Increasingly Unequal. Electoral Participation and Political Inequalities in a Context of Decreasing Unionization in Italy (1983-2018)

This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record.


Introduction
Democratic theory postulates that citizens should be given the same opportunities to voice their own interests and preferences, and that the latter should be equally weighted by the political system (Dahl 1971(Dahl , 2008)).At least formally, the electoral process guarantees this democratic ideal to be fulfilled.
The very basic principle of one person-one vote gives each eligible citizen the chance to express her own preferences in the political process, while at the same time allowing the political system to give full consideration to citizens' stances.More realistically, however, not all citizens take part to the electoral process.To be true, this might not be necessarily a problem (actually, some degree of abstentionism could also be desirable for democratic systems, see Rosema 2007), to the extent abstentionism is evenly distributed across different segments of society.
Alarmingly enough, however, turnout in many advanced democracies has grown unequal in the last decades, thus meaning that specific social groups (with specific characteristics and interests) are increasingly failing to show up to the polls (e.g., Gallego 2015).This is in particular true when considering citizens with a lower socio-economic status (Gallego 2010(Gallego , 2015)).This trend has been often associated to the weakening of those mobilisation institutions which traditionally appealed to lower strata of society (Alford 1963;Verba & Nie 1972;Rosenstone & Hansen 1993;Verba et al. 1978).Amongst these, trade unions have been given a special attention, since these organisations have traditionally worked to integrate and mobilise lower social classes (Verba et al. 1978;Gray & Caul 2000;Radcliff & Davis 2000;Leighley & Nagler 2007;Flavin & Radcliff 2011), in fact playing as turnout equalisers.Accordingly, if trade unions are strong enough to carry out successful mobilizing strategies, their activities should favour the participation of socio-economically disadvantaged groups and thus reduce the turnout gap.
To the best of our knowledge, no study so far has systematically tested this argument in Italy, a country which experienced a steady and dramatic decrease of electoral turnout since the end of 1970's. 1 This paper attempts to fill this void by primarily analysing whether voter turnout is becoming increasingly unequal along the lines of socio-economic status (SES) in Italy.Second, it explores whether and to what extent the SES based turnout gap is affected by the strength of trade unions.
Finally, it assesses whether, in a context where trade unions are losing members as well as legitimacy (e.g., Visser 2006;Culpepper & Regan 2014), they are still able to play a role as turnout equalisers.
The primary hypothesis is that, as turnout declines, socioeconomically disadvantaged individuals are participating less compared to their better-off counterparts.If this was the case, then this should imply that a relevant share of the turnout decline should be in fact driven by increasing political inequalities.Furthermore, borrowing from mobilization theories of voter turnout, it explores the extent to which this trend might be mitigated by the role played by trade unions.
Understanding whether in this context turnout decline is unevenly distributed across different social strata is extremely relevant for several reasons.First, if socio-economic disadvantaged groups vote increasingly less compared to those who are better off, this means that what Lijphart referred to as unequal participation (1997) is on the rise.Political inequalities harm the health of a democratic system, as they imply that alienated groups would not receive a fair representation of their interests by the political system (Lijphart 1997;Verba et al. 1995).And in fact, political systems tend to be more responsive to the participative sectors of society, that is to say those groups which regularly show up to the polls (Franko et al. 2016;Bennet & Resnick 1990;Martin 2003).Second, despite the limited generalisability, the focus on the Italian case offers the chance to test existing theories in a country which has suffered from an intense downward trend of turnout, and where the underpinnings of this trend have not been sufficiently investigated yet.This is all the more relevant, given that the same kind of explanations might not be equally applicable in different contexts.The reasons behind declining turnout in one country, might not be the same as the ones found in other countries.As an example, generational explanations of turnout decline perform quite poorly in accounting for the turnout decline in Italy (see Tuorto 2018), although they are considered extremely valuable in other contexts (see e.g., Blais & Rubenson 2013;Kostelka & Blais 2021).This calls for the empirical testing of different theories within specific countries.Finally, understanding the reasons at the base of the decline of voter turnout is a necessary first step which might help policy makers to elaborate policies which might reverse the trend.
Relying on survey data collected by the Italian National Election Study (ITANES) from 1985 to 2018, our empirical analysis shows that the turnout gap between low and high socio-economic groups has dramatically increased already starting from the '80s.Most relevantly, this increasing gap is almost exclusively driven by decreasing levels of turnout among non-unionised socio-economically disadvantaged groups.Furthermore, it shows that, on the whole, de-unionization is negatively associated with the turnout gap.
