1What is a couple? This question formed the focus of discussions about how to design the EPIC survey on individual and conjugal trajectories (Étude des parcours individuels et conjugaux, INED–INSEE, 2013–2014), the most recent national survey on couple formation in France. In demographic and sociological surveys, the conjugal status of individuals constitutes an essential unit of analysis, an aspect of life that is included in every questionnaire. On the surface, the notion seems self-evident: everyone knows what a couple is. However, a closer look at these surveys shows they are based on not only variable but variably explicit definitions that often reveal how a particular historical context views certain forms of conjugality as legitimate. For Alain Girard, who designed the first French survey on couple formation (Le choix du conjoint, INED, 1959), it was necessary to study the unions of two never-married persons to understand their ‘choice of spouse’ (the title of the survey). This allowed him relatively broad coverage in a time when few individuals in a couple failed to marry, and remarriage was rare. Following the changes that arose in the 1970s with the growth of ‘juvenile cohabitation’ (Roussel, 1978) and then of de facto unions, both of which reflected a growing rejection of marriage, Michel Bozon and François Héran broadened the spectrum of analysis when they designed the second survey on the subject (La formation des couples, INED, 1983–1984). The couple could no longer be reduced to marriage, and the survey sample thus integrated persons who were not married (never married, widowed, or divorced).
2Surveys now approach and define the concept of the couple differently. An overview of French surveys on the family before the EPIC survey speaks directly to this point. In the Family History Survey (Étude de l’histoire familiale, or EHF; INSEE, 1999), a retrospective history of past unions covered only cohabiting unions that lasted at least six months; whereas the French version of the Generations and Gender Survey (Étude des relations familiales et intergénérationnelles, or ERFI; INED–INSEE, 2005, 2008, 2011) covered those that lasted at least three months. The latter also investigated non-cohabiting conjugality by asking respondents living alone at the time of the survey about any ‘stable non-cohabiting intimate relationship’ (Beaujouan et al., 2009). The Family and Housing survey (Famille et logements, INSEE, 2011) distinguished persons who were ‘in a couple with someone living in the same residence’ from those who were ‘in a couple with someone living in another residence’. It also integrated same-sex unions and used more inclusive terminology: the module on persons who reported being ‘in a couple’ referred not only to the possible ‘conjoint∙e’ (‘spouse’ or ‘partner’)—a term suggesting an official status that could lead to excluding less formalized relationships—but also to a possible ‘boyfriend/ girlfriend’ to better capture the diversity of relationship situations.
3As the couple concept has come to be dissociated from that of marriage, with the rise of de facto unions and, more recently, the PACS (civil union),  questions now emerge about whether it can be associated with cohabitation and heterosexuality. This shift reflects the fact that between living in a couple under the same roof and having a ‘strictly’ single status runs a continuum of lifestyles that surveys struggle to capture and characterize. Some, notably surveys on sexual behaviour, nonetheless include configurations that are overlooked by surveys focused on the household. Many other configurations exist besides the ‘classic’ couple, which is defined by criteria such as cohabitation, legal formalization, and duration. Although they are less visible, these configurations may prove important from an individual’s perspective, particularly when asked to describe their personal trajectory. These ‘fuzzy states’ (GRAB, 2006) will develop further when we consider that ongoing demographic changes continue to weaken both the norm and the sustainability of having one partner over a lifetime. It is not only age at first cohabitation that has continually increased over recent decades,  but above all the rates of divorce and separation, so individuals are ever more likely to experience multiple unions and relationships over their lifetimes (Cassan et al., 2001; Beaujouan, 2017). While less than 5% of couples formed in the 1950s had dissolved after 10 years, this figure rose to 1 in 5 couples formed in the 1980s (Vanderschelden, 2006). Periods of life spent without a partner have automatically grown, as has repartnering. This calls for new research on the process of couple formation.
4These transformations stimulated the development of the EPIC survey. The aim was not so much to reopen questions on the mechanisms highlighted in the previous Girard and Bozon-Héran surveys as it was to explore the effects of a new context on couple formation. The elements mentioned above led the design team to choose a broader approach to the couple (Section I) and to extend the spectrum of analysis by investigating not only the formation but the ‘de-formation’ of the couple, as well as periods of life spent outside of any serious intimate relationship. The survey did not, however, break with the spirit of the two surveys carried out on the subject. The project followed explicitly in their footsteps and updated the data where necessary (Section II). This double objective can also be seen in the methods used to carry out the survey, which in some respects borrow from the previous surveys on the subject; but they also reflect the new possibilities opened up by the major transformations in survey logistics since the 1980s (Section III).
I – The diversity of couples
1 – The ‘couple relationship or serious intimate relationship’
5The EPIC survey had two objectives for renewing the study of intimate relationships. First, it was important to move beyond having an overly restrictive focus on the cohabiting couple and relying on any pre-established definition that could exclude some configurations. The second aim was to capture the entire sequence of conjugal and affective stages within individual trajectories. To do this, the survey’s coverage was extended to all persons in an age range (26–65 years), regardless of their situation at the time of the survey. Instead of centring on a particular union (whether the first or the current one), the survey sought to capture the diversity and complexity of individual trajectories. A central module was thus constructed to collect information on the respondents’ complete intimate and conjugal relationship histories. While some surveys have covered this type of information, none have done so in such a detailed fashion, and none have collected information on the social characteristics of the respondents’ previous partners.
6Adhering to a restrictive definition often limits existing sources to describing periods when the respondent was in a couple relationship. The difficulty of defining what a couple is today suggests that it may be best instead to let individuals describe their conjugal and intimate relationship histories according to a broad and deliberately subjective description, thus setting aside all preconceived notions: no fixed criterion of duration and no restrictions based on living together, but simply the stories that make sense in the respondents’ own eyes. EPIC therefore retraced the full set of the respondents’ ‘couple relationships or serious intimate relationships’, which was specified for interviewers: ‘this means a relationship that, with or without marriage, PACS, or cohabitation, counts or counted in the past, even if it is no longer the case today.’
