CAIRN-INT.INFO : International Edition

1In France, the census is the main data source for studying migration. This contrasts with countries, such as Denmark, Finland, Iceland, Norway and Sweden, where population registers are traditionally used. Other countries, such as Austria, Belgium, Switzerland and Luxembourg have switched from a conventional census to a register census [1] – or are planning to do so – as a means to cut costs (Laihonen, 2000). The exhaustive French censuses conducted up to 1999 produced data at very fine spatial levels – neighbourhood, street block (notably to monitor priority urban development zones), etc. – and provided a means to observe the mobility of small sub-populations. Studies of recent migration trends, which aim, among other things, to describe changes in population distribution across France, focus on arrivals and departures and on net migration at specific administrative levels, i.e. commune (municipality), département, region, metropolitan France. They require data based on large samples, so generally make use of census material which also includes detailed sociodemographic information. For this reason, the French census is the main data source used by geographers, demographers, statisticians and planners interested in population mobility, local urban development or more general regional studies.

2The things and events of the past tend to acquire the aura of a paradise lost, of a golden age that has gone forever, as the rough edges of reality become smoothed over with time. Concerns that censuses are too widely spaced out, that the information grows stale over the years, that the intercensal period has become longer or more irregular since 1968, or that census data is inaccurate, are all supplanted by the rose-tinted memory of an exhaustive set of statistics all collected at the same time. Moreover, the French intercensal periods are meaningless in themselves. They are not set by convention but depend solely on the budgetary constraints of the moment. The unequal length of these intervals (they have ranged from six to nine years over the last half century) has partially invalidated [2] the temporal and international comparisons based on census statistics.

3With the dual aim of producing fresh information and of spreading the government cost burden more evenly, the French census has been totally re-designed. Since 2004, it has been radically transformed from a traditional exhaustive census of individuals to a sophisticated annual sample survey [3]. The actual “census” now comprises a compilation of data collected over five consecutive years. As France is the only country to have adopted this census method to date, the international scientific community is following its implementation with particular interest, to identify both its advantages and its drawbacks.

4With new responsibilities devolved to local government under the recent French decentralization laws, the state budgets allocated to local authorities depend partly on the size of the populations concerned, so the accuracy and freshness of census statistics is crucial. Likewise, accurate knowledge of the social fabric is vital for local decision-makers. The development of the European Union and its targeted regional development policies are also increasing demand for accurate local data.

5Certain statistical information sources rely directly on the French census. One such example is the permanent demographic sample (échantillon démographique permanent, EDP), a longitudinal data source based on successive censuses and civil records covering persons born in the first four days of a particular month in the year. The new EDP project, built around the new census, is currently being developed and will considerably broaden in scope. The sample size will be multiplied by four and new administrative data will be included, taken from the annual registers of private-sector employees (déclarations annuelles des données sociales, DADS) and probably other administrative datasets [4]. However, the fact that the census data is no longer collected simultaneously or exhaustively raises new problems for this longitudinal sample (see below).

6In the context of radical change in the census methodology and in the associated data sources, this paper first presents the original formula adopted by the French National Institute for Statistics and Economic Studies (INSEE, Institut national de la statistique et des études économiques), then examines the strengths and weaknesses of the new census with respect to the study of residential mobility. Last, the consequences for the permanent demographic sample, a longitudinal dataset based directly on the census, will be analysed.

I – The new census: a sophisticated tool or a complex one?

7Until now, the old-style census provided a measure of residential mobility at different spatial levels – national, regional, departmental, communal and infra-communal – during the intercensal period. Once a first cycle has been completed, the new census should enable us to measure mobility over the previous five years on an annnual basis via the answers to the question “Where were you living on 1 January n–5?”. These data will not be available at all geographical levels however, since not all communes are covered by the census each year (see below). Moreover, to study a specific sub-population, it may be necessary to compile data collected over the last five years so that the sample is large enough to provide a robust estimation. It is therefore important to distinguish clearly between results based on the “annual census survey” whose indicators are representative at national and regional levels only, and those based on the “new census” comprising aggregate data from a complete cycle of five annual surveys Data analysis is more complex than might be imagined at first glance, since indicators are produced by combining two different levels, i.e. a geographical breakdown and a time dimension, given that the indicators drawn from the census are not all representative at the same levels and at the same times (Table 1).

Table 1

New census sampling system (rate in %)

Table 1
Rotation group 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Small commune (fewer than 10,000 inhabitants) 1 20 20 2 20 20 3 20 20 4 20 20 5 20 20 Large commune (10,000 inhabitants or more) 1 8 8 2 8 8 3 8 8 4 8 8 5 8 8 Overall P P P, T P, T P, T P, T P, T P, T P, T P, T Representativeness P = partial France and regions for the annual census survey. Representativeness T = all geographical levels at the end of a data collection cycle (the “new census”): national, regional, communal and infra-communal. The indicators produced for year n are calculated in n+2 so that the entire population of the five annual surveys are included. Interpretation: in 2004, an 8% sample will be drawn in all communes of 10,000 inhabitants or more. The first of the five groups of communes with fewer than 10,000 inhabitants, representing 1/20th of all small communes, will be fully surveyed. The complete set of indicators does not become representative until the end of a five-year cycle.

