The fragmentation of the UK party system was dramatically illustrated in the discussion of the 2019 European Parliament Elections in Chapter 4, but this process has been going on over a much longer period of time. In the first fully peacetime general election in 1950, Labour and the Conservatives received 89 per cent of the popular vote on a turnout of 84 per cent. In contrast, in the general election of 2019 these parties obtained 76 per cent of the vote on a turnout of 67 per cent.Footnote ¹ So fewer eligible people are voting, and when they do so, they are less likely to support the two major parties. The share of the vote for the two main parties recovered to some extent in the 2017 and 2019 elections compared with 2015, as previous chapters have shown. But the growth in support for minor parties has substantially reduced their share of the vote compared with fifty years ago.

As we have seen, a major contributor to this development has been differences in age-related electoral turnout and support for the major parties, trends which have been growing over time. We have known for a long time that young people are less likely to turn out than the middle-aged and elderly in British elections.Footnote ² Moreover, as discussed earlier, age was not only significantly related voting to Leave in the 2016 referendum on the country’s EU membership but was also a strong predictor of voting Labour in the 2017 and 2019 general elections.Footnote ³ Young people were very likely to vote Remain in the referendum and then to vote Labour. After the 2017 election, this surprisingly strong age effect was described as a ‘Youthquake’.

This chapter investigates why the process of party fragmentation has occurred over a long period of time with a particular focus on the role of age. We will explain why electoral participation among different age groups is so varied, why this variation has grown over time and how this has contributed to the fragmentation of the party system. We begin by focusing on turnout, and then subsequently look at voting for parties over a period of fifty-five years using British Election Study (BES) surveys as well as our own more recent national surveys. This exercise uses every BES survey from 1964 through 2017.Footnote ⁴

At first sight, it appears easy to analyse if young people participated in greater numbers in recent elections compared with the past simply by looking at turnout and party support by different age groups over time. But it is a more complicated exercise than it first appears, since there are three different aspects to the relationship between age and voting behaviour that we need to examine. These are what are known as life-cycle, period and cohort effects, each of which makes a separate contribution to the relationship between age and voting.Footnote ⁵

To consider each of these in turn, life-cycle effects are associated with individuals getting older and therefore having a variety of life experiences which can affect their attitudes, beliefs and rates of political participation. For example, 18-year-olds are unlikely to have full-time careers, or high incomes; few will own properties and have mortgages and they are less likely to be in a stable relationship with a partner or to have children than are older people. All these experiences are likely to influence the way they look at politics and the world more generally in comparison with the middle-aged or elderly, most of whom will have had a number of politically relevant experiences over the course of their lives.Footnote ⁶

One way of explaining lower turnout among the young is to note that they have less of a ‘stake in the system’. If they do not pay taxes, have full-time jobs, own property, are relatively healthy and do not have children, they are less likely to be concerned about politics because, while issues related to these characteristics are central to policy, they are peripheral to young people’s concerns. Youth are less focused on issues such as taxes, employment, health and housing, in comparison with older groups because their experiences in the labour market and in society differ from their older counterparts. On the other hand, when young people get into the world of work, acquire a partner, a mortgage and have children, their circumstances change. The more they acquire a stake in the system the more they will pay attention to government, politics and voting.

Period effects, in contrast, arise from the state of society and the political context at the time when an election takes place. Each election is to an extent unique and so the specific context in which it takes place can influence participation. For example, the 2015 election occurred at the end of the Conservative–Liberal Democrat coalition government during which the worst effects of the Great Recession had eased. This allowed the government to claim success in managing the economy, while at the same time blaming New Labour for ‘crashing the economy’ prior to the Great Recession. Circa 2015, the ideological distance between the parties was not especially large and the date of the election was known well in advance, such that the parties were well prepared for the campaign.Footnote ⁷

Things were very different in the 2017 general election, which came as a surprise to everyone, had a longer than usual campaign and was dominated by the Brexit debate, as we discussed in Chapter 3.Footnote ⁸ In addition, the ideological divisions between Theresa May’s Conservatives and Jeremy Corbyn’s Labour Party were much wider than two years earlier. Equally, the effect of the campaign was much larger in 2017 than it had been in 2015, with significant increases in public support for Labour occurring as polling day approached, something which had not happened in the previous election. Clearly, period effects were quite different in the two elections. A similar point can be made about the 2019 general election as the analysis in Chapter 5 makes clear.

The third aspect of the relationship between age and participation involves cohort effects, an idea originally introduced by the social scientist Karl Mannheim.Footnote ⁹ Cohort effects are based on the idea that the values, attitudes and beliefs that determine people’s voting behaviour are formed in late adolescence and early adulthood. As this process happens it is influenced by the economic, social and political circumstances of the time. Once formed, the evidence suggests that, unlike life-cycle effects, cohort effects remain relatively stable over time as people grow older, even if their social and economic circumstances change.Footnote ¹⁰ In this respect higher education is an important socializing agency. For example, consider an 18-year-old student who became politically aware in 2003 and joined in several protests against the UK’s participation in the war in Iraq. There is evidence suggesting that these experiences are likely to stay with this person for many years after graduation, working to shape their subsequent political attitudes and behaviour.

At a collective level, these early-life socialization processes can produce ‘generational’ or cohort effects among voters. This means, for example, that individuals who ‘came of age’ politically during the Second World War and who faced physical dangers and wartime deprivation are likely to look at the world differently from those brought up in the relatively peaceful and affluent 1960s. Moreover, this will continue to be true for the wartime generation even when they live comfortably in the changed circumstances of later decades. The claim is that early experiences can mark a generation of people for life, and this is also true of their politics and voting behaviour.

One prominent researcher, Ronald Inglehart, developed these ideas in a series of publications about changing cultural values across the world.Footnote ¹¹ In a series of widely cited studies, Inglehart argued that successive cohorts of people coming of age in affluent post–Second World War societies increasingly have ‘post-materialist’ values focusing on ‘self-realization’ and ‘personal development’. In contrast, older cohorts who reached maturity during periods of economic hardship and war were much more likely to have ‘materialist’ values that prioritize ‘economic successes’ and ‘personal security’. Inglehart’s central argument is that the evolution of political orientations is primarily driven by the successive replacement of cohorts with different socialization experiences rather than by age or period effects.

