1The spring 2017 election cycle was characterized by a rise in abstention. Fewer people voted in the presidential election compared to 2012, with abstention increasing by 1.7 percentage points in the first round. [3] This increase is noticeably more marked when compared with the first round of 2007: an increase of 6 points. The second round saw an even sharper level of disengagement. In an atypical development, abstention increased between the two presidential rounds, reaching its highest level since the 1969 clash between Alain Poher and Georges Pompidou: 25.4%. This is an increase of 5.6 points in comparison with the second round of 2012, and 9.4 points in comparison to that of 2007.

2This trend of diminishing voter turnout was strikingly confirmed during the June 2017 legislative elections. For the first time in the legislative elections, among those on the electoral roll, abstainers were more numerous than voters, amounting to 51.3%. And participation further decreased between the two rounds, leading to a record rate of abstention: 57.4%. This is an increase of 12.8 points in comparison with 2012, and 17.3 points in comparison with 2007. Even the 2015 departmental elections brought more people to the ballot box. The abstention rate for the 2017 legislative elections is similar to that typically recorded during a “second tier” election, such as the European elections.

Is the “Sociological” Model of Abstention Still Relevant?

3These high abstention rates form a part of certain long-term developments in electoral behavior. The amplitude of the turnout differential between the presidential and legislative elections (an increase of 35 points between the first round of the presidential election and the second round of the legislative elections) can be explained by a sharp reduction of the group of consistent voters, within a context where irregular turnout is increasingly becoming the norm in elections. While around half of all registered voters participated in all rounds of voting in both 2007 (50.9%) and 2012 (47.8%), this number fell to 35.9% in 2017 (Insee première 2007, 2012, 2017). [4] Systematic abstention saw a slight increase: plus 4.8 points in comparison with 2007, and plus 2.5 points in comparison with 2012, standing at 15.6%. In this context, intermittent voting has become predominant, with 50.8% of voters displaying this behavior. If confirmed by future elections, this significant increase in intermittent voting will constitute a major development in citizens’ relationship to voting.

4Beyond these preliminary observations, abstention levels for spring 2017 raise questions about the relevance of the analytical and interpretive categories habitually employed to account for nonvoting. The widespread abstention and the fact that consistent voters account for no more than a third of those voting in the presidential and legislative elections cast doubt on the validity of “sociological” explanations of participation. Are we witnessing the propagation, or even the generalization, of new abstention behaviors that are more “political” and therefore less demographically and socially determined than in the past? Are we seeing the development of a form of abstention that is more socially diverse and therefore less representative of electoral disparities? In short, have variables such as age or level of educational attainment, hitherto the most effective predictors in terms of abstention not only in France but also elsewhere, lost their statistical weight and explanatory power? [5]

5These questions can be answered by way of a certain number of empirical tests. This is firstly a matter of verifying whether or not the traditional factors of abstention, beginning with age and level of educational attainment, but also including professions and socio-professional categories (professions et catégories socioprofessionnelles, PCS), as well as contextual factors such as marital status and type of cohabitation, still retain significant explanatory power regarding the behaviors observed during the most recent presidential and legislative elections. It is also a matter of testing, from both a synchronic (between the presidential and legislative elections of 2017) and a diachronic perspective (between 2007, 2012, and 2017), whether or not the increase in abstention is indicative of a weakening of the sociological determinants of participation, and therefore of a reduction in disparities in voter turnout between different age groups and different social categories.

INSEE’s Voter Turnout Survey

6The databases of the voter turnout survey conducted by INSEE during each presidential election since 1988 provide an excellent and internationally unparalleled resource for tackling these questions. This type of survey is based on a representative sample of voters registered on the electoral rolls in each metropolitan area. It boasts five methodological advantages over traditional surveys.

7First of all, the sample is ensured to be representative of the electoral population by the fact that individuals are directly selected by lottery. The voter turnout survey is based on the Permanent Demographic Sample (Échantillon démographique permanent, EDP), a sociodemographic sample group that includes all individuals present on French territory born between the second and the fifth of January, between the first and the fourth of April, between the first and the fourth of July, and between the first and the fourth of October. The EDP therefore constitutes a sample of 4% of the French population. [6] The individuals present in the EDP are stratified (by region and municipality size) by lottery in such a way as to create a representative sample of voters registered on the metropolitan electoral rolls. Since this sampling procedure does not require consent, or even contact with the respondents, the samples of the voter turnout survey are not subject to the selection or self-selection biases that typically affect surveys of electoral behavior. [7] Note that this year, the sampling process has been slightly modified to better represent small municipalities comprising fewer registered voters. The goal was to more precisely measure misregistration, which had been underestimated by earlier survey methods.