New census sampling system (rate in %)

8The main characteristics of the census survey are given in Table 2. The survey is conducted in January and February each year, and two complementary surveys are organized simultaneously, depending on the commune size. One-fifth of communes with under 10,000 inhabitants are exhaustively surveyed each year, so that each commune is surveyed once every five years. In communes with more than 10,000 inhabitants, 8% of the population is surveyed each year. In order to spread the sample of respondents and limit the risk of re-enumerating the same individuals, each large commune is divided into five groups of addresses and the rotating survey is conducted each year on a different group. After five years, around 40% of households in large communes have been enumerated.

9The sampling frame of communes of 10,000 or more inhabitants is the regularly updated register of identified buildings (Répertoire des immeubles localisés, RIL), from which a sample of addresses is drawn. As this type of sampling may produce a cluster effect [5], the addresses in each group are scattered over the entire commune to avoid bias due to socially differentiated population distribution in the various neighbourhoods [6].

Table 2

Main characteristics of the new census compared with the traditional census

Table 2
Characteristics New census (from 2004) Traditional census (up to 1999) Periodicity Annual Every 6 to 9 years Survey period January and February February for institutional households, March for private households Sample type Divided into 2 parts Sample of addresses for communes of more than 10,000 inhabitants Sample of communes for communes with fewer than 10,000 inhabitants, balanced at regional level Samples then balanced by infra-communal census zone (IRIS), by sex, age group, etc. No sample, exhaustive survey Mean number of respondents 8.5 million people per year (6 million in communes of fewer than 10,000 inhab., 2.5 million in others), i.e. 42.5 million individuals after 5 years – Mean sampling ratio 14% per year (8% in large communes, 20% in small communes), i.e. around 70% of the entire population after 5 years
tableau im3
Characteristics New census (from 2004) Traditional census (up to 1999) Questionnaire administration method Self-administered Self-administered Collection method Printed forms Printed forms Questionnaire filled in by another household member Authorized Authorized Population All individuals. Persons living in institutional households (soldiers, hospital patients, prisoners, etc.) are surveyed by INSEE. Homeless persons and persons living in mobile homes are surveyed by the communes. All individuals. The census is conducted by INSEE with the assistance of the communes concerned. Representativeness Individuals and households in private and institutional households Individuals and households in private and institutional households Finest level of representativeness Infra-communal census zone (a) Neighbourhood (a) Non-response rate Around 4-5% for 2004 and 2005 No systematic estimates (b) Wording of question on place of residence “Where were you living on 1 January 1999?” The question is then updated each year to maintain a 5-year interval “Where were you living on 1 January 1990 (year of previous census)?” Response categories 0. In the commune or arrondissement to which the respondent is attached (mobile home occupiers, homeless, sailors and prisoners) 1. in the same dwelling 2. in another dwelling in the same commune 3. in another commune of the same département 4. in another département in the same region 5. in another region of metropolitan France 6. In an overseas département (DOM) or territory (TOM) 7. In another EU country 8. In another non-EU country Ditto Additional question “In what year did you move into this dwelling? If all current occupants of this dwelling did not arrive at the same time, give the date when the first current occupant arrived. If this person has always lived in this dwelling, give his/her date of birth.” Ditto (a) In densely urbanized zones, the “neighbourhood” is often a block of buildings. The infra-communal census zones (IRIS) comprise a contiguous group of street blocks forming a small district. (b) The most recent estimate was 2% for omissions and 1% for duplications (Coëffic, 1993).

Main characteristics of the new census compared with the traditional census

10In small communes, institutional households and the homeless are enumerated at the same time as private households. In large communes, the homeless are enumerated every five years from 2006, and one-fifth of institutional households listed in a specific directory are enumerated every year. INSEE has pointed out that due to the heterogeneity of these institutions – retirement homes, nursing homes, prisons, etc. – their annual representativeness will be poor.

11The complete set of variables will be fully representative from 2008 for the year 2006 (n–2), then every year on a rolling basis, by adding the data for the most recent year and removing data for the oldest year.

12Users are therefore in a hiatus period, with data from the last traditional census growing stale, while data from the new census is not yet fully representative and will not become so until the first census cycle has been completed. For 2004, several indicators have already been published by INSEE, notably the estimated populations of 7,000 communes with fewer than 10,000 inhabitants, and of around a hundred large towns, along with various population and housing characteristics at national level (Borrel and Durr, 2005). For the period 1999-2004, the new census has already given a measure of residential migration at national level and by five-year age group, and of the overall rates of change of residence to a new dwelling, commune or region. Net migration between regions and annual mobility trends have also been published more recently (Baccaïni, 2005). In line with CNIS working group recommendations [7], only migration between the largest regions can be calculated from the annual census survey.

II – Advantages and drawbacks of the new census [8]

13To date, the data are not yet available outside INSEE. Our discussion of the advantages and drawbacks of the new census is an ex-ante judgement, expressed before any data from this new source have actually been analysed. Certain remarks may prove unfounded, while other problems, not mentioned here, may emerge at a later stage.

1 – The advantages of the new census

14a) As well as giving annual resident population figures, the new census will provide annually updated data on a sample that will ultimately be the largest of any French survey (census survey). Ongoing trends and recent trend reversals can therefore be monitored closely. This was not possible with the traditional census, whose observation points were spaced six or nine years apart. In fact, once a full cycle has been completed, France will have a “census” every year, providing a much enhanced data source for the study of population change.

