1The main aim of a census is to count the population of a given area by sex and age, and to describe its main characteristics. This entails asking an (often limited) series of simple questions about the occupants of each surveyed housing unit. More ambitious aims, which are only partially met by the French census in its current form, are to characterize the kin relationships between these occupants and to identify the different family forms observed in each household. Do some family forms remain undetected in the census? How are children in shared physical custody and three-generation families counted and categorized? Loïc Trabut, Éva Lelièvre and Estelle Bailly used the additional information on family kinship ties collected by the “Famille et Logements” (Family and Housing) survey coupled with the 2011 census to analyse the family types identified by these sources and look for possible discrepancies between them.
2For the first time since the new French population census was introduced in 2004, based on annual surveys of samples selected on a geographical basis (Desplanques, 2008), an updated version of the Family and Housing survey (Enquête Famille et Logements, EFL) was conducted in 2011. Traditionally coupled with the census, the aim of the EFL survey is to provide a detailed description of the diversity of family forms in France.  The previous survey of its kind dated back to 1999 (Lefèvre and Filhon, 2005). The self-administered questionnaires are distributed at the same time as the census forms and the completed documents are collected by a census taker. To simplify the collection process, the 2011 census takers assigned to the EFL distributed questionnaires to either women or men only, depending on the sampling area. A total of 359,770 questionnaires were thus collected from a regionally representative sample of men and women aged 18 years and over. This large sample size made it possible to study some of the less common family configurations.
3The four-page EFL questionnaire focuses on the family territory, asking about the residential location of family members across three generations (respondents and their parents and children; Rault et al., 2011). As the survey is coupled with the census, additional information about the respondents can be obtained from the housing form of the annual census survey (Enquête annuelle de recensement, EAR) and from the individual forms (IF) filled in by (or for) each member of the household.
4The aim of the present article is to examine the family types brought to light by the census by comparing the census data with those of the associated EFL survey. This direct comparison reveals divergences between the two sources and enables us to identify more clearly the composition of the family categories produced by the census. This comparison looks at how the census, using its own definitions, constructs families based on household composition, bearing in mind that the resulting statistics on family structures are routinely supplied to researchers, decision-makers and the general public.
I – Counting and describing families
5Although the comparison we describe here is part of a broader debate on how the family is defined and how it functions (Bonvalet and Lelièvre, 1995; Bonvalet and Lelièvre, 2016), our aim is not to offer an alternative definition of the family or even to discuss whether observations based on co-residence can adequately capture the complexity of family configurations. Rather, we want to examine how well the census currently measures different family forms, with the aid of complementary data yielded by another survey (EFL) that was conducted at the same time and completed by the same persons. We begin by describing these two sources in order to set the terms of the comparison.
6The census is based on housing units and their occupants (Lefranc, 1997). Information about the nature of each housing unit is collected via the census housing form. This form lists the occupants under three headings: “Persons living in the household most of the year” (List A); “Children over 18 who are currently living away from home for educational reasons […] but who return home” (List B); and “The household’s other occupants” (List C). The housing form is accompanied by individual forms, one for each of the “household’s permanent occupants” (i.e. List A), whatever their age. This description goes well beyond the family per se, as all the housing unit’s occupants are included in the lists, and for each one their “kin relationship or connection to the first person on List A” is recorded. As household members are described solely in relation to this reference person, it is not always possible to establish a clear and unambiguous picture of the ties between each of them. These individual forms provide additional information, including the person’s sex, date of birth, conjugal status (“Are you living in a couple?”) and marital status.
7The prime purpose of the census is to enumerate the population. Each individual is therefore recorded as residing in a single place and, where applicable, allocated a single position within his or her family, as defined by the census. In line with United Nations recommendations,  which are based on the notion of identifying families within a given household and on the concept of the nuclear family, the information collected is used to establish a typology that distinguishes between persons living alone, couples with or without children, lone-parent families (within the household) and complex households (Toulemon, 2012).  The census is thus also used to describe the family structures of the French population.
8The EFL aims to describe respondents’ families without necessarily determining the household composition and without restricting its scope to household members only. Carried out in the homes of individuals surveyed by the EAR, it provides data on the respondent’s family, both inside and outside the household, and no-one else. Based on information provided by all the male or female adults living in the housing unit, the EFL describes three generations and indicates whether there are any grandchildren (fourth generation). It asks about the respondent’s (ego’s) conjugal status (“Are you currently living with a partner?”) and, if so, the type of union. It also asks about ego’s brothers and sisters. There are questions about the presence of children in the dwelling or elsewhere, their relation to ego and his or her partner, and the amount of time they spend in the household. Lastly, EFL collects information about ego’s parents, whether they are dead or alive, where they live, and their conjugal status, via the question “Does your father live with your mother?”. The description of most family members includes their place of residence, so we know whether they live in ego’s home, and if so how often.