Taken together, the preceding discussion implies that if we wish to understand the relationship between age and political participation, we need to identify age, period and cohort effects using a longitudinal study. However, there is a serious problem in doing this because the following relationship holds between the three effects.

$\text{Period}-\text{Age}=\text{Cohort}$

If we know the age and period effects, then the cohort effects are fixed and cannot be independently estimated. To illustrate this, consider an 18-year-old student who was a first-time voter in the 2019 general election. We know her age and the period or election in which she participated, so she must be a member of the cohort who became politically aware at some point between the 2016 EU referendum and the 2019 election. This assumes that she had to be at least 15 years of age before she started to take an interest in politics just prior to the referendum. Clearly, she cannot be part of a cohort who came of age politically in the 1960s or 1970s, so age and periods perfectly predict her cohort membership. Technically this is referred to by statisticians as an ‘identification problem’. This means that it is impossible to estimate the three effects in a linear regression model of electoral turnout.Footnote ¹²

Recent debates have divided the methodological community on the issue of overcoming this problem and estimating trends in turnout among age groups over time. All are agreed that it cannot be done with a survey conducted at a single point of time, but it is argued that the availability of longitudinal data collected over time may make this possible.Footnote ¹³ We will begin the data analyses in this chapter by examining turnout before going on to look at support for the parties, utilizing these three age-related measures to examine how voting behaviour has changed over time. Each of these aspects plays an important role in explaining electoral participation and voting for the major parties, and they help to explain the observed fragmentation of the British party system.

Periods, Cohorts and Turnout in Britain 1964–2019

We begin our investigation of age-related influences on electoral participation by looking at the effects of periods and cohorts on turnout in surveys conducted between 1964 and 2019. This is the longest period over which national survey data on voting behaviour in Britain are available. We first examine trends in periods and cohorts without separating them out in order to provide an overall picture of what happened across this fifty-five-year time interval. Subsequently, we distinguish between the different age-related effects after introducing hierarchical age-period-cohort modelling. Altogether, there were sixteen different election studies conducted between 1964 and 2019, and so we start by examining them. The cohort and life-cycle effects are discussed more fully below.

The analysis of turnout utilizes self-reported voting by the survey respondents. This presents a potential problem, since there is abundant evidence indicating that individuals tend to exaggerate their electoral participation when they are questioned in surveys.Footnote ¹⁴ However, we can examine the extent to which this is a problem by calculating an average figure for self-reported turnout in each of the sixteen surveys and then compare these figures with official turnout statistics from the actual elections. The relationship between the two appears in Figure 6.1, where each dot is an election, and the regression line summarizes the association between the two measures over time. The correlation between them in these 16 elections is very strong (r = +0.93). Thus, self-reported turnout is an excellent guide to what actually happened in an election.

Figure 6.1 The Relationship between Reported Turnout in Election Surveys and Actual Turnout in General Elections, 1964–2019.

Having established that self-reported turnout is a suitable proxy measure of actual turnout, Figure 6.2 shows levels of electoral participation by age groups in all sixteen general elections. The figure confirms the point that younger age groups have always voted at lower rates than their older counterparts. It is also evident that the differences in turnout among age groups have grown significantly over time. These age-related differences reached an all-time high in the 2005 election, after remaining fairly stable between the 1964 and 1992 elections. The years when Tony Blair was prime minister were associated with the largest changes in age-related turnout over the entire period. Electoral participation in the Labour landslide victory in 1997 declined rather sharply among those below the age of 40 compared with 1992 and subsequently fell even more in the 2001 election. The latter was undoubtedly influenced by the fact that the Conservatives had little chance of eliminating Labour’s massive majority, and this clearly had an impact on participation by all age groups, except for the very old. If people think elections are a ‘done deal’ then they are less likely to participate.

Figure 6.2 Turnout in General Elections by Age Groups (Periods), 1964–2019.

One of the most striking observations in Figure 6.2 is the rather dramatic rise in turnout during the 2017 general election among young voters, something which amply justifies the description ‘Youthquake’. This change had a lot to do with the referendum on EU membership which took place a year earlier and which mobilized many young people to vote.Footnote ¹⁵ In addition, as we pointed out in Chapter 3, in 2017 Jeremy Corbyn ran a very successful barnstorming campaign in comparison with Theresa May, whose campaign was largely ineffective.Footnote ¹⁶

Corbyn was a new face at the time and held a series of very successful open-air rallies attended by many young people, including at the Glastonbury festival. In addition, much of Labour’s campaign was conducted online and this was also very effective.Footnote ¹⁷ However, Figure 6.2 also shows that this surge in turnout among the young did not continue through to the 2019 general election. Many young people failed to vote on this later occasion when their counterparts over the age of 60 turned out in even greater numbers. This helped to produce a very different result, helping to clear the way for the rise of Boris Johnson.Footnote ¹⁸

Figure 6.2 shows the relationship between elections and turnout when these occur at well-defined periods of time, but what about cohort effects? While period effects are relatively easy to identify the same cannot be said for cohorts. Becker was one of the earliest political scientists to examine cohort effects, and he defined them as ‘a grouping of a number of cohorts characterized by a specific historical setting and by common characteristics’.Footnote ¹⁹ The important point about defining cohorts is to locate them in a political and historical context. If they are defined too broadly this will bundle together rather different political eras, and so we will not be able to distinguish between them. On the other hand, if they are defined too narrowly this will treat rather similar political contexts as if they were uniquely different from each other and give a misleading picture.

Researchers often have defined cohorts more or less arbitrarily as time intervals of five- or ten-years’ duration. This may be acceptable in some contexts, for example in medical research where it is difficult to pin down when a problem like obesity started to be a serious issue. In this case it makes sense to divide the time into equal periods.Footnote ²⁰ But in electoral politics there are clear differences between political eras, and they are not all of the same duration. For example, there are good reasons to expect that individuals who reached political maturity during the years of consensus politics in the 1950s are likely to look at the political world differently from those who came of age in the divisive Thatcher era of the 1980s.