15b) Though not exhaustive in all communes, the new census comprises a very large sample of around 42.5 million individuals enumerated over five years. According to INSEE, from the end of the first cycle (2008), estimations at infra-communal level (IRIS) will be possible.

16c) The shorter and more consistent mobility estimation period is also a real improvement. It is now set at five years (rather than the previously variable intercensal period), making period comparison simpler and more reliable. The period is also shorter than before, thus reducing memory bias;

17d) It should be possible to obtain detailed information for clearly defined sub-populations (young people, etc.);

18e) In the old census, certain variables were available in the “one quarter sampling” only. This was the case for socio-occupational category (SOC), which was processed manually if the automatic coding failed. This task was deemed too labour-intensive to be applied to the complete sample each year. In the new census, the quarter-sampling principle is now used only for communes with fewer than 10,000 inhabitants, and processing of SOC variables is exhaustive for communes of 10,000 inhabitants or more. This relative advantage is countered by the drawback that the data varies by type of commune, and hence between communes that may form part of the same urban agglomeration.

19f) The census remains exhaustive for communes of fewer than 10,000 inhabitants.

2 – The drawbacks of the new census

20Despite its undeniable advantages, the new census also has many drawbacks. Not all are equally important, and their effects depend on the way the data are used. The following problems are foreseeable:

21a) The census has become a “census survey”. It is no longer exhaustive, so some of the infra-communal data analyses are less reliable;

22b) Data is no longer processed at street block level [9] (generally a block of houses or buildings);

23c) For this reason, it will probably no longer be possible to follow migration patterns in priority urban development zones – urban empowerment zones (zone franche urbaine, ZFU), vulnerable urban (zone urbaine sensible, ZUS), etc. – or to understand population dynamics in these districts. This change is occurring at a time when the public authorities appear to be focusing renewed attention on these districts. INSEE will nevertheless officially enumerate the inhabitants of priority neighbourhoods. Some of the sociodemographic information on the ZUSs and ZFUs will be taken from analyses of various geocoded administrative datasets to partly make up for the loss of data.

24d) Alongside the problem of priority development zones, it will not always be feasible to produce data on the basis of sub-divisions defined ad hoc for specific local needs, and when such data are produced they will have a high margin of error (see Léger and Raulot, 2005).

25e) Variations in indicators and comparisons between different geographical areas must be interpreted on the basis of indicators of accuracy, i.e. confidence interval or variation coefficient. In other words, the transition from an exhaustive census to a survey system generates a loss of accuracy that may result in errors of interpretation on the part of users who are unfamiliar with statistical tools.

26f) The two groups of communes (fewer than / more than 10,000 inhabitants) raise specific data analysis problems, notably because of the potentially desynchronized survey dates (see below). In fact, there is a time lag of four years between the enumeration of the first and last rotation groups, both for the exhaustively enumerated “small” communes and for the “large” communes surveyed annually. The data collected may thus be staggered over four years. For example, people in the first rotation group will be asked in 2004 where they were living in 1999, while those in the fifth and last group will be asked the same question in 2008 for 2003. Mobility will therefore be measured for the five-year periods preceding the census surveys, which themselves cover a data collection period of five years. INSEE has chosen to attribute the census estimate thus obtained to the median year (2006 for the first cycle 2004-2008).

27g) Overall data accuracy will not be guaranteed until the end of the survey cycle, the first of which ends in 2008. Data will then be updated annually on a rolling basis.

28h) Despite annual data collection, annual mobility is not known. The period of mobility considered is five years (see Table 1). Though less acute than with the traditional census, memory bias thus remains a problem.

29i) We will have to wait until the end of the survey cycle to analyse the full potential of this new census. Outside INSEE, individual-level data from this first series are unlikely to become available before 2009-2010.

30j) Some of the flaws of the old census are also present in the new one. For example:

  • the date of mobility is unknown;
  • individual characteristics are known at the time of the census but not at the time of mobility;
  • data is less detailed than survey data, notably with regard to housing, income etc.;
  • the census under-estimates migration flows and migrant numbers: first, persons who leave metropolitan France or who die between two censuses are not counted and second, return journeys made between two censuses are not taken into account;
  • the parents’ country of origin is not known, making it impossible to study discrimination and spatial polarization of immigrants’ descendants.
Is the collection method used liable to introduce bias? A person living in a rural commune enumerated in year n and who moves to another rural commune enumerated in year n+1 would be enumerated twice. In fact, mobile individuals will be enumerated with varying frequency, depending on the rotation group into which they move, since migration is measured in relation to place of arrival. As these destination groups are random, there should be no resulting bias.

31Though this problem is new with respect to the traditional census, it is a classic difficulty of all sample surveys (Baccaïni, 1999a). The sampling frame of each of the five annual census surveys is designed to ensure that each year the enumerated population is representative of the population as a whole, as is the case for all surveys. The aggregation of five years of data forms a new entity which differs from a sample survey because of the staggered timing of data collection. However, it would be necessary to assess the impact of mobile individuals who, over the five-year data collection cycle, have the greatest likelihood of being enumerated several times. The probability of the same person being observed at times t and t+1 is low, but far from negligible since migratory flows are large. Note that the probability of an individual moving to a different census zone lies somewhere between the annual rate of change of dwelling (10%) and the annual rate of change of commune (6%) (Courgeau, 1980).