9The purpose of the comparison undertaken here is to take the family types as defined by the census (i.e., nuclear families identified within the household) and examine how they are described by the census and the EFL. 
10There are several possible types of divergences between the census and EFL data, each potentially requiring its own type of intervention or correction. Some disparities stem from the fact that the two frames of reference are not directly comparable. For example, family relations that extend beyond the household are, by definition, excluded from the census in its current form. Incoherencies may also arise from the way the data are collected, as well as from the algorithm used by the National Institute for Statistics and Economic Studies (INSEE) to automatically code positions within the family. Lastly, respondents may give inaccurate answers to the EAR or EFL. Our aim is not to examine all these divergences between these two sources, but to shed critical light on the census typology of families which serves as a primary reference in public debate. To this end, we restrict our field of investigation to families that can be meaningfully compared using census and EFL data.
II – Comparing information on families from the population census and the Famille et Logements survey
11Although they have the same respondents, the two surveys differ not only in the wording of their questions, but also, and above all, in their objectives and their frames of reference (occupants of the housing unit vs. the family). Accordingly, some of the disparities that emerge when we compare the co-resident family constructed from the census (designed to count each household’s occupants) and those described by the family-centred EFL arise specifically from this difference in positioning.
12The differing scopes of the two surveys must therefore be examined to compare families described on the basis of these two sources. We will first restrict our scope to co-resident families, as defined by the census, to compare the two samples. We must also take account of unrelated individuals who usually reside in the household (who are not included in the EFL).
13Figure 1 provides one illustrative example of how the census and EFL describe the usual occupants of a household and the members of a family. We will examine the available data concerning persons in households (1) and (2) in both sources, bearing in mind that the 2011 EFL was administered alternately to men or to women in the surveyed households, depending on the sampling area. Our ability to compare numbers and to compare the two sources of information for a given individual is governed by this sampling plan.
General population census and Famille et Logements survey. Description of families and Households (1) and (2)
General population census and Famille et Logements survey. Description of families and Households (1) and (2)
14Let us take the rather unusual example of a mother living with her adult son and daughter and who is in a living apart together (LAT) relationship  with their father. The EAR counts three persons in Household 1 and codes them (see below) as a lone-parent family with two children where the parent is a woman.
- If the EFL is administered to the male members of the households surveyed by the EAR in that area, a single questionnaire is collected here. It describes the family of a young male adult with no children, living with his mother. It also tells us that his father lives elsewhere (Household 2). We know that he has a sister. We therefore have information about three members of Household 1 (and one person living elsewhere).
- If the EFL is administered to the female members of the households surveyed by the EAR in that area, two questionnaires will be collected:
- That of the older woman describes her non-cohabiting couple (partner living in Household 2), her two adult children living in the same household, and her parents living together in Household 3.
We therefore have information about all six related persons, independently of their place of residence: three persons living in Household 1 and three residing elsewhere;
- That of the younger woman describes her childless non-cohabiting couple (partner living in Household 4) and her parents (mother in Household 1 and father in Household 2). We know that she has brother.
We thus have information about three members of the household (and two persons living elsewhere).
- That of the older woman describes her non-cohabiting couple (partner living in Household 2), her two adult children living in the same household, and her parents living together in Household 3.
15In Household 2, the EAR counts one person, coded as a male person living alone.
- If the EFL is sampling female occupants of households surveyed by the EAR, no questionnaire will be collected. 
- If, however, the EFL targets male occupants, one questionnaire will be collected, and the data will describe a non-cohabiting couple (partner in Household 1) and their two adult children, also living in Household 1, as well as the respondent’s parents living together in Household 5.
16We thus have detailed information about one person living in Household 2 and five residing elsewhere.
17The divergences generated by these different frames of reference mean that several precautions must be taken when comparing the two sources. As mentioned earlier, the EFL is designed to describe respondents’ family relationships, so does not record unrelated individuals living in the same household. Similarly, the census does not describe people living elsewhere, which is why its lone-parent families are more like lone-parent households (Théry, 1993, 1996; Lefaucheur, 1987, 1991, 1993). Just because the other parent does not live under the same roof does not mean that he or she does not exist.
Numbers of housing unit occupants counted in the two sources
18We begin our comparison of the two sources by looking at the numbers  of people living in the housing units surveyed by both the EFL and the EAR in 2011 (Table 1). The latter counted “every person who usually lives in this housing unit”, while the former counted respondents, cohabiting partners, children described as living there at least half the time, and parents residing permanently in ego’s home. Out of the 359,770 EFL questionnaires collected, 313,098 (i.e., 87.03%)  can be used to make direct comparisons with the family structures described by the census, as the number of cohabiting family members (EFL) matches the number of household occupants (EAR).