The paper is structured as follows: the next section specifies the theoretical framework of the analysis and lay out the empirical hypotheses of the paper; data and methodology are presented in the third section, while empirical findings are discussed in the fourth section; conclusions follow.

Do low SES groups vote progressively less than high SES groups?
Classic studies on political participation have extensively focused on examining the association between social stratification and voter turnout, widely confirming the relationship between socioeconomic status and individual voter turnout.This connection was first postulated by the SES (i.e., socio-economic status) model of political participation (Verba & Nie, 1972;Verba et al., 1978;Rosenstone & Hansen, 1993;Verba et al., 1995) and subsequently confirmed by numerous studies worldwide, until recently (e.g., Milbrath et al., 1977;Rosenstone & Hansen, 1993;Bartels, 2008;Gallego, 2009;Nevitte et al., 2009;Gilens, 2012;Anderson & Beramendi, 2012;Armingeon & Schädel, 2015): less-educated individuals with lower incomes and occupational status tend to participate significantly less compared to their more affluent counterparts.
While the SES model quickly became the standard for predicting political participation, more uncertain have been the various arguments that have attempted to explain why a different distribution of material and symbolic resources among different social groups account for significant disparities in political participation.A fundamental (and, by now, widely accepted, see e.g., Nevitte et al., 2009) contribution in this regard came from Brady et al. (1995), who supplemented the traditional SES model with what is known as the resource model of participation.
The fundamental idea of the model is that political participation (broadly defined) is a challenging endeavour that necessitates individuals to allocate various resources, encompassing cognitive, economic, and time-related aspects.These resources are stratified based on the socio-economic status of individuals (Brady et al. 1995).Individuals with higher socio-economic status (i.e., those who enjoy high levels of income, better occupational status, and higher levels of education) have been consistently found to be more likely to vote, donate money to political campaigns, be involved in political groups and associations, and have their interests better represented, either in conventional politics or through interest groups and lobbies (e.g., Schlozman et al. 2012).This is because individuals with high SES possess a greater number of resources that reduce the costs of political participation.In brief, high SES citizens are more likely to be endowed with those material, cognitive, and symbolic resources that allow them to follow the complexities of politics, understand better the political process, gather and process autonomously political information, and develop a stronger sense of internal and external efficacy.All these elements together facilitate the participation into politics, reducing the costs of voting.For low SES people, instead, the costs of voting are relatively higher, something that traditionally hampered their involvement into politics.
In the heydays of mass politics, the adverse impact of resource scarcity among lower social groups was partly alleviated by the consolidation of mass organizations and parties (in particular, class-based parties such as socialist, social-democratic, and communist parties).These parties, in fact, directly appealed to socioeconomically disadvantaged groups.(Alford 1963;Lipset & Rokkan 1967;Bartolini & Mair 1990[2007]; Gallego 2010).And indeed, at least until the 80's, electoral participation in Western Europe was only partially influenced by the socio-economic position of voters (Armingeon & Schädel, 2015).This is because lower social groups were mobilized by specific agents of mobilization (Verba et al., 1978;Gray & Caul, 2000).
In their attempt to mobilize disadvantaged social groups, these organizations (in particular the parties of the left) also realised a network of collateral associations which, beyond the political arena, acted as agents of socialization for the lower social classes, providing a sense of belonging and a common ideological terrain which served as an engine for mobilisation (Duverger 1954;Lipset & Rokkan 1967;Kirchheimer 1966).Amongst the others, trade unions have played a major role in defending the interests of the working class (Gallego 2010), also promoting voter turnout of less advantaged social groups (Leighley & Nagler 2007).
In line with the mobilisation approach to political participation (e.g., Verba et al. 1978), the role of these organisations is then key to explain individual turnout, in particular among lower social classes: individuals participate when they are asked to, when they are mobilised by organisations which provide them with information and cues facilitating the act of voting, and thus reducing the barriers to participation (Brady et al. 1995).And all this should be even more relevant for lower social classes, as the latter are the ones that depends on cues more than higher social classes (Armingeon & Schädel, 2015).At the same time, mass organizations can facilitate (as in the case of the working-class movement) the participation of lower social strata by providing a sense of common identity, solidarity, and shared interests which can produce strong incentives for group mobilization.The participation gap between lower and higher social groups should be then reduced when the costs of voting for lower classes is subsidized by this kind of organizations.