8The second methodological advantage afforded by the voter turnout surveys arises from the large size of the samples they provide: 39,434 individuals for the study of the presidential and legislative elections of 2007; 39,728 in 2012, and close to 45,000 in 2017 (representing roughly one thousandth of the electoral body). These dimensions allow for very fine stratifications while conserving statistical power, which is invaluable when studying a phenomenon that can often be quite subtle during certain rounds of the election: an abstention rate of only 16.2%, for example, during the first round of the presidential election in 2007.

9The third advantage is that INSEE’s voter turnout survey stands out in terms of the wealth of information available for each individual in the sample. For every respondent, the database includes information on their civil status (birth, marriage, divorce, etc.), financial data, the most recent population census data, and information from the national electoral roll.

10The fourth advantage is that the voter registration status of individuals, and more importantly their participation history, is derived from administrative records, as opposed to the declarative information collected by traditional surveys. Turnout is taken directly from the attendance records in each prefecture during the ten days following the election. As such, information that is crucial to the study of abstention is not subject to self-selection bias and/or to “sayability”, which influences participation studies produced from the statements of politically invested subjects. [8] Consequently, INSEE’s voter turnout survey provides a level of realism and an exceptional objectivity that is particularly valuable for exploring a phenomenon such as abstention, which today remains stigmatized, and therefore potentially under-declared. [9]

11Finally, the voter turnout surveys are especially well adapted to longitudinal approaches such as the ones we are applying to the decade 2007-17. Because the sampling principles from one election cycle to another are identical overall, and because the inputs are free from selection bias, self-selection, and declaration, the information is not exposed to the usual hazards of cross-sectional studies. In particular, developments that are measurable from one election cycle to another are less at risk of being artifacts of circumstantial effects or changes in survey methods.

12However, like all databases, those of the voter turnout survey suffer from a certain number of flaws. Chief among these is the temporal discrepancy between the production of certain items of information provided by the EDP and the carrying out of the voter turnout survey. This discrepancy concerns variables arising from the population census: profession and socioprofessional categories, marital status, level of educational attainment, the type of cohabitation within a household, residential mobility, and so on. For the 2007 survey, the information provided by the EDP came from the census of 1999. For the 2012 survey, it was taken from the 2010 Annual Census (Enquête annuelle de recensement, EAR), and that of 2017 came from the 2015 EAR. This discrepancy constitutes the principal risk of error linked to the use of the database. It is likely that some individuals will have changed profession, marital status, or type of cohabitation between the time they were surveyed for the census and the time of the elections for which their participation was studied. From this perspective, it is the 2007 voter turnout survey that presents the highest risk of containing artifacts of this type, since the data comes from a census carried out eight years previously. Nevertheless, the specific parts of the data we use are generally stable across a lifespan, such as, for example, the level of educational attainment, which is less likely to alter after the age of 25. Age, recalculated for each election, is not affected by this bias. In any case, we must emphasize that this temporal discrepancy can only weaken the impact of social determinants on abstention. If, for example, the group of supposedly non-tertiary-educated individuals in fact includes a number of individuals who have attained a tertiary level of education, or if the group of supposedly unmarried individuals in fact includes some married individuals (and vice versa), this can only reduce the significance of these different factors for the statistical results that we are going to present. As a consequence, the relationship between social factors and voter turnout in what follows are probably stronger in reality than the statistical results produced via the INSEE database allow us to see.

Strong Social Disparities in Voter Turnout

13From the first round of the 2017 presidential election, despite the abstention rate remaining relatively limited (22.2%), significant differences in turnout are still observable depending on age, level of educational attainment, or “profession and socioprofessional category” and standard of living (Figure 1). As is common in the sociology of abstention, disparities in turnout are most prominent between different age groups. The elderly abstain the most by far: 42.7% among those aged 85 or over. Abstention is also much higher than average among the youngest groups, especially those aged between 25 and 29: 31.6%. Conversely, abstention is reduced in the groups aged between 40 and 69 (less than 14%) and is lowest among those aged between 50 and 54 years: 12.3%. As a result, there is a turnout difference of 19.3 points between the 50-54 age bracket and the 25-29 age bracket. This disparity exceeds 30 points between 50-54-year-olds and those aged 85 and over.