3 – Communes of fewer than 10,000 inhabitants

32When the geographical level of interest is the commune, things become more complicated, especially when comparing the mobility level with another group of communes. An interval of four years separates the data collected for the first group of communes from that collected for the fifth and last group (Table 1). It is therefore impossible to compare residential mobility statistics for one particular year in communes not belonging to the same group. Indeed, for the first group enumerated in 2004, the mobility reference date is January 1999, while for the fifth group enumerated in 2008, the reference date is January 2003. This time lag makes it difficult to compare communes, even when they are geographically close.

33One way to get around this problem is to calculate the difference between the mobility rate of a commune in a particular rotation group and the mobility rate of the large communes, or the difference with respect to the national mobility rate, which is more reliable, and then to compare this difference with that observed for a commune of a different rotation group. This comparison implicitly assumes that the spatial disparities are structural and hence stable over time (Jugnot, 2005). Such stability is difficult to justify and does not take account of sudden structural changes (bankruptcy, relocation or arrival of a large employer, demolition or construction of buildings). In practice, INSEE intends to interpolate the indicator of the first group in n+2 and retropolate the indicator of the fifth group in n–2, using data from the tax on furnished accommodation (taxe d’habitation) database to make comparisons (see remarks below).

34The problem is less acute when the aggregation level is higher, for example when comparing all small communes with all large ones. Nevertheless, geographic divisions such as urban areas and urban units, agglomeration communities or employment zones also raise problems, since these various entities may include a certain number of small communes not evenly distributed among the different rotation groups at this level. Indeed, the sample of small communes is balanced each year at regional level only. Taking this reasoning to the extreme, this means that a zone comprising large communes and small communes in the first group, surveyed in 2004, might be compared in 2008 with a zone comprising large communes and small communes in the fifth group, surveyed, for its part, in 2008.

4 – Migration flows

35Migration flows are also more difficult to measure for small communes (Baccaïni, 1999b). For example, in the case of a small commune enumerated in 2004, entrants between 1999 and 2004 will be identified exhaustively if they are still in the same commune on the census date in 2009. Persons who move to a different commune will be enumerated between 2004 and 2008 with reference periods for the prior place of residence ranging from 1999-2004 to 2003-2008, depending on the rotation group. Migration flows cannot be directly broken down by year, as they can with a population register.

36An alternative option is to keep the same comparison dates, whatever the rotation group of the communes compared. In this case, the data concerning a particular commune are obtained by linear interpolation. This is done simply by drawing a line between the last two points observed and reading off the information given by this line for the date of interest. Take the example of a commune of the first group that we want to compare in 2009 with a commune in the fifth group. For the fifth group commune, we have two census dates: 2008 and 2013. We simply read off the mobility level for 2009 and compare it with the level in the first commune. Clearly, this method gives good results for indicators with strong inertia that follow a regular trend. Residential mobility trends change quite slowly and appear to lend themselves to this approach. However, such calculations may be unreliable if sudden unexpected changes occur (natural or industrial disaster, factory closure, construction or demolition of residential zones, etc.), especially if the population in the zone concerned is small (see examples given by Bertrand et al., 2002). Moreover, the above example does not becomes valid until sufficient census surveys have been conducted to provide two different observation dates for the same group of communes. For the intermediate period it is not applicable. Another method recommended by INSEE is to extrapolate certain indicators from external data sources, such as the tax on furnished accommodation database. But as yet, INSEE has given no precise information about this option which, in any case, will not be available until a full census cycle has been completed.

5 – The inter-regional mobility matrix

37According to the conclusions of the CNIS working group on new census data analysis (2004), the inter-regional mobility matrix of 22x22 regions, tracking the arrivals and departures of individuals between each region of metropolitan France, will be calculated once a full census cycle has been completed [10]. But it is important to remember that the five annual sets of data collected do not have the same time reference. After a full five-year collection cycle, we will have data for mobility measured between 1999 and 2003, i.e. a measure of five-year mobility over a period of five years. In our example, the median observation date would be 2006. For users without specialized knowledge of the data source, the results presented and, above all, the reasons for interpreting them with precaution, may not be readily understood. Sub-modules of this matrix could be produced, paying particular attention to the regions selected and to the cross variables used (age groups for example) so that the sample size is not reduced too drastically.

6 – Mobility measured by the new census

38As in most French datasets, it is migrants rather than migration flows that are counted. The chosen timespan (five years) facilitates international comparison and enhances the estimation accuracy of flows between small or widely spaced zones (Baccaïni, 2001) [11]. Mobility can be recorded in two ways. Five-year mobility can be measured from the annual sample survey concerning year n. The sample size limits this type of analysis to inter-regional migration flows between the main migration regions (CNIS, 2004). The second measure is based on the complete sample built up over five years of observation. No specific recommendations were issued by the CNIS working group. This is a new approach in France, representing an average measure over a five-year survey period and calculated for a reference period of five years [12]. Once the first cycle has been completed, the new census will thus provide an annual estimate of five-year migration based on a smoothed and centred moving average. This is destined to become the key data source for French studies of current migration trends.