In our earlier book, Political Choice in Britain, we identified five different age cohorts in an analysis which looked only at age and cohort effects in the 2001 election.Footnote ²¹ We assumed that individuals would reach political maturity by the age of 21 in that analysis, although it is now more common to assume that this happens by the age of 25.Footnote ²² Our original classification of the different cohorts was:

1. (1) ‘Post–Second World War’ cohort who achieved political maturity before 1950;

2. (2) ‘Macmillan’ cohort who achieved it between 1951 and 1964;

3. (3) ‘Wilson-Callaghan’ cohort who achieved it between 1964 and 1979;

4. (4) ‘Thatcher’ cohort who achieved it between 1979 and 1992;

5. (5) ‘Blair’ cohort who achieved it after 1992.

That analysis focused only on one election and so did not consider period effects, and therefore did not encounter the statistical identification problem discussed earlier. There may have been period effects, but they could not be identified using data from a single survey. The findings suggested that both the ‘Thatcher’ and ‘Blair’ generations were less civically minded than their earlier counterparts, and this was manifested by their lower participation.

More recently, Maria Grasso has suggested that different researchers working on age-related effects in politics have identified rather similar political cohorts in Britain, indicating that there is something of a consensus about them in practice.Footnote ²³ Her focus was on studying age-related political participation across several European democracies and for this reason her definition of cohorts is rather broad. She identified five cohorts in her study: ‘Pre–Second World War’, ‘Post–Second World War’, ‘Baby-Boomers’, ‘80s generation’ and ‘90s generation’.

Since the present focus is solely on Britain, we can be somewhat more specific about the definition of cohorts, linking them to periods of post-war political history which are clearly different from each other. Accordingly, we use the original cohorts introduced in the 2001 analysis but add another four to bring it up to date. Specifically, we define members of a cohort as people between the ages of 15 and 25 living during a given political era, at which point they are assumed to have reached political maturity.Footnote ²⁴ The nine cohorts are defined as follows:

1. (1) ‘First World War’ cohort who achieved political maturity before 1919;

2. (2) ‘Post–First World War’ cohort from 1919 to 1929;

3. (3) ‘Great Depression’ cohort from 1930 to 1939;

4. (4) ‘Second World War’ cohort from 1940 to 1949;

5. (5) ‘Macmillan’ cohort from 1950 to 1963;

6. (6) ‘Wilson/Callaghan’ cohort from 1964 to 1979;

7. (7) ‘Thatcher’ cohort from 1980 to 1997;

8. (8) ‘Blair’ cohort from 1998 to 2010;

9. (9) ‘Austerity’ cohort from 2011.

These cohorts are not equal in duration, but rather reflect the circumstances of different periods of British political history. It is useful to sketch out some of the key differences. The First World War radically changed the country not merely because of the large number of deaths and wartime shortages, but also because the existing Conservative–Liberal two-party system was shattered in the 1918 general election.Footnote ²⁵ This meant that the wartime cohort faced a different political landscape from the post-war cohort. The latter came of age in a period of turbulent party politics and conflicts over public spending and industrial relations, the latter exemplified by the general strike of 1926. This era ended with the election of the first minority Labour government in 1929.

The Great Depression cohort went through the severe economic disruption in which the unemployment rate exceeded 15 per cent in 1932, at a time when welfare benefits were meagre.Footnote ²⁶ This was followed by a period of recovery that was largely associated with rearmament and increased defence spending in the face of the growing threat from Nazi Germany. It was also a period when the National Government was formed by Labour leader Ramsey Macdonald. The Labour Party split and was subsequently heavily defeated in the 1931 general election.Footnote ²⁷ This was a turbulent period in British politics.

The Second World War cohort experienced physical dangers, civilian and military losses, and major shortages as the country was placed on a wartime footing. Children were evacuated from cities and thousands of adults were drafted into the armed forces, many of whom did not see their families for years. However, victory in 1945 did not bring an end to austerity, since rationing continued up to the end of the decade and beyond. Elected in 1945, Clement Atlee’s post-war Labour government started the huge task of reconstruction, while at the same time laying the foundations of the modern welfare state and creating the National Health Service.Footnote ²⁸

Conservative governments, in power in the 1950s, were exemplified by that of Prime Minister Harold Macmillan, who presided over a period of growing prosperity as post-war reconstruction and full employment finally brought an end to wartime deprivation. It was an era of broad consensus politics described at the time by the epithet ‘Butskellism’, i.e., a combination of ‘One Nation’ Conservatism associated with prominent Conservative R.A. Butler, and the centre-left politics of Labour leader Hugh Gaitskell.Footnote ²⁹ Macmillan captured the mood of the era in a 1957 speech in which he claimed: ‘Our people have never had it so good.’Footnote ³⁰

The Macmillan era came to an end when Labour leader Harold Wilson won a narrow victory in the 1964 general election. Wilson captured the desire for change in the country at that time with a speech in 1963 in which he talked about using the ‘white heat of technology’ to invigorate the British economy.Footnote ³¹ He won a large majority in the subsequent 1966 election, but then lost power in 1970 following an economic crisis, the devaluation of the pound and a recession. Edward Heath’s Conservative Party won that election, but his administration was largely a political failure beset by industrial relations problems and the emergence of rising inflation. Labour subsequently returned to power in the February 1974 election and stayed in office until 1979 under James Callaghan who took over from Wilson as prime minister in 1976. By then the government’s majority was precarious and the loss of a vote of no confidence in the House of Commons precipitated the 1979 election, won by Margaret Thatcher.

The Thatcher government marked a new era in British politics, following a policy characterized by Andrew Gamble as ‘The Free Economy and the Strong State’.Footnote ³² Initially it created a period of austerity with the aim of quelling rampant inflation and became involved in serious industrial conflicts exemplified by the miners’ strike of 1984. Austerity proved unpopular, but the victory in the Falklands War and a split in Labour in 1981 enabled the Conservatives to win a large majority in the 1983 election. Mrs Thatcher’s government then embarked on a policy of extensive privatization of state assets and deregulation of the City of London. In turn, these actions helped to produce a stock market rally and consumer boom in the economy. However, the seeds of divisions over Brexit were sown in the Conservative Party at this time by a growing conflict between the prime minister and the European Union over Britain’s budgetary contributions, reflected in her phrase: ‘I want my money back!’