Figure 1

14Again, as is typical, we observe a very strong disparity in turnout depending on the level of educational attainment (Figure 1). The higher the level of educational attainment, the lower the risk of abstention: the rate of abstention goes from 11.4% among holders of a tertiary degree (baccalaureate plus five years of study) to 37.6% among those who have no qualification and who ceased their studies before the completion of senior secondary school. The fact that this relationship between higher levels of educational attainment and lower chances of abstention follows a linear trend (Figure 1), even though those with professional qualifications are younger on average, already illustrates the significance of this variable for the likelihood of an individual’s electoral participation (cf. Table 1 below).

15Different socio-professional categories are also associated with notable disparities in turnout. Indeed, the abstention rate is more than twice as high among manual and industrial laborers as among executives and the higher intellectual professions: 24.1% compared to 11.6%. As Camille Peugny has shown, the “professional position” variable enables an increase in precision on this point by allowing for finer distinctions internal to certain types of professions and socio-professional categories (PCS). [10] Among laborers, manual laborers and specialized laborers are slightly more likely to abstain than qualified industrial laborers and workshop technicians: 21.6% compared to 18.4%. In a similar fashion, category A civil service staff are even less likely to abstain than executives: 9.1% versus 10.6%.

16The “standard of living” variable, which was not included in previous INSEE voter turnout surveys, yields results that follow this same trend: the higher a voter’s standard of living, the lower the probability of abstention. The disparity in turnout is as much as 15.4 points between the first and last quartile. Similarly, property owners have a turnout rate that is 10 points higher than tenants (excluding social housing). In sum, the results of the 2017 survey are entirely congruent with sociological models of electoral behavior: the turnout probability is perfectly aligned with the amount of economic, social, and cultural resources voters have at their disposal. The first round of the last presidential election demonstrates once again that members of the popular classes (low level of educational attainment, manual and/or less-qualified professions, limited financial resources) are far more likely to abstain than members of the upper classes.

Family Affects Voting

17The attributes of voters’ immediate families tend to reinforce or, conversely, to mitigate these differences. Contrary to what might be inferred from its normative definition, voting is not an individual act. It has long been known that voter participation is affected by marital status. [11] The data from the 2017 voter turnout survey allow us to verify anew that voters vote more frequently if they are part of a couple (Figure 2). During the first round of the presidential election, abstention was close to 7 points higher among divorcees than among married couples. Similarly, widowers are the most abstinent category (32.4%). This is partly explained by their higher average age, but likely also owes to the loss of the family encouragement that benefits individuals in a couple. [12] Conversely, the least abstinent are individuals in a civil union, for whom participation rates are slightly higher than those of married couples (Figure 2).

Figure 2

18The 2017 voter turnout survey allows for a closer measurement of contextual effects than previous versions, thanks to two variables that facilitate the identification of the level of educational attainment and the PCS of partners. These two variables allow us to verify that the effect of being in a relationship is further accentuated when a partner has attributes that favor electoral participation. Without yet being able to unravel the part played by homogamy in these associations, it is noteworthy that voters in a couple that includes an executive only have an 8.8% likelihood of abstaining in the first round of a presidential election. When the partner is a laborer or an employee, abstention increases by 6 points to 17%, which however is still 5 points below the average. The level of educational attainment of a partner also influences electoral behavior within couples. This could only be confirmed by differentiating on the basis of the level of homogamy, but it appears upon first assessment that individuals in a relationship with a holder of a higher tertiary qualification have around three times less chance of abstaining in the first round of the presidential election than those in a relationship with an unqualified individual.

19Taking “type of cohabitation” into account enables close measurement of the effect of the position held within the family, and more broadly, across different types of households. Here too, our results confirm the insights of the existing literature. Unsurprisingly, individuals living “outside of the household” — in a hall of residence, a shelter, or a retirement home, for example — constitute the most abstinent section of the entire electoral body: 53.7% of them did not vote in the first round of the presidential election. Similarly, people who live “with non-relatives” or live “alone” prove to be considerably more abstinent than those living “in a couple”: they are around twice as likely to abstain from voting (Figure 2). Finally, in accordance with what has already been established in the United States, the survey con-firms that single-parenthood reduces familial influence in electoral matters. [13] Adults in a single-parent family show a rate of abstention almost twice as high as that of adults in a couple with a child or children: 23.2% compared to 13%. This situation is further reflected in the electoral behavior of their children: a child from a single-parent family is more often abstinent than a child living with two parents: 32.7% against 24.6%. These are undoubtedly the combined effects of family influence and parental example working in tandem.

