Since the 2014 independence referendum, a debate over the relationship between age and support for Scotland seceding from the United Kingdom has flared up regularly. Polls consistently show that the youngest voters are much more likely than older voters to support Scottish secession. The most recent Norstat poll for the Sunday Times found that among likely voters, 67% of those aged between 16 and 34 would vote Yes to independence, compared to less than 40% among those aged 55 and over. It is not unusual for polls to find even greater gaps between the youngest and oldest voters, and these figures are routinely deployed in arguments over whether Scottish secession is only a matter of time.
These arguments tend to break down along partisan lines. Supporters of secession often make what might be called the ‘actuarial’ argument: as older, less secessionist voters pass away, and as younger, more secessionist voters come of age and replace them in the electorate, the level of overall support for secession will rise until a persistent majority are in favour of Scottish secession. Their opponents typically respond with an argument rooted in prospect theory: as younger voters age, they will become more risk averse – due to having a mortgage, pension, savings, children, and so on – and thus will become less likely to support radical policies in general and secession from the United Kingdom in particular. This is, essentially, a variation on the argument that voters become more conservative as they age.
This is a slight oversimplification – it is, of course, possible for both mechanisms to be in effect – but it captures the general state of the debate. Until now, these arguments have been made with little empirical evidence in favour of either, but they present us with clear hypotheses that can be tested.
At the heart of this debate is a single key question: is the relationship between age and support for Scottish secession a cohort effect, a lifecycle effect, or a cohort effect mediated by a lifecycle effect? Cohort effects refer to differences between birth cohorts that persist over time, while lifecycle effects refer to changes that occur among a birth cohort as they age. A third type of effect, period effects, refers to events that shift every birth cohort at the same time.
The modelling I will now set out suggests that there has been a consistent cohort effect since 1999 – that is, voters in younger birth cohorts are more likely to support secession than voters in older birth cohorts, and this is a persistent finding over time. It also fails to find evidence of a lifecycle effect – birth cohorts do not become less likely to support secession as they age. We also see period effects around the 2014 and 2016 referendums, when events reshaped Scottish politics.
These findings need to be carefully interpreted. Firstly, with only 24 years of data and only a few waves of data for the youngest voters, it is impossible to say that no lifecycle effect will ever kick in; there is just no evidence of one yet. Secondly, this does not mean that independence is a matter of when, not if – a week is a long time in politics, and the generational timescales we are discussing here can contain many unpredictable changes that reshape public opinion. However, it does suggest that the actuarial argument outlined above is better supported by empirical evidence.
Modelling Support for Scottish Secession by Birth Cohort
To begin to answer this question, I used data from the Scottish Social Attitudes Survey[1] collected between 1999 when the study first launched, and 2023, when its most recent wave was conducted, providing 20 waves of trend data over the course of 24 years (no survey was conducted in 2008, 2018, or 2020, and I exclude the 2021 wave for the reasons outlined in the Scottish Government’s publication of the 2023 wave[2]).
The Scottish Social Attitudes Survey asks respondents ‘which of these statements comes closest to your view?
- Scotland should become independent, separate from the UK and the European Union.
- Scotland should become independent, separate from the UK but part of the European Union.
- Scotland should remain part of the UK, with its own elected parliament which has some taxation powers.
- Scotland should remain part of the UK, with its own elected parliament which has no taxation powers.
- Scotland should remain part of the UK without an elected parliament.’
The first two answer options signal support for Scottish secession from the United Kingdom, and the latter three support for Scotland remaining part of the Union. I used this question, rather than the 2014 Scottish independence referendum question, for two reasons. Firstly, the Scottish Social Attitudes Survey has not consistently asked the referendum question over time, and publicly available datasets that do and match the sample quality of the SSAS, such as the Scottish Election Study, are less frequent, with far fewer years of trend data. Secondly, the 2014 referendum question will not necessarily be the question used the next time Scotland holds an independence referendum (if it ever does). It is more appropriate to use a measure that allows respondents to pick specific governing arrangements, a preference that would underpin voting intention in a future referendum.
I combined the first two answer options and the latter three answer options to create a binary measure of support for Scottish secession. This is the dependent variable I will model over time to assess whether there is a cohort effect on support for Scottish secession and whether the probability that a member of a given cohort’s support or opposition to secession changes as they age.
