In its June 2016 referendum, the UK voted to leave the EU. One clear pattern in the returns was the positive correlation between age and the probability of voting Leave – 25% of those aged 18-24 voted Leave, compared to more than 56% of those 50 and over (Figure 1).
Figure 1 Referendum breakdown by age
How should we interpret this pattern? Does it mean that voters become more Eurosceptic as they age? Or does it indicate that recent cohorts are more pro-European and likely to stay that way? If cohort effects dominate, can they be explained by changes in the characteristics of different generations, such as their education, in ways that enable us to reliably anticipate how the attitudes of future cohorts will evolve?
These interpretations have very different implications for the development of public opinion. If the observed pattern reflects seasoning, then the electorate is likely to become more Eurosceptic as it ages. If the pattern reflects a cohort effect, on the other hand, the UK will eventually become more Europhile as recent pro-European cohorts supplant their Eurosceptical forebears.
Data and methods
Disentangling seasoning and cohort effects is not straightforward when there also exist electorate-wide swings in sentiment. Given the available data, one can’t observe the same cohorts at two different ages at the same time, or at the same age at two different times. Put it another way, if one knows the age of a voter or survey respondent and when the vote or survey took place, one can back out when the voter or survey respondent was born. Age, cohort, and period effects are perfectly collinear, in other words.
Researchers therefore must make assumptions about the form of age, period, and cohort (APC) effects. Some authors have attempted to proxy for period effects with observable variables such as GDP growth, for example, backing out the associated age and cohort effects on this basis (e.g. Tilley and Evans 2014, Dohmen et al. 2017).
For our analysis, we assemble panel and repeated cross-section datasets of UK sentiment toward the EU from the British Election Survey, corroborating our findings using the British Social Attitudes and European Social Surveys (Eichengreen et al. 2018). In all we have more than 100,000 observations spanning the period 1963-2017.
We separately identify age, period, and cohort effects by adapting the Kalman filter to a panel context. Specifically, we assume that age and cohort effects evolve relatively smoothly, while this is not necessarily the case of period effects.
With this approach, we establish five main findings.
First, both ageing and cohort effects are present in the data.
- Second, UK voters grow more Eurosceptical as they age. This is consistent with studies showing that individuals as they age become less ‘open’ or more ‘conservative’ (e.g. Donellan and Lucas 2008).
- Third, the most UK recent cohorts are more pro-EU than their immediate (baby boomer) predecessors. This has been referred to, variously, as the Ryanair, Bologna Process, and Italian barista effect. However, the earliest cohorts, which lived through WWII as adults, are also relatively pro-EU. Thus, the attitudes of cohorts trace out a U-shaped pattern.
- Fourth, education is positively associated with pro-Europeanism. Because average educational attainment has increased noticeably over the period, this effect is important. Most of the pro-Europeanism of relatively recent cohorts is accounted for by greater educational attainment.
- Fifth, there have also been large nationwide swings in sentiment toward the EU over time that have little do with seasoning or cohort effects.
Figure 2 shows how the age distribution of the electorate is forecast to change. Population ageing is forecast to continue over coming generations. Given our estimates, this is likely to have a negative effect on aggregate sentiment towards the EU.
Figure 2 Age distribution of the UK population aged 18-80
In terms of the attitudes of cohorts, a neutral assumption would hold future cohort effects at the level of current young cohorts. Our finding that a large part of cohort effects is accounted for by differences in education levels, which are unlikely to change radically, lends plausibility to this assumption.
Figure 3 shows the contributions of age and cohort effects to average sentiment. The solid lines show 1971-population-weighted averages, while the dotted lines show voting-population-weighted averages, where we obtain the age-turnout gradient from the 2016 British Social Attitudes Survey.
The figure shows that cohort effects have affected sentiment towards the EU over the past half-century in increasingly negative ways, as the relatively pro-European pre-war generation has been progressively replaced by more Eurosceptic baby boomers. This contribution is now about to turn positive, although it will be partly offset by an increasingly negative contribution from the ageing effect. The net impact of these two effects has been to reduce pro-European sentiment by about three percentage points over the past half-century, a change that will be reversed by the middle of this century.
Figure 3 Time series of estimated age and cohort effect contributions
In Figure 4 we illustrate the contribution of time effects. We linearly interpolate the estimates for survey years to obtain an annual dataset, fit a second-order autoregressive model to the interpolated data, simulate draws from this series, and add the 5thand 95thpercentiles to the baseline forecast (denoted ‘low time effect’ and ‘high time effect’ in the figure). If past volatility is a guide to future volatility, then uncertainty about these effects dominates the predicted changes in the age and cohort effects.
Figure 4 Past uncertainty and future volatility in the change in weighted average yitdue to sampling variability, model uncertainty, and volatility of time effects
In the UK’s 2016 Brexit referendum, young voters were more likely than their elders to support remaining in the EU. We have shown that this pattern reflects both seasoning and cohort effects. Recent cohorts are more pro-European than their predecessors, but voters also become more Eurosceptical as they age. Much of the pro-Europeanism of recent cohorts is associated with greater education. Our baseline estimates suggest that age and cohort effects together have reduced pro-EU sentiment by three percentage points over the past 40 years. This effect will be reversed by the middle of the present century, assuming that the attitudes of future cohorts resemble those of the currently young.
But uncertainty about future time effects dwarfs all of the above. Understanding what lies behind them – macroeconomic conditions, surges of immigration, or other factors – should be at the top of the agenda for those concerned to understand the evolution of UK public opinion toward the EU.
Donellan, M B and R E Lucas, (2008), “Age Differences in the Big Five Across the Life Span: Evidence from Two National Samples”, Psychological Aging 23(3): 558-566.
Dohmen, T, A Falk, B H H Golsteyn, D Huffman and U Sunde (2017), “Risk Attitudes Across the Life Course”, Economic Journal 127(605).
Eichengreen, B, R Mari and G Thwaites (2018), “Will Brexit Age Well? Cohorts, Seasoning and the Age-Leave Gradient, Past, Present and Future”, CEPR Discussion Paper 13288.
Tılley, J and G Evans (2014), “Ageing and generational effects on vote choice: Combining cross-sectional and panel data to estimate APC effects”, Electoral Studies 33: 19-27.