A Low Level of Educational Attainment and Age Are the Most Predictive Factors of Abstention in the First Round of the Presidential Election

20We carried out a logistic regression analysis with the aim of better understanding the respective impact of factors of abstention in the first round of the presidential election. The evaluation of the disparities presented above is not sufficient to prove that a variable has a clear effect on voter turnout. For example, if widowers and widows are equally likely to abstain, this is perhaps because they are overrepresented in the older age groups. Similarly, it is not necessarily the PCS that counts most, but the educational attainment or income levels that are unevenly associated with each of the socio-professional groups. Our regression model thus needed to be able to establish the marginal effects on the abstention of each of the factors studied while controlling a series of variables. It also needed to make it possible to combine these factors via the regression equation so as to better draw standard profiles of non-voters or, conversely, of voters.

21Our model is therefore based on a logistic regression analysis in which the variable to be explained is a binary variable: participated (coded 0) vs. abstained (coded 1) in the first round of the 2017 presidential election. It is based on eight explanatory variables. Six of these are the “heavy” variables examined above: gender, age organized in five-year brackets (with the exception of the first bracket, which comprises 18-24-year-olds, and the last, 85 years and over), PCS, level of educational attainment, marital status, and type of cohabitation. Our model also includes a variable indicating immigration status and a variable permitting us to distinguish between those properly registered to vote and those who are misregistered in various ways (intra-departmental, intra-regional, extra-regional). Misregistration remains very high in 2017: 17.3% of the individuals in our database are registered to vote in a different municipality to the one in which they were surveyed. This misregistration strongly affects participation. During the presidential and legislative elections of 2012, it constituted the most predictive factor for consistent abstention. [14] Its inclusion in our model has two objectives: to allow us to measure its impact on the probability of abstention, but also to better establish the impact of the other explanatory variables given the same voter registration status.

22We have chosen the following modalities as reference (intercept): male, executive, between 60 and 64 years old, master’s level qualification (baccalaureate plus five years of study), married, in a couple with children, properly registered on the electoral rolls, and not of immigrant background. The regression analysis establishes that an individual with such a profile would only have a 6.1% probability of abstention in the first round of the most recent presidential election. It also enables us to observe that the different variables have, “all things being equal”, a clear and highly uneven impact on voter turnout (Table 1).

Table 1

Marginal Effects of Variables on the Probabilities of Abstention

Marginal effect in relation to the constant (%) Presidential Ist round Presidential 2nd round p Legislatives Ist round ’ Legislatives 2nd round p GENDER Female vs male -0,58 O.OOOl 0,86 <0001 0,01 0,9735 0,33 0.4302 SOCIO-PROFESSIONAL CATEGORY Farmers vs executives and the higher intellectual professions -0,08 0.9199 1.25 0,1952 0,32 0,8432 -1,69 0,3664 Artisans, commercial traders and small business owners vs the higher intellectual professions 1,38 0,013 2,87 <0001 6,7 <0001 6,39 <0001 Intermediary professions vs executives and the higher intellectual professions 0,41 0,2945 0,89 0.0462 3.97 <0001 3,92 <0001 Employées vs executives and the higher intellectual professions 0,7 0.0842 1,61 0,0007 6,23 <0001 5,8 <0001 Manual and industrial laborers vs executives and the higher intellectual professions 1.67 0.0002 2.07 <0001 8,47 <0001 9,25 <0001 Retirées vs executives and the higher intellectual professions 0,35 0.4882 0,84 0.1561 3,81 0,0003 4,24 0,0004 Other persons without a professional activity vs executives and the higher intellectual professions 2,02 <0001 2,79 <0001 6,73 <0001 5,71 <0001 AGE GROUPS 85 years or above vs 60-64 years 15,44 <.0001 16,88 <0001 13,56 <0001 12,3 <0001 80-84 years vs 60-64 years 6,09 <0001 7,4 <0001 0,67 0,5357 0,31 0.8052 75-79 years vs 60-64 years 2,42 <0001 2,05 0,0016 -2,04 0.0422 -3,9 0.001 70-74 years vs 60-64 years 1.14 0.0224 0,56 0,3186 -3,28 0.0003 -6,24 <0001 6S-69 years vs 60-64 years 0,72 0,1043 0,47 0,3488 -1.31 0.1137 -2.3 0,0185 55-59 years vs 60-64 years -0,34 0.3797 0,29 0.S42S 1.29 0.1185 3.15 0,0016 50-54 years vs 60-64 years -0,4 0.3166 0,37 0.4467 3,35 0.0001 6,19 <0001 45-49 years vs 60-64 years -0,17 0.6774 0,4 0,423 7,3 <0001 9,45 <0001 40-44 years vs 60-64 years 0,1 0.8226 0,73 0,1627 10,97 <0001 13,21 <0001 35-39 years vs 60-64 years 1,24 0.0093 2,9 <0001 16,29 <0001 19,93 <0001 30-34 years vs 60-64 years 2.73 <0001 4,14 <0001 20,99 <0001 25,97 <0001 25-29 years vs 60-64 years 3,68 <0001 5,46 <0001 27,4 <0001 31,36 <0001 19-24 years vs 60-64 years 1,05 0.0285 2.52 <0001 24,49 <0001 30,74 <0001 18 years -0,51 0.S85 1,57 0,208 20,58 <0001 25,83 <0001

