Context, research question, and literature review

The Brexit vote was followed by commentaries on young people’s low turnout at the poll. Even though the link between age and lower vote turnout has been established in the literature, the low turnout surprised as young people’s future was more likely to be affected by the outcome of the referendum and as a majority of them supported the Remain side.Footnote ² Some Op-Eds emerged, with titles in the vein of the FT’s “Young people feel betrayed by Brexit but gave up their voice” (Reference Khalaf2016).

In parallel, estimates of youth turnout varied and were revised over time, as no exit poll was conducted. Sky Data, Sky News polling branch, initially evaluated the 18–24 turnout at 36%, but polls conducted later by IpsosMori and Opinium estimated, respectively, that 60% and 64% of registered voters aged 18–24 voted. While not as striking as earlier figures, these numbers are still well below the average turnout of 72% for all age groups. The BBC further documented that counties with a higher share of young people had lower voting turnout during the referendum (BBC 2021).These numbers are disconcerting given the expected repercussions of Brexit for young people: those aged 18 to 24 will have to live with the consequences of the Brexit vote for an average of 69 years, compared with 16 years for those over 65 (Generation Citizen 2016).In light of this evidence, one could wonder whether some economic factors played a role in the lower than anticipated turnout for young people. In particular, substantial welfare cuts took place before the Brexit vote that seems to have played a role in the referendum’s results (Fetzer Reference Fetzer2019).

A wealth of research examines how political attitudes are developed. A number of studies focus on childhood factors and in particular the effect of early forces in the political socialisation of individuals. They highlight the role played by parents’ political orientation on a child’s political attitude formation (Jennings Reference Jennings1996; Jennings and Niemi Reference Jennings and Niemi1968; Jennings et al. Reference Jennings, Stoker and Bowers2009; Maccoby et al. Reference Maccoby, Matthews and Morton1954) as well as the educational system and peers’ (Campbell Reference Campbell1980; Jennings and Niemi Reference Jennings and Niemi1974; Tedin Reference Tedin1980), or even genetic influences (Alford et al. Reference Alford, Funk and Hibbing2005). Another area of research focuses on the influence of recent or contemporary events on political attitudes. Specifically, several articles have demonstrated how economic circumstances can affect political attitudes and electoral outcomes (Durr Reference Durr1993; Fiorina Reference Fiorina1978; Kinder and Kiewiet Reference Kinder and Kiewiet1981; Kramer Reference Kramer1983; Lewis-Beck and Stegmaier Reference Lewis-Beck and Stegmaier2000)

Margalit (Reference Margalit2019) lays out four possible outcomes of the impact of economic shocks on political behaviour:

• they can lead to an increase in support for left parties and redistributive policies

• the losers of the economic shock can embrace anti-establishment contenders and far-right parties

• they can trigger a vote against the incumbent

• they can lead to a reduced interest in politics and lower the voting turnout

The rise of populism that followed the Great Recession led to considerable research on the second point (see, e.g., Algan et al. (Reference Algan, Guriev, Papaioannou and Passari2017), Dustmann et al. (Reference Dustmann, Eichengreen, Otten, Sapir, Tabellini and Zoega2017), and Rhodes-Purdy et al. (Reference Rhodes-Purdy, Navarre and Utych2020)). However, recent articles also highlight the role of abstention in the rise of populist and radical right parties (Guiso et al. Reference Guiso, Herrera, Morelli and Sonno2017). Yet, out of the four outcomes Margalit (Reference Margalit2019) lists, abstention and a decrease in the interest in politics have been the least developed topic. In fact, evidence on the effect of economic shocks on turnout appears to be still limited and so far points to an interaction with the life cycle.