39The dwelling data sheet identifies second or occasional residences, and the individual questionnaires are not administered unless the respondents are present in these dwellings in the census year n. There are no details concerning the type of dwelling in the census at year n–1; it is simply stated that for members of institutional households, only the address of the “personal residence” counts. This notion of “personal residence” is distinct from that of residence in an institutional household but remains ambiguous, since main and second residences may both be considered as personal. Consequently, it is not exclusive for this population category. However, we can assume that in most cases, it concerns the primary residence, but only when this notion is meaningful for the respondent [13].

III – A longitudinal dataset directly dependent on the census: the permanent demographic sample

40The analysis of French administrative files for statistical purposes is tending to develop, as it presents many advantages: lower data collection costs, large samples, data free from individual self-reporting bias. On the other hand, these data, not originally intended for this purpose, are often incomplete and may be biased by their origin (tax returns, wage declarations etc.). The new census data will be integrated with administrative data [14], making it possible improve the quality of statistical studies at little additional cost. This is already the case for the permanent demographic sample (EDP), and will be so for the new permanent demographic sample, known as EDP++, whose aim is to supplement the census data, in a rational and controlled manner, by field of research [15]. For researchers, access to this detailed data is more restricted than for survey data. It is nevertheless possible, with prior authorization, provided that the database is queried from the premises of INSEE.

1 – A system in the making

41From 2008, the EDP++ will be coupled with the new census. Though the project has not yet been finalized and is likely to evolve, it has major potential and deserves to be presented here. The new EDP sample will be four times larger than before. It will include all persons born on four specific days of each quarter, in order to follow individuals whose births are spread across the year. In 2006, the designers of the new EDP were hoping to link EDP++ to other databases: causes of death, panel of the annual register of private-sector employees (DADS), panel of civil service employees, panel of schoolchildren and of the electoral register, subject to approval by CNIL. As these bodies of data are very large, the EDP++ will be divided into modules (Table 3). One module will include data from the new census, from the DADS panel and from the civil service employees database. It will contain detailed data on household composition, sociodemographic characteristics (excluding income), employees and their employers, along with information on housing. The other modules will be created by data matching with the causes of death database, the panel of schoolchildren, the electoral register and the civil records.

Table 3

Main characteristics of the permanent demographic sample (EDP++)

Table 3
EDP++ Periodicity Same as the census. But individuals will not always be followed, since the sample is no longer exhaustive (see below) Survey period Same as census. Sample 1/25th Number of respondents Around 2.5 million drawn from the new census Population Depending on the module: – Employees (DADS panel and civil service employees) – Causes of death – Panel of schoolchildren – Electoral register – Civil records Representativeness Depending on the module: Persons in employment Schoolchildren Finest level of representativeness To be determined Mobility detected Five-year for all (new census) Annual for persons in employment (DADS) Remark Secondary employment will also be included

Main characteristics of the permanent demographic sample (EDP++)

42The potential advantages of the new EDP will be as follows:

43a) The datasets combine the advantages of the new census and of the administrative datasets. One provides wide-ranging information on ego, his/her household and dwelling, while the others give valuable data in the areas covered by administrative files, such as the characteristics of ego’s occupation and of the company where he/she works (DADS database) [16];

44b) Combining data on private-sector employees with those on civil service employees, a practically exhaustive dataset of persons in employment will be obtained. However, among the working population, the self-employed will still not be accounted for;

45c) The new census data and some administrative data will be available annually.

46The following problems are foreseeable:

47a) The datasets also inherit the particularities of the new census. The first problem is how to maintain a panel configuration when respondents are not systematically enumerated at regular intervals;

48b) After several years of data collection, and despite the modular structure, the datasets will soon become enormous and statistical analysis will be difficult;

49c) Certain data useful for studying mobility, such as interpersonal relationships and the amenity of places of residence, are not available, as was already the case previously;

50d) But the main difficulty is the deformation of the sample over time, with the most mobile individuals becoming over-represented while others leave the sample (see below).

2 – Measuring mobility

51Mobility will be measured in different ways. For example, the panel of schoolchildren will be used to measure the annual mobility of schoolchildren, the DADS panel will serve to calculate annual mobility rates of working-age people, and both will be complemented by information taken from the census. The EDP++ raises many questions however, since the new census, which forms the backbone of the system, does not provide for the systematic and regular follow-up of inhabitants from one census to the next, even at the end of a five-year cycle. All types of situation are liable to arise: both that of persons enumerated every year (if they move each year to a commune in a different rotation group) and that of persons who are never enumerated because they are among the 60% of residents of large communes who have not been enumerated at the end of 5 years. These two cases are rare of course. The most commonly encountered situations are as follows. Persons eligible to join the panel and enumerated in year n in a small commune will, in principle, be enumerated again in the year n+5, if:

  1. they do not move;
  2. they move to a new dwelling but return to or remain in their commune of origin;
  3. they move to a small commune in the same rotation group;
  4. they move to a large commune and are among the 8% of households enumerated that year.
Persons living initially in a large commune, may be enumerated again after five years if:
  1. they do not move and are included again in the 8% of households enumerated;
  2. they move but return to the commune of origin and are again included in the initial rotation group of addresses;
  3. they move to a large commune and are among the 8% of households enumerated that year;
  4. they move to a small commune enumerated that year.
All other cases and frequencies of enumeration are possible. Consequently, following a panel of individuals at the same dates will be purely hypothetical, especially since the sample of persons present will doubtless become deformed, firstly by the selection effect of immobile persons, tendentially more advanced in age, enumerated exhaustively every five years in the small communes, and secondly by the over-representation of younger, more urban individuals who, due to their high mobility, are more likely to be enumerated. Data matched from other sources (DADS, panel of schoolchildren) will provide intermediate observation points for the individuals concerned, via the information contained in these sources. At the same time, the independence of the observation date with respect to the event in question (Courgeau, Lelièvre and Wolber, 1998) will be verified with the DADS, panel of schoolchildren and electoral register modules. This was not the case for the former EDP integrated with data from civil records, as the latter relate family events correlated with migration (marriage, birth of a child, divorce). There are many unanswered questions, but as the project has not yet been finalized, it is too early to assess the exact potential of this data source.