It follows that the incentives for participation for lower socio-economic groups should be reduced to the extent the mobilising agents of these social groups are in decline (i.e., no longer able to appeal to this specific segment of the society).Evidence, in this respect, has shown that both parties of the left and trade unions are in fact facing hard times in the Western world.The literature on class voting has by now demonstrated that in many Western European countries class-bloc parties have lost their appeal towards the working class and that, more generally, the left is no longer able to massively mobilize its traditional electorate (Clark & Lipset 1991;Evans & Tilley 2012, 2017).Along the same lines, trade unions' membership in Europe has declined over time (Ebbinghaus & Visser 2000) and, as unionized labour force declined, the capacity and legitimacy of trade unions as important players in the policy making process has declined as well (e.g., Culpepper & Regan 2014: 723).
We thus expect that political inequalities based on socio-economic status should be on the rise, with lower social groups turning out progressively less than their better off counterpart over time.More formally: H1: The turnout gap between higher and lower social groups should be on the rise.
Are trade unions still able to mobilise lower social groups?
While based on the existing literature we expect that socio-economically based political inequalities should be on the rise, this expectation explicitly assumes that this trend is, at least in part, due to the weakening of those mass organizations which appealed to lower social groups.In this respect, a particular attention should be given to the role of trade unions, given their traditional role as mobilising institutions of lower social classes (Flavin & Radcliff 2011, p. 633).
There are at least two different perspectives that one should consider when looking at the relationship between trade unions and turnout: one referred to unionisation strength, intended as an aggregatelevel variable; and the other, instead, referred to individual unions' membership.Unionisation strength concerns the aggregate association between turnout and the strength of trade unions in a given context, as measured, for example, by trade union density.In this respect, several scholars have shown that turnout in the aggregate is higher in those countries where unionisation rates are higher (e.g., Gray & Caul 2000;Radcliff & Davis 2000).This implies that unionisation has an effect on turnout that goes beyond the mobilisation of trade unions members.This might be due to the fact that, for example, during an electoral campaign, strong trade unions might be able to target and mobilise also non-members whose interests are however aligned with those of trade unions' members, and which are advocated for by trade unions.Or, the defence of specific interests made by trade unions could trigger a counter-reaction on the side of those who are opposed to the advocated interests and who might decide to go to the polls to avoid that pro-union parties could win the elections.However, since the bulk of trade unions' members (and sympathisers) was drawn in the past especially from lower socio-economic groups (Flavin & Radcliff 2011), the shrinkage of trade unions' membership should affect turnout more severely among lower social groups.
In relation to unions' membership, instead, we refer to the individual-level effect on turnout which might be produced by the individual membership (due, for example, to socialization processes within the organisation).Traditionally, trade unions' members have been found to vote more compared to non-union members.One of the mechanisms underpinning this pattern is the one postulated by the civic voluntarism model, suggesting that membership in associations (be them political or not) allows citizens to gain those skills which enhance political involvement (i.e., knowledge about politics, political interest, etc.) and consolidate a habit of voting.In addition, trade unions favoured the political involvement of members by defending and voicing their interests.In this sense, trade unions worked as participation equalizers, as they compensated the lack of politically relevant resources of low SES members.In this regard, recent studies have shown that individual membership in trade unions is in fact still a good predictor of political attitudes as well as political behaviours.Trade However, the association between unionization and equal turnout was particularly relevant in the past, when the bulk of trade unions' members was made up by working class people and when trade unions were mostly focused on the defence of class interests.However, the internal composition of trade unions has changed in the last decades: there is evidence showing that the decline of unionization in Europe has been mostly concentrated among the working class (Visser 2006;Gallego 2015) and that, today, unions' members are, on average, as educated as the general population (Gallego 2015).If this is true, the mobilizing efforts of trade unions might be no longer directed to low SES citizens only; from a rational point of view, they would rather try to please the interests of a more heterogeneous membership.
The question then is whether, in a context of declining and changing composition of unions' membership, trade unions are still able to play a role as participation equalizer.On the one hand, one could expect that, given the broader range of interests expressed by a more heterogenous base, trade unions should be willing to mobilize the interests of both low and high SES groups.If then trade unions' membership had a positive effect on turnout, this should be the same across different social groups (that is to say, a relatively small capacity to close the turnout gap) (Leighley & Nagler 2007;Gallego 2015).On the other, however, we could still expect that any effect of trade unions' membership on turnout should be stronger among low SES groups in the first place, as these are the groups that, more than any other, need information and politically relevant skills to get engaged in politics.
If it is true, then, that there are good reasons to suppose that de-unionisation might have contributed to the decline of turnout among lower classes, this does not mean that trade unions have automatically ceased to mobilize their own members (especially low SES ones).On the one hand, de-unionization might have reduced the overall pool of voters potentially mobilised by trade unions in the aggregate (especially among lower social classes); on the other, trade unions might have maintained their mobilising capacity among their own members (and, maybe, especially among low SES members, who are more in need of politically relevant resources).