Household size of Famille et Logements survey respondents as indicated by the census enumeration and the EFL count
Household size of Famille et Logements survey respondents as indicated by the census enumeration and the EFL count
19We also find that 10.68% of households are larger in the EAR than in the EFL, which probably did correspond to 38,435 cases with missing individuals that were not immediate family members, i.e. not the respondent’s parents, children or partner. We do not expect these situations to affect the household’s family structure coded by the census.
20Lastly, we find 8,237 cases (2.29%) of missing family members in the EAR. Even though some family members were declared in the EFL as living in the household for at least half the time, no IFs were collected for them. There are several possible reasons for this. The first is linked to a difference in the reference dates. The EAR (and hence the EFL) was conducted between 20 January and 26 February 2011, but whereas the EFL reflects the situation on the actual survey date, the EAR’s reference date was 20 January 2011. EAR procedures therefore “eliminate” children born after that date, and no IFs were filled in for them, whereas the EFL counts respondents’ children who were already born on the date when the EAR was actually administered. The second is linked to the definition of children in shared physical custody, who were systematically counted in the EFL, but not necessarily in the EAR if they spent half their time in one household and half their time in the other, as the EAR instructions clearly state that “[…] where equal amounts of time are spent with each of the parents: the child shall be surveyed in the housing unit where he or she is present on Thursday 20 January”. However, data from the French Justice Ministry, (Guillonneau and Moreau, 2013) place the frequency of these situations into perspective: just 16.7% of custody proceedings following the separation of a child’s parents result in equally shared physical custody.
III – Reconstructing families from census data
21The lengthy and costly task of coding the data yielded by the annual EAR is now largely automated. It is carried out in accordance with EU recommendations intended to permit inter-country comparisons. These recommendations take the nuclear family as their basic unit, and give priority to describing children’s daily living situations.  Individuals are assigned a single and unambiguous position within the family, in order to avoid the risk of counting them twice. To describe the family structures of the French population, INSEE constructs the families living in each household from the information recorded in the census forms, thus identifying people living alone, couples with or without children, lone-parent families and complex households (Toulemon, 2012). Individuals are classified under family types according to the structure of the household, their family and conjugal status, and their age.
22The EFL is not designed to describe the household composition, but collects detailed information about each respondent’s family. It is this information that will be used to assess the census typology and pinpoint divergences between the two sources.
Categories of the L1 and mode of cohabitation (MOCO) variables
1 – first person on the list
2 – first person’s partner (husband, wife, consensual union, in a couple, etc.)
3 – child of the first person or his/her partner, or child’s partner
4 – grandchild
5 – ascendant relative
6 – other relative
7 – friend
8 – boarder or subtenant
9 – live-in domestic worker or employee
MOCO: mode of cohabitation
11 – child of a couple
12 – child of a lone-parent family
21 – adult of a childless couple
22 – adult of a couple with child(ren)
23 – adult of a lone-parent family
31 – person not in a family, living in a household with other persons
32 – person living alone
Information collected and data used
23The EAR’s housing form lists the number of people occupying the housing unit, distinguishing between different levels of occupancy. It records in plain text the relationship of each person “living in this housing unit for more than half the year”, to the first person on List A. Separate IFs are filled in by (or for) each household member on that list.
24As a general rule, while the information contained in the IFs is coded and made available for use, the data provided in Lists A, B and C (i.e. persons living in the household most of the year, adult children living elsewhere, and other occupants) are not recorded in full, especially those, recorded in plain text, on kin relationships of household members with the first person on list A. Some of these data may be used, however, to resolve complex cases such as the ones described below. In fact, the nine possible relationships with the reference person (see Box) are mainly generated by the algorithm, and the coding is only very marginally based on respondents’ actual statements on the housing form. The declared relationships between the various household members and the reference person recorded on List A of the housing form are used to describe only around 6% of EAR respondents’ households. 
The first stage in processing the individual forms (IFs)
25INSEE begins by processing the variables in the IFs for a given household, using an algorithm that “automatically describes the composition of around three households in four. This algorithm is based on the occupants’ age, sex, marital status and response to the question ‘Are you living in a couple?’”  (INSEE, 2011) and assigns the following definitions:
- Empty housing unit if there is no IF for the housing unit;
- Person living alone if there is a single IF for an adult;
- Childless couple if there are two IFs for different-sex individuals who correspond to a couple;
- Couple with child(ren) if there are at least three IFs: two corresponding to a couple and the remainder to children (only those under 18 at this stage in the automatic coding);
- Lone-parent family if there are at least two IFs, one for an adult and the other(s) for minor(s).