7Each relationship was described while considering several temporal markers (where applicable): the date when the relationship began; the chronological position of first sexual relations (before or after the relationship was begun, coinciding with it, or no sexual relations with the person); the date of moving in together; conclusion of a PACS, civil marriage, or religious marriage; end of cohabitation; divorce; and dissolution of the PACS. Besides the dating of these events, several further details were gathered for each relationship: place of meeting, first impressions, the employment status of the two partners when they met, their respective places of residence, a subjective assessment of their similarities (economically, their upbringing and leisure activities), and general characteristics of the conjugal and reproductive history of each of the respondent’s partners (having lived with a partner, been married, or had children).
8The choice to establish such a detailed description of the respondent’s relationship history in a survey was not self-evident for at least two reasons. The first relates to the very notion of a ‘couple relationship or serious intimate relationship’, a description that is simultaneously subjective and highly inclusive, and which it was thought might lead to listing too many relationships for the questionnaire. Given the level of detail requested for each relationship, this could have led to fatigue in the respondents and resulted in a very lengthy data collection process. The second reason relates to the objective of collecting a variety of dates that specify the nearest month or, if this was impossible, the nearest season. One could reasonably expect that this may excessively tax the memories of respondents.
9An initial test performed in 2010 (see Box) was preceded by a 2009 pretest, which specifically aimed to validate this module. No major problems were encountered. The notion of the ‘couple relationship or serious intimate relationship’ made sense to respondents: they did not react with surprise, pose specific questions, or offer extended reflections to answer. The respondents easily decided which relationships to include—that is, which relationships made sense to them. Concerns that the number of relationships covered by this formulation would be too large proved to be unfounded: respondents reported a mean of two relationships (and a maximum of six). However, the interview time for the initial version of the retrospective component alone did prove to be too long (26 minutes), which led to our decision to pose certain questions pertaining only to the respondent’s current relationship at the time of the survey.
10The full-scale data collection process confirmed these observations: the mean number of reported relationships was 1.9 (4% reported not having had any relationships; 43% described one, 28% two, 17% three, and 8% four or more). Among the 14,699 relationships described, a third did not include a period of cohabitation; 4% involved no sexual relations. The choice to broaden the description of partnership histories beyond the narrow frame of cohabiting relationships thus proved to be well founded.
Box. Summary of preparatory steps for the EPIC survey
1. Pre-test of the ‘History of couple relationships or serious intimate relationships’ module, December 2009
This pre-test aimed to test the (non-predefined) notion of a ‘couple relationship or serious intimate relationship’ in the field and to assess the possibility of retracing a dated, detailed history (characteristics of partners, types of relationship, etc.). This test took place by telephone, with the participation of two male and two female interviewers, each charged with collecting seven questionnaires and taking care to interview people with diverse characteristics in terms of sex, age, and social background.
2. First test for the questionnaire, November–December 2010
This first test for the questionnaire was performed by telephone (CATI) on a sample of around 100 individuals aged 25 to 65 years (including approximately 20 partners) and using the quota method (by sex, age, and socio-occupational category). It had two principal goals: to test both the notion of ‘couple relationship or serious intimate relationship’ and the means for contacting respondents. After a day and a half of training, data collection took place over three weeks, with the survey team members present each day for the interviews.
3. Second test of the principal questionnaire, November–December 2011
A second test took place in 2011 with the dual aims of testing the full questionnaire and of comparing two administration methods (by telephone and face-to-face) to estimate the participation rate. A white pages phone book was used to randomly draw a sample of 500 addresses from 10 municipalities in the Île-de-France, Nord-Pas-de-Calais, and Pays de la Loire regions. Respondents were contacted, half by telephone and half face-to-face. A notification letter was first sent to each household. Interviewers were required to make at least 10 attempts at contact and one visit in person (with face-to-face interviews) at different times and on different days. Households that initially refused to participate (citing lack of time, etc.) were contacted one further time; those who refused more categorically were not. Survey team members listened to most telephone interviews, the face-to-face interviews were sometimes recorded (for later listening), and team members sometimes accompanied the interviewers. In total, 116 people were interviewed.
4. Test of the partners’ questionnaire, September 2012
This test in paper format aimed to evaluate a first version of the self-administered questionnaire for persons in a couple with a respondent in the main survey. Its principal goals were to verify the quality of the answers and the respondents’ understanding of the questions. This questionnaire was sent by mail to 71 persons selected through the ‘snowball effect’; 57 answered.
5. Pilot survey, spring 2013
At the end of this preparatory phase, the questionnaire and the survey protocol were adjusted. The pilot survey was implemented in two regional divisions of INSEE in spring 2013 with the aim of precisely estimating the average duration of interviews and the participation rate. It also served to check for any programming errors (filters, etc.) and to decide on whether it was necessary to use headphones for self-administration of the few questions about sexuality (number of partners and age at sexual debut). Finally, the pilot allowed us to validate the data collection tools (notification letter, brochure presenting the survey, and data collection instructions) and the protocol for transmitting the self-administered questionnaire to the respondent’s partner. The test took place over seven weeks. In total, 560 households were drawn from the 2011 annual census survey; 313 interviews were performed (complete survey).
2 – How accurate is the monthly dating of events?
11Having respondents describe all their diverse ‘serious relationships’ had a major consequence: the collection of many dates. This raised the question of memory in the dating of events. The initial tests had shown that it was possible to have respondents date the identifying moments collected in their relationship history to a particular month, although the precision of these dates varied depending on the type of event (and notably whether it was anchored to an external marker). The possibility of indicating a season instead allowed a portion of the imprecision to be dealt with.
12These initial observations were confirmed during data collection. The year of an event is generally known, the month less systematically. For example, the year that a relationship began is lacking for less than 0.3% of the relationships described (in this case, the respondent’s approximate age was collected); however, in 23% of cases the precise month was not collected. The proportion of ‘don’t know’ responses depended strongly on the degree of institutionalization of the relationship. The month was not reported in 17% of cases when the relationship involved marriage, in 21% of cases of unmarried cohabitation, and in 30% of cases where there was no cohabitation. In the case of a missing month, ascertaining the season made it possible to refine the date in two-thirds of cases (regardless of the degree to which the relationship was formalized). The precision of the date when a couple moved in together or the relationship ended was of the same order as for the beginning of the relationship. For other events, such as a marriage, memories were far more precise (the month was known in 98% of cases). Given the significance of this history (notably the study of durations), missing dates (year or month) were imputed based on (i) additional information collected from the questionnaire (approximate age to impute the year, season to impute the month); (ii) survival probability in a given state according to individual characteristics; and (iii) respecting the chronology of the relationship to other events dated by the respondents (ensuring, for example, that the date of divorce did not precede the date of marriage). However, the crude variables (before imputation) were not deleted from the database, as the missing information may sometimes provide meaning; moreover, imputations may be useful for certain research questions and not for others. 