On this basis, we advance two expectations.First, if the SES-based turnout gap is a consequence of the weakening of trade unions, we should then expect that unionisation strength (in the aggregate) should be negatively associated with the turnout gap between higher and lower socioeconomic groups.Second, despite the increasing weakness of trade unions, the latter should be still able to influence and, possibly, mobilise at least their own members (in particular, among low SES individuals), thus acting as turnout equalisers between lower and higher socioeconomic groups.More formally: H2: The higher the de-unionisation rate, the larger the gap between higher and lower socioeconomic groups.
H3: The turnout gap between higher and lower socioeconomic groups should be moderated by trade unions' individual membership.

The Italian context
Since the first democratic election after WWII and up to the end of the 70s, electoral participation in Italy has been among the highest in the Western world (Figure 1) (Corbetta & Schadee 1982, Mannheimer & Zajczyk 1982;Caramani 1996).With an average of over 90% throughout this period, the turnout even reached almost 100% in some territories (Tuorto 2018).On the one hand, these high levels of turnout for a country that has always scored poorly in terms of civic culture (Almond & Verba 1963), is to be attributed to the diffused perception of the vote as a moral duty among citizens (e.g., Mannheimer & Sani 2001).On the other, high turnout levels were a product of the specific features of the Italian political system, such as the structure of the political competition, the presence of compulsory voting (although only formally), the proportional nature of the electoral system (e.g., Corbetta & Parisi 1987, 1994;Corbetta & Schadee 1982;Caramani 1996).In particular, the period between 1946 and 1992, also known as the First Republic, was marked by extreme levels of polarization of the party system (Sartori 1976).Party competition between the two major parties of the Italian political system (i.e., the Christian Democratic Party (DC) and the Communist Party (PCI)) reflected a system of social fractures in which the class cleavage was flanked by (and to a certain extent overlapped with) the religious cleavage.The profound lines of divisions between the Christian Democrats and Communists (which clearly reflected deep societal conflicts) had favoured the consolidation of Catholic and Communist subcultures in specific areas and regions of the country, something that undoubtedly had relevant implications in terms of mobilization (e.g., Tuorto 2018).
Both parties were able to consolidate strong socio-political allegiances with specific social groups which were massively mobilized at each election.This also allowed to bring to the polls those segments of the society which were traditionally more difficult to be mobilized (e.g., lower SES groups).the massive corruption scandal of Tangentopoli (i.e., Bribesville), which discredited the established political parties and broke down the party system which ruled Italy since the end of WWII.The delegitimation of the Italian political system that followed, produced disaffection towards conventional politics among voters, something that contributed to accelerate the decline of electoral turnout.Since the beginning of the Second Republic turnout has then declined interruptedly until the last general election of 2022, when turnout reached its lowest level (63.9%).
In part, this negative trend has been considered as the product of changes occurred on the demand side.Some scholars have pointed to the demographic changes of the Italian society, arguing that as the Italian population got older, the proportion of voters who might have been more prone to abstention (due, for example, to illness) has increased (e.g., Mannheimer & Sani 2001).Others, instead, have mostly referred to cultural changes which have invested the Italian society (e.g., Mannheimer & Sani 2001;Raniolo 2007).Generational approaches, for example, postulate that younger generations, socialised in a context of increasing affluence and well-being, might have developed post-materialist values, something that should predispose them to prefer more fluid and less hierarchical forms of political participation, such as turnout (Inglehart 1977;Dalton 2006Dalton , 2007)).
At the same time, they might be characterized by a more cynical and detached approach to institutional politics and more conventional forms of political participation.However, empirical evidence has shown that the generational approach has only a limited explanatory capacity of the negative trend of turnout in Italy (see Tuorto 2018).
A second line of thought, instead, has prevalently focused on the supply side of politics, positing that the decline of turnout should be mostly imputed to the fact that parties and other traditional agents of political mobilisation have progressively lost their capacity to remain in touch with voters (Corbetta & Parisi 1987, 1994), something that spurt increasing levels of apathy and discontent among citizens (Cerruto 2012).
In this paper, we borrow from this latter approach, in fact hypothesising that the decline of turnout is, at least in part, due to the demobilisation of lower socioeconomic groups, and that the latter is associated with the weakening of mobilising agents which traditionally appealed to these segments of the society.In Italy, this weakness is well visible from the decline of class voting and the reduced appeal of the class left on lower social classes (e.g., Bellucci 2001) as well as from the decline of unionisation rates.The latter are unequivocally depicted in Figure 2, which reports the trend of trade union density in Italy for the period 1983-2018 (the period covered by our study, see below).