26This first stage only takes account of the respondents’ age and conjugal status, as described in the IFs. No use is made of the relationships indicated in plain text on the housing form. This means that the relationships constructed between parents and children do not necessarily correspond to legal or even de facto relationships. “Children” are assigned to one or other partner, and in rare instances, may even be associated with a person who is not necessarily a parent (e.g., uncle, older brother). In this first stage, 75% of the population are automatically classified.
27For the remaining 25%, either the data needed for the algorithm are incomplete, or the situations are deemed to be more complex (e.g., presence of children over 18, more than one couple). In these cases, INSEE proceeds to a second coding stage that uses information contained in the three lists of the housing form and has a less restrictive age limit for children.
Second coding stage: potentially more complex and genuinely more complex households
28After the first automatic classification, the remaining 25% of households, labelled as potentially more complex, are dealt with in a second coding stage. As the enumeration system only allows for a maximum of two families per household, once one family has been identified (based on the presence of a couple or a lone parent and their child(ren) in the housing unit), the remaining household members are all classified in relation to this first family. If one of these is living in a couple, then that person automatically forms a second family, even if he or she is the adult child of the first family (note that an adult cannot be described as the offspring of the first family nucleus if he or she has a cohabiting partner or child).  Lastly, “household members who do not belong to a family are categorized as ‘non-related persons’”.
29For the coding of these households, the age restriction is lifted, meaning that, for instance, children aged 18 or above can be assigned to their parents if they share the same household. Priority is given to identifying couples (even if the two IFs do not indicate that two individuals are living as a couple), based on the occupants’ relationships with the first person on list A. “The relationship indicated in the lists can be that of spouse, partner, friend, cohabitee, civil partner, fiancé, etc. If the context suggests a consensual union, then a couple will be recorded.” Only at this juncture is information from List A of the housing form used to code the relationships (in addition to all the information contained in the IFs).
30In the case of genuinely more complex households, i.e. those that cannot be classified in the second coding stage, the occupants’ IFs are examined again. The rules now are: “the partner […] of the child is also coded as a child (son-in-law, daughter-in-law, stepson, stepdaughter of the first person on the list and/or his or her partner, child’s cohabitee). The partners of any grandchildren are coded in the same way. A child can also be adopted or under guardianship”. Regarding the occupants’ relationship with the reference person (L1 variable; see Box), an ascendant relative can be the reference person’s father, mother, stepfather or stepmother, but also a grandparent (no distinction made between grandmother and grandfather). A friend is neither a relative nor a work colleague, the same being true for any boarder or subtenant. Lastly, a live-in domestic worker or employee cannot be a relative. If such a person is related to the reference person, then this relationship must always take priority.
31Once these relationships have been coded, the data are processed to construct the first family, that of the first person on list A of household members. The information about individuals living in a couple (and thus potential parents) is also re-examined. To ensure that the same person is never counted twice, each individual can only occupy one position within the household. At the end of the coding process, he or she is therefore placed in one of seven categories, using the MOCO mode of cohabitation variable (see Box): child of a couple; child of a lone-parent family (regardless of age); adult in a couple without children; adult in a couple with child(ren); adult of a lone-parent family; person not in a family living in a household with unrelated persons; person living alone. Note that while the L1 variable is only coded using List A information for the 6%  of households that are found to be genuinely more complex, all surveyed individuals are assigned an L1 category by imputation.
32As a result, the census can produce some very peculiar family structures. In Figure 2, the diagram on the left shows a household containing three generations of the same family. The representative of the pivotal generation does not live in a couple, but shares the household with her two parents and two children. In the classification yielded by the algorithm, the mother is associated with her two children, and can therefore no longer be coded as the daughter of her parents, given that each individual can only be assigned one type of relationship. The pivotal individual cannot, therefore, be a daughter because she is a mother, and as a result, her two parents form a family of two childless adults in a couple. The example on the right features a genuinely complex family situation. This household would classically be described as containing three couples: a couple with three (adult) children, two of whom are living with a partner. However, because of the rule that says there can only be two families at the most, the census describes the occupants as forming two families, the first based on the parental couple, the second on the first child and his or her partner. The household’s remaining members are all treated as the children of the first couple.
Coding of the family structures of two surveyed households
Coding of the family structures of two surveyed households
IV – Comparing individual respondents’ positions in the family
33The way the two surveys are constructed and the principles applied by the census for coding family types are responsible for several divergences. For a start, we have seen that the data do not systematically cover the same frame of reference, and whereas respondents are asked to describe their families in the EFL, the census uses a set of very rigid rules to produce its family typology. There are other factors of divergence, too, which we attempt to identify in the comparisons below. For instance, the EAR does not collect all possible information about family relationships, and its methods for constructing families within each household can lead to coding simplifications. Furthermore, respondents’ own descriptions of their kin relationships may be ambiguous, and errors can occur when individuals have more than one home, most notably children in shared physical custody; not forgetting the inevitable missing values and respondents failing to follow the instructions they are given.