In November 1990, Mrs Thatcher was ousted as prime minister, largely as a result of her determination to push through a highly unpopular poll tax to finance local government.Footnote ³³ She was replaced by John Major, who won a surprise victory in the 1992 general election, but his term in office was very much an extension of the Thatcher era. The ‘Black Wednesday’ currency crisis in September 1992 blighted his administration from the start and intra-party divisions over the European Union grew stronger among Conservatives, symbolized by the political rebellion over the signing of the Maastricht Treaty. The Thatcher era finally came to an end with Tony Blair and New Labour’s landslide victory in the 1997 general election.

Tony Blair served as prime minister for ten years, during a period of growing prosperity and rapid increases in public expenditure on health, education and welfare described as the ‘New Labour’ era.Footnote ³⁴ But Blair’s popularity was blighted by his support for the Iraq War of 2003 and infighting with the Chancellor of the Exchequer, Gordon Brown, who took over as prime minister in June 2007. Brown’s premiership was undermined by the financial crash which started with a run on the Northern Rock bank in October 2007. The crash precipitated a deep recession that had major effects on the economy and the course of British politics.Footnote ³⁵

As a result, Labour lost the 2010 general election and was replaced by David Cameron’s Conservative–Liberal Democrat coalition government which chose to respond to the crisis with a long period of austerity. Voters who came of age in this period also experienced increasingly fractious politics associated with the debate about whether or not to continue membership of the European Union, which served to exacerbate age-based political divisions and had also been reflected in the rise of the new populist UK Independence Party. The latter drew most of its votes from older Britons. As discussed in Chapter 4, in the 2016 referendum on EU membership many young people voted to remain while their older counterparts largely supported Brexit, with these divides over age among the largest reported at the referendum.Footnote ³⁶

Using the nine distinct political cohorts defined above, Figure 6.3 displays the relationship between electoral turnout and cohort membership over time. The figure differs from Figure 6.2 because older cohorts die out and younger ones take their place, so cohorts appear and disappear at different points in time. Turnout in the various cohorts remained similar up to the start of the Wilson/Callaghan era in the 1970s. At that point electoral participation among the youngest cohort started to decline and fell precipitously during the Thatcher era. To a lesser extent turnout also decreased among the under-40s making electoral participation among cohorts much more variable. Participation by the youngest cohort then increased during the austerity years, reflecting the rise in youth voting culminating in the 2017 election.

Figure 6.3 Turnout in General Elections by Age Cohorts, 1964–2019.

These are interesting findings which we explore more fully below, but it is important to remember that they do not separate age, period and cohort effects, which are all mixed together. Table 6.1 shows the interaction in turnout between the sixteen general elections and nine cohorts over the 1964 and 2019 election surveys. If we read the table by rows, we can see that people in the ‘Blair cohort’ who first appeared in the 1992 election were less likely to vote in all subsequent elections in comparison with their predecessors. In Labour’s landslide, for example, only 60.7 per cent of the Blair cohort voted in comparison with 72.3 per cent of the Thatcher cohort. In the 2001 election, which had the lowest turnout of any election in the post–Second World War period, the equivalent figures were 48.4 per cent and 66.8 per cent.

This decline in electoral participation continued into the era when the Austerity cohort began. The members of that group who first appeared in 2001 were less likely to vote than members of the Blair cohort in every single subsequent election after 2001. For the Austerity cohort, the low point of participation was reached in the 2005 election when only 44.3 per cent of them cast a ballot. It is not an exaggeration to say that these two cohorts are increasingly abandoning electoral participation in a way not observed in the past. The only exception to this trend occurred in the 2017 general election when young people turned out in numbers equivalent to that of their older counterparts. But as mentioned earlier, the 2019 election showed that this was an exception.

If one reads Table 6.1 by columns this shows how particular cohorts behaved in different elections over time. For example, 84.3 per cent of the First World War cohort turned out in the 1964 general election but by 1979, the last survey to include enough of them for meaningful analysis, their turnout was down to 70.6 per cent. Turnout among the really elderly tends to be lower than among their younger counterparts, but electoral participation by the elderly in 1964 stands in stark contrast to the figure of 53.3 per cent for the Austerity cohort in the 2019 election. Again, the biggest contrast in the columns of Table 6.1 is between the average turnout of 89.5 per cent among the Second World War cohort in fourteen separate elections and 61.0 per cent among the Austerity cohort in six elections.

A third way of reading Table 6.1 is to look at it diagonally from left to right. Unlike the row figures which classify voting in each election by successive cohorts, or the column figures which look at each cohort in different elections, the diagonal identifies the interaction between cohorts and elections. For example, it shows that voting by the First World War cohort was still at 81.2 per cent during the Wilson/Callaghan era and fell to 73.8 per cent in the last (1979) election in which enough of them could be counted.Footnote ³⁷ Generally, diagonal comparisons show that the pattern of declining turnout began in the Thatcher years and continued until the present era.

In this section we have focused on the relationship between periods and cohorts and electoral participation over time. But it is equally interesting to look at the relationship between these variables and party support. We examine this next.

Periods, Cohort and Party Support 1964–2019

Table 6.2 shows vote shares for the Conservative, Labour and Liberal/Liberal Democrat parties in each of the sixteen elections.Footnote ³⁸ These figures are consistent with the story sketched earlier that Labour won the 1964 and 1966 elections, but narrowly lost the 1970 election. Labour then regained power in February 1974, but with only four seats more than the Conservatives. This subsequently led Harold Wilson to call the second election in October of that year. In that election, Labour increased its majority by eighteen seats, producing a lead over the Conservatives of forty-two seats, but with a narrow overall majority.

Mrs Thatcher’s victory in 1979 set the stage for nearly two decades of Conservative dominance. As the previous discussion indicated, this dominance continued through to1997, the start of the ‘New Labour’ era. After winning in 2001 and again in 2005, Blair stepped down as Labour leader in 2007 and Gordon Brown took over only to be engulfed in the financial crisis and the subsequent Great Recession. This was followed by the coalition government of 2010–2015, at which point David Cameron’s Conservative Party won a narrow majority. This subsequently disappeared in the 2017 election called by Theresa May in the middle of the Brexit negotiations. Finally, May’s successor, Boris Johnson, won a decisive majority of eighty seats in 2019. Overall, the table illustrates the well-known story of the two major parties alternating in office, a pattern interrupted by the Coalition government interlude from 2010 to 2015.