13Because the data was collected in a structured but not restrictive manner, the timing and durations of couple formation and separation open up innovative research prospects, not only on the unfolding (succession, overlap, and concomitance) of individuals’ different intimate relationships  and their impact on the processes of couple formation and separation, but also on representations associated with the couple (fidelity, exclusivity, etc.).
3 – Not being in a couple: A little-studied situation
14The explicit broadening of the notion of the couple combined with the retrospective approach invited making other notable additions, such as an exploration of life sequences that remain outside an intimate or couple relationship but within individuals’ conjugal trajectories. Just as a diversity of situations underlies the general notion of the couple, the survey design team hypothesized that life outside of a couple should also be heterogeneous, as highlighted by surveys on sexuality (Beltzer and Bozon, 2009). The increase in separations has automatically had an effect on family structures, with single-parent households and reconstituted families becoming more common (Chardon and Vivas, 2009; Breton and Prioux, 2009) and more people living alone (Daguet and Niel, 2010). The research literature on periods of life without a partner has been growing (in particular through a focus on the situation of women, and using qualitative methodologies: Kaufmann, 1999; Flahault, 2009; Singly, 2011), but there have been few quantitative studies in France. The EPIC questionnaire aimed to fill this gap. Between individual life choices and unwanted situations, life outside the context of a couple presents multiple facets, each of which sheds some light on contemporary couple formation. What does it mean not to have a partner? Does it mean having no relationships at all? Is it a chosen situation? Is it accompanied by aspirations to meet someone? Does it present difficulties or advantages in social life? On this point, the EPIC survey innovated a module devoted to persons who reported not being in a couple at the time of the survey. It thus offers a portrait of both the diversity among the individual situations under this general categorization and how they are experienced.
4 – Understanding separation as a process
15The survey data provide further information on the ‘de-formation’ of the couple, which is in contrast to what is found when treating separation as a single event through a reductive, homogenizing approach. Such data is preferable because numerous qualitative surveys have highlighted the variably progressive and heterogeneous nature of separation (Collectif Onze, 2013). The dissolution of couples merits consideration from two perspectives. First, it can be viewed as a process, beginning with enquiries into the first moments when the respondent considered separation, the implications for other spheres (work and cultural), sharing the news with family and friends, consulting professionals, and finally the decision to continue living under the same roof (or not) after choosing to separate. Second, we can take a multidimensional look into various practical aspects of the separation (children’s residence and the division of property) and the relational dimension. In addition to the relationship history, the survey also included a more specific module aimed at describing the last separation of a co-residential union, if applicable. Different constraints on the conjugal couple, notably material constraints, can lead to a prolonged period of separation, as well as the norm of placing value on the endurance of a parental couple. This prolongation can sometimes reach the point where separated partners continue to cohabit, a phenomenon called ‘living together apart’ (Martin et al., 2011). Although observing such situations is complex, they can be studied using this module.
5 – Surveying both members of the couple
16Finally, the diverse individual trajectories led the survey team to implement another questionnaire to be answered by the partner of respondents in a couple or an intimate relationship at the time of the survey. The formation of a couple depends on the characteristics and conjugal histories of both partners.
17Because collecting details from individuals on their partners’ relationship trajectories seems difficult and subject to uncertainty, the respondents’ partners were directly provided a shorter, self-administered questionnaire centred on information that could not be satisfactorily collected through the main survey interview. The partners were asked about their previous history, aspirations, opinions, and depiction of the couple; more intimate and personal items were included as well, such as the number of sexual partners over their lifetime and age at sexual debut. The questions for this part of the survey were formulated as identically as possible to those on the main questionnaire. This complementary component can be used to place the current relationship within the conjugal history of each member in the couple. What is more, it allows us to test the hypothesis of homogamy not only in terms of social characteristics and age but also in regard to previous conjugal history.
II – In the footsteps of the 1959 and 1983–1984 surveys
18Despite its innovations, this third survey on couple formation in France did not simply start from scratch. In many respects, EPIC follows in the footsteps of INED’s two previous surveys on the subject, the 1959 Girard survey (Girard, 1964), and the 1983–1984 Bozon-Héran survey (Bozon and Héran, 2006). EPIC maintained several lines of questioning and approaches from these surveys to allow exploration of changes over time. Like the two previous surveys, EPIC investigates couple formation as a process that is characterized by several key steps in which the views of family and friends may play a significant role. Particular attention was once again given to the context of the couple’s meeting and its consequences and to markers in building the relationship, some of which—like marriage—have generally come to be considered optional. In addition, the survey again looked at criteria for partner choice, particularly the partners’ physical and social characteristics when they met. However, in light of recent transformations, we revised our approaches to three ‘classic’ areas of study.
1 – The emergence of new meeting places
19The 1983–1984 survey focused particularly on the couple’s meeting place, a key element in the formation of a union. However, the landscape of possibilities for meeting partners has changed. Educational situations and friendship circles have become important spaces for couple formation, with large variations depending on social background (Bozon and Rault, 2013). An additional possibility that has emerged in recent years is meeting a partner through the Internet, whether on online dating sites or social networks (Bergström, 2011). The survey not only allowed respondents to report whether they met their partner online, it also explored this phenomenon in more depth. The initial results of a survey on the context of sexuality in France (Contexte de la sexualité en France, INSERM–INED, 2006) indicated that going online was becoming commonplace. The EPIC survey gave us the opportunity to develop a set of ad hoc questions on this subject, such as whether respondents had ever used these sites and, if so, what prospects they associated with their use. This allows several hypotheses to be tested. Do these spaces open up new horizons and facilitate more heterogamous meetings? Or are they structured by mechanisms of co-optation (for social networks) and selection (for online dating) in ways that instead tend to reinforce homogamy? Have they become spaces where individuals are likely to meet a partner? One of the first analyses of the survey data found that the answer to the latter question is in fact no (Bergström, 2016): few couples are formed by means of the Internet.