34Now that we know the census coding for the family status of each EFL respondent, in order to pinpoint the divergences between the two sources we will compare situations where the EFL provides a precise description of the cohabiting family, making it possible to use ego’s position given by the MOCO variable.
35To this end, we examine situations where the number of household occupants and persons described by the EFL is the same, or lower than that given in the census, i.e. household configurations that should not produce divergences in the description of the co-resident family. This limits differences arising from the two surveys’ dissimilar frames of reference. Also for simplicity’s sake, we eliminate all adult respondents classified by the EFL as children. The census coding of adults living with at least one of their parents depends on a large number of variables (in particular, the number of families in the housing unit, the number of occupants, the adult children’s family status, and whether any children are attributed to them), and would therefore render our analysis unnecessarily complex. Children (under 18) are counted as residents in the EFL if they are declared as living in the household at least half the year.
Comparison of sources when the census and the EFL survey count equal numbers of persons in each household
36Excluding adults counted as children  in the EFL slightly reduces the proportion of households of equal size that we can compare, from 87.03% of the sample to 83.43% (Table 1).  Table 2 summarizes the family position of respondents in households of equal size in both sources.
Family status attributed to census respondents in relation to four family situations identified in the Famille et Logements survey for households of equal size (83.43% of cases)(a) (%)
Family status attributed to census respondents in relation to four family situations identified in the Famille et Logements survey for households of equal size (83.43% of cases)(a) (%)(a) Comparison of cases where the number of persons in the household counted by the EAR was the same as the number of cohabiting family members counted by the EFL.
37For these four simple cohabiting family configurations, at least, the census coding places the vast majority of surveyed individuals in the category that they had described in the EFL.
38All the individuals who stated in the EFL that they were single are coded as persons living alone, while 98.38% of individuals who said they were living with a partner are coded as an adult living in a childless couple. Similarly, 97.58% of EFL respondents living in a couple with child(ren) are correctly classified as an adult living in a couple with child(ren), and 95.08% of individuals with child(ren) but without a partner as an adult in a lone-parent family. The rare mismatches concern the census category person not in a family living in a household with unrelated persons. Most of the 4.83% of EFL adult respondents who identified themselves as lone parents but who were placed in this category were over 50 (many were considerably older) and their cohabiting child(ren) were over 18.
Comparison of sources when the census counts more individuals in the household than the EFL survey
39Cases where more household occupants were counted by the EAR than cohabiting family members by the EFL should not affect the MOCO variable, which codes the person’s position within his or her family. In the event, however, we do find substantial differences in classification between the two sources (Table 3). These are cases where the family described by the EFL respondent cohabits with non-relatives or another family within the household.
Family status attributed to respondents in the census in relation to four family situations identified in the EFL survey for households containing non-relatives (7.07% of cases)(a) (%)
Family status attributed to respondents in the census in relation to four family situations identified in the EFL survey for households containing non-relatives (7.07% of cases)(a) (%)(a) Comparison of cases where the number of persons in the household counted by the EAR was greater than the number of cohabiting family members counted by the EFL.
40It is in these households that include persons who are not members of the family described by the EFL respondent (approximately 7% of our sample) that we find the greatest divergences.
41Situations where ego described him- or herself in the EFL as living in a couple with child(ren) are successfully identified by the census coding process in 84% of cases. However, 56% of childless EFL individuals living in a couple are classified as an adult living in a couple with child(ren), and 27% of those not in a family as an adult in a lone-parent family. Furthermore, 42% of EFL respondents who stated that they were lone parents are classified as not in a family in the census. Differences therefore mainly concern the coding of filiation, with parent-child relationships being either artificially constructed or ignored in the semi-automatic census coding process. 
42In more than half (55.64%) the 5,004 cases where ego did not report any children who cohabited for at least half the year, he or she is classified as an adult living in a couple with child(ren). The census therefore codes a parent-child relationship either with an unrelated member of the household or with a child reported on List A as habitually living in the household. Similarly, even though the majority of EFL respondents who did not declare any cohabiting children or partners are placed in the corresponding category in the census (person not in a family living in a household with unrelated persons), 26.61% of them are labelled by the MOCO variable as an adult of a lone-parent family. These are situations where the EFL respondents did indeed state that they were parents, but in 67% of cases, all their children were in situations of multiple residence, and cohabited for less than half the time. This could potentially mean they were counted twice, once in the home of each parent.
43By contrast, 42% of EFL respondents who stated that they were lone parents are classified by the census as a person not in a family living in a household with unrelated persons. These are mostly older individuals whose children are themselves adults, but despite the relaxation of the age limit, their parent-child relationship nonetheless went unrecognized in the second census classification phase.
