Getting the most from political panel data

At the end of May, a small group of presenters gathered at the University of Essex to discuss the state of longitudinal methods in the discipline. Through a series of substantive papers demonstrating the utility of different techniques, a consensus grew that renewed awareness and engagement with longitudinal data can help us make real substantive discoveries – even in questions that appear to be settled with cross-sectional analysis.

Political science has a long history of panel studies – from Lazarfeld’s study of political communication in the 1940 presidential election, through the panel elements form the earliest American National Election Studies, and the tradition of political socialisation studies which follow young people into adulthood. Researchers interested in longitudinal data in Britain and Europe today have many options – as well as the prominent British Election Study Internet Panel, there are household surveys that collect political information over long time periods – Understanding Society and the British Household Panel Survey in the UK, the German Socio-Economic Panel in Germany among others. More recently there are more frequent panels that carry political variables in the form of the LISS panel in the Netherlands, the GIP in Germany or ELIPPS in France.

Nevertheless, it is our belief that this data is frequently not fully exploited, despite the potential insights it offers. For instance, of the 11 published papers currently indexed in Google Scholar as using the “British Election Study Internet Panel”, only two use any kind of panel data method. This is indicative of a tendency to rely on cross-sectional methods, when panel data is available that would improve the inferences that can be made. A further problem was described by one of the papers at the workshop; Fairbrother did a systematic review of multi-level models published in the European Sociological Review, finding that many models used an incorrect (and anti-conservative) specification of the random effects. It is surely only a slight consolation to scholars of politics that these papers were taken from a prominent sociology journal, and not one with ‘political’ in the title.

The aim of the workshop then was to showcase longitudinal data and methods, and to build a network of people interested in promoting them. Two papers showed that longitudinal data can provide evidence that calls into question a long-standing theory of political behaviour, where cross-sectional data in generally supportive. Martin Kroh and Peter Selb look at the question of whether unemployment reduces political participation – in line with a resource model – or not. Using difference-in-difference estimation with propensity score matching to identify a causal effect, they argue that unemployment has both mobilizing and demobilizing effects depending on the person in question. Similarly, James Dennison looks at whether decreases in financial wellbeing decrease (in line with a resource model) or increase (in line with a grievance model) political participation, showing that in fact it depends on people’s partisanship. Those who support the governing party are demobilized, but those who do not are mobilized in line with the grievance model. Looking purely at cross-sectional data, neither of these studies would have accurately identified the full story.

Relying on data from the British Household Panel Study, Anja Neundorf and colleagues used Mixed Latent Markov Models to address another core debate – to what extent partisanship is stable or updated. Their research not only identifies both stable and less-stable partisans – it examines the characteristics of different groups. It shows that New Labour’s supporters were both less stable partisans and demographically different from previous Labour partisans, illustrating the success of New Labour in attracting new supporters. Meanwhile, Jon Mellon presented work with Geoff Evans, which looks how leadership evaluations led to vote intention switching over the course of the British Election Study Panel. The longitudinal perspective in this analysis overcomes the tricky problem of identifying what comes first of leadership evaluations or vote intention. It turns out that both David Cameron and Ed Miliband were assets to their party when it came to retaining voters, but only David Cameron gained new voters for the Conservatives. Nigel Farage on the other hand was good at gaining voters – but very bad at retaining them. Ann Kristin-Kolln and Kees Aarts showcased latent growth curve modelling, using panel data from the Netherlands, and Nicole Martin considered how detrimental (or not as it turns out) panel conditioning is for the measurement of key political attitudes in Understanding Society.

Following substantive discussions of the papers, the conversation turned to the future of political panel data in the UK. Compared to some other European countries, we are not spoilt for choice when it comes to short-term panel measurement of political issues (the BESIP and Understanding Society notwithstanding). Other disciplines have succeeded in making the case that their research is politically and policy relevant, thus securing space for their key questions on panels. Political scientists have been somewhat less successful at this – but as our discussion revealed, there are many reasons that political questions should be considered just as urgent for policy makers as economic or sociological ones. Competing national and cultural identities are likely to stay at the heart of politics for a good time to come, just as responses to major events such as terrorist attacks or economic shocks will be crucial for our elected representatives (and those who wish to influence them) to understand. How low turnout and political disengagement of young people develop and respond to a changing political environment will have big consequences for the shape of future governments.

With the state of the art tools and analysis demonstrated by this workshop, we should not be shy in advocating for political variables in any new UK panel study.

The workshop was generously funded by Understanding Society, the ESRC Centre for Micro Social Change (MISOC) and the Political Studies Association.

List of papers:

Why are the Unemployed Politically Quiescent? Unemployment and Civic Action in a Longitudinal Perspective
Martin Kroh (Humboldt) and Peter Selb (Konstanz)

Financial Well-Being and Voter Turnout
James Dennison (EUI)
Download (Word)

The Random Effects in Multilevel Models: Getting Them Wrong and Getting Them Right
Malcolm Fairbrother (Bristol)
See full paper:

Will Britons vote to leave the European Union? A multi-level dynamic analysis
Harold Clarke (UT, Dallas) and Paul Whiteley (Essex)

Using panel data to estimate the impact of party leaders in the 2015 election
Geoffrey Evans (Oxford) and Jonathan Mellon (Oxford)
Download (PowerPoint)

The Dynamics of Party Identification in Turbulent Times: The Case New Labour in England, 1991-2008
Daniel Stegmueller (Mannheim), Anja Neundorf (Nottingham) and Thomas Scotto (Essex)
Download (pdf)

Satisfaction with democracy and electoral behaviour in the Netherlands: findings from the 2007-14 LISS Panel
Ann-Kristin Kölln (Gothenburg) and Kees Aarts (Twente)
Download (pdf)

The effect of panel conditioning on support for political parties and political interest
Nicole Martin (Essex)
Download (PowerPoint)