2 – New forms of union, new indicators
20Particular focus was given to the diversity in formalizing unions, which is linked mainly to the creation and growth of the PACS. While few different-sex couples made use of the PACS at the time of its creation (in 2000, there were around 15,000 PACS versus 300,000 marriages), it grew considerably in the 2000s, as its provisions were revised in ways that made it more similar to marriage. In 2010, the year of the first EPIC tests, different-sex couples contracted around 195,000 PACS versus 250,000 marriages. Nonetheless, a paucity of data made it difficult to perform detailed analyses of this phenomenon. The Family and Housing survey (INSEE, 2011) did allow PACS to be identified and thus made up for this information being absent from the annual census survey until 2015. However, it did not provide fine-grained information that could be used to understand the growth of the PACS: neither the reasons for choosing it, its place either within a marital dynamic or as an alternative to marriage, its treatment as a public or private event, nor the associated celebration (or lack thereof). These different markers can be used to distinguish the meanings given to this form of union.
21Reflecting on the forms of union also led to experimenting with new indicators. Participants were asked not only about the year and the month of their marriage or PACS but also the day. Although this information in itself was of no use to us (and the responses have thus not been disseminated), it allowed us to examine the ‘don’t know’ responses. During in-depth interviews, attention is generally given to silences and hesitations, which can reflect a specific relationship to an event or situation. Qualitative research on the PACS (Rault, 2009) led to the hypothesis that socially differentiated uses of these forms of union could be revealed by not only whether the respondent knew the precise date of the marriage and/or PACS but also by gauging the spontaneity of his or her response to this question. In the pilot survey, the interviewers were instructed not to pose follow-up questions, which did not prevent the respondents from finding the information by consulting administrative documents (PACS certificate), asking their partner, or looking at the date engraved inside their wedding ring. For collecting data on this matter, a new protocol was devised. Interviewers had to indicate whether the date had been provided spontaneously: ‘Question for the interviewer: Did the respondent… 1) answer spontaneously, 2) need to think, 3) go and look up the answer in documents, ask their partner, etc.’ This procedure worked well during data collection, and the results offer promising information. Respondents were less likely to know the date of a PACS than of a marriage; and when they did give a response, it tended to be much less spontaneous. Moreover, the spontaneity of the response correlated highly with the respondents’ reasons for formalizing their union and how they celebrated it (Rault and Régnier-Loilier, 2016). A specific study on this subject is under way.
22Each conjugal configuration was described in detail. At the time of the survey, among those aged 26–65 years, 51% of respondents were married (35% religiously, 16% only in a civil ceremony), 5% were in a PACS, 16% cohabiting, 5% in a non-cohabiting relationship, and 23% were in neither a couple nor a serious intimate relationship. Among the aims of the specific questions on each form of union were, first, to determine why individuals were in one or another situation and, second, to describe each one of them. Which member of the couple had the strongest wish for this or that type of relationship? How were officially registered unions (PACS and marriage) celebrated and organized? Was there a party? How many guests? Who paid the costs? What sort of attire was worn? Were intentions announced to family and friends? One must study rituals to capture the diversity of the meanings of different forms of union in private life, as these meanings are expressed outwardly in how a couple celebrates or does not celebrate their union. More broadly, studying rituals allows researchers to investigate the forms of union as well as their place within couple formation and in today’s society more generally (Maillochon, 2016). The survey also looked at premarital rituals like engagement celebrations, which continue to occur (Rault, 2018), and ‘stag’ or ‘hen’ parties, which are falsely understood to be ‘traditions’ and are now unprecedentedly popular.
3 – Better accounting for the role of sexuality
23Among the principal transformations in couple formation, Bozon noted in 1991 the ‘new place of sexuality in couple formation’ (p. 69), as sexual relations increasingly came to serve as markers indicating the beginnings of a relationship. These now take place before becoming partners, cohabitation, and a fortiori marriage; whereas these three events were still frequently linked in couples formed in the 1960s. Since then, not only has this new role been confirmed, but sexual debut and the beginning of conjugality have become increasingly dissociated (Toulemon, 2008).
24Sexuality was already present in the 1959 survey, but it was only approached through general representations and not as a stage in union formation. These representations were revealing of the context of the time, when women still experienced strong pressure to remain virgins until marriage. The survey thus asked respondents: ‘In your opinion, is it important for a young woman to save herself for marriage?’ EPIC included a richer set of questions on sexuality than did previous surveys. In addition to determining the date of the couple’s first sexual relations (which was already introduced in the 1983–1984 survey), the questionnaire asks about age at sexual debut and other aspects of the respondent’s sexual trajectory, such as the use of online dating sites to meet sexual partners, number of male and female partners over their lifetime, and representations of the connections between sexuality and conjugality.
III – EPIC in the field
25The scientific objectives of the survey were accompanied throughout by methodological reflections on how they were to be pursued. The population coverage of the sample, the sensitivity of the respondents to the topics in the survey, methods of administration: all were examined in regard to the main aims of the project.
1 – Determining the coverage
26Insofar as the objective of the survey was to collect information on conjugal and intimate relationship histories, the EPIC team surveyed individuals aged at least 25 years (on 1 January 2013). At this age, most individuals have had their sexual debut, but a large proportion of them have not yet lived with a partner.  Given the retrospective record of relationships aimed at in the survey, it thus made sense to begin at this age. Furthermore, collecting histories from earlier ages allowed us to study the beginning of conjugality. The upper age limit was set at 64 years, which was high enough to maintain continuity between EPIC and the 1983–1984 survey on couple formation, which in turn had covered a population that was continuous with the unions from the 1959 survey. The three surveys combined thus allow researchers to investigate 100 years of couple formation (under certain conditions; see Appendix). Separations and repartnering in late life remain rare. Because the survey took place in late 2013 and early 2014, the age range covered by EPIC is 26–65 years. Only private households constituting a principal residence in metropolitan France were surveyed. 
2 – Size of the survey and sampling frame
27The sample was drawn from the OCTOPUSSE master sample from the last available annual census survey (2012), with the coverage restricted to residences including at least one occupant aged 25 to 64 years on 1 January 2013 (thus, an occupant born between 2 January 1948 and 1 January 1988).