Indeed, two articles, Finseraas (Reference Finseraas2017) and Emmenegger et al. (Reference Emmenegger, Marx and Schraff2017), highlight that economic shocks that take place when young affect political interest and participation. Exploiting the discovery of oil outside the Norwegian county of Rogaland, Finseraas (Reference Finseraas2017) finds that cohorts that experienced a positive shock in family income during their childhood are more likely to vote. Emmenegger et al. (Reference Emmenegger, Marx and Schraff2017) find a symmetrical effect when analysing whether being unemployed interacts with the life cycle stage in depressing political interest. They use German panel data to show that unemployment spells among young adults trigger a drop in political interest, which is not visible later in life. They explain this finding by young people being more malleable by the economic conditions they live in while forming their opinion. This is consistent with the “impressionable years” or “formative years” theory from the psychology literature, according to which adolescence and early adulthood are more sensitive to major life events and can lead someone to revise their preferences (Krosnick and Alwin Reference Krosnick and Alwin1989; Sears Reference Sears1983). That susceptibility decreases thereafter and remains low during the rest of the life cycle.

Hence, this article adds to a small but growing literature on how the experience of economic shock during formative years increases young people’s detachment from politics. In particular, I test the following hypotheses:

• H1: The welfare cuts led to decreased political efficacy/sense of political representation among young people.

• H2: Such effect was not visible in other age groups.

I treat the welfare cuts as a negative income shock and interpret my results in line with the cited literature. However, I should note that the decrease in political efficacy I observe among young people could also reflect a political disappointment following the government’s decision to cut benefits. I am not able to disentangle these two effects.

Another limitation of my article is its focus on attitudes towards politicians and voting, which I measure with survey data. Therefore, I can only see how declared political beliefs evolve and do not evaluate any attitudinal feedback on real-life political outcomes. Nevertheless, although political interest and attitudes towards politicians are conceptually different from political behaviour, they remain important determinants of political participation and views of the political system (Brady et al. Reference Brady, Verba and Schlozman1995; Powell Reference Powell1986).

Finally, this article contributes to two additional strands of literature. The first one is in labour economics and suggests economic shocks have heterogeneous effects depending on when they occur in a person’s life. Negative labour shock when young have been shown to have long-term consequences in terms of earnings (Oreopoulos et al. Reference Oreopoulos, Von Wachter and Heisz2012; Schwandt and Von Wachter Reference Schwandt and Von Wachter2019), the likelihood of facing poverty (Schwandt and Von Wachter Reference Schwandt and Von Wachter2019), many social outcomes ranging from fertility decisions, marriage, and divorce to criminal activities, risky alcohol consumption, and lower health outcome (Von Wachter Reference Wachter2020), as well as the choice of major and line of job (Cotofan et al. Reference Cotofan, Cassar, Dur and Meier2023). Secondly, it adds to the debate on the political cost of austerity. While some articles find no link between austerity and a fall in popularity of governments and/or electoral defeat (Alesina and Ardagna Reference Alesina and Ardagna2010; Alesina et al. Reference Alesina, Perotti, Tavares, Obstfeld and Eichengreen1998; Arias and Stasavage Reference Arias and Stasavage2019), others observe a relation between fiscal consolidations and social instability (Ponticelli and Voth Reference Ponticelli and Voth2020; Vegh and Vuletin Reference Vegh and Vuletin2014). Finally, and as cited above, Fetzer (Reference Fetzer2019) shows that austerity in the UK caused a rise in the popularity of UKIP, the pro-Brexit political party. My results support the latter side of the debate.

Data

The post-recession period saw the implementation of severe austerity measures in the UK, triggered by the election of a Conservative-led coalition government in 2010. Fetzer (Reference Fetzer2019)describes three implementation phases. The first one consisted of budget cuts across most Westminster departments in 2010. Local governments experienced a decrease in funding to conduct their daily activities. Nominal wage freezes were then implemented for public sector employees from 2011 to 2013, with a public sector wage growth cap set at 1 per cent in 2014. Finally, a reform of the welfare state took place, with the 2012 Welfare Reform Act. This reform is the bulk of the austerity plan advocated by the ruling government and consisted of substantial welfare cuts among several dimensions.