44Regarding childless couples, many were assigned a child in the semi-automatic coding stage (55.64%), while in a few cases (5.96%) their conjugal union went unacknowledged. Of these, 18% were cohabiting same-sex couples, so the divergence between the two sources was quite logical, insofar as the census coding system is not designed to construct same-sex relationships (even if the IFs contain information about living with a same-sex partner).
45To sum up, several factors contribute to the differences observed. Some are quite predictable, arising from the nature of the definitions used by the census,  but some stem from shortcomings in the automatic coding of complex households. For example, the fact that children over 18 are only assigned to their parents in the second phase seems to have resulted in an under-estimation of parent-child relationships between adult children and their cohabiting parent(s). Lastly, inaccurate statements made by both EAR and EFL respondents heightened the difficulty of identifying family situations.
46The first conclusion we can draw from these comparisons is that the simplest and most frequent forms of the nuclear family are correctly coded by the semi-automatic census coding algorithm, which does not use any of the information provided in List A about occupants’ relationships with the reference person. Nevertheless, given that the census is supposed to provide information about family structures, we would expect this exhaustive source, which is widely used as a reference, to be better at identifying new, emerging, rare or more marginal situations. Indeed, the coding of family ties upon which the MOCO variable is based, mainly built upon information from the IFs, is not good at describing families in more atypical households – those comprising several families, a group of unrelated occupants, or a family living with non-relatives.
47To explore this issue in greater depth, while remaining within the framework of the relationship types identified by the census, a different perspective is needed. Building on research by Toulemon and Pennec (2011), we use the detailed information provided in the EFL to examine two of the more marginal census categories, namely persons not in a family living in a household with unrelated persons (a total of 15,011 individuals, i.e., 4.17% of the sample) and lone-parent families, the latter being of particular interest to policymakers on account of their higher poverty rate.
V – Overestimation of persons not in a family and of lone-parent families
48By comparing the two sources, we can show that the census typology is good at identifying the more usual forms of cohabiting nuclear family, i.e. couples with or without children and people living alone, providing that the household does not include any other family type. However, for more atypical configurations, the divergences between the two sources are larger. We will now describe two of these more marginal categories of the typology, namely persons not in a family living in a household with unrelated persons (a total of 15,011 individuals, i.e., 4.17% of the sample) and adults in a lone-parent family, using the detailed information reported in the EFL survey.
Unidentified family ties: the shortcomings of the census definitions
49As shown in Figure 3, 15.6% of the individuals classified by the census as persons not in a family living in a household with unrelated persons stated in the EFL questionnaire that they actually lived with at least one of their parents, while 12.1% of them had children who lived with them on a permanent basis, and 23.8% lived in a couple and cohabited with their partner. Hence, 44.1% of the individuals classified in the census as not in a family stated in the EFL that they actually lived in a family group of some kind.
Family situations reported in the EFL by persons classified by the census as not in a family living in a household with unrelated persons
Family situations reported in the EFL by persons classified by the census as not in a family living in a household with unrelated personsNote: As respondents could be in more than one situation at the same time, the figure of 44.1% is lower than the sum of the three separate situations.
50Although we cannot exclude the possibility of inaccurate reporting in the EFL, INSEE’s algorithm seems unable to cope with the problems that arise at the complex families processing stage, which deals with households containing several families, one family living with non-relatives, or a group of unrelated persons. In this second stage, the age constraint is removed, meaning that a child can be over 18, and optical recognition technology is used to take account of the occupants’ relationships with the reference person recorded on the list of household members.  However, the fact that only the relationships with the first person on list A are taken into account, and that the terms employed may take several forms (masculine, feminine, plural) means that the construction of the MOCO variable cannot be corrected. Furthermore, the census only allows each individual to have one family relationship, making it impossible to adequately describe the more complex family structures.
Difficulty of declaring children in shared physical custody
51We showed earlier that some of the lone parents identified by the EFL were ignored by the census (4.92% in Table 2; 49.12% in Table 3, i.e. 1,778 cases). We postulated that a failure to take parent-child relationships between adults into account was mainly responsible for this difference.
52This time, therefore, we looked at the family situations identified in the EFL for 20,685 individuals classified by the census as an adult of a lone-parent family, 10.5% of whom were men and the remaining 89.5% women (Table 4). This coding (MOCO = 23) is based on the presence in the household of an adult not in a couple and his/her child(ren). Parents who are separated have to include their children on one of three lists, depending on their situation (see Section I and Appendix):  List A for “persons living in the household most of the time”, which also includes (with no minimum presence requirement) “children under 18 who are currently living away from home while attending school”; List B for “adult children who are currently living away from home while attending tertiary education […] but who return home at weekends and during the holidays, for example”; or Sublist C2 of the “Other occupants” list C for “children in joint custody who live with the other parent most of the time”. In the case of children who spend exactly half their time with each parent, their inclusion on Sublist C2 depends on where they are on a given date (20 January for EAR 2011).  The EAR does not request any details about the precise frequency or duration of the child’s presence in the housing unit, relying instead on the respondent’s choice of list, and the coding is based on children who are declared as “usually living in the household” and for whom there is an IF. No other information about, say, shared physical custody, is taken into account (Toulemon and Pennec, 2010).