28Next, 14,434 households were selected for data collection and assigned to the 400 interviewers, the aim being to survey around 8,000 individuals. The sample size had been chosen to provide enough observations to study certain phenomena that remain uncommon in the population as a whole because they have recently emerged (use of online dating sites, the PACS, non-cohabiting couples, etc.). In the end, 7,825 persons were surveyed (completed questionnaires), a response rate of 62%.  Non-participation was generally due to refusal or avoidance (25%), to persons being unreachable or absent during the data collection period (9%), and to interviews that proved impossible to perform (3%).
29As in most surveys, despite random selection within households, certain categories of individuals are over-represented or under-represented. This is true, for example, of women, who are more likely to participate in surveys than men. EPIC is no exception: women make up 57% of the corpus. After cleaning the data, a weighting variable (at the individual level) was thus created. This was a two-step operation: first, correction of total nonparticipation based on information available in the sampling frame, for respondents and non-respondents; and second, calibration based on variables from the continuous Employment Survey (2013 ): the combination of sex and age, size of the urban unit, level of education, region of residence, socio-occupational category, nationality, residence in a sensitive urban area, type and size of household.
3 – Motivating individuals to participate in a survey on an intimate topic
30To ensure the representativeness of the data, particular attention was given to how the survey was presented to potential respondents. How could individuals be motivated to participate in a non-obligatory  survey on a topic that touches on intimate matters in a way that does not ‘hide’ the topic? One objective of the first test (2010; see Box) was to identify the best way to present the survey so that the letter of notification and associated brochure would be as effective as possible.
31The chosen strategy was to present the survey as the latest in a series, following two historic INED surveys on the subject, which would allow researchers to retrace the history of the couple going back to the early 20th century. This choice proved fruitful. The persons contacted saw the survey as an opportunity to participate in an ‘unusual’ survey. However, the choice to explicitly define the topic as ‘intimate [amoureux] and conjugal histories’ posed a problem, as the connotations of the term amoureux (love) are too personal and not sufficiently factual. It seemed necessary to immediately go beyond the personal and intimate nature of the survey on initial contact and instead emphasize the need to update knowledge, the survey’s sociohistorical depth for researchers, and its ‘heritage’ value. The trials on contacting potential respondents showed that associating the conjugal dimension of the survey with the family dimension was beneficial. The formulation ultimately chosen for the EPIC survey was: ‘Following two surveys on this topic in 1959 and 1983–1984, this new study will allow us to update our knowledge on conjugal and family life while also allowing us to retrace the evolution of the couple over the last century, from the early 20th century to the present’ (excerpt from the letter of notification).
32Despite this approach, the pilot survey (2013) found that EPIC had more difficulty making contact than the other INSEE surveys did, first because it was not obligatory, but also because of its topic. Individuals’ responses to contact were often clear-cut: they offered either unhesitating and even enthusiastic participation or outright refusal due to the subject being too personal and intimate. The survey design team thus devoted an entire portion of interviewer training to making contact, using role-playing games as exercises. A set of arguments was also provided in the instructions on data collection to help interviewers be persuasive when faced with a variety of refusals.
33In the second test (2011), the survey was simply titled the ‘Couple Survey’, which proved to be problematic when approaching people not in a relationship. Although the presentation brochure that accompanied the notification letter emphasized that the survey was also addressed to them, the title interfered with this message. People without partners did not feel that the survey concerned them. A few months before data collection, the name for the survey was finally found: the Study of Individual and Conjugal Trajectories (EPIC).
4 – Choice of administration method
Using an alternative administration method to recruit people who initially refuse
34Although EPIC is not a survey on sexuality (in which case administration by telephone has proven more appropriate for some topics; Bajos and Spira, 1991), recounting one’s relationship history to a stranger is not necessarily a straightforward decision. One main objective of the second test (2011) was to decide on the most appropriate data collection method, given the topic of the survey. The participation rate that could be expected also remained an open question.
35While the data transmitted by the polling company administering the test did not allow for precise calculation of a net participation rate, it was clear that the figure was unsatisfactory, particularly by telephone (around 20% by telephone compared to 26% face-to-face). This was due to several factors: the reputation of the company, whose image is more often associated with opinion polling than with scientific surveys; the way in which the sample of addresses had been established, even though selecting from the white pages for the same year suggested that the name and address information would be fresh and up to date; and the interviewers’ experience in conducting studies of this type.
36This test showed higher face-to-face participation, but 58% of the respondents said that they would not have agreed to respond through any administration method other than the one used to interview them (meaning those who had responded by telephone would not have responded face-to-face, and vice versa). This led us to choose face-to-face data collection as a first option; for people who declined, they also had the option to respond by telephone, thus allowing us to optimize the participation rate. This protocol was tested in the pilot survey (2013), which confirmed certain advantages to giving them the option to respond by telephone, such as being able to contact respondents at atypical times and obtaining more confidential responses (since it was determined that a partner’s presence interfered less over the telephone than during an in-home interview).
37Overall, 9% of the interviews were conducted over the phone. However, this proportion does not only reflect the ‘efficiency’ of attempts to interview persons who were not inclined or available to respond face-to-face; it is also due to the fact that, during the extension of the data collection period (January and February 2014) in certain regions where the process had fallen behind schedule,  telephone interviews were offered at once, rather than as an alternative option if refusal occurred. Table 1 thus illustrates a double effect: as data collection advanced through 2013 (the initial period in the field), repeated attempts at interviewing respondents bore fruit with increased numbers of telephone interviews; while in 2014 (during the extension), the proportion of telephone interviews became high due to adapting the protocol, but their number remained limited.
Administration method by month of data collection
Administration method by month of data collectionInterpretation: Of the interviews carried out in October 2013, 96.4% were performed face-to-face.
In face-to-face interviews, a few self-administered questions
38The decision to favour face-to-face interviews led to some discussions on how to pose three questions related to sexuality.  These questions were met with some a priori reservations by INSEE, which is not used to posing questions of this type.  During the pilot survey, the decision was made to try two administration methods in random fashion: in half of the cases, these questions were asked explicitly, while in the other half they were self-administered through audio-CASI (that is, the respondents entered the responses directly into the computer after listening to the question through headphones). The objective of this experiment was to determine the most appropriate administration method. When the questions were asked through headphones, a self-administration question was added: ‘If this question had been posed not through headphones but by the interviewer directly, would you have answered in the same way? Yes, you would have responded in the same way / You would have answered it, but not with the same sincerity / No, you would not have answered at all.’ The responses were divided. Some considered the headphones unnecessary, with trust already established between interviewer and respondent at this point in the interview; but others welcomed them, particularly when a third person was present for the interview. The number of refusals proved slightly higher when the headphones were not used; but when they were used, surprising answers were sometimes given, which might indicate disguised refusal or errors while entering the response. Moreover, the self-assessment question showed that most respondents would probably have given the same response to a direct question from the interviewer.