Beatty and Fothergill (Reference Beatty and Fothergill2016)Footnote ³ provide a quantitative estimate of these welfare cuts, focusing on ten measures. These measures encompass changes in housing benefits, non-dependent deductions, household benefit caps, council tax benefits, disability living allowances, incapacity benefits, child benefits, tax credits, and the reduction in annual up-rating of working-age benefits. While the figures are estimates, they are “deeply rooted in official statistics” and come, for example, from “the Treasury’s own estimates of the financial savings, the government’s Impact Assessments, and benefit claimant data” (beatty2016, beatty2016, p.6). Overall, they calculate that these measures yielded savings of almost £14 billion a year by 2016. Taken as a share of the benefit claimants, this amounts to an average welfare loss of £345 per working-age adult, per annum.Footnote ⁴

This average hides considerable variation between districts. Indeed, the estimated annual financial losses per working-age adult range from £100 in the City of London to £720 in Blackpool, with a standard deviation of £84. Figure 1 displays the geographical variation of the austerity shock, per annum, across districts in the UK. I use this estimate as a proxy for the regional intensity of the austerity measures voted in 2012. In particular, I compare individuals across districts subject to different austerity shocks.

Figure 1.

Distribution of the austerity shock in the UK - Source: Beatty and Fothergill (Reference Beatty and Fothergill2016).

There are two important caveats to my analysis. First, the austerity shock’s regional variation is driven by the heterogeneity in the distribution of the benefits claimants over the country. This could imply a bias in my estimation as the districts most affected by the welfare cuts are mainly populated by vulnerable households. To address this issue, I include individual and district-level fixed effects in my estimation.

Second, the austerity shock provided by Beatty and Fothergill (Reference Beatty and Fothergill2016) contains a set of measures implemented before the election of the Conservative-led government. In particular, the Labour party introduced the Employment and Support Allowance (ESA). The Conservatives-led coalition then added new elements to it, such as the time-limiting and non-means-tested ESA. Therefore, my austerity shock is partly “contaminated” by additional measures that pre-date the 2012 Welfare Reform Act. Nevertheless, the vast majority of the austerity cuts took place as part of the Welfare Reform Act. In addition, I expect the austerity cuts to increase negative attitudes towards politicians and politics in general. Therefore, an overestimation of the fiscal shock will likely lead to underestimating the elasticity of political attitudes to austerity. Hence, my estimates should be interpreted as lower bounds.

I combine this regional austerity shock with survey data from the Understanding Society (UnderSoc) database (University of Essex 2019). The UnderSoc is a longitudinal survey run by the Institute for Social and Economic Research at the University of Essex, which replaced the British Household Panel Survey (BHPS) in 2009. It follows a representative sample of 40,000 British households across nine waves, from 2009 till 2019, and interviews them on socio-economic issues. One particularity, compared to BHPS, is that the interviews are conducted over 2.5 years and that the waves overlap. Table 1 maps each wave to its corresponding years. In general, each individual is interviewed around the same time every year, yet I structure my dataset as a wave-individual panel rather than a year-individual panel. The UnderSoc database provides the local authority district of residence of each respondent. Therefore, I can map each individual to its corresponding regional austerity shock, my main explanatory variable.

Table 1.

UnderSoc waves and corresponding years

I use five dependent variables of political attitudes from this survey:

1. 1. Public officials don’t care stems from the question “How far do you agree or disagree with the following statements? Public officials don’t care much about what people like me think.” Respondents can strongly agree, agree, neither agree/disagree, disagree, or strongly disagree. It is coded 1 if the respondent answered strongly agree or agree.

2. 2. No say in what government does follows the same format as above: people are asked whether they agree with the statement “People like me don’t have any say in what the government does.” It is coded 1 if the respondent answered strongly agree or agree.

3. 3. Interest in Politics is created from the question “How interested would you say you are in politics? Very interested, fairly interested, not very interested, or not at all interested.” It is coded 1 if the respondent answered very interested or fairly interested.

4. 4. Personal benefit from voting asks respondents whether they strongly agree, agree, neither agree nor disagree, disagree, or strongly disagree with the following statement “I feel a sense of satisfaction when I vote.” It is coded 1 if the respondent answered strongly agree or agree.

5. 5. Perceived political influence asks respondents “On a scale from 0 to 10, where 0 means very unlikely and 10 means very likely, how likely is it that your vote will make a difference in terms of which party wins the election in this constituency at the next general election?” and was entered as is in the regression.