LEVEL OF EOUCATIONAL ATTAINMENT No schooling vs tertiary studies graduate (master’s or PhD) 14,71 <.0001 12.03 <.0001 23,53 <0001 19,15 <0001 No qualification but schooling to primary or junior secondary level vs tertiary studies graduate (master’s or PhD) 16,85 <.0001 11,88 <0001 26,4 <0001 20,79 <0001 No gualification but schooling beyond senior secondary level vs tertiary studies graduate (master’s or PhD) 13,55 <0001 10,39 <0001 20,11 <0001 17,11 <0001 Primary éducation certificate vs tertiary studies graduate (master’s or PhD) 8,24 <.0001 7 <.0001 16,87 <0001 14,36 <0001 BEPC (junior secondary leavinq certificate), elementary certificate, secondary school leavinq certificate vs tertiary studies qraduate (master’s or PhD) 3,88 <.0001 2,83 <0001 9,95 <0001 8,63 <0001 Vocational or apprenticeship certificate vs tertiary studies graduate (master’s or PhD) 5.28 <0001 4.33 <0001 12,49 <0001 11,21 <0001 Certificate of professional studies vs tertiary studies graduate (master’s or PhD) 2.01 <0001 2,08 <0001 6,77 <0001 6,38 <0001 General baccalaureate, senior secondary school certificate vs tertiary studies graduate (master’s or PhD) 3,83 <0001 3,66 <0001 10,51 <0001 9.72 <0001 Technoloqical or vocational baccalaureate; vocational or technoioqical certificate; commercial, industrial, or service certificate (BEC. BEI, BEH), or law school admission certificate vs tertiary studies qraduate (master’s or PhD) 1,49 <0001 1.57 <0001 3.92 <0001 3.14 <0001 MARITAL STATUS Civil union vs married -0.9 0,0306 -0.7 0,136 -0.8 0.3194 -0,67 0.5046 De facto relationship vs married 2.4 <0001 2.84 <0001 3,45 <0001 3.7 <0001 Widower, widowed vs married -0,22 0,6447 •0,07 0,9049 0,18 0.8799 0,94 0.5183 Divorced vs married 0,71 0,1655 0,47 0,4279 2,76 0.0192 3,14 0,0275 Single vs married 1.93 0,0001 0,6S 0,2252 0,43 0,6727 -1.1 0,385 TYPE OF COHABITATION Children of a couple vs adults in a couple with a child/children 2,73 <0001 4,61 K 1.4 0,2855 2.3 0.1573 Children of a single-parent family vs adults in a couple with a child/children 7.13 <0001 9,62 <0001 6,02 <0001 5,46 0.0025 Adults in a couple without children vs adults in a couple with a child/children -0,58 0,0111 0,13 0,653 -2,09 <0001 •0,73 0,2146 Adults of a single-parent family vs adults in a couple with a child/children 4,24 <0001 4.37 <0001 4,66 0.0002 6,88 <0001 Persons living In a household of several (non-related) others vs adults in a couple with a child/children 3,34 <0001 4,09 <0001 2,69 0.0333 4,01 0.0098 Persons living alone vs adults in a couple with a child/children 1.82 0.0004 3,2 <0001 1.34 0.2115 3,08 0.0211 Persons living in a résidence, retirement home vs adults in a couple with a child/children 4,39 <0001 6,74 <0001 5,19 0.0026 10,25 <0001

VOTER REGISTRATION STATUS Extra-régional misregistration vs properly registered 12,71 <0001 12,28 <0001 9,55 <.0001 9,95 <0001 Intra-departmental misregistration vs Properly Registered 7,47 <.0001 7,69 <0001 8,88 <0001 9,61 <0001 Intra-regional misregistration vs properly registered 6,42 <0001 1 7.15 <0001 5,91 <0001 9,12 <0001 IMMIGRATION Immigrant vs non-immigrant 1 0,0235 6,77 <0001 Abstention rate of the constant 6,1 7.56 19,19 25,23

Marginal Effects of Variables on the Probabilities of Abstention

23An examination of the marginal effects associated with the modalities of each of these variables (Table 1) reveals that level of educational attainment and age, followed by voter registration status, are the factors likely to have the greatest impact on voter turnout in the first round of the presidential election.