39To avoid having to implement a complex and costly protocol using headphones that deliver uncertain benefits while avoiding discomfort for the interviewers or participants, an intermediate solution was devised. These three questions were self-administered, but without headphones: the computer was turned toward the respondent, who was invited to answer these questions directly on the keyboard. To ensure confidentiality, a locking system then prevented anyone from looking at the responses.
5 – Quality of the collected information
Effect of mixing data collection methods on responses
40Using multiple modes of data collection raises questions of possible response biases. The rate of refusals to answer the question on age at sexual debut, for example, was markedly higher when the interview was carried out by telephone compared to face-to-face interviews in which this question was self-administered. This refusal took two forms: directly (12% of respondents refused to answer over the telephone versus 7% in the face-to-face/self-administered situation) and indirectly through a markedly higher proportion of ‘don’t know’ responses (7% by telephone versus 2% in the face-to-face/self-administered situation). The same tendency appeared in responses to the question on monthly income, which was posed explicitly in both data collection modes: 8% of respondents refused to answer over the telephone versus 4% in face-to-face interviews, and the proportions of ‘don’t know’ responses in the two cases were 16% and 11%, respectively. Preliminary multivariate analyses aimed at measuring the net effect of the administration method (controlling for the effects of other factors) confirmed these trends. However, caution is required in interpreting these patterns, as telephone interviews were used mainly as a second option in case of refusal or inability to give a face-to-face interview. The population of those who responded by telephone was thus intrinsically different because it was made up of persons who participated less readily in the survey and who may thus have been more inclined to refuse to answer certain questions. The data thus confirm that giving potential respondents the option of responding by telephone allowed the survey to reach a specific population.
Are questions that are legally considered ‘sensitive’ also sensitive for respondents?
41The above observations on the rate of refusal to answer certain questions (age at sexual debut and income) raise further questions about their ‘sensitivity’ for respondents—and notably those classified as such by the French data protection authority (Commission nationale de l’informatique et des libertés). Several questions in the EPIC survey meet this criterion, including those on sexuality (‘Over your lifetime, and including your current situation, how many women have you had sexual relations with?’; ‘And how many men?’), religion (‘What is your religion, if you have one?’), and political opinions (‘Politically, would you say that you tend to be… very left-wing / left-wing / in the centre / right-wing / very right-wing / no political opinion?’). The survey introduction informed the participants from the outset that it included questions of this type and, further, it explicitly stated that the respondent could choose not to answer them. To what extent do respondents consider these questions ‘sensitive’? The rate of refusal in answering them depended heavily on the subject matter: a little over 8% for questions on politics and sexuality versus under 2% for the question on religion. Other questions not classified as sensitive under the French data protection act proved more sensitive for respondents, such as income (refusal rate of 5%). The tendency to refuse to answer certain questions varied not only with the data collection method but also with the respondents’ social characteristics (Rault et al., 2014). The conditions under which the interview takes place can also have an effect; for example, one-on-one situations versus having a third person present.
The effect of a third-party presence during the interview
42During the first test phase (2010), 40% of interviews took place in the presence of a third person (most often the respondent’s partner). A quarter of these respondents indicated that—although interviews in this test phase were conducted over the telephone—the presence of a third person affected them.  The survey design team thus included an indicator for the presence of other persons at different points in the interview. Furthermore, the instructions for the pilot survey (2013) specified that a quiet space should be sought in order to ensure confidentiality and avoid any interference from others present. Despite these recommendations (which were included in the data collection instructions and which were pointed out during the interviewers’ training), the interview was not performed one-on-one in 1 of every 5 cases. One reason for this is that it is not always possible for the interviewer to impose the conditions for the interview; second, it is commonplace for there to be significant coming and going during an interview that has an average duration of one hour (Régnier-Loilier, 2007). Respondents did not necessarily experience any awkwardness concerning the presence of a third person, which they sometimes accepted in terms such as ‘I have nothing to hide from my spouse.’ However, it did seem to change the tone of the interaction in some cases. For example, one interviewer noted in her log book from the pilot survey that she ‘felt that the responses came with more difficulty, less spontaneously.’ The telephone interviews, in contrast, seem to have limited this problem.
43In the survey itself, an indicator for the presence of other persons was distributed throughout different parts of the questionnaire to obtain the most precise possible information on the conditions under which the respondent was answering the questions (which could change during a single interview). In the face-to-face version, it took the form of a simple question to be answered by the interviewer: ‘Are there, at this moment in the interview, other persons present in addition to the respondent? Yes / No; if yes, indicate the person(s) currently present: spouse, partner / children, stepchildren, grandchildren / parents or parents-in-law / brothers or sisters / friends.’ Over the telephone, the question was explicitly posed. In 27% of interviews conducted face-to-face and 19% of telephone interviews, a third person was present during some or all of the interview. The other person was most often either the respondent’s partner (7 in 10 cases) or children (3 in 10 cases), with other persons in rare cases. Responses to certain questions proved to be related to these interview conditions. In particular, respondents were more likely to refuse to respond to certain questions when their partner was present (age at sexual debut, see Figure 1; lifetime number of sexual partners). Similarly, certain situations were less likely to be reported when a third person was present (for example, men were less likely to report that they had been in love with someone with whom nothing or little had happened). Additionally, men’s answers to questions on opinions concerning fidelity also varied in this situation. More specifically, answers obtained from telephone interviews seem less subject to variations than those obtained in face-to-face interviews (Régnier-Loilier and Rault, 2016). An in-depth methodological study is currently under way for examining the effect of a third-person presence on responses.
Proportion of refusals to respond to the question on age at sexual debut according to conditions of administration, by sex
Proportion of refusals to respond to the question on age at sexual debut according to conditions of administration, by sexInterpretation: Of the men interviewed in the presence of their partner, 9.7% refused to give their age at sexual debut.