Less familiar is Table 6.3 which shows the parties’ support for nine different cohorts over time. It is striking how support for the Conservatives has declined markedly across the cohorts. Almost 60 per cent of the First World War cohort voted Conservatives and only about a third of them supported Labour. In contrast, nearly 55 per cent of the Austerity cohort voted Labour, with only just over a fifth of them supporting the Conservatives. Liberal support in the First World War cohort was below 8 per cent and it reached a high point in the Thatcher cohort at 17 per cent, before falling to 12 per cent in the Austerity group. Interestingly, there were no respondents to be found in the First World War cohort who voted for minor parties, compared with 12 per cent who did this in the Austerity cohort.

This long-run perspective on generational change shows that the Conservatives had a consistent advantage over the other parties among the pre–Second World War cohort, but Labour caught up by the time of the Second World War and thereafter moved ahead of its main rival – this gap widening to a chasm by the time of the Austerity cohort. Having presented this overview of the relationship between voting and cohort membership since 1964, we next discuss how we can separate age, period and cohort effects on turnout and party support.

A Hierarchical Age-Period-Cohort Analysis of Turnout

As discussed above, the problem of trying to separate age, period and cohort effects is that, when we know the person’s age and the elections in which they voted, that automatically fixes their cohort membership. This ‘identification problem’ is analogous to trying to solve a single equation which contains two unknown variables. The problem has been known about for many years and some researchers have argued that it is impossible to separate the three effects.Footnote ³⁹ However, other analysts contend that the problem arises from a linear relationship between the variables and if the model can be recast in a non-linear form, then the difficulty can be circumvented.Footnote ⁴⁰

A number of political scientists have attempted to identify separate age, period and cohort effects using various approaches.Footnote ⁴¹ Clearly, cohort and period effects have very different implications for electoral behaviour. If cohort effects predominate then it means that change will be quite slow, since it relies on younger groups replacing older ones over a relatively long period of time. This is also true but to a lesser extent for life-cycle effects. On the other hand, if period effects dominate, then change is likely to be much faster but at the same time voting behaviour will be more volatile. Since any given election can be held in very different circumstances from the previous one, period effects can produce large short-term changes.

The Hierarchical Age-Period-Cohort (HAPC) approach divides the estimation procedure into two parts which are examined at different levels of analysis. One equation is estimated at the level of individual survey participants, and it delineates the relationship between age and turnout. This is known as a ‘fixed’ effects model, and it is used to examine life-cycle effects. The second or aggregate level of analysis examines cohorts and periods as the cases rather than individuals and is known as the ‘random’ part of the model. This looks at cohort and period effects by predicting changes in the intercept of the individual-level model. This intercept represents what is left over after accounting for various factors such as a respondent’s age, educational level, occupational status and gender. If the second level equation contains variables that significantly influence the intercept of the first-level equation, it means that contextual factors such as an individual’s cohort membership are influencing turnout.

This approach enables us to determine how age influences turnout over time while controlling for cohorts and periods. We also can identify if individuals who became politically aware at particular points in time, e.g., the Blair cohort were less likely to vote in the 2015 election than individuals who came of age in the 1980s during the Thatcher cohort. If true, this would mean that there are cohort effects that operate independently of age or life-cycle effects when they are included in the analysis.

Yang and Land are leading advocates of the HAPC approach, and they explain why they think this works in the following terms: ‘An HAPC framework does not incur the identification problem because the three effects are not assumed to be linear and additive at the same level of analysis.’Footnote ⁴² In their view, this solves the identification problem.Footnote ⁴³ However, their approach has been challenged, particularly by Bell and Jones in a series of papers based on simulating the results of HAPC modelling.Footnote ⁴⁴ They create a model with known parameters and then try to estimate it using the HAPC procedure. They conclude: ‘For us, the key critique of the HAPC model lies in its inability to accurately represent data generating processes (DGPs) in simulations.’Footnote ⁴⁵ In other words, their statistical analyses are unable to accurately delineate the model they constructed to produce the data. In addition, they find that period effects tend to dominate the cohort effects, although the reason for this is unclear.

This suggests that it may be unwise to try and estimate all age, period and cohort effects at the same time. One solution to this problem is to focus just on estimating life-cycle effects at the individual level and cohort effects at the aggregate level and ignore period effects. It is important to know the extent to which individuals change their party support as they get older (life-cycle effects) as opposed to remaining loyal to the parties they supported when first becoming aware of politics in their teens and early adulthood (cohort effects). However, if period effects are important, then the life-cycle and cohort effects alone will produce inaccurate estimates which fail to take into account differences between elections.

We respond to the Bell and Jones critique not by ignoring the period effects but by changing their definition in the modelling. In a full HAPC analysis, the period effects would represent all sixteen elections between 1964 and 2019, but this is what gives rise to the identification problem. An alternative is to look only at a limited number of elections which are chosen on theoretical grounds as being important turning points in electoral politics. In the subsequent modelling we do not attempt to assess the impact of all elections or periods, but only those which meet a specific criterion which we expect to be important. This is done by focusing only on ‘turnover’ elections that produced a change in government and ignoring the others.

A ‘turnover’ election is defined as one in which an incumbent party or coalition government was removed from office by a challenger party. There were seven elections between 1964 and 2019 when this happened. Labour replaced the Conservatives in power after the 1964, February 1974 and in the 1997 elections, and the Conservatives replaced Labour after the 1970 and 1979 elections. In addition, the Conservative–Liberal Democrat Coalition government replaced Labour after the 2010 election and the Conservatives replaced the coalition government in 2015.

When incorporating these period effects as controls in the statistical analysis, they are treated as significant political ‘shocks’ to the system, rather than as recurring events covering all elections from 1964 to 2019. This approach removes the identification problem referred to earlier, since the periods are no longer systematically linked to changes in the respondent’s age defined by their cohort membership and life-cycle experiences. Rather they are defined by political events which we have good reason to expect will influence voting because they have a major political consequence.

Turnout 1964–2019

We begin by examining turnout in the individual-level model. We use a logistic regression specification which is appropriate when the dependent variable is a dummy variable scoring one if people voted and zero if they did not.Footnote ⁴⁶ We have met this model already in Chapters 3 and 5. The Austerity cohort is treated as the reference category with which to compare the others, and therefore is omitted from the model specification. This means we are evaluating how turnout has changed over time in the different cohorts in comparison with the most recent one.