24The regression analysis thus fully confirms the decisive impact of an individual’s level of educational attainment: the probability of abstention increases steadily and significantly with the decline in the level of educational attainment. In the first round of the presidential election, under controlled variables, the shift from master’s level (baccalaureate plus five years of study) to general baccalaureate level increases the chances of abstention by around 4 points. This rises to 8 points when educational attainment drops to the level of a primary education certificate. And it reaches the remarkable level of 17 points when the voter in question does not have any qualification and has ceased their studies before senior secondary school. In the context of the first round of the presidential election, these 17 points represent the most pronounced marginal effect associated with a change in the values of a variable. The fact that it is the level of educational attainment that produces such disparities seems to us to present a preliminary confirmation of the continued decisive impact of social determinants on participation. It also reinforces the pattern according to which the level of educational attainment remains one of the indicators of the volume of cultural capital that determines the chances of politicization and thus also the probability of an individual’s electoral participation. The over-determining character of level of educational attainment is also visible in the limited impact of other sociological factors, particularly PCS. When we control for the other factors, the shift from executive to laborer only increases the risk of abstention by 1.7 points. Differences in turnout between social classes are first and foremost consequences of educational disparities and are therefore likely to be disparities in politicization. [15]

25A more complex interpretation is required for the factor of age. It only exerts a strong impact on individuals in the highest age brackets, particularly those in the 80-84 and the 85 years plus categories. In comparison with the 60-64 age bracket, being a member of this latter age group increases the probability of abstention by more than 15 points. Among the youngest, the effect of age remains very limited, including in segments that are strongly abstinent: only 1 point of marginal effect among 24-29-year-olds, and even a slightly negative effect among 18-24-year-olds who are therefore, with certain variables controlled, more likely to participate than 60-64-year-olds.

26This is a significant result for at least two reasons. First, it demonstrates that in a context of high intensity such as the first round of the presidential election, when other variables are accounted for, being young is not a factor of abstention. Second, it compels us to look elsewhere for the principal reasons behind the strongest levels of abstention among young people, and in particular to the category of registration status. Indeed, our model confirms that misregistration is one of the principal factors behind abstention: when we control for the other variables, being registered in a region other than that in which one has taken the survey increases the chances of abstaining by close to 12 points (Table 1). It is the under-30s who are by far the most affected by misregistration: 51% of 25-29-year-olds were misregistered in 2017. This is the principal reason why they were clearly more abstinent than average in the presidential election (Figure 1). Contextual factors also add to this explanation. [16] This is especially true for type of cohabitation, the effects of which are stronger than those of marital status, which only possesses limited influence once other variables are controlled for. The analysis thus enables us to confirm, for example, that single-parenthood is heavily linked to higher abstention rates: in comparison to adults in a couple with a child/children, the probability of abstention for adults of a single-parent family is 4 points higher—a difference, incidentally, identical to that between children living with their parents and children from single-parent households.

27In summary, the regression analysis confirms that demographic, social, and contextual factors continue to have a decisive influence on the probability of voter turnout. Changing the value of a variable (level of educational attainment or age, for example) can be enough to increase the chances of abstention by 17 points. Furthermore, certain social profiles present almost no chance of abstention. For example, a female senior executive aged between 50 and 54, with a master’s or equivalent, in a civil union, living in a couple without children, properly registered, and not of immigrant background, has only a 4% chance of having abstained in the first round of the last presidential election. In contrast, a male laborer without any educational qualification who has ceased schooling before senior secondary school, aged between 25 and 29, single, living alone with his children, properly registered, and not of immigrant background has a 61% chance of having abstained in the same round of elections.

The Impact of Heavy Variables Increases with the Rise in Abstention

28Abstention reached unprecedented levels in the first round of the 2017 legislative elections, and even higher levels during the second round. The fact that abstention appears to have become a majority response, even if we limit the analysis to those listed on the electoral rolls, gives rise to a certain number of questions. What effects does this popularization of abstention have upon the traditional social logics of abstention? Does it imply a weakening of social determinants in the face of a more socially and culturally intersectional abstention? Are we witnessing a change in the nature of abstention that would see it become more inter-classist and less determined by traditional factors favoring politicization, starting with the level of educational attainment?