I categorised the Scottish Social Attitudes Survey respondents into six birth cohorts:
Born before 1958
Born between 1959 and 1968
Born between 1969 and 1978
Born between 1979 and 1988
Born between 1989 and 1998
Born between 1999 and 2007
I then carried out two types of modelling. Using the full set of data from 1999 to 2023, I built a mixed-effects logistic regression model with support for secession as the dependent variable and birth cohort as our key independent variable. I also controlled for sex and home ownership, bearing in mind that men were more likely to support secession in 2014 and home ownership is a commonly cited factor in encouraging greater risk aversion (I control for relative prosperity and whether a respondent has children in year-specific modelling below, but such data is not available for all years).
The findings of this model are presented in the table below and can be interpreted by examining the odds ratios. For example, the odds that a member of the youngest birth cohort supports Scottish secession are 3.12 times higher than the odds for a member of the oldest cohort. Likewise, the odds that men support Scottish secession are 1.28 times higher than the odds for women. In contrast, the odds that homeowners support secession are 42% lower than those for non-homeowners. All of these relationships are statistically significant at the most stringent conventional level (p < 0.001).
Mixed Effects Logit Model for Secessionist Support | ||
| Supports Scottish | |
secession | ||
Predictors | Odds Ratios | CI |
(Intercept) | 0.49 *** | 0.43 – 0.56 |
Birth Cohort (Reference: born before 1958) | ||
1959-1968 | 1.52 *** | 1.42 – 1.63 |
1969-1978 | 1.56 *** | 1.44 – 1.68 |
1979-1988 | 1.69 *** | 1.55 – 1.85 |
1989-1998 | 2.25 *** | 1.97 – 2.58 |
1999-2007 | 3.12 *** | 2.19 – 4.46 |
Male | 1.28 *** | 1.21 – 1.35 |
Home Owner | 0.58 *** | 0.55 – 0.62 |
Random Effects | ||
σ2 | 3.29 | |
τ00Wave | 0.08 | |
ICC | 0.02 | |
N Wave | 20 | |
Observations | 27032 | |
Marginal R2 / Conditional R2 | 0.045 / 0.069 | |
* p<0.05 ** p<0.01 *** p<0.001 | ||
To further investigate these relationships over time, I also conducted year-specific logistic regressions. It would consume far too much space to present each model individually[3], but these models find that the cohort effect on Scottish secessionist support was persistent for much of the period between 1999 and 2023, and it was in 2014 that the contemporary gap in constitutional preferences between younger and older birth cohorts became cemented.
The best way to demonstrate this is by looking at the modelled probabilities that a member of a given birth cohort supports Scottish secession over time. That is what the chart below does. There is a persistent and clear cohort effect but no decline in secessionist support within birth cohorts. Indeed, events surrounding the 2014 and 2016 referendums appear to have grown support across cohorts but not to have tightened or changed the gaps between cohorts.
Conclusion
We started with a clear question: is the relationship between age and support for Scottish secession a cohort effect, a lifecycle effect, or a cohort effect mediated by a lifecycle effect? This analysis should lead us to reject the hypothesis that there is a lifecycle effect at work, at least until the passage of time, the collection of more data, and a reappraisal of the data convinces us otherwise.
This is not the same as saying that there is no lifecycle effect at work or that no such effect could emerge in the future. But there is no evidence for such an effect right now.
In turn, it would be reasonable for advocates of the actuarial argument outlined above to take this analysis as supportive of their argument. Still, it should not be taken as evidence that independence is inevitable. Public opinion is mutable, not set in stone. As we can see above, period effects can shift opinion across the population – they are also not necessarily predictable. If we were looking at the chart above in 2010, without any subsequent data, we would have been unlikely to predict the 2014 independence referendum or the 2016 EU membership referendum, never mind black swan events like the COVID-19 pandemic.
We should also bear in mind that campaigns can change opinion, too. It’s perfectly possible that younger birth cohorts would be more susceptible to risk aversion if the prospect of independence was real, with an upcoming vote, and the unionist side focused on such arguments.
In short, this analysis offers some early evidence to provide a greater empirical underpinning of debates around the relationship of age to support for Scottish secessionism, but it is far from the final word.