Note: The share of refusals to respond (dot) are represented with 95% confidence intervals.
Coverage: Individuals aged 26–65 years with a partner, cohabiting or not.
What effect does the sex of the interviewer have on the interview situation?
44Other paradata were included in the EPIC database, notably the interviewer’s sex and age. During the survey’s development, especially the initial tests, the design team considered the question of the best possible configuration for the interviews. This question was motivated by previous methodological studies, notably those based on the ACSF study analysing sexual behaviour in France (Analyse des comportements sexuels en France, 1991–1992). These had yielded evidence of interviewer effects (sex and age) on certain behaviours. For example, women interviewed by another woman seem to have under-reported the practice of masturbation (Béjin, 1993). However, while EPIC did cover intimate topics, overall its subject was less personal than that of sexual practices. Moreover, respondents in the testing phases did not demonstrate any particular discomfort about having to respond to either a man or a woman. While interviewers and respondents were not matched according to their gender, the interviews were conducted mainly by women (83%), who make up the bulk of the INSEE network.
45No in-depth study has been conducted on interviewer effects, but certain differences emerged in the descriptive analyses.  For example, women interviewed by a man were more likely to refuse to answer the question on age at sexual debut; no effect on this question appeared among male respondents. For other questions, notably those on opinions (degree of agreement with the statement ‘It is possible to love someone and have affairs on the side’ or ‘It is possible to have sex with someone you do not love’), both male and female respondents were more likely to disagree when being interviewed by a woman.
6 – The partner questionnaire
46The partners of the respondents to the main questionnaire were consistently invited to respond to a brief questionnaire (around 20 minutes) aimed mainly at collecting information on their conjugal and intimate relationship history. At the end of the principal questionnaire, respondents with a partner were asked to pass on to them a letter along with a paper questionnaire (if the partner was present—which was true in 41% of cases—the partner was given the questionnaire directly; Figure 2). No age restriction was set for respondents’ partners. They were asked to complete the paper questionnaire and return it in a pre-addressed stamped envelope or to respond over the Internet using a personal identifier and password indicated in the letter. A response-monitoring procedure was implemented to follow up with persons who had not responded. It stipulated two reminders: one by telephone a week after the interview and the other by mail (accompanied once again by a questionnaire and personal identifiers) after two weeks. In total, 3,216 persons responded to the complementary questionnaire, a third over the Internet.
47Figure 2 recapitulates the configurations that led to partners’ (non-) participation. The crude participation rate was 57%. This rate was markedly higher when the interviewer handed the questionnaire to the respondent’s partner (72% versus 47% otherwise). This rate was much lower when the main interview was conducted by telephone (making direct contact with the respondent’s partner more difficult), in specific conjugal situations (where love had declined since the beginning of the relationship, for example), and most of all when the respondent was in a non-cohabiting relationship (Régnier-Loilier et al., 2014). The low participation rate among non-cohabiting partners (24%) led in the end to our decision not to include these responses in the database. Only the 3,005 questionnaires collected from cohabiting partners were included.
48The weighting of the data from the partner questionnaire was calculated in two steps: first, we corrected for total non-response based on the information collected through the main questionnaire; and second, we calibrated the same variables as those in the principal component of the survey. The variable used to weight the partner component does not, however, make the data representative of couples in France (see Appendix).
Configurations involved in the partner participation component of the EPIC survey
Configurations involved in the partner participation component of the EPIC survey
49As in any survey of this type, the scientific and methodological choices that guided the EPIC survey are not without their limitations. Some relate to the choice of coverage. Because of the objectives of the survey, which involved an important retrospective dimension as well as budgetary constraints, the age range was restricted (26–65 years). This choice means that EPIC cannot be used to study separation and (re)partnering at more advanced ages even if these have specific characteristics, as documented by Caradec (1996). These topics probably merit a specific survey, which should explore the relationships between the end of working life and the possible redefinition of conjugal, family, and intergenerational relationships in this context. Nor can EPIC be used to study couple formation among young people in the 2010s due to the minimum age of 26 years. And yet, union formation has undergone major transformations. The desynchronization of the beginning of sexual activity, life as a couple, and the arrival of children has been intensifying over several decades, accompanied by the emergence of more or less unprecedented relationship forms: short-term cohabitation, non-cohabiting intimate relationships, ‘polyamory’ (which explicitly challenges exclusive monogamy), encounters resulting from the use of geolocation applications, increased reporting of relations with persons of the same sex, etc. Here again, the nature of these transformations is such that they merit a specific survey that focuses on the youngest age groups and that follows in the footsteps of previous surveys on both sexual behaviours and couple formation. An INED research team has thus begun to prepare a national survey on the emotional and sexual lives of young people.
50Every survey is made from choices. This article presents some innovations, such as establishing a history of serious intimate relationships by integrating questions on separation and time outside relationships, as well as exploring in greater depth certain topics like the role of the Internet, the PACS, and non-cohabiting relationships. What is more, it represents a continuation of previous investigations on the process of couple formation, the importance of different criteria in partner choice, and the public celebration of marriage. All these factors together resulted in a need to leave other aspects aside. This was notably the case for certain economic and legal dimensions of couple formation (Who pools their resources? Which resources? Beginning when?). These questions can be explored to varying degrees by using other sources such as the ‘decisions in couples’ module of INSEE’s Time Use Survey (2009–2010), tax data, or the Household Wealth Survey (Frémaux and Leturcq, 2013).
51The EPIC survey follows on from two previous surveys on the same subject while taking an updated approach to different topics, thus offering an innovative portrait of the formation of contemporary couples. It also creates the conditions for a historical analysis, as the couples studied by Girard and then later by Bozon and Héran were formed between 1914 and 1983. Now, with the EPIC survey, it is possible to study a century of couple formation in France.
AcknowledgementsThe EPIC survey was carried out by INED and INSEE. We wish to extend our sincere thanks to Jean-Luc Dedenon and Tania Gluminski (INSEE, Direction régionale de Nancy) for their steadfast involvement in this project, to the INED and INSEE teams that accompanied us all the way to the project’s completion, and to all the survey respondents for their time, trust, and participation.