29An examination of Figure 1 already suggests a preliminary answer to these questions. It is within those categories that were already the most abstention-prone in the first round of the presidential election that the increase in abstention is most notable. The gradients of the participation curves measured by age group, standard of living quartiles, or level of educational attainment are steeper for the legislative elections than they are for presidential elections (Figure 1, right-hand panel). This shows that the increase in abstention throughout spring 2017, and especially between the presidential and legislative elections, occurred within the framework of an increase in demographic and social disparities in voter turnout. The differences in turnout between age groups and social categories, far from having been reduced, in fact increased with the broad swing toward abstention in the electoral body. To take just a few examples, the participation gap between the 25-29-year age bracket and the 70-74-year age bracket was 19 points in the first round of the presidential election; it rose to 37 points in the second round of the legislative election. Between laborers and executives, this gap rose from 13 to 23 points. The same increase is measurable in relation to the standard of living quartiles. The election cycle begins with a difference of 15 points between the first and the fourth quartile, which rises to 25 points by the close of the cycle.

30These developments appear to indicate that demographic and social determinants of abstention have an even greater impact in the case of an abrupt fall in electoral intensity, such as the one that characterized the passage from the presidential to the legislative elections. Here too, regression analyses should make it possible to better identify the social logics at play behind the heightening of disparities in voter turnout. In particular, they should allow us to verify that the impact of the different variables, grasped via their marginal effect on abstention, increases between the presidential and the legislative elections. This would constitute an empirical confirmation of the even more decisive nature of an individual’s social characteristics in a context of low electoral intensity. Over and above the descriptive statistics given in Figures 1 and 2, these regression analyses should also make it possible to identify the influence of each variable on abstention, and thereby to show more clearly how each of these factors specifically contributes, “all things being equal”, toward the developments we have just described.

31To achieve this, we have supplemented our first model with three logistic regression analyses aiming to explain abstention in the second round of the presidential election, as well as the first and second rounds of the legislative elections. We have kept the same explanatory variables, and for each of these variables, the same reference values to allow for comparison. In order to better understand the marginal effects of each value of the different variables included in the model on abstention, at the bottom of the table we have added the predicted rate of abstention for each reference value. Thus, it would appear that a male executive between 60 and 64 years of age, married, in a couple with children, properly registered on the electoral rolls, and not of immigrant background has a 7.6% chance of having abstained from the second round of the presidential election, 19.2% for the first round, and 25.2% for the second round of the legislatives.

32This modelling confirms that abstention increases in the legislative elections for all sections of the electorate, including those profiles that gather together attributes favoring participation, such as that of our “constant”. But as Table 1 also shows, the marginal effects associated with the values of our different variables are still much higher for the legislatives than for the presidential election. This provides our first important observation. If, because abstention was so widespread during the legislative elections, it was less demographically and socially determined and its composition had become more “mixed”, we should see the opposite result to the one we observe here. However, the coefficients associated with the different values of our variables, and therefore their marginal effects, are clearly stronger across the board for the legislatives than they are for the presidential election. For example, the marginal effects associated with the different age groups did not exceed 17 points for the presidential election; they rise to 31 points in the second round of the legislative elections. Similarly, for level of educational attainment, they climb to 26 points for the legislatives against a maximum of 17 points for the presidential election. The same pattern occurs with the PCS, which had only a weak impact on participation during the presidential election and whose influence on the probability of abstention is far greater in the legislative elections.

33A detailed examination of each of the variables shows that it is the effects related to age group that undergo the most marked development between the presidential and the legislative elections. While in the case of the presidential election, as we have seen, being young alone is not a factor of abstention, we observe the opposite for the legislative elections. In comparison with 60-64-year-olds, and with all other variables accounted for, the likelihood of abstention is far higher among the youngest sections of the electorate: up to 31.4 points higher for 25-29-year-olds (Table 1). Similarly, from this point onwards, seniors, and in particular the 65-79 bracket, are the most likely to participate; and as for the participation differential, it is henceforth close to 40 points for the youngest voters. With the fall in electoral intensity, age becomes the most predictive variable of the likelihood of voting or abstaining.

34The relationship between the level of educational attainment and abstention operates in the same way as it did in the presidential election. Overall, in the legislatives, an increase in the level of educational attainment leads to a reduction in the chances of abstention. The impact specific to educational capital is even stronger in the case of the legislative elections. Thus, in the second round of the ballot, descending from the level of master’s (baccalaureate plus five years of study) to the level of a vocational qualification (CAP), the probability of abstention increases by 11.2 points. This increase reaches its maximum (20.8 points) among voters with no qualifications who have ceased schooling before senior secondary school.

