011 Joint Sessions of the ECPR in St. Gallen, Switzerland. I would like to thank the participants of the workshop ‘Issue Congruence and Policy Responsiveness in European Governance’ and the convenors, Mark Franklin and Christine Arnold. I am also grateful to Sara Hobolt, Geoff Evans, Steve Fisher, Mark Franklin, Paul Warwick, and Chris Wlezien for comments on previous drafts. All remaining errors are my own.

Notes

 1. For overviews, see Burstein (2003), Weakliem (2003), and Wlezien and Soroka (2007).

 2. For the sake of completeness we note that the actual ‘thermostatic model’ consists simultaneously of two equations, one concerning the effects of opinion on public spending, the other concerning the effects of public spending on opinion (e.g. Soroka and Wlezien 2010). We present these as separate hypotheses here, since they entail different constellations of the policy–opinion nexus, but this does not mean that these hypotheses are contradictory in nature. Quite the contrary, Soroka and Wlezien would argue that this keeps the representative system in equilibrium.

 3. Speech to Small Business Bureau Conference, 8 February 1984, Frimley.

 4. Quoted in The Times, 3 November 1977.

 5. Anthony Downs is notoriously ambivalent on the matter of preference exogeneity. On the same page of An Economic Theory of Democracy, he states that ‘at the beginning of our study we assumed that voters’ tastes are fixed, which means that the voter distribution is given’, but then admits that ‘though parties will move ideologically to adjust to the distribution in some circumstances, they will also attempt to move voters towards their own locations’ (Downs 1957: 140).

 6. To be fair, these adoptions of the thermostatic model differ from the original in terms of the precise model specification. Erikson et al. (2002), Kellstedt (2000), Stevenson (2001) and the present paper include a measure of policy activity on the right-hand side of the equation. In Wlezien's thermostatic model relative public preferences depend on what the public gets and what the public wants. We therefore refer to this hypothesis as a ‘counter movement hypothesis’, since it does not provide an actual test of the thermostatic model as developed by Wlezien and his co-authors. However, in terms of the theoretical foundations of counter movement, all these accounts cover common ground.

 7. On average the Budget Speech is almost 10 times as long as the Queen's Speech, which is likely to contribute to the reliability of position estimates (Laver et al. 2003).

 8. The conservative turn in mass attitudes has also been documented by Sarlvik and Crewe (1983) and Heath et al. (1991).

 9. The swing to the left is also described by Crewe and Searing (1988) and Heath et al. (1991).

10. DeBoef and Keele (2008) illustrate this point with simulated examples in which they compare an Autoregressive Distributed Lag (ADL) model with an Error Correction Model (ECM). The ADL and ECM are functional equivalents and fit the data equally well, yet the ADL uses a dependent variable in level form while the ECMs dependent variable is differenced. Accordingly, the ADL explains 94 per cent of the variance, while the ECM ‘only’ explains 45 per cent.

11. Popularity data are taken from Gallup (pre-1980) and Ipsos-MORI. These polls contain monthly data on party support so the annual mean is a more reliable measure than using one item from the Eurobarometer.