Wednesday, 31 May 2017

Put an X in the Box

One of the lesser noted features of the 2016 EU referendum was that the purity of the exercise - a simple binary choice on a nationwide basis - was a godsend to polling organisations seeking to segmentally divide the population. This allowed a fundamental political opinion, which could be treated as a proxy for a whole range of preferences, from commodities to cultural capital, to be mapped against multiple dimensions from geography through education to age. General elections are less useful in this regard because voters are often either loyal to family tradition, so dimensional changes over time may not be reflected in a changing vote, or they are floating voters whose choices are close to random in the aggregate data. When you add in the local constituency factor and the often arbitrary reasons people have for their choice (underplayed in general elections and overplayed in the referendum - e.g. bendy-banana woman), it is clear that eliciting meaning from a parliamentary election vote is not straightforward.

This applies even to aggregate indicators that avoid local variability and noise, such as national voter turnout. The propensity to vote has become a prominent topic of debate in recent years, hence there's a lot of discussion on the subject this week as polling day looms, partly because there is a real issue - the aggregate decline in turnout since the 1990s - but also because the differential analysis by age, showing higher rates of turnout the older the age group, has fed into the popular trope of intergenerational conflict. The story goes that the elderly have become the pivotal voters of modern times due to a combination of higher turnout and a larger cohort - i.e. the baby-boomer generation. This has given us the pensions triple-lock, a reluctance to build houses that would undermine property prices, and an obsession with inheritance, not to mention the "dementia tax". The supposed corollary of this is that millenials only have themselves to blame. Not only do they insist on buying avocados rather than saving for a deposit on a house, but they can't be arsed to register to vote because they're too busy circulating memes. In reality, simple logic tells you that the pivotal voters must be in the middle of the age range, with differential propensity at the ends simply shifting the fulcrum a little. In other words, the true pivotal voters are the 40 and 50-somethings who moan about pensioners and millenials.

The way that turnout by age is usually presented is by voter age group at the time of the election, rather than by birth cohort. For example, the turnout of 25-34 year-olds in 2017 is based on a population that is wholly different to the same age group in 1997. That earlier population is now the 45-54 year-old group. In contrast, a cohort analysis tracks those born in the same period, showing how their behaviour varies over time. The recent fall in total turnout is often presented in the media as a result of young people "drifting away" or "losing interest", though this language implies a persistent population with changing attitudes. This generational framing is found across the political spectrum, with those on the right tending to dismiss the young as trivial and irresponsible while those on the left talk of disillusion and a lack of choice. In fact, the data shows nothing of the sort. The issue is that more recent birth cohorts have failed to start voting at the same rate as older cohorts. This truth is obscured by the routine presentation of contemporary voting behaviour as a continuation of historic data.

The following table and chart shows how voter turnout is typically presented in the media. The numbers show the percentage of each age group that voted in every UK general election between 1964 and 2015. (This is based on data from the British Electoral Survey up to 2010 and Ipsos MORI in 2015 - the latter is rounded. The 18-24 age group for 1964 and 1966 is actually 21-24, as the voting age wasn't lowered to 18 until 1970. There are variations on this data available, but they have broadly the same profile). There are multiple problems with this approach. The age-groups differ in relative size (i.e. the number of years covered by each); the group memberships vary in absolute numbers (e.g. the baby-boomers are a population bulge that isn't evident); and the time sequence isn't consistent because elections don't happen regularly (as an aside, elections are less frequent nowadays, which makes the case previously made for the Fixed Terms Parliaments Act even more nonsensical).

There are some questions about the accuracy of the 1964 data, with the youngest apparently voting at a higher rate than the middle-aged, but the broad impression is of relatively high turnout across most groups, with 18-24 year-olds lagging slightly from 1966 onwards, until the mid-90s. At that point turnout drops across all age groups and more steeply among the younger, leading to an age-correlated spread and a widening gap by 2015. Cohort analyses aren't easy to come by (see pages 9-14 of this study, which interprets the BES data differently), but using the same data as employed for the above chart and rearranging it into approximate cohorts (assuming median ages relative to a 21 year-old in 1964) gives a different picture.

The dataset is incomplete (the data-points fade away) because all cohorts are eventually merged into the 65+ age group data, but it suggests that turnout by cohort does not vary that much, which makes intuitive sense: people get into the habit of voting. Clearly there will be fluctuations, particularly in elections considered to be watersheds or where the expectation was for a close contest, such as in 1979 and 1992 respectively, but the overall picture is closer to consistency with the one notable variation being the drop off in turnout in 2001, an election that was considered as much of a foregone conclusion as next week's poll.