35The case of the PCS is particularly interesting. The effect they have on voter turnout remains very limited, even in the heightened context of the presidential election (cf. Table 1 above). It is much stronger in the case of the legislative election. While the differential in terms of marginal effect between executive and laborer was only 1.7 points for the presidential election, it increases as the election cycle progresses, culminating at 9.2 points for the second round of the legislative elections (Table 1).

36In summary, the regression analyses fully confirm that a context of very low turnout, such as that of the 2017 legislative elections, further accentuates the influence of traditional demographic and social factors. In the case of an election of this kind, being aged under 40 or over 85, not highly qualified, belonging to the working-class categories and, to a lesser extent, living alone or in a single-parent family, renders participation highly improbable, even if one is properly registered on the electoral roll. Using our model, it can be estimated that a female manual laborer aged between 25 and 29, unqualified and having ceased schooling before senior secondary school, divorced, living alone with her children and properly registered has an 89.4% chance of abstaining from voting in the second round of the legislative elections. In contrast, a female executive aged between 70 and 74, with a master’s (baccalaureate plus five years of study), married, living in a couple without children and properly registered only has an 18.7% chance of abstaining in the same election.

The Rise in Abstention and the Increase of Social Disparities in Voter Turnout between 2007 and 2017

37To conclude, it is interesting to observe the developments in abstention beyond a single election cycle such as that of spring 2017, and to look at them throughout the ten-year period, beginning with the 2007 presidential election. Over the course of this last decade, abstention has increased very significantly: by 9.4 points in the second round of the presidential election, and by 17.3 points in the second round of the legislative elections (Table 2).

38Has this increase been accompanied by an “intersectionalization” of abstention, which is to say, by a trend toward the reduction of disparities between age groups and social categories? Or has it, in accordance with the logic established in the 2017 election cycle, been brought about as a result of a rise in disparities in voter turnout?

39In order to answer these questions, we have measured the development of the differences in turnout between the age group presenting the highest levels of abstention in the legislative elections, the 24-29 age bracket, and the age group with the highest levels of participation, the 65-69 age bracket. We have also measured the development of the differences in turnout between laborers and executives, and between secondary school graduates and graduates of a tertiary institution. [17] The results are presented in Table 3. They confirm that the logics identifiable in 2017 are not just short-term, because they can be observed in operation across the whole of the examined period.

40Reading across the rows of the table shows that the swing toward greater overall abstention, which henceforth characterizes the passage from the presidential election to the legislative elections, continues to exhibit the effect of a very significant growth in inequality in voter turnout. This mechanism can be seen in 2007, likewise in 2012, and, as we have examined in detail, in 2017. In 2007, the difference in abstention between 25-29-year-olds and 65-69-year-olds is 3.7 times higher in the first round of the legislative elections than it is in the first round of the presidential elections. The growth in disparity between laborers and executives and between secondary school graduates and graduates of a tertiary institution is comparable. Here we observe dynamics very similar to those identified in the spring 2017 election cycle: the lower the intensity of the election, the more influential the social attributes of voters become in determining the likelihood of their electoral participation.

41Reading the table down its columns, one observes that the development of abstention over time produces identical effects. This is very clear in the case of the presidential election: between 2007 and 2017, the differences in abstention between 25-29-year-olds and 65-69-year-olds, between manual laborers and executives, and likewise between secondary school graduates and tertiary level graduates have, on the whole, doubled (Table 3). This development is slightly less for the legislative elections, which began in 2007 with a very high abstention rate and thus with already significant demographic and social differences in turnout. Although between 2007 and 2017 the difference between the 25-29 age bracket and the 65-69 age bracket remained relatively stable, the difference between laborers and executives has, on the other hand, clearly increased (up 36% in the first round), as has that between secondary school graduates and tertiary level graduates (up 41% in the first round).

42The increase in abstention over the last decade functions like an amplifier of disparities in voter turnout. The electoral body was less representative of the electorate as a whole in 2017 than in 2012, and less representative again in 2012 than in 2007.

43The drop in voter turnout is therefore not indicative of a process of equalization of citizens before the ballot box. It is in fact the opposite: the demographic and social determinants of turnout were reduced to their minimum in the context of the 2007 presidential election’s high turnout rates. They exerted their maximum influence in the context of the very low turnout rates of the 2017 legislative elections. Thus, it is first and foremost the disengagement of those categories presenting properties least conducive to electoral participation that explains the continuous increase in abstention in the presidential election and the legislative elections over the last ten years. This rise in abstention was borne out by an increase in the demographic and social disparities of voter turnout.