Provisional conclusions

52Like all radical reforms, the new census has aroused many misgivings and legitimate reservations on the part of its traditional users [17]. The considerable and undeniable gains in terms of data freshness and reliability (once the first cycle is complete, France will have an annual census) have been achieved at the cost of abandoning exhaustiveness and simultaneous data collection, two characteristics that made it possible to analyse and compare data at all spatial levels. For this reason, the census was a valuable and practically unique data source for many local-level studies. Moreover, with the disappearance of street block sub-divisions, it will no longer be possible, for example, to predict, at catchment area level, whether additional capacity will be needed in schools (Léger and Raulot, 2005), and the scope for infra-communal analyses of small population groups and for precise breakdowns will be limited.

53The new census does not pose problems for studying current mobility trends at national, regional and communal levels. However, it is still unclear whether residential mobility will be comparable at infra-regional level (employment zones for example) and infra-urban level (the geography of school catchment areas for example) due to time lags in the collection of data in the small communes making up these zones, or to the elimination of street block sub-divisions. A major methodological effort will be required to compare infra-regional migration, at a time when such information is of increasing importance for local governments entrusted with ever wider responsibilities. The new census also entails a loss of information for the study of priority urban development zones. Here too, despite a need for information that has been heightened and institutionalized by the Borloo Act [18], the new census will no longer provide a detailed picture of each vulnerable urban zone [19]. The more extensive use of data from administrative files will only partially offset this lack of information. Indeed, these data are scattered between a variety of sources and concern specific populations: persons declared unemployed or receiving unemployment benefit, taxpayers, etc. This rules out multivariate analysis, since data cannot be taken into account on a simultaneous basis.

54The quality of the permanent demographic sample depended on the census and its capacity to follow individuals from one census to the next. Now that the census is no longer exhaustive, survey follow-up on the basis of the new census would appear to be compromised.

55Of course, all these remarks concerning the study of mobility via the new census in no way detract from the dedication of the design teams involved in creating and implementing the new census. Their achievement, invisible to the untrained eye, is remarkable. The financing problems also affecting other European institutes (Laihonen, 2000), are obliging the countries concerned to envisage new modes of data collection other than the traditional census. Such a change is bound to bring problems in its wake.

56All in all, the gains and losses for the census users are not evenly distributed: the advantage of annual data refreshment is counteracted by a narrowing of scope. This imbalance in the nature of gains and losses provides legitimate grounds for dissatisfaction. We need to wait until a full five-year cycle of the new census has been completed to identify all the gains and losses of the new system. The creation of a procedure for assessing the census results and for developing dialogue with the persons and institutions who traditionally make use of census data would provide a means to improve data quality and enhance communication with users.