Comparison of how the children of lone-parent families are counted in the annual population census and the EFL survey
Comparison of how the children of lone-parent families are counted in the annual population census and the EFL surveyNote: The difference in numbers between the EAR and EFL can be explained by 136 missing values in the EFL for the amount of time spent by one or more children in the household. Note that when no time was given, the default duration assigned in the data cleansing process was “all the time”.
53Unlike the census, the EFL provides precise information about the respondent’s number of children and their living arrangements. Respondents are asked whether their children live in the household “all the time”, “at least half the time”, or “less than half the time”. Assuming that the instructions for filling in both the EAR and EFL  are correctly followed, children who are declared in the EAR and for whom there is an IF should match those recorded in the EFL as living in the household either “all the time” or “at least half the time”.
54Table 4 allows us to compare the two sources, as it sets out in detail the situations of individuals identified as an adult in a lone-parent family by the census, according to the number of children reported by him or her in the EFL as cohabiting “all the time” or “at least half the time”.
55First of all, 19.2% of persons classified as an adult of a lone-parent family in the census lived with their child(ren) for less than half the time. More specifically, 32% of lone fathers (MOCO variable) stated in the EFL that none of their children cohabited for at least half the time (this figure was 18% for women). These children were therefore either potentially (and legitimately) declared in the EAR as belonging to another household as well (that of the other parent), and thus counted twice at the national level, or else were confused with another individual in the household by the EAR.
56Moreover, when we compared and contrasted the two sources of data, we found a major discrepancy in the counting of children in lone-parent households (Table 4). Whatever the number of children declared in the EAR, the EFL almost systematically produced a lower number. For example, there were 45.5% fewer lone-parent households occupied by a father and his two children in the EFL than in the EAR, and 22.6% fewer occupied by lone mothers living with their two children.
57We can surmise that these gaps arise from incorrect statements in the EAR housing form. Because of the delivery and collection system used to administer both surveys, the forms need to be kept simple. However, the rules for filling in the lists on the housing form are actually quite complex. An additional factor, we believe, is that parents are liable to want to include their children even if they do not live with them all or most of the time. It is clear from face-to-face surveys and from tests carried out for the EFL  that it is important for respondents, especially those in more atypical situations, to be able to describe their children in a satisfactory manner (Lelièvre and Trabut, 2013). For some parents with shared physical custody, who have chosen and arranged their post-separation residence to accommodate their children, the fact that their child(ren) can only be mentioned in a small box at the bottom of the page (Sublist C2), or even not at all, could well explain the additional declarations, especially by men.
58The census in its current form fully meets its remit to enumerate the population living in France. Its observation unit is the household, but its counting mechanism relies on the construction of family types, defined via the MOCO variable, and it was precisely this typology and the way it is constructed that we examined here. Unlike the EFL, the census is not designed to describe families, but the data it produces are used in numerous studies of families in France. The aim of this article was to examine the family categories it produces in terms not of their relevance to the population count, but rather of their ability to reveal contemporary trends. Coupled with the EAR in 2011, the EFL allows us to specify the typologies based on the EAR data and to confirm the family categories produced. This comparison of the two sources led us to investigate the reasons behind the discrepancies observed.
59In situations where a census household contains just one family, as defined by the EFL, the algorithm for coding the MOCO variable was found to be quite adequate, and there is a good match between the four family situations  described by the EFL and the census.
60However, the automatic coding of family ties, based mainly on the information provided in the individual forms, proves less satisfactory for describing more unusual households containing several families, several generations, a group of unrelated people or a family living with non-relatives. The coding of parent-child relationships is particularly problematic, as the semi-automatic system tends either to invent or ignore relations between parents and children, particularly when the latter are over 18.
61The current system whereby respondents only indicate relationships with the reference person on the housing form, and the rules for filling out the various lists, are clearly not suited to describing the family within the household. Moreover, respondents’ statements recorded in plain text may be ambiguous (use of polysemous terms such as beau-père, which can mean either stepfather or father-in-law in French, or ami [friend]). We showed that some of the divergences between the two sources concerning the parent-child relationship stem from the complexity of the rules for filling in the various lists on the housing form. This is especially the case for parents whose children live elsewhere part of the time, and results in considerable divergences in the counting of lone-parent or stepfamilies. Children may either be counted twice or left out altogether, while the number of people not in a family may be overestimated.