These variables correspond to standard measures of political efficacy and sense of political representation. In particular, question 1 has been used as a measure of external political efficacy in the past, while questions 2 and 5 are usually used to measure internal political efficacy (Craig and Maggiotto Reference Craig and Maggiotto1982; Craig et al. Reference Craig, Niemi and Silver1990; Miller and Traugott Reference Miller and Traugott1989; Niemi et al. Reference Niemi, Craig and Mattei1991).Footnote ⁵ Question 1 and 4 are additional measures of political attitudes that correlate with voting.

While the first question is asked at all waves, the second and third ones are asked at waves 3, 6, and 9, and the latter two at waves 2, 3, 6, and 9Footnote ⁶ I choose to exclude the waves that occurred after wave 6 for two reasons. First, the Conservatives voted for new austerity measures in 2015, and hence, the austerity shock is different starting in 2016. In addition, the Brexit vote took place in June 2016 and likely led to significant adjustments in political attitudes. Indeed, intense political turmoil took place following the Brexit vote, and polls highlighted that some Brexiters regretted their voting choice.Footnote ⁷ Therefore, I choose to focus on the period pre-Brexit and post-Welfare Reform Act to estimate how austerity affected political attitudes in the UK.Footnote ⁸

I use the longitudinal nature of the survey and run a DiD where 2013 is the start of my treatment.Footnote ⁹ I include the longitudinal weights provided by UnderSoc to ensure that my analysis remains representative of the British population over the different waves. Further details about my empirical strategy are provided in section 4.

Table 2 provides summary statistics of my five dependent variables for waves 2, 3, and 6, for the whole sample and young people. On average, half of the respondents declare they are interested in politics, agree that public officials don’t care and that they have no say in what the government does, and feel a sense of satisfaction from voting. The mean perceived political influence is 3.22 (out of 10). Those numbers are lower for younger people, implying that their sense of political representation is lower than the population’s average. This is consistent with lower turnout rates and lower electoral registration among young people. The standard deviations in Table 2 are large as the variables are dummies or categorical.

Table 2.

Dependent variables - summary statistics

Estimated with longitudinal weights.

I construct indices following a principal component analysis to reduce the risk of multiple hypothesis testing and because my dependent variables are correlated (see Table 3). I group the variables according to their level of correlation and create a first index that measures the level of low political efficacy in my sample with the variables Public officials don’t care and No say in what government does, and a second one that measures the level of satisfaction with politics with the variables Interest in Politics, Personal benefit from voting, and Perceived political influence. Grouping the variables this way will also make the interpretation easier as the first and second indices correlate negatively.

Table 3.

Correlations between the outcome variables

Empirical strategy

DiD

I first run a standard DiD specification (equation 1) for young people and “non-young” people separately, using the austerity shock as treatment.

(1) $${y_{i,d,w,t}} = {\alpha _0} + {\alpha _1} \times {\bf{1}}(Year \ge 2013) \times Austerit{y_d} + {\lambda _i} + {\gamma _t} + {\nu _d} + {\varepsilon _{i,d,w,t}}$$

where ${y_{i,d,w,t}}$ corresponds to one of my outcome variables of political attitudes for individual i, living in district d, at wave w and time t; ${\alpha _0}$ is a constant intercept across all individuals, districts, wave, and periods; ${\lambda _i}$ is an individual fixed effect; ${\gamma _t}$ is a year fixed effect; ${\nu _d}$ is a district-level fixed effect. I cluster the standard errors at the year and district levels.

My main explanatory variable, $Austerit{y_d}$ , takes the value of the austerity shock in district d (described in Figure 1) when the year is superior or equal to 2013. This is equivalent to having it take the value of the austerity shock for wave 6 only. One particularity of my estimation is that the treatment is a continuous variable that applies to all individuals after 2013. My results come from the variation in the intensity of the treatment and imply linearity in reaction to the treatment.