What this suggests is that the "decline" in voter turnout is made up of two elements. First, there has been a fall in the ratio of people who first start voting between the ages of 18 and 24. This begins between 1992 and 1997 when turnout for that age-group drops from 67.3% to 54.1%. While this could be attributable to irresponsibility or disillusion, it is just as a likely to be due to increased progress to tertiary education delaying voter registration (which in turn reflects the hassle both of getting registered under current regulations and then changing address). This then ripples through older age groups in subsequent elections. A representative 20 year old in 1995 would contribute to the following age group turnout rates: 54.1% in 1997 (age 23); 45% in 2001 (age 26); 47.7% in 2005 (age 30); 64.4% in 2010 (age 35); and 64% in 2015 (age 40). If you discount the general drop in turnout after 1997 and slight recovery in 2010 for a moment, this series suggests a pick up since the early 2000s among this cohort. If this is related to going to college - i.e. graduates starting to habitually vote in their late-20s or 30s - then it suggests that the positive correlation of voting and educational level is simply kicking in later, amplified by the tendency for older groups to vote at a relatively higher rate.

The second element is a falling away of voting rates across all age groups over the last twenty five years, from the high of a national turnout of 78% in 1992 to a low of 59% in 2001 and then a partial recovery to 66% in 2015. While the media have enjoyed pointing the finger at millenials, it is clear that the fall-off cannot be exclusively attributed to one generation (and the idea of attitudinal generations is dubious anyway, despite their ubqiuity among think-tanks and thus media). Not only are they a small minority of the total (representing 6 birth-years as opposed to the 10-plus of other groups), but they could only have a significant impact on the aggregate figure if they started from a very high level, which is clearly not the case. Less of a small number isn't significant. It starts to become significant when you add in earlier cohorts now appearing in the numbers for older age groups. While turnout among the 18-24 group increased in the 80s from 62% to 67%, it fell markedly in the 90s, hitting a low of 38% in 2005. Since then, these earlier cohorts have been playing catch-up (e.g. the 25-34 group recorded 62% in 1997 and then, as the 45-54 group, 72% in 2015), but they remain hampered by their low starting base (it's a compounding problem). In summary, not only has the demographic bulge of the baby-boom amplified the voting power of the over-50s, but the steep fall in turnout among younger groups in the late 90s through to 2008 (essentially the New Labour years) has discounted the voting power of the under-50s. Basically, blame Generation X and Britpop.

So what does this mean for next Thursday? In truth, I have no idea, but I am prepared to make three guesses. First, I think that turnout among the young will increase. This is due to a combination of factors: guilt by many over sitting-out the EU referendum; the shift from household to individual voter registration in 2013, which has perversely made voting more salient among the young; and perhaps even a little bit of Corbyn-mania, which may prove more substantial that Milifandom. Second, I suspect the baby-boomer generation are not as rightwing as is generally assumed, partly because the vanguard Thatcher vote was actually made up of the back-end of the "silent generation" (born 1926-45). A typical representative would have been 21 in 1964 (too old for the counterculture), 36 in 1979 (young family, keen to buy their council house), and 49 in 1992 (obsessing about property prices). This cohort went from a turnout high of 82% in 1992 to a low of 64% in 2001. It came back to 78% in 2015, but the members are in their 70s now and their shrinkage in numbers is accelerating. Third, I suspect overall turnout will be closer to the EU referendum's 72% than 2015's 66%. While this could mean a reactionary bonus for Theresa May, it could also mean a recovery among 40-something Gen-Xers (despite the best efforts of John Harris and others to convince us that there is nobody worth voting for), and that might benefit Labour.

The one thing I can be confident of is that the pollsters will be issuing a leaver/remainer breakdown of the national vote, segmented by party, age and region, within hours of the voting booths closing on the 8th of June. Inevitably, this will be taken as evidence that the country will swing to remain given a few more years, which will mean absolutely zilch if Theresa May has made it back to Number 10.


  1. "If this is related to going to college - i.e. graduates starting to habitually vote in their late-20s or 30s - then it suggests that the positive correlation of voting and educational level is simply kicking in later"

    The most detailed treatment of the subject matter of this post in political science is Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945 by Mark Franklin. He makes many points that are very similar to those you make here, but would seem to disagree on the above. According to Franklin's data, young adults basically have to learn the habit of voting already in the very first election in which they are legally enfranchised. This is in fact true to such an extent that there is no statistical variable that better predicts whether someone in their forties or fifties will vote in the next upcoming election than knowing whether they voted in their first one 20–40 years earlier! And if we know the population pyramid of a country, as well as the turnout for each election that was the first one for each birth cohort with living members, the turnout for an upcoming election can be calculated directly from that, so accurately that it's almost eerie.

    So a mid-life pick-up of voting seems to need some active countervailing cause strong enough to offset the declining trend of youth voting. I wonder what that may have been in this case.

    A second point Franklin makes is that the lowering of the voting age to 18 may have had a counterproductive negative effect on youth turnout, basically because around age 18 many people are in a phase of life that makes them exceptionally cynical and socially atomised (much more so than, say, 8-year-olds on the one hand and 28-year-olds on the other). And because of the lifelong effect of the first election, Franklin suggests, "the well-intentioned decision to enfranchise young adults one election earlier than previously had the unanticipated consequence of giving rise to a lifetime of disenfranchisement for many of the intended beneficiaries" (p. 213).

    1. The countervailing factor I'm suggesting is the increase in graduates due to the expansion of tertiary education since the 90s. Of course, this assumes the correlation between education level and voting (i.e. grads vote more) continues.