62For households comprising occupants with more than one status (child of, parent of, partner of) the categories produced show up several weaknesses in the system, which can be attributed to the method used by the census to describe family structures. Limiting each individual to just one family tie may be essential for ensuring an accurate population count, but this simplification makes it impossible to properly identify situations where a dwelling contains more than one family (multigenerational situations), as it attributes a single family position to individuals who serve as a pivot between two generations.
63There are two potential ways of improving the description of family structures. One option would be to modify the way that the relations between household occupants are recorded in the EAR’s housing form. Alternatively, the EFL survey could be conducted on a regular basis alongside the EAR. We believe it is important to do this, as the published census figures, including those for family types, are widely cited in public debate. Persons living alone, lone-parent families and children in shared physical custody are all examples of cases where a more fine-grained analysis – such as that provided by the EFL is required.
64In the light of the findings presented here, it is clear that the census needs to be modified to take account of relationships between individuals that extend beyond the housing unit. Like many other national censuses (e.g. United Kingdom, Portugal), it needs to look at the relationships between all the household members and not just with the reference person, as we have already indicated.  Digital census data collection has not yet been tested in France, although there have been some successful experiments in countries such as England and Canada. Its introduction would provide an opportunity for an overhaul of the census design to include these changes. This would enable the census to move from being a good enumeration tool to being a genuine demographic tool capable of describing household family structures in a rigorous and systematic manner.
Population census - Housing form
Population census - Housing form
French Institute for Demographic Studies, Paris, France.
Correspondence: Loïc Trabut, Institut national d’études démographiques, 133 boulevard Davout, 75980 Paris Cedex 20, France. email: firstname.lastname@example.org
ANR-LiLi team (ANR-10-ESVS-004): Laurent Toulemon, Wilfried Rault, Magali Mazuy, Catherine Bonvalet, Célio Sierra-Paycha, Julie Chapon, Elisa Abassi, Christophe Imbert and its partner MIGRINTER.
A family is defined as part of a household comprising two or more people who are either:
- a couple living in the household with any child(ren) they have;
- or an adult living with his or her child(ren) (lone-parent family).
The EFL provides fuller descriptions of family configurations because it takes account of persons whose main place of residence is not the housing unit being surveyed (Breuil et al., 2015).
To explore this notion further, see Régnier-Loilier et al. (2009).
The EFL does not provide any data for this type of situation, making additional comparisons impossible.
These numbers do not take account of the sex or age of the persons listed in the housing unit.
All the comparisons here are based on raw, non-weighted and non-imputed data, as this allows us to work on all the responses given by a single individual to the EAR and the EFL questionnaire collected by the census taker at the same time. The respondent’s number of children is based on the declarations made in the EFL, regardless of whether the corresponding IFs were filled in.
An L1 category (see Box) was only assigned to 22,423 individuals in our sample.
For the definitions of an adult, a child and a couple in this first processing stage, see the following INSEE document: http://www.insee.fr/fr/bases-de-donnees/default.asp?page=recensement/resultats/doc/traitementdonnees_rp.htm.
A person ceases to be a child under the census definition if he or she is identified as a (potential) parent.
Proportion observed in the EAR sample associated with the EFL survey.
These are cases where divergences between the two surveys can easily arise, as the census only assigns one family position to each individual.
In our analysis of households where more individuals were counted by the EAR than by the EFL, excluding adults counted as children in the EFL reduces the proportion of comparable households from 10.68% to 7.07% (Table 1).
If a child is present in the EAR (IF and therefore on List A) but absent from the EFL, the outcome will be the same. This is an unusual situation, however, as EFL respondents are asked to give details about all their children, regardless of their age and the amount of time they spend in the household, whereas the EAR applies restrictions to its list of household occupants based on age, labour market status (student or not), and place of residence.
Not taking same-sex couples into account in the census analysis of families living in the household, and not considering people living in a couple or with their own children as children both fall into the conceptual difference category.
The relationships to the reference person are written in plain text on the form. This software “reads” the terms used and automatically codes them.
Census definition: “Children in the custody of the other parent or who must be included in Sublist C2 of the housing form and who do not fill in an IF”.
Providing the respondent has read through the instructions for filling in the forms.
Here, we used the raw data, with 136 missing values.
Couple with or without child(ren), lone-parent family, adult living alone.
See research by the “Changes in the Population Census Questionnaire” CNIS working group, http://www.cnis.fr/cms/site/Cnis/lang/fr/Accueil/activites/Organisation/ Groupes_de_travail/Anciens_groupes_de_travail/GT_evolution_questionnaire_ recensement.