Equation 1 is similar to Fetzer (Reference Fetzer2019)’s, with the exception that my explanatory variable is at the district level while my measure of political attitude is at the individual level. Hence, I aim to capture how macroeconomic conditions influence one’s views on government rather than individual conditions. Nevertheless, my measure of austerity stems from the heterogeneous distribution of benefit claimants over the UK. This implies that districts with more welfare beneficiaries will encounter a higher austerity shock and might be more prone to adverse consequences on political attitudes as their inhabitants rely more on the welfare state. To address this issue, I use fixed effects at the district and the individual level. The former controls for regional specificities that might explain why some regions react differently than others to the austerity shock, while the latter controls for the same characteristics for individuals. In particular, my district fixed effects allow me to control for the impact of secular structural shocks at the regional level, such as import competition or automation, that could impact political attitudes. The use of individual fixed effects enables me to compare the evolution of an individual’s reaction to the austerity shock, controlling for all the individual fixed characteristics that determine her/his preferences.

One should note that the fixed effects control for individual and community characteristics that may influence political engagement and may be correlated with the austerity shock, but not for the heterogeneous impacts of the austerity shock. Indeed, the impact on political attitudes could come from (i) districts with larger cuts having more people affected by the cuts, (ii) districts with larger cuts experiencing larger cuts in benefits, or (iii) people living in districts with larger cuts reacting more intensely to the cuts. I cannot distinguish between the three channels here.

I define young people as individuals that are 25 years old or younger.Footnote ¹⁰ Running equation 1 on the sample of young people will give us an estimate of how young people updated their political attitudes following the austerity shock. I can then compare it to the estimate I obtain when running the same equation on the sample of individuals that are older than 25.

For my results to be credible estimates, I need to assume that the parallel assumption holds between people living in different districts. This means assuming that the expected variation in political attitudes of people living the most affected districts and those living in the least affected ones would have been the same. I cannot test for the parallel trend assumption here and return to this issue in section 4.3. Comparing the estimates I get for young and non-young enables me to relax this assumption, providing that the bias is the same for the two groups living in the same district.

DiD with interaction terms

Another way to look at the effect of austerity on young people is to add interaction terms. I run a new specification, similar to equation equation 1, where I interact the continuous treatment variable interacted with a dummy for young people. In practice, I add the terms ${\beta _2} \times {\bf{1}}(Year \ge 2013) \times Youn{g_i} \times Austerit{y_d}$ and $Youn{g_i} \times {\gamma _t}$ to equation 1. It takes the following form:

(2) $$\matrix{ {{y_{i,d,w,t}} = {\beta _0} + {\beta _1} \times {\bf{1}}(Year \ge 2013) \times Austerit{y_d}} \hfill \cr { + {\beta _2} \times {\bf{1}}(Year \ge 2013) \times Youn{g_i} \times Austerit{y_d}} \hfill \cr { + {\lambda _i} + {\gamma _t} + Youn{g_i} \times {\gamma _t} + {\nu _d} + {\varepsilon _{i,d,w,t}}} \hfill \cr } $$

where $Youn{g_i}$ is a dummy for individuals younger than 25. I also cluster the standard errors at the year and district levels.

This second equation enables me to check whether young people have been more affected by the welfare cuts than their older peers. If significant, my coefficient ${\beta _2}$ would provide additional evidence of an heterogeneous effect of austerity on the British youth. This would be while controlling for the impact of the austerity shock in general with the term ${\beta _1} \times {\bf{1}}(Year \ge 2013) \times Austerit{y_d}$ and for the “usual” evolution of young people’s political attitudes over time with $Youn{g_i} \times {\gamma _t}$ .

While the specification is close to a triple DiD, it displays several important distinctions. First, a triple DiD would also include the interaction of each of the three components with each other and each component as a fixed effect. In this case, it means adding the interaction $Young \times Austerit{y_d}$ and a fixed effect for young people. Yet, including these items risks saturating my model as the individual fixed effects already control for these two items. Second, the treatment here affects everyone, with a different intensity. Ideally, with a triple DiD, the causal effect estimate will come from changes in the treatment units, and not changes in the control units. For this identification strategy to be a perfect triple DiD, I would need to have an austerity treatment that only affects young people, and not older cohorts.

Hence, specification 1 gives me an estimate of how people updated their attitudes following the austerity shock by comparing them to individuals from the same age category that lived in a different location. On the other hand, specification 2 provides an estimate of how young people reacted differently than elder peers subsequently to the welfare cuts, while controlling for a number of time variant specificities.