  • [*]
    Institut national d’études démographiques.
    Translated by Catriona Dutreuilh
  • [1]
    Michel Poulain had already assessed European migration statistics in an article in 1994. He outlines the concepts, methods and sources used (Poulain, 1994).
  • [2]
    Daniel Courgeau’s “migrants-migration” model has nevertheless made it possible to estimate annual migration rates under certain hypotheses (Courgeau 1973; Courgeau and Lelièvre, 2004).
  • [3]
    For more information, an explanatory document is available from INSEE at the following address: legales.pdf
  • [4]
    A new permanent demographic sample project, known as EDP++ is currently being developed. The decision to include data from administrative files will not be finalized until approval is obtained from the French data protection agency (CNIL). The purpose of the CNIL is to protect privacy and individual freedom against the threat of uncontrolled use of personal or public data stored in computer files.
  • [5]
    The cluster effect occurs when the survey respondents live in the same locality, building or group of buildings, assuming that these localities are correlated with the individuals’ social characteristics. This results in a non-random concentration of these characteristics. On this point and on the sampling frame, see Grosbras et al. (2001, 2002).
  • [6]
    Following discussions with statisticians of the Société française de statistique (SFdS), two major changes were made. All occupants of buildings were surveyed and high-rise buildings were taken into account when stratifying the sample. Transcriptions of the discussions with the census team are given in the Journal de la société française de statistique, 142(3), and 143(3-4).
  • [7]
    The French National Council for Statistical Information (CNIS) liaises between the producers and users of public statistics. Documents concerning the use of new census data can be viewed on the CNIS website ( Document III-1 concerns the various uses of the census and, in particular, the analysis of regional migration via a migration matrix.
  • [8]
    We omit here the problem of duplication resulting from the dual residence of students, executives and retirees, which contributes to the discrepancy observed between the demographic balance published by INSEE and the population estimated by the 2004 census survey (Héran and Toulemon, 2005).
  • [9]
    The “street block” was the basic statistical information unit in the most recent traditional censuses. The infra-communal census zones (IRIS) are the new basic statistical units. An IRIS corresponds to a small neighbourhood of between 1,800 and 5,000 inhabitants, or the entire commune in the case of small communes with no sub-divisions.
  • [10]
    In fact, this matrix has already been calculated by INSEE from an annual census survey and is available for the year 2004 (Baccaïni, 2005). However, inter-regional flows are very small, so some of these data may be inaccurate.
  • [11]
    However, the United Nations recommendation to also take one year as a reference, supported by INED researchers (notably Baccaïni, 1999b), was not followed.
  • [12]
    Daniel Courgeau’s “migrants-migrations” model is still useful for estimating annual migration, though to obtain pertinent results, the model coefficients need to be regularly updated, and an appropriate dataset must be available (L’Hospital, 2001; Courgeau and Lelièvre, 2004).
  • [13]
    Daniel Courgeau (1975, 1980, 1988) has pointed out that the notion of primary residence is not always clear for nomads, children living apart from their parents or at boarding school, soldiers and persons in long-term hospital care. The same applies to retirees and working people with two homes.
  • [14]
    After approval by the CNIL.
  • [15]
    Note that the EDP is a longitudinal dataset combining census data and administrative data via the data sources that are included in it, i.e. the civil records and, at certain times, the electoral register. The new version of the EDP will replace the old one (Couet, 2006). Issue 113-114 of Courrier des statistiques (2005) describes the setting up of the EDP++.
  • [16]
    An initial exploratory study of residential mobility based exclusively on the DADS dataset was carried out by the Rhône-Alpes regional division of INSEE (Brun, 2000). Information from the new census should facilitate in-depth studies of the mobility of working-age people by using wage levels to approximate types of employment.
  • [17]
    Which have been hotly debated (cf. Cybergeo).
  • [18]
    Loi d’orientation et de programmation pour la ville et la rénovation urbaine (no. 2003-710), 1 August 2003.
  • [19]
    Apart from the so-called “de jure” population, currently being determined.


In France, the transition from the traditional census to the “new census” has answered certain needs while raising a number of concerns. INSEE has devised an original methodology based on the aggregation of five annual census surveys to produce this so-called “new census”. Once the first five-year cycle has been completed, data from the most recent annual survey will be added and those of the oldest year removed so that a complete cycle is obtained on a rolling basis each year. This radically new approach is attracting the attention of demographers, statisticians and geographers beyond the frontiers of France. The advantages of an annual census and its large body of data are countered, however, by the drawbacks of desynchronized data collection, the loss of exhaustiveness and the absence of spatial data at highly localized levels. The new census is assessed here primarily from the viewpoint of residential mobility. After a presentation of its complex methodology, its potential advantages and drawbacks are reviewed in detail. Last, the forseeable impact on the permanent demographic sample, a rich longitudinal body of data based on the census, is assessed.



En France, le passage du recensement traditionnel au recensement rénové a répondu à des attentes tout en suscitant des inquiétudes. Cette façon, radicalement nouvelle, de produire des données est l’objet de l’attention de démographes, statisticiens et géographes au-delà des seules frontières de l’hexagone. L’Insee a élaboré une méthodologie originale qui repose sur l’agrégation de cinq enquêtes annuelles de recensement pour parvenir au « recensement rénové » proprement dit. En rythme de croisière, l’adjonction de la dernière enquête annuelle et le retrait de la plus ancienne permettront en glissement d’obtenir chaque année un cycle complet. La disponibilité d’un recensement chaque année et de ses nombreuses données a pour contreparties la désynchronisation temporelle de la collecte ainsi que l’abandon de l’exhaustivité et de données spatiales disponibles à des niveaux très fins. Le recensement rénové est plus particulièrement observé ici sous l’angle des mobilités résidentielles. Après l’exposé de sa méthodologie complexe, un tour d’horizon des avantages et des limites qui lui sont adressés est effectué en détail. Enfin, l’impact prévisible sur l’échantillon démographique permanent, riche source longitudinale utilisant les données de chaque recensement, est examiné.



En Francia, el paso del censo tradicional al censo renovado ha respondido a expectativas suscitando al mismo tiempo preocupaciones. Esta manera, radicalmente nueva, de producir datos es objeto de atención de los demógrafos, estadistas y geógrafos más allá de las únicas fronteras de Francia. El Insee (Instituto nacional de estadística francés) ha elaborado una metodología original que se apoya en la agregación de cinco encuestas anuales de censo para llegar al « censo renovado » propiamente dicho. En ritmo de crucero, el añadir la última encuesta anual y el retirar la más antigua permitirán en deslizamiento obtener cada año un ciclo completo. La disponibilidad de un censo cada año y de sus numerosos datos tiene como contrapartida la desincronización temporal de la colecta así como el abandono de la exhaustividad y de datos espaciales disponibles a niveles muy específicos. El censo renovado está observado de modo más particular aquí bajo el ángulo de las movilidades residenciales. Tras la presentación de su metodología compleja, se elabora con precisión un panorama de las ventajas y de las limitaciones que le son destinadas. Por último, se examina el impacto previsible sobre la muestra demográfica permanente, nutrida fuente longitudinal que utiliza los datos de cada censo.