In 2018 the unionised labour force in Italy constituted the 32.6 of the total number of employees, a figure that is 13 percentage points lower compared to what is observable in 1983, when the unionisation rate was 45.5%.

Data and methods
We test our hypotheses combining 10 waves of the Italian National Elections Studies (ITANES), starting from 1985 until 2018.On the whole, this longitudinal, individual-level dataset covers 10 national elections, starting from the general elections of 1983 (ITANES wave 1985) until the general elections of 2018 (ITANES wave 2018), with an overall number of observations which is 21,524.We excluded from the dataset the ITANES waves conducted prior to 1985 for two reasons.First, from a substantive point of view, turnout decline in Italy became relevant starting from the '80s, while, as we have already argued above, in previous decades it was extremely high and without significant oscillation over time.Therefore, we take into considerations those waves which can reliably map the declining trend in the aggregate.Second, and from an operational point of view, we could not include in our analyses the ITANES waves collected before 1985 (and referring to the Italian general elections held before the 1980s), because of the numerical instability in some key variables for our analyses (for example, in 1983 the number of respondents falling in the higher education category (e.g., university degree) is problematically small, especially when including the variable in a multivariate model).
Typically, ITANES conducted post-electoral surveys in the aftermath of each general election.In some cases, however, the study has been enriched by a panel design, including both a pre-and a postelectoral survey.For all those ITANES waves which feature both a pre-and a post-electoral component, we have always kept the post-electoral one.A prospect of all the ITANES waves included in this study (with reference to the specific election covered by the ITANES wave and the relative sample size) is reported in Table 1.Our dependent variable is the self-reported individual turnout at the general election.In all the waves included in our dataset, individual turnout is measured by the classical vote recall question.The variable is simply coded as a dichotomy, with 1=voted and 0=did not vote.No answers and those who did not recall whether they voted or not are excluded from the analysis.Given the dichotomous nature of our dependent variable, a series of logistic models are estimated to assess the effect of our independent variables on individual turnout.Self-reported turnout, however, is all but not unproblematic.Individual-level surveys, in fact, tend to overreport turnout, a bias that is mostly due to social desirability (Karp & Brockington, 2005).As a consequence, to avoid severe distortions in our estimations, we weighted our data based on official turnout figures.
Our focal predictor is the socio-economic status of voters, as captured by their education level.We measured education as a dummy variable, with 0 including low educated respondents (holding primary education) and 1, instead, middle or high educated ones (holding secondary or tertiary education).Although SES is usually measured relying on different variables (i.e., education, income, and occupational class), in this paper we mainly relied on the educational level of the respondents, as this is in fact amongst the most powerful predictor of political participation (e.g., Schlozman et al. 2012) and, most relevantly, a structuring factor of the socioeconomic status of individuals, as both income and occupational class are, to some extent, dependent on educational attainment.
We are aware, however, that controversy does exist among political scientists in relation to how to interpret the positive association between education and turnout.While some scholars refer to education as a direct cause of political participation (i.e., education "teaches specific skills and knowledge" that foster political participation, Willeck & Mendelberg 2022: 90; see also Verba et (Willeck & Mendelberg 2022).Up to now, the studies investigating increasing participatory inequalities based on socio-economic conditions and employing education as a measure of SES are a multitude: Verba et al. (1995) and Brady et al. (1995) already conceived education as a fundamental component of SES, and they used education to assess the association between socio-economic status and political participation.Along the same lines, Gallego too (2010;2015) relied on education to measure socio-economic disparities and their impact on political participation.More recently, Armingeon & Schädel (2015) explored the participatory gap between social groups in Western Europe using education as a measure for social position.As this paper directly contributes to this stream of literature, we preferred to maintain consistency and to use education as a measure of SES in our main analyses.However, we also replicated our models by relying on a measure of SES as derived from the occupational class of respondents. 2The results of these analyses, reported in Appendix A, are consistent with our main findings.