Figure 6.4 depicts the impact of both the individual and aggregate variables on turnout over the 1964–2019 period. The individual-level or fixed effects appear in the bottom of the table in a lighter shade than the aggregate-level random effects which appear further up in the figure. The middle dot shows the size of the effect of the variable on the probability of voting.Footnote ⁴⁷ The left-hand and right-hand bars linked by a horizontal line measure the uncertainty associated with the estimate, i.e., the confidence interval.Footnote ⁴⁸ When the confidence interval crosses the vertical line at zero, it means we do not have a statistically significant effect, since the true coefficient may be zero rather than positive or negative. This happened in the case of the Blair cohort, for example, which means we cannot be sure that this cohort differs from the Austerity cohort when it comes to voting.

Figure 6.4 Hierarchical Age Cohort Model of Turnout with Period Controls, 1964–2019.

Note: Aggregate random effects are first followed by individual fixed effects.

Unlike the Blair cohort, all the other cohorts had significantly higher turnout than the Austerity cohort, since they all show statistically significant positive effects compared with this most recent cohort. In these cases, the uncertainty bars do not cross the zero line, so we infer that they voted in greater numbers than the Austerity and Blair cohorts. This suggests that there was a significant ‘break point’ in socialization processes that encourage young people to vote after the Labour government won the 1997 general election. Cohorts that became politically aware prior to that period consistently voted in greater numbers than cohorts emerging after that election.

The implication of this finding is that the Blair era was a key turning point for electoral participation over this lengthy sixty-year period. Those who became politically aware during and after the Blair years were less likely to vote even when their occupational status, age, education and the controls for turnover elections are considered in the analysis. This is particularly true of the Second World War and Macmillan cohorts, who were much more likely to vote than their Austerity counterparts. Those years represented something of a golden age in electoral participation in Britain. This is illustrated by the fact mentioned earlier that the highest turnout in any post-war election was 84 per cent in 1950, at a time when the post–Second World War cohort was coming of age politically. That compares with a turnout of only 67 per cent in the 2019 election.

The age or life-cycle effects in the individual-level model reinforce this conclusion, since age is a strong predictor of turnout throughout the entire period,Footnote ⁴⁹ with older people voting in greater numbers than the young. But as Figures 6.2 and 6.3 show, the gap between the old and the young has grown significantly wider over time, and the modelling confirms this point. However, the quadratic specification of the age variable in the analysis shows that the effect of age on turnout tends to decline as individuals get really old, which is not surprising given the physical infirmities that accompany ageing.Footnote ⁵⁰

The election ‘shocks’ in the figure should be compared with the average turnout in those elections which did not lead to a change in government. In general, these period effects are weak, with two clear exceptions. The February 1974 election saw an upsurge in turnout following the troubled Conservative government of Edward Heath which had come to power in 1970. The second exception was the turnout in the 2015 election, when David Cameron’s Conservative Party almost wiped out their coalition partners, the Liberal Democrats. Participation in this election was below average. Apart from these exceptions, the long-term decline in turnout is clearly driven by cohort replacement and persistent age effects rather than by specific period shocks.

This finding carries a disturbing implication, namely that decline in electoral participation over time is unlikely to be easily reversed in the future.Footnote ⁵¹ The growing gap between the young and the old apparent in Figure 6.2 is being ‘baked in’ to the system in the long run, as the low participation age cohorts are subsequently joined by new cohorts as the future unfolds. Unless something unforeseen happens, these new cohorts are likely to continue voting in lesser numbers than their older counterparts, leading to a continuing decline in overall electoral participation.

Party Choice, 1964–2019

We turn next to support for the two major parties. The results of applying the HAPC analysis to the Labour share of the vote are presented in Figure 6.5. Starting with the fixed effects or individual-level part of the analysis, age, occupation and education are all statistically significant predictors of Labour voting. The confidence intervals around age are very narrow, indicating that we are estimating effects with little uncertainty. Unfortunately, this means that the effects are hard to see in the figure, but suffice it to say that they are very strong.Footnote ⁵² As expected, the coefficients indicate that older working-class people have been more likely to support the party and the highly educated professionals have been less likely to do so. Of course, these relationships are very familiar to students of British party politics. In addition, and similar to the turnout model, the analysis does not indicate any significant gender effects on party support.

Figure 6.5 Hierarchical Age Cohort Model of Labour Voting with Period Controls, 1964–2019.

Figure 6.5 shows that cohort effects on Labour voting have been rather different from those on turnout in Figure 6.4. None of the cohorts were significantly different from the Austerity cohort, with the sole exception of the Second World War cohort, members of which were more likely to support Labour in successive elections. That said, the post–First World War and the Macmillan cohorts are both close to being statistically significant, implying that they contain more Labour supporters than the Austerity cohort. But these effects are rather weak evidence of a preference for Labour. Just as the Second World War cohort have been more likely to vote in elections, they are more likely to vote Labour, suggesting that wartime experiences provided a turning point in support for the party.

The turnover election effects for Labour also are rather sparse. Clearly, the 1964 election under leader Harold Wilson mobilized additional support for the party after thirteen years of Conservative rule. Equally, and not surprisingly, the 1997 election victory produced considerable additional support for Labour reflecting the fact that it was such a landslide win for the party. But in other respects, the impact of election-related shocks on Labour support has been rather modest.Footnote ⁵³ The general conclusion from this analysis is that neither cohorts nor electoral shocks have had large effects on support for the party over this long period of time.

There has been a recurring argument which surfaces every time Labour loses an election ever since Mark Abrams and Richard Rose first raised it over a half-century ago.Footnote ⁵⁴ It is the proposition that Labour can never win again because of a lost generation or cohort which was very supportive of the party in the past but is not being renewed. However, there is little statistical evidence to support this idea. Rather the analysis suggests that the party fights each election as it comes and, depending on the circumstances, it will do well if the context is favourable and badly if it is not. Examples of this are the 2017 general election discussed earlier, in which the party did very well, although it did not win. Similarly, the 2019 election produced the worst result for Labour since its 1935 election defeat. These are very different outcomes in elections held only two years apart.