  • Baccaïni B., 1999a, “Recensement traditionnel et recensement rénové pour l’étude des migrations résidentielles”, Rapport n° 99239, Service études et diffusion, Direction régionale Rhône-Alpes de l’Insee, 101 p.
  • OnlineBaccaïni B., 1999b, “Analyse des migrations internes et estimation du solde migratoire externe au niveau local à l’aide de données intercensitaires”, Population, 54 (4-5), pp. 801-816.
  • Baccaïni B., 2001, “Première approche de l’étude des migrations avec le recensement rénové”, Journal de la Société française de statistique, no. 3, pp. 31-39.
  • Baccaïni B., 2005, “Enquêtes annuelles de recensement : résultats de la collecte 2004. Des changements de région plus fréquents qui bénéficient aux régions du sud et de l’ouest”, Insee Première, no. 1028.
  • Bertrand P., Chauvet G., Christian B., Grosbras J.-M., 2002, “Données produites par le recensement rénové de la population”, Actes des journées de méthodologie statistique 2002,
  • Borrel C., Durr J.-M., 2005, “Enquêtes annuelles de recensement : premiers résultats de la collecte 2004. Principales caractéristiques de la population et des logements”, Insee Première, no. 1001.
  • Brun C., 2000, “Étude des migrations résidentielles dans la zone d’emploi de Saint-Étienne à partir du fichier DADS 1997”, internal memo, Mission Migrations, Direction régionale Rhône-Alpes de l’Insee.
  • Cnis, 2004, “Utilisation des données produites par le recensement rénové de la population et leur diffusion”, (
  • OnlineCoëffic N., 1993, “L’enquête post-censitaire de 1990. Une mesure de l’exhaustivité du recensement”, Population, 48 (6), pp. 1655-1681.
  • Couet C., 2006, “L’échantillon démographique permanent de l’Insee”, Courrier des statistiques, no. 117-119, pp. 5-14.
  • OnlineCourgeau D., 1973, “Migrants et migrations” Population, 28(1), pp. 95-129.
  • Courgeau D., 1975, “Le concept de migration”, Actes du quatrième colloque de démographie africaine, Ouagadougou, pp. 27-33.
  • Courgeau D., 1980, Analyse quantitative des migrations humaines, Paris/New York/Barcelona/Milan, Masson, pp. 14-15.
  • OnlineCourgeau D., 1988, Méthodes de mesure de la mobilité spatiale. Migrations internes, mobilité temporaire, navettes, Éditions de l’Ined, pp. 11-14.
  • OnlineCourgeau D., Lelièvre É., Wolber O., 1998, “Reconstruire des trajectoires de mobilité résidentielle, éléments d’une analyse biographique des données de l’EDP”, Économie et Statistique, no. 316-317, pp. 163-173.
  • OnlineCourgeau D., Lelièvre É., 2004, “Estimation of French internal migration in the period 1990-1999 and comparison with earlier periods”, Population-E, 59(5), pp. 703-710.
  • Grosbras J.-M., Allain D., Durr J.-M., Godinot A., 2001, “Recensement rénové de la population. Compte rendu des échanges”, Journal de la Société française de statistique,142 (3), pp. 41-49.
  • Grosbras J.-M., Allain D., Durr J.-M., 2002, “Séminaire méthodologique SFdS-Insee. Compte rendu des échanges sur la partie plan de sondage”, Journal de la Société française de statistique, 143 (3-4), pp. 125-128.
  • Héran F., Toulemon L., 2005, “What happens when the census population figure does not match the estimates?”, Population and Societies, no. 411, 4 p.
  • Jugnot S., 2005, “Analyser les disparités spatiales sur un indicateur conjoncturel avec le recensement : l’exemple du taux de chômage”, Document de travail, Journées de méthodologie statistique de l’Insee, ( jugnot_rrp_article.pdf).
  • Laihonen A., 2000, “2001 Round population censuses in Europe”, Paper presented at the Insee-Eurostat seminar on censuses after 2001, Paris, (
  • Léger J.-F., Raulot J.-Y., 2005, “Mieux tenir compte de chacun de nous : l’objectif du recensement rénové français sera-t-il tenu à l’échelle infra-communale ?”, (
  • L’Hospital F., 2001, “Les migrations internes en France. Estimation des paramètres du modèle “migrants-migrations” de Daniel Courgeau”, Document de travail, Direction régionale Rhône-Alpes de l’Insee.
  • OnlinePoulain M., 1994, “La mobilité interne en Europe. Quelles données statistiques ?”, Espaces, populations, sociétés, pp. 13-30.
Jean-Louis Pan Ké Shon [*]
Institut national d’études démographiques, 133 boulevard Davout, 75980 Paris, Cedex 20, tel: 33 (0)1 56 06 22 63
  • [*]
    Institut national d’études démographiques.
    Translated by Catriona Dutreuilh
Translated by
Catriona Dutreuilh
Latest publication on cairn or another partner portal
Uploaded on on 03/03/2014
Distribution électronique pour I.N.E.D © I.N.E.D. Tous droits réservés pour tous pays. Il est interdit, sauf accord préalable et écrit de l’éditeur, de reproduire (notamment par photocopie) partiellement ou totalement le présent article, de le stocker dans une banque de données ou de le communiquer au public sous quelque forme et de quelque manière que ce soit.
Loading... Please wait