As for our measure of trade union membership, we employed a dummy variable, coded as 1 if the respondent has ever been part of a trade union; the value of 0 is instead assigned to those who do not belong to any trade union.As for the overall strength of trade unions in the country, we used the trade union density in each election, as retrieved from the OECD data.This measure reports the proportion of the unionized labour force on the total number of employees. 3e also include controls for a set of standard predictors of individual turnout (see Smets & Van Ham 2013 for a review).We include gender as a dummy variable (1=Woman), to account for the fact that usually men are found to turn out at higher rates compared to women (Verba et al., 1995).Age as well is a standard predictor of voter turnout.In particular, younger people are more likely to abstain, compared to mid-age voters.At the same time, the probability of turning out declines again in old age.To account for the curvilinear effect of age, we thus included a variable distinguishing different age-cohorts (18-34; 35-54; 55+).The variable is plugged into the models as a set of dummies (with the category 18-34 serving as a baseline), thus allowing us to control for the non-linear effect of age on turnout.
Following the mobilization model of political participation, beside the effect of trade unions' membership, we also consider the effect of church attendance, a variable which is hypothesised to be positively associated with individual turnout.Church attendance is measured on a 5-point scale, with 1= "Every week" and 5= "Never".A dummy variable is then included to control for political interest (1= "Very interested/Interested"; 0= "Not interested/Not at all interested"), as more politically interested people are more likely to turn out compared to not-interested ones.
As turnout in Italy varies considerably across regions (with northern regions usually turning out at higher rates compared to southern ones), we included as a control variable the geographical area in which the respondents live (North; Centre; South).This variable has been included as a set of dummy variables, with North serving as a baseline.We also took into consideration the delicate transition from the First to the Second Republic, and the political consequences produced by it, including a dummy variable distinguishing the elections held during the First Republic from those instead held during the Second Republic.Finally, the effect of time on turnout is captured by a linear term measuring the year of each election covered by our data.

Assessing the turnout gap between low and high SES groups
Our analysis starts by first assessing the average effect of education (our indicator of SES) on turnout.
Model 1 in Table 2 reports the bivariate effect of education on turnout considering the pooled dataset. 4ot surprisingly and in line with the SES model of participation, we found that better educated people, on average, participate more than those who are poorly educated.The logit coefficient for those holding a secondary or university degree indicates in fact that the log odds of turning out at the elections is significantly higher compared to those holding just elementary education (logit=0.504,p<0.001).This effect is further confirmed in Model 2, where besides education, we plugged in our models all control variables.Once again, results are consistent with existing evidence showing that better educated turn out at higher rates as compared to poorly educated people.5 Furthermore, in line with the mobilization model of political participation, we find that the effects of trade unions' membership, church attendance, and political interest are all significant and in the expected direction.
Specifically, members of trade unions, churchgoers, and politically interested individuals are more likely to go to the polls.We also confirm that people living in southern Italy are less likely to turn out compared to those living in northern Italy, and that turnout in the period of the Second Republic is, on average, lower compared to the First Republic.Finally, it is interesting to note that the effect for gender is negative and statistically significant, indicating that, on average, women participate less compared to men.As puzzling as this result might seem in light of a growing literature showing that the gender-related gap in turnout has decreased in many advanced democracies (e.g., Carreras 2018), our results are consistent with the existing literature demonstrating that, in the specific case of Italy, women are still less likely to vote compared to men (see, e.g., Tuorto and Sartori 2021).
While these results are overall reassuring about the quality of our data, they do not tell us anything about the evolution of the turnout gap between higher and lower social groups.To assess whether and how this gap has changed over time, we estimated the effect of education on turnout as moderated by the effect of time.The results of these interactive models are presented in Models 3 and 4. In Model 3 we let the effect of education interact with a linear term for time, without including control variables (something that allows us to leverage the information coming from all the ITANES waves).
In Model 4, instead, we include control variables.On the one hand, this allows to have a more accurate estimation of the effect of education on turnout over time, net of possible confounders.On the other, the inclusion of control variables reduces the number of available ITANES waves included in the analysis (as some controls are not consistently covered in all ITANES waves, see fn. 5).To ease the interpretation of the findings, we also present the results of Model 3 in graphical form in Figure 3.
The latter displays the predicted probability of turning out for different education levels over time.
Looking at the figure, it is more than clear that starting from the '80s, turnout has decreased across all social groups.However, the decline for lowly educated people is staggering and much more pronounced compared to better educated people.While still in the 1983 election, low and high educated people voted at the same rates, the turnout gap between these two groups has steadily increased in the last three decades.In 1987, the probability of turning out of high educated people was 3 percentage points higher compared to low educated people; in the general election of 2018 it increased up to 12 percentage points.It is also interesting to notice that this trend started well before the collapse of the First Republic, thus suggesting that, although the transition from the First to the Second Republic might have accelerated the process, the latter was already unfolding in the years preceding the decomposition of the Italian political system in 1992.Furthermore, this finding is fully confirmed in Model 4, when control variables are included in the analysis.Overall, these results clearly lend support to our first hypothesis: the turnout gap between higher and lower socio-economic groups has significantly increased over time.2. Note: A replication of this analysis, employing occupational class as a measure of SES, is reported in Table 1A and Figure 1A of Appendix A.