Results of the Conservative HAPC analysis displayed in Figure 6.6 are very different from those for the Labour model. Once again, in the individual-level model, age has a highly significant impact on party support, with older people more likely to vote Conservative than the young.Footnote ⁵⁵ The occupational status and education effects are the opposite of Labour’s, with highly educated individuals in professional occupations being more likely to vote Conservative than the rest of the electorate. Again, these effects have been known about for many years. Also, unlike Labour, gender has a significant effect, with women being more likely to vote Conservative than men.

Figure 6.6 Hierarchical Age Cohort Model of Conservative Voting with Period Controls, 1964–2019.

The really striking feature of the Conservative vote model is the importance of the cohort effects. They show that all cohorts, with the sole exception of the Blair cohort, are more likely to vote for the party than the Austerity cohort. These strong cohort effects for Conservative support mirror what we observed in Figure 6.4 in relation to turnout. In a reflection of the results depicted in that figure, there has been a significant decline in support for the Conservatives among young people who came of age politically during and after the Blair years. These most recent cohorts in British politics have abandoned the party in a way not seen before. The effect associated with the Blair cohort is strikingly different from the previous Thatcher cohort, whose members support the Conservatives to a much greater extent.

We will examine this further below, but a clue to the causes of this loss of support lies in the period effects. The first significant period effect occurred in the 1979 election, when Mrs Thatcher won a comfortable majority and achieved a significant boost in Conservative support. However, it is evident that the negative impact of the loss of support for the party in the 1997 election was considerably greater than the positive effect of winning the election in 1979.Footnote ⁵⁶ Again, this suggests that the election which brought Tony Blair to power was a turning point in the Conservative Party’s history. In effect, the party has never fully recovered from Labour’s 1997 victory, since the socialization processes which renewed Conservative support in the years prior to that election were wiped out during and after the Blair era.

To summarize, cohort effects are strong for the Conservative Party and point in the direction of it losing support among younger voters. As regards period shocks, the Conservatives did well in 1979 and badly in 1997, and since then, specific elections have had no impact on Conservative support, despite winning in 2010 and again in 2015 and 2019. This shows that electoral success for the party is now dependent on the context of each election, much as it is for Labour. Labour lost the 2010 general election in the context of the aftermath of a serious financial crisis and subsequent Great Recession. But as analyses in Chapter 9 will show, although their effects are still playing out, the large-scale economic and social disruptions caused by Brexit and the Covid-19 pandemic have the potential to be more serious than those associated with the financial crisis of a decade ago. If so, as new voters join the electorate and old voters fade away, the Conservatives may find it increasingly difficult to win majorities in general elections. That said, the population in Britain is ageing and this could offset these trends to some extent, with the party retaining strong support among the elderly via life-cycle effects. Overall though, the results show that electoral politics is becoming more volatile and unpredictable over time.

Explaining the Dynamics of Turnout and Party Voting

Up to this point we have examined HAPC effects on turnout and party support for the two main parties and established that there are important differences between them with respect to the age-related voting. But we have not fully explained these differences except in rather general terms. In this section we probe the reasons for these differences in more depth. To do this, it is useful to draw on the three prominent theoretical explanations of electoral participation and party choice. These are the social class, valence and spatial models of voting behaviour, which we introduced in earlier chapters. We begin by briefly summarizing all three.

Extended Modelling Results

The results of the extended modelling of turnout appear in Figure 6.7. For technical reasons we use a single variable to identify period effects while retaining different variables for the cohort effects.Footnote ⁸⁵ To examine the individual-level fixed effects first, all variables in Figure 6.7 are statistically significant predictors of turnout with the sole exception of gender. Thus, older people are more likely to vote, but again the quadratic specification of age shows that the effect declines as people get even older. Occupational status and education have strong positive impacts on turnout whereas unemployment has a strong negative effect. This replicates the findings in the earlier analyses.

Figure 6.7 Hierarchical Age Cohort Extended Model of Turnout with Cohort and Period Controls, 1964–2019.

Turning next to the aggregate-level variables, the class-related measures, namely the proportions of professionals and workers in the workforce both influence turnout. However, the latter effect is much weaker than the former. Thus, the growth of middle-class professional occupations stimulates turnout much more than the decline in working-class occupations reduces it. Clearly, the individual level effects of occupational status described above are reinforced by these aggregate contextual effects, although it is important to remember the effect of relative status on turnout discussed earlier. The implication is that changes in the occupational structure will not have a large impact on turnout.

The valence measures are interesting in the turnout model, since there is evidence to suggest that both period inflation and period unemployment stimulate turnout, so in the latter case the contextual effect offsets the individual-level effect. This is largely because voters who are personally unemployed are likely to be demobilized by the experience, whereas national rates of unemployment are a symptom of recession which mobilizes some people to vote against the government.

Once again there are striking cohort effects in the modelling as was evident in Figure 6.4. In this extended model, turnout among the Blair, First World War and post–First World War cohorts is no different from those of the Austerity cohort. But all the other cohorts record higher participation. The First World War–related effects are likely to be a product of age, with the very old being less likely to vote than their middle-aged counterparts. But the Blair cohort effect, which is the same as the Austerity cohort effect, is likely to be influenced by the growing insecurity in the labour market, which young people have experienced since the Great Recession.

Macropartisanship is not a significant predictor in the turnout model, and nor is the difference between the ideological positions of Labour and the Conservatives as measured using the Manifesto data. Of course, both are period effects, since pre–Second World War data are not available to identify cohort effects associated with these measures. But it seems likely that cohort effects might be important if they were available, in view of the decline in partisanship observed in Table 6.3.

Finally, residual period and cohort effects were included in the analysis to pick up anything missed by the modelling. In addition, a dummy variable was included identifying if Labour was in office at the time of the election to determine if party incumbencies make a difference. The results show that there are additional period effects in the data, and a Labour incumbency effect, but there are no additional cohort effects. The period effect charts a long-term decline in turnout in successive elections, and the Labour incumbency effect denotes lower electoral participation when Labour was in power. The latter is exemplified by the turnout of 59 per cent in the 2001 election, the second election won by Tony Blair and the lowest turnout since the Second World War.

Figure 6.8 illustrates the impact of the different variables on voting for the governing party, a specification which allows us to examine valence effects on party support. To repeat an earlier point, the valence model emphasizes the performance of the governing party, with opposition parties gaining support from poor performance and losing it from good performance. This means that the incumbency status of a party is controlled in the modelling to avoid missing this distinction.