EPIC benefited from the financial support of the French National Research Agency (CECHIC Project, Corpus pour l’Étude de cent ans d’histoire du couple en France, reference: ANR-12-CORP-0016-01), the Caisse nationale des allocations familiales (CNAF), the research and statistics department in the Ministry of Health (DREES), and the iPOPs Laboratory of Excellence (Individuals, Populations, Societies).
Finally, we would like to thank Géraldine Vivier (INED) for her careful review and constructive comments.
52Due to the large retrospective component of the EPIC survey, a few notes on data analysis and interpretation are warranted.  It can be tempting to try and use this type of data to study the evolution of particular behaviours across generations or to quantify the number of couples with particular characteristics. Such studies can be undertaken, but precautions must be taken. First, the cross-sectional nature of the EPIC survey raises questions about how far back through past cohorts one can go to follow changes over time. Second, as the survey’s unit of observation was the individual and not the couple, appropriate weighting is needed to count couples.
How far into the past can one go with the retrospective EPIC data?
53To track changes over time, we can look at an indicator for multiple consecutive years. However, indicators cannot be calculated for years too distant from the time of the survey. Figure A.1 presents a Lexis diagram with time on the x-axis and age on the y-axis. In 2013, the year of the survey, the women and men who took part in the survey (born between 1948 and 1987) were between the ages of 26 and 65 years. These individuals were at most 39 years old in 1987 and at most 27 years old in 1975. Thus, with right truncation, the further back into the past we go, the younger the observed population.
54As an example, let us take the evolution that occurred between 1987 and 2013 for the proportion of persons who have never lived with a partner. Using the answers to the retrospective questions, we can calculate whether each respondent had already lived in a couple in a given year. For 2013—the year of the survey—the proportion of persons aged 26–65 years who had never lived with a partner can be calculated. This proportion can also be calculated for 1987; but in this case, the coverage will be much more restricted because the respondents were then aged between 0 and 39 years. Thus, when comparing prevalences at different times, it is essential to calculate them for the same age range to avoid bias due to the ages of the respondents. In our example, it is only possible to calculate change in the proportion of persons aged 26–39 years who had never lived with a partner between 1987 and 2013 (Figure A.1, coloured square). Similarly, if we wish to determine change in this variable since 1975, we can only describe the population aged 26–27 years (Figure A.1, hatched rectangle).
Lexis diagram for respondents to the EPIC survey (age and cohorts)
Lexis diagram for respondents to the EPIC survey (age and cohorts)
How to count couples
55The weighting of the survey is individual. Because a couple consists of two individuals, their inclusion requires that at least one of them must fall within the population covered by the survey (i.e. between the ages of 26 and 65 years at the time of the survey). Thus, in all couples described by the respondents to the EPIC survey, at least one of the members was between the ages of 26 and 65 years at the time of the survey. Certain couples could therefore have been described twice if both partners belonged to this population. Others would be described only once if only one of the partners met this age criterion. Finally, certain couples were not included in the survey. In giving indicators on couples, it is important to take these facts into account and alter the weighting correspondingly.
56For example, if we wish to calculate the number of couples in metropolitan France in 2013, we can establish the number of couples only by including at least one person who fell within the coverage of the survey. In Figure A.2, couple (A) is counted twice, as each partner is included in the target population; the weighting of the individual who described this couple should thus be divided by two. In couple (B), one partner is included in the population and the other is not; in this case, we maintain the individual weighting of the person who described the couple. Couple (C) was not observed (because neither member of the couple was between the ages of 26 and 65 years at the time of the survey) and will thus not be counted.
Inclusion of couples in the scope of the survey
Inclusion of couples in the scope of the surveyInterpretation: (A) Both partners are in the scope. (B) Only one partner is included in the scope. (C) Both partners are outside the scope, as neither of them are 26 to 65 years old.
57To calculate weighting at the couple level, we must know whether the partner of the respondent to the principal questionnaire fell within the age range for the survey. In the example described here, there were 21.6 million persons in couple situation (A) (cohabiting with a person aged 26 to 65 years at the time of the survey) and 1.9 million persons in couple situation (B) (in a cohabiting relationship with a person under the age of 26 years or over the age of 65 years). There were thus 23.5 million persons aged 26 to 65 years in a cohabiting relationship in 2013. In the same year, there were 12.7 million couples that included at least one person aged 26 to 65 years (21.6 million/2 + 1.9 million).
The civil solidarity pact, or PACS, is a contractual form of civil union created in 1999 and is available to both same-sex and different-sex couples. It has grown continuously since its creation. In 2016, four PACS were concluded for every five marriages celebrated.
However, the data suggest this trend is slightly slowing down among the youngest cohorts (Rault and Régnier-Loilier, 2015).
For example, the month of June is over-represented in the imputed monthly variables, making them inappropriate for studying the seasonality of events. In this case, the crude variables are preferred (see, for example, Breton et al., 2018).
For example, out of the 14,699 relationships on which information was collected, 8% of second relationships began at least a month before the end of the previous relationship, 4% at least one year before.
On the basis of the ERFI survey (INED–INSEE, 2005), we had estimated that nearly three-quarters of those aged 18–24 years had never lived with a partner.
As in most surveys conducted in France, only French-speaking individuals participated because the questionnaire was not translated into any other language. If the potential respondent did not speak French, the respondent was classified as ‘impossible to interview’ but included in the denominator for calculating the participation rate.
Households that were out of scope consisted mainly of persons outside the age range for the survey and of empty residences.
2012 for the socio-occupational category and the size of the urban unit due to problems encountered in the 2013 Employment Survey for these variables.
Certain public statistical surveys are obligatory, which is emphasized in order to convince individuals to respond. The design team did not wish to request this status due to the topic of the survey (a personal subject, which can be sensitive in some situations).
Due to management difficulties linked to a lack of interviewers in these regions at the time of the EPIC survey. The extension of data collection applied only in certain interviewer action areas (ZAE) of INSEE’s four regional divisions.
Age at sexual debut, number of women and number of men with whom the respondent had had sexual relations.
Although some had already been asked in past INSEE surveys (for example, age at sexual debut in the 1994 Family Situation and Employment survey).
At the end of the interview, the respondents were asked about their experience of the tested questionnaire.
Chi-squared test in descriptive statistics and logistic regression controlling for the data collection method (telephone or face-to-face), age, and level of education.
We thank Vianney Costemalle (INSEE) for having made these two issues explicit.