Does unionization affect the turnout gap?
As anticipated above, one of the factors behind this trend might be linked to the weakening of those mobilizing agents which traditionally appealed to lower classes.Amongst these, trade unions have certainly had a prominent role.If this assumption is correct, we should find that the turnout gap between low and high educated people should be associated with the strength of trade unions.Specifically, we should expect that the gap between high and low educated voters should be lower when trade unions are stronger.
We tested this hypothesis using as our dependent variable the predicted turnout gap, as obtained by the estimation of a series of bivariate logistic models in each single ITANES wave.We then plotted the turnout gap against our measure of trade unions' strength (i.e., trade union density).The association between the two variables is reported in Figure 4.The y-axis reports the gap in turnout as obtained by the estimated log-odds of turning out of high educated people vs low educated people in each election.The x-axis, instead, reports the trade unions density in correspondence to each election.
The relation between the two variables, as expected, is negative, thus showing that as trade-union density decreases, the socio-economic gap in turnout significantly increases R correlation is equal to -0.8, p<0.001).In particular, the turnout gap between high and low educated people remains a significant one as long as trade unions appear relatively weaker, while it becomes not significant, from a statistical point of view, when the trade unions density increases.We are aware that these results do have limitations.First, the number of observations we are using here is small (N=10); second, and perhaps most importantly, the association between trade union density and the turnout gap might be the result of the common trending over time of the two variables (i.e., trade union density decreases over time while the turnout gap increases over time, following a similar, although in different directions, trend).We tend, however, to rule out this possibility.First, when looking at Figure 4, we notice that the association between trade union density and the turnout gap does not follow a linear trend over time (this is true in particular after the 90's); second, the association between time and the turnout gap is lower and less significant (R=0.7;p<0.05) than the association between trade union density and the turnout gap (R=-0.8;p<0.01); finally, when we regress the turnout gap on trade union density and time separately, we observe an explanatory capacity of trade union density which is higher as compared to time (65% and 54% of variance explained respectively).
Of course, these results are far from demonstrating any causal relationship between unionization and the turnout gap, and other factors should be controlled for, which cannot be included in this analysis (e.g., institutional factors, party strategies, social transformations, just to name a few).However, while cannot establish a causal relationship between unionization and the turnout gap, we interpret these results as providing some evidence which corroborate our hypothesis: the turnout gap between low and high educated people tend to increase as unionization decreases.

And yet it moves…?
While the turnout gap between low and high educated people seems to be connected to deunionization, it is still to be verified whether trade unions are still able to favour turnout equalization.
Following our H3, we expect that trade union members should participate more as compared to nontrade union members and that the decline of turnout should be less pronounced among unionized people belonging to lower social classes.In other words, membership in trade unions should reduce the turnout gap between lower and higher social classes.
We test this hypothesis estimating a logistic regression model, where we let education interact with both time and trade union membership in a three-way interaction (Model 5, Table 3).We estimate this model including control variables, exactly as we did in Table 2, Models 3 and 4. To make the results more easily understandable, we plot the predicted probability of turning out for different levels of education sorted by trade union membership (Figure 5).From Figure 5, we first observe that the probability of turnout among non-unionized and low-educated people declines significantly more over time compared to non-unionized and high-educated individuals (left-hand panel).This results in an increasing participatory gap between non-unionized citizens with high and low education levels.
The same pattern does not emerge when considering the unionized SES groups (right-hand panel).In this case, there is no clear divergence in the turnout trends between lower and higher social groups, indicating that the decline in turnout for unionized low SES groups follows the same pace as that of unionized high SES groups.Second, we can see that turnout declines more rapidly among nonunionized and low SES individuals, not only compared to non-unionized and highly educated people but also in comparison to both unionized high and low SES individuals.
All in all, these results confirm our third hypothesis, showing that the decline of turnout among lower socio-economic groups is in fact moderated by their union membership.In other words, unionization is still able, according to our data, to provide incentives for participation among lower social classes, something that in fact tends to favour a more equal political participation.

Conclusion
While voter turnout has been decreasing in Italy for nearly four decades, it has remained unclear whether this decline has been accompanied by a rise in participatory inequalities.In this paper, we directed our efforts toward understanding two key aspects: first, whether unequal participation is indeed increasing, and second, the extent to which this trend can be (at least in part) attributed to the influence of trade unions.