Figure 6.8 Hierarchical Age Cohort Extended Model of Voting for the Governing Party with Linear Period Controls, 1964–2019.

The results in Figure 6.8 indicate that in the individual-level model occupational status and unemployment are significant predictors of support for incumbent governments, but none of the other demographics are significant. This shows that, when indicators of the valence model are included, the relationship between individual demographics and support for the governing party are no longer important, with the exception of the workforce measures. The fact that age differences cease to be predictors in this extended model highlights the importance of including economic variables in the analysis. Put differently, the age effects we have observed are a lot to do with the impact of the economic circumstances and the experiences of young people.

With regard to the percentage of professionals and workers in the aggregate contextual model, the former boosts support for the incumbent party, while the latter is non-significant. This should be understood in the context of the Conservatives being in office for thirty-one of the fifty-five years between the 1964 and 2019 elections. The party has consistently been supported by middle-class professionals to a greater extent than Labour, although the gap is narrowing in recent elections. This means that class is still important in British politics, but the class divide favours the Conservatives in the long run. This is because the middle class are more likely to support them, than the working class are to support Labour.

The period inflation and period unemployment variables are both significant and negative in the analysis, which is consistent with the valence model, since economic problems reduce support for incumbent governments. In addition, the estimates show that unemployment has a bigger impact on incumbency support than does inflation.Footnote ⁸⁶ British governments have more to fear from rising unemployment than they have from increased inflation. However, it should be noted that inflation was only a serious problem in the 1970s and has been at relatively low levels ever since, which reduces its salience. That has changed in the post-pandemic period, but overall in the past half century, rising unemployment is more of a problem for successive British governments than price rises.

Once again, the most striking feature of the government support model in Figure 6.8 is the very strong cohort effects. All cohorts, including the Blair cohort, were significantly more likely to support the government than the Austerity cohort. This finding is the same as earlier findings in Figure 6.6, which show how much the younger generation has deserted the Conservative Party in office since 2010. The party may have won three elections in a row after Labour left office in 2010, but it is vulnerable to generational replacement in the long run, as these large cohort effects demonstrate.

Finally, macropartisanship clearly plays an important role in influencing support for the governing party. The finding implies that there is greater support for the governing party when macropartisanship moves in favour of the Conservatives. We observed in Table 6.3 that Conservative partisanship fell from 39.9 per cent in 1964 to 29.8 per cent in 2017. The decline for Labour is slightly greater, with 42.1 per cent identifying with the party in 1964 and only 28.4 per cent doing so in 2019.

Clearly, both parties have experienced a sizeable erosion of their partisan bases. The long-term implications of this depend on the extent to which partisanship is a largely static product of socialization in early adulthood or, alternatively, a dynamic ‘running tally’ of updated performance-related evaluations. Existing research suggests that in Britain the latter is more important than the former.Footnote ⁸⁷ So the long-run effects of declining partisanship at any point in time are mitigated by the possibility that increases in parties’ partisan shares are ongoing possibilities. That said, the weakening of early-life political socialization creates problems for both major British parties.

Conclusion: The Dynamics of Turnout and Party Support

This chapter has investigated both the short-run and long-run influences of age, period and cohort effects on turnout and party support in Britain. It is clear that the class and valence models help to explain the relationship between these different aspects of age-related voting in Britain over the fifty-five-year period between 1964 and 2019. With regard to turnout, there have always been differences in rates of electoral participation between age groups, but these became far more pronounced after the Labour landslide of 1997 when Tony Blair’s New Labour became the dominant party. This resulted in lower rates of voting among members of the Blair cohort, and this has subsequently carried over to the Austerity cohort.

Another important finding is that over this long period of time the Labour and Conservative HAPC models are very different. In the case of Labour, changes in support are related to different periods or elections with little evidence of enduring cohort effects. In contrast, the Conservatives appear to have relied heavily on socialization processes which strongly influenced their support prior to the New Labour era. These have subsequently largely disappeared, such that ‘socialization politics’ have become very much weaker.

The fragmentation of the class system as measured by changes in the occupational structure has weakened the relationship between social status, partisanship and electoral behaviour. The decline of traditional working-class occupations and the rise of managers and professionals have weakened support for Labour more than for the Conservatives. Labour won a lot of support from middle-class voters during the Blair years and subsequently the party has done better than expected in a number of relatively middle-class constituencies, even in the most recent 2019 election.Footnote ⁸⁸ At the same time, as Chapter 5 shows, the party lost a number of traditional working-class constituencies in the so-called ‘Red Wall’ of Labour strongholds in the North West in that election, illustrating the continuing fragmentation of the association between class and party affiliation.

Another finding is the lack of support for the spatial model of party competition. It is well established that divisive issues can play an important role in electoral politics, exemplified by the 2019 general election, which, as we have seen, was influenced by the issue of Brexit. However, that chapter revealed that the impact of Brexit on the vote was rather modest, with the possible exception of support for the Liberal Democrats. In the present context there were no significant contextual left–right ideological effects in the modelling, either in relation to turnout or support for the governing party. That said, macropartisanship is very influential in general elections, and this is clearly associated with ideology. But partisanship is much more strongly related to political identity and policy delivery than long-term support for left-wing or right-wing policies. The implication is that either the Manifesto data do not capture ideological differences very well, or more likely their influence on electoral behaviour has been exaggerated by the popularity of the spatial model.

It also bears emphasis that prosperity increases support for incumbents regardless of the party they represent, and recession reduces their support, in line with the valence model. This has important implications for future elections relating to the longer-term consequences of the protracted Brexit crisis, the ongoing Covid-19 pandemic and the cost of living crisis. This topic is discussed in more detail in Chapters 9 and 10. The responses to these shocks will vary between the parties and will be influenced by ideological differences between them over the next decade. The valence model suggests that these events will make life difficult for incumbents until prosperity returns and the economy recovers. There is an argument that the fragmentation of party support in Britain may make government less effective in the future, but the counterargument is that valence voting which focuses tightly on performance rather than ideology can actually improve accountability, a topic to be examined in Chapter 8.

This chapter has been a story of the fragmentation of party support and the strength of the relationship between partisan attachments and demographics such as age and social class over time. However, there is a second perspective on these political changes which relates to the geography of the vote – the spatial dimension in British politics. This is examined in the next chapter.