Relying on survey data taken from the Italian National Election Studies, our empirical analyses showed that indeed the decline of turnout in Italy is, at least in part, driven by a disproportionate decline of turnout among lower social classes.In other words, turnout is becoming increasingly unequal in Italy.While in the '80s, low and high social classes participated in the elections at almost the same rate, as time passed, the gap between these two groups has significantly increased.
In this perspective, Italy is not different from other advanced democracies where turnout has been shown to have grown unequal over time (see e.g., Armingeon & Schädel, 2015).
In trying to understand the factors that might explain this increasing gap, we focused our attention on the role played by trade unions, which are traditionally considered key actors for the mobilization of low SES groups.In this respect, we provided some evidence supporting the idea that de-unionization is among the factors behind increasing turnout inequalities.At the same time, we have shown that, even though de-unionization is associated with an increase in the turnout gap, trade unions are still able to provide incentives for electoral participation that are apparently higher for low SES groups compared to high SES ones.In other words, in a context where trade unions are losing members as well as their legitimacy, they are still able to play a role as turnout equalizers.
Taken as a whole, these results lead us to emphasize four key points for reflection.First, electoral participation in Italy is becoming increasingly unequal, as demonstrated by the growing influence of socioeconomic status (SES) in explaining electoral participation.This rise in political inequalities might, in turn, result in a decreased representation of the interests of lower social classes.This is alarming, considering that, at least normatively, representative democracy is built on the principle of equal participation and representation for all citizens in the democratic process.
Second, the increasing weight of SES in determining electoral participation suggests that one of the reasons for the dramatic decline in voter turnout in Italy is the growing and relatively greater reluctance of lower classes to take part in the electoral process.Part of the story is related to the weakening of those mobilization agents that traditionally appealed to lower social classes.
However, further research is certainly needed to understand other factors that might explain why turnout is becoming more unequal.
Third, and related to our previous point, a potential check on the unequal growth of participation comes from the mobilization role played by organizations and associations active in society.In line with the mobilization approach to political participation, our data clearly show that if individuals are members of a trade union, citizens from lower social classes participate at the same level as those from higher social classes.This data suggests that associations such as trade unions are still able to provide information and skills that are functional for mobilizing their members, particularly when these members come from lower social groups.It is indeed the lower social classes that have a greater need for (and are more receptive to) the mobilization efforts of trade unions.
This also explains why, and this is our fourth point of reflection, the decline in trade union density might translate into a greater participatory gap between lower and upper classes.Because it is the lower classes that are most affected by the lack of cues, information, and skills coming from associations, the decline in membership in these organizations will have an asymmetric effect on different social classes, particularly affecting those classes with a greater need for guidance in the political process.
More in general, and besides the direct implications for the Italian context, our findings testify to the critical importance that political and non-political associations can have for the well-being of a democratic regime.The formal right to vote alone does not prevent the consolidation of patterns of participatory inequalities that clearly contrast with the crucial democratic principle that citizens' preferences should be equally weighted by the political system.As long as specific social groups lack those relevant skills necessary to activate them politically, the role played by associations that can compensate for the lack of these skills remains a crucial ingredient for the well-functioning of a democracy.From this perspective, future research as well as political decision-makers should focus their efforts not only on identifying formal mechanisms that can curb the detachment from democratic life (clearly epitomized by the decline of political participation in many advanced democracies) but also on identifying courses of action that can revitalize the associational life of a democratic society.
unions' members are, for example, still more likely to support redistributive policies (e.g., Mosimann & Pontusson 2022), although this depends on the type of trade union.Along the same line, working class voters have been found to be less likely to abstain, if members of a trade union (e.g., Renwald & Pontusson 2021).

Figure 1 -
Figure 1 -Electoral turnout in Italy.Source: Italian Centre for Electoral Studies.Note: In the period 1979-2001 Italian voters abroad were included in the computation of turnout.

Figure 3 -
Figure 3 -Predicted probabilities of turning out by education level (95% CIs).Note: estimates are derived from Model 3, Table2.Note: A replication of this analysis, employing occupational class as

Figure 4 -
Figure 4 -Association between trade union density and the predicted turnout gap between high and low educated people (95% CIs).

Table 1 -
Prospect of the ITANES waves employed in this study.

Table 3 -
Logistic regression.DV: Self-reported turnout Predicted probabilities of turning out at election by education level and trade unions' membership (95% CIs).Note: A replication of this analysis, employing occupational class as measure of SES, is reported in Table1Aand Figure